TRACE: An Operational Reasoning Schema for Auditable Agentic Commitments

# TRACE: An Operational Reasoning Schema for Auditable Agentic Commitments
Source: [https://arxiv.org/html/2607.12480](https://arxiv.org/html/2607.12480)
###### Abstract

This paper defines and proposes TRACE \(Typed Reasoning And Commitment Evidence\): a typed, versioned schema for recording reasoning traces, one reference procedure for writing records against that schema, and one operating discipline, no durable state change without a record\. The paper first states goals and non\-goals, then argues in three layers that reasoning is not in the language model: the autoregressive mechanism natively computes association; chain\-of\-thought and reinforcement learning inherit the mechanism’s limits; and the formal constructs of reasoning theory, from Socratic procedure to Pearl’s ladder, are absent as machinery\. The TRACE schema answers the absence with fields and tests: the TraceRecord and its causal specialization, an eight\-stage reference writer, a gate\-first measurement regime, the TRACE\-Bench protocol, and the consumers, memory admission, plan gating, temporal regret, and verdict reuse, whose improved and more auditable decisions are the measure by which the record is judged\. A record\-consumer contract states the guarantees a record makes to a consumer and the obligations a consumer accepts in return, so the schema is an operational interface rather than a passive document\. Two worked examples run in the main text: a music\-lessons argument traced from raw sentence to typed verdict, which shows the writer separating association, intervention, and prescription; and a flood search\-and\-rescue vignette in which a predictive world model reports a confident plan success that its own support and out\-of\-distribution scores contradict, so the record defers the commitment, requests a bounded observation, revises the claim append\-only, and clears a different plan branch\. The vignette is illustrative, not an empirical result; a closed\-loop evaluation is specified but left to future work, so the paper’s contribution is the schema and its consumer contract, not a performance claim\. Appendices carry the full schema, the reference writer’s algorithms and cost model, the contested anatomy, worked illustrations at clinical and policy stakes, the benchmark protocol, the convergence\-validity metrics and imported propositions, and usage scenarios for memory admission and plan gating\.

## 1The Paper’s Aim: TRACE, Its Goals, and Its Non\-Goals

This paper defines and proposes TRACE \(Typed Reasoning And Commitment Evidence\): a typed, versioned schema for recording reasoning traces, one reference procedure for writing records against that schema, and one operating discipline that makes the records consequential\. The artifact is the TRACE record; everything else in the paper either writes it, measures it, or consumes it\. So that the reader’s expectations are set before any machinery arrives, the paper’s goals and non\-goals are stated here, first\.

##### Goals\.

The paper pursues three\.

1. 1\.*Record\.*Define one unified schema, concretely a versioned JSON type definition, that captures the typed structure of a reasoning episode: what was claimed, what kind of reasoning the claim required, what evidence and warrants were offered, what was tested, what failed, what is missing, and what verdict followed\.
2. 2\.*Protocol\.*Make the same schema the interchange format between an agentic system’s services: writers emit records, consumers act on them, and no service passes reasoning state to another in any other form\. The schema is simultaneously a log format and an exchange protocol\.
3. 3\.*Improvement, through the callers\.*Improve reasoning quality where it can actually be improved: in the systems that call TRACE and consume its records\. SocraSynthChang \([2023b](https://arxiv.org/html/2607.12480#bib.bib77)\)and EVINCEChang \([2024](https://arxiv.org/html/2607.12480#bib.bib78)\)elicit against recorded claims; RCAChang \([2026a](https://arxiv.org/html/2607.12480#bib.bib184)\)localizes failures to recorded stages; ERMChang and Geng \([2026](https://arxiv.org/html/2607.12480#bib.bib185)\)prices recorded errors; RLERChang \([2026b](https://arxiv.org/html/2607.12480#bib.bib183)\)revises policies from recorded verdicts; TriviumChang \([2026c](https://arxiv.org/html/2607.12480#bib.bib189)\)computes temporal regret over recorded steps; MnemosyneChanget al\.\([2027](https://arxiv.org/html/2607.12480#bib.bib190)\)gates memory and MACIChang \([2025](https://arxiv.org/html/2607.12480#bib.bib182)\)gates plans on recorded outcomes\. The record is what makes each of these operations possible, and consumer value is how the record is judged\.

##### Non\-goals\.

Three disclaimers bound the proposal\.

1. 1\.TRACE does not itself optimize or improve reasoning\. It is infrastructure in service of its callers: it records and adjudicates so that the systems above it can improve reasoning, the way a write\-ahead log recovers no transaction by itself\.
2. 2\.The eight\-stage reference writer is not claimed to be optimal; it is one conforming implementation, and every stage is replaceable without moving the record format\.
3. 3\.TRACE is not a universal reasoning\-quality metric\. Its scores measure compliance with the declared, versioned TRACE policy; no objective scalar of reasoning quality exists, and nothing essential in the design depends on one\.

##### Two precedents\.

Systems engineering has met this situation twice before\. A write\-ahead log does not make a database faster or smarter, and a flight recorder does not fly the aircraft\. Yet recovery, replication, and audit are impossible without the first, post\-incident analysis is impossible without the second, and in both cases the value of the log is proven entirely by what its consumers can do\. The TRACE record plays the same role for reasoning in an agentic system\. It does not make the underlying model more correct\. It makes every durable consequence of the model’s reasoning attributable, contestable, and improvable\.

##### The TRACE discipline\.

The schema comes with one operating rule, which the companion implementation enforces: no durable state change without a TRACE record\. A memory is not committed, a plan is not cleared, and an obligation is not updated unless a TRACE record has been written against the schema and carries a verdict\. Section[7\.1](https://arxiv.org/html/2607.12480#S7.SS1)makes this closure concrete in a flood search\-and\-rescue plan: a high prediction is held when its support fails, a drone observation repairs the record, and only the dependent route branch is revised\. The companion volume’s implementation chapter closes this loop; Appendices[G](https://arxiv.org/html/2607.12480#A7)and[H](https://arxiv.org/html/2607.12480#A8)sketch the two admission consumers\.

The paper’s design claim is therefore:

> The unit of contribution is the record, not any stage algorithm\. Reasoning evaluation should be decomposed into typed, auditable responsibilities that read and write one versioned TRACE record; a claim must be typed before it is tested, and tested at the standard its type demands; and the record is judged not by a quality scalar but by what its consumers, admission control, plan gating, and temporal learning, can do with it\.

Two consequences of this claim are worth stating before the machinery arrives\. First, the paper’s foundational claim is not that TRACE measures reasoning quality; it is that TRACE records the typed structure of reasoning sufficiently for admission, audit, regret, repair, and downstream learning\. Throughout this paper and its companion volume, TRACE names this whole contract: the TRACE schema, the TRACE writer contract, and the TRACE discipline, so that a TRACE, in the ordinary sense of the word, is exactly what the system leaves behind\. TRACE still performs evaluation\-like functions, but its verdicts are fields of the record, and the stages presented below are one implementation of the writer contract, every one of them replaceable without moving the record format\. Second, the scores defined later measure compliance with the declared, versioned TRACE policy, not objective reasoning quality\. Section[7](https://arxiv.org/html/2607.12480#S7)states both points precisely, together with what would falsify the design\.

##### Contributions\.

The paper makes five contributions\. First, a*schema*: the TraceRecord and its causal specialization the TraceCausalRecord \(Section[3](https://arxiv.org/html/2607.12480#S3)\), versioned like an interface, carrying evaluation state, including rung, identification status, bounds, missing evidence, and regret, rather than argument structure alone\. Second, a*type system*for the schema’s fields: four layers of typed evaluation that route claims across formal disciplines \(identification theory, proof checking, decision theory, law\), where prior argument\-mining pipelines and scheme\-based systems such as Carneades classify and evaluate within the argumentation layer; an explicit subsumption rule governs which layer adjudicates when they overlap\. Third, a*reference writer*\(Algorithms[1](https://arxiv.org/html/2607.12480#alg1)and[2](https://arxiv.org/html/2607.12480#alg2)\), one conforming implementation of the writer contract, presented for auditability and cost discipline rather than claimed as optimal\. Fourth,*two complementary worked illustrations*: one ordinary sentence followed through every writer stage, and one flood search\-and\-rescue plan followed from predictive evidence through defer, evidence\-seeking repair, append\-only revision, and plan\-gate consumption\. Fifth, a*benchmark and ablation plan*, TRACE\-Bench, which judges the schema on expressiveness, the writer on stage\-level and end\-to\-end quality, and the record on consumer value, and whose baselines include the single\-pass writer that serves as the null hypothesis for the staged decomposition\.

##### Organization\.

The paper proceeds as aim, critique, schema, hierarchy, stages, measurement, consumers, illustrations, debate, guarantees, and falsification\. Section[2](https://arxiv.org/html/2607.12480#S2)states the layered critique\. Section[3](https://arxiv.org/html/2607.12480#S3)fixes the terminology, the schema and its key fields, and the writer API \(Section[3\.3](https://arxiv.org/html/2607.12480#S3.SS3)\)\. Section[4](https://arxiv.org/html/2607.12480#S4)builds the reasoning hierarchy that gives the typing fields their semantics: the CRIT roof \(Section[4\.1](https://arxiv.org/html/2607.12480#S4.SS1)\), the four evaluation layers \(Section[4\.2](https://arxiv.org/html/2607.12480#S4.SS2)\), the reasoning families with the Toulmin, Walton, and Dung toolbox and the enthymeme\-reconstruction algorithm \(Section[4\.3](https://arxiv.org/html/2607.12480#S4.SS3)\), and Pearl’s ladder as the causal plug\-in \(Section[4\.4](https://arxiv.org/html/2607.12480#S4.SS4)\)\. Section[5](https://arxiv.org/html/2607.12480#S5)depicts the writer’s stages: the formulation gate \(Section[5\.1](https://arxiv.org/html/2607.12480#S5.SS1)\), one claim traced from sentence to verdict \(Section[5\.2](https://arxiv.org/html/2607.12480#S5.SS2)\), and the stopping rules \(Section[5\.4](https://arxiv.org/html/2607.12480#S5.SS4)\)\. Section[6](https://arxiv.org/html/2607.12480#S6)defines the TRACE policy’s quality metrics and hard gates; the benchmark that operationalizes them, TRACE\-Bench, is Appendix[E](https://arxiv.org/html/2607.12480#A5), and the convergence theory and validity metrics of the debate stage are Appendix[F](https://arxiv.org/html/2607.12480#A6)\. Section[7](https://arxiv.org/html/2607.12480#S7)states what the paper claims, including the consumer claim that TRACE is a better admission explainer rather than a better gate\. Section[7\.1](https://arxiv.org/html/2607.12480#S7.SS1)then gives the operational consumer illustration before the paper states what it does not claim, what would falsify it, and its limits\. Worked illustrations at clinical and policy stakes are in Appendix[D](https://arxiv.org/html/2607.12480#A4)\.

## 2Why Reasoning Is Not in the Language Model

Reasoning is the capability on which the current era of AI turns\. The decisive advances of the last five years were not simply advances in knowledge storage\. They were advances in*elicited computation*: training models to emit intermediate steps or scratchpadsNyeet al\.\([2021](https://arxiv.org/html/2607.12480#bib.bib29)\), prompting them to produce chains of thoughtWeiet al\.\([2022](https://arxiv.org/html/2607.12480#bib.bib30)\); Kojimaet al\.\([2022](https://arxiv.org/html/2607.12480#bib.bib31)\), sampling and voting over diverse reasoning pathsWanget al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib32)\), decomposing problems from least to most complexZhouet al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib33)\), searching over branching thoughtsYaoet al\.\([2023a](https://arxiv.org/html/2607.12480#bib.bib34)\), and interleaving reasoning with actionYaoet al\.\([2023b](https://arxiv.org/html/2607.12480#bib.bib35)\)\. The industry then made reasoning the explicit scaling axis: allocating inference\-time compute to deliberation can outperform scaling parametersSnellet al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib36)\), and reinforcement learning with verifiable rewards produces models that plan, self\-check, and revise before answeringOpenAI \([2024](https://arxiv.org/html/2607.12480#bib.bib37)\); Guoet al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib38)\)\. Surveys of this trajectory now describe a shift from prompting techniques to “large reasoning models”Chang \([2023a](https://arxiv.org/html/2607.12480#bib.bib76)\); Chenet al\.\([2025a](https://arxiv.org/html/2607.12480#bib.bib40)\); Huang and Chang \([2023](https://arxiv.org/html/2607.12480#bib.bib131)\); Qiaoet al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib90)\)\.

Before the schema can be motivated, the critique it answers must be stated in an organized way, because the problems are of different depths and are too often listed as one undifferentiated complaint\. The critique here has three layers, each deeper than the last\. Layer one examines the mechanism: what an autoregressive model natively computes\. Layer two examines the two current patches, chain\-of\-thought elicitation and reinforcement learning, and shows that both inherit the mechanism’s limits\. Layer three brings in the formal constructs of reasoning theory and asks, construct by construct, where each one lives in a language model; the answer, nowhere, converts the critique into the specification that the TRACE schema answers\.

### 2\.1Layer One: The Mechanism Is Association

Elicitation, however, changes what the model emits, not what the model computes\. Maximum\-likelihood training makes association the native operation: an autoregressive model is optimized to continue text as its corpus would, and correlation, not causation, is what corpus statistics carry\. When a model’s output appears causal, it is because causally generated text was absorbed associatively, the corpus contains the sentences that causal processes in the world produced, not because the autoregressive mechanism can distinguish a causal relation from an association\. Models accordingly deploy causal vocabulary without satisfying causal semanticsKıcımanet al\.\([2024](https://arxiv.org/html/2607.12480#bib.bib104)\); Zečevićet al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib105)\): the words*because*,*causes*, and*therefore*are high\-probability tokens, not inference steps\.

The claim should be stated at its defensible strength\. Mechanistic interpretability has found nontrivial intermediate structure inside trained transformers, including world\-model\-like state representations and evidence of forward planningLiet al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib193)\); Lindseyet al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib194)\), so the layer’s argument is not that no internal structure exists\. The argument is narrower and sufficient: the training objective neither requires nor certifies causal semantics, and whatever structure emerges is not exposed as checkable machinery, no typed claim, no identification status, no auditable warrant\. A schema is needed precisely because the mechanism offers no interface at which such properties could be verified, whatever is happening inside\.

##### The maximum\-likelihood trap\.

The point has a distributional statement\. Autoregressive decoding often favors high\-probability continuations underPθ​\(continuation∣context\)P\_\{\\theta\}\(\\text\{continuation\}\\mid\\text\{context\}\), a distribution shaped by corpus frequency, instruction tuning, and preference optimization rather than by evidential support\. Decoding strategies vary, but decoding alone does not distinguish evidential support from high\-probability continuation unless evidence is explicitly represented, weighted, and audited\. Call the*modal basin*the region of the model’s conditional distribution containing familiar, high\-probability continuations that tend to reinforce one another during generation\. Reasoning that remains inside this basin tends to reproduce the familiar answer rather than the best\-supported conclusion: the majority view, the standard explanation, the safest hedge\.

Conditioning is the steering wheel out of the basin:Pθ\(⋅∣context,role,contentiousness\)P\_\{\\theta\}\(\\cdot\\mid\\text\{context\},\\allowbreak\\text\{role\},\\allowbreak\\text\{contentiousness\}\)is a different distribution, and its high\-probability continuations may have low probability under the unconditioned distribution\. Adversarial role assignment therefore surfaces continuations that are coherent yet unlikely to be produced by an unconditioned passChang \([2023b](https://arxiv.org/html/2607.12480#bib.bib77)\)\. Whether an exchange actually left the basin, and whether it returned for the right reasons, is a measurable question, monitored by information\-theoretic dials and claim\-level auditsChang \([2024](https://arxiv.org/html/2607.12480#bib.bib78),[2023a](https://arxiv.org/html/2607.12480#bib.bib76)\)\. The trap also explains why purely self\-directed reasoning has no reliable mechanism for escaping it: chain\-of\-thought, self\-evaluation, and self\-consistency all sample from closely related conditionals, so their internal agreement may be a property of the basin, not of the world\.

### 2\.2Layer Two: The Patches Inherit the Mechanism

Two patches dominate current practice: elicit a chain of thought, and train with reinforcement on outcomes\. Both help; neither changes the mechanism, and each fails in the way the mechanism predicts\.

##### Chain\-of\-thought\.

A model can produce a plausible chain because the words and steps are strongly associated in its training distribution, not because the chain faithfully caused the answer\. The empirical record supports this caution\. Chain\-of\-thought explanations can misreport the factors that actually drive a model’s answerTurpinet al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib41)\); answers can survive truncation or corruption of the stated chain, showing that the chain is often post hocLanhamet al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib42)\); and reinforcement\-trained reasoning models reveal exploited hints in their visible chains only a minority of the timeChenet al\.\([2025c](https://arxiv.org/html/2607.12480#bib.bib43)\)\. Measured competence is also brittle: accuracy degrades under superficial perturbations of familiar problemsMirzadehet al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib44)\)and under counterfactual variants of routine tasksWuet al\.\([2024](https://arxiv.org/html/2607.12480#bib.bib45)\)\. Reported collapses of reasoning models at high problem complexityShojaeeet al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib46)\)are themselves disputedOpus and Lawsen \([2025](https://arxiv.org/html/2607.12480#bib.bib47)\), and the dispute is revealing: the field lacks agreed instruments for measuring reasoning quality\. Read against the families of reasoning developed later in this paper, a chain of thought at its best mimics one family only, abduction: it produces a plausible account after the fact\. It performs no validity check, consults no identification condition, and surfaces no defeater, which is why fluency and correctness so often part company\.

##### Reinforcement learning\.

Outcome reward addresses*what*answer was accepted while leaving unclear*why*the answer was correct, why another answer was wrong, or which hidden assumption failedChang and Geng \([2026](https://arxiv.org/html/2607.12480#bib.bib185)\); Chang \([2026c](https://arxiv.org/html/2607.12480#bib.bib189)\)\. The reward is also universal: one scalar per episode, blind to the user, the stakes, the claim type, and the evidence standard the claim required, so the same gradient that fixes one context’s error installs another context’s bias\. The training signal itself is contested: reinforcement learning with verifiable rewards may mostly re\-weight reasoning paths already latent in the base modelYueet al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib51)\), may expand them under prolonged trainingLiuet al\.\([2025b](https://arxiv.org/html/2607.12480#bib.bib52)\), and can register “gains” under spurious rewardsShaoet al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib53)\)\. In the worst case, the model learns to satisfy the evaluator or the user rather than to repair its own reasoning: models defer to users against their own evidencePerezet al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib50)\); Sharmaet al\.\([2024](https://arxiv.org/html/2607.12480#bib.bib49)\), the route from apparent learning to sycophancyChang \([2026a](https://arxiv.org/html/2607.12480#bib.bib184)\)\. The layer’s conclusion is blunt: if the mechanism does not support reasoning, patches that elicit more reasoning\-like text bear no fruit, and rewarding the text’s outcomes teaches the mechanism to imitate success rather than to reason\.

### 2\.3Layer Three: The Formal Absence

The deepest layer of the critique uses the formal constructs of reasoning theory as instruments\. The object of evaluation is the reasoning relation: a typed claim, the reasons offered for it, the hidden warrants that connect premises to conclusion, the evidence standard that applies, the counter\-reasons that would defeat it, and the context in which the claim will be used\. Centuries of work, from argumentation theory through causal inference, have made each part of that relation a formal construct with checkable conditions\. The hammer questions are then direct\. Where, in a language model, is elenchus? Where are the argument roles, the warrant, the qualifier, the rebuttal? Where are the argument schemes and their critical questions, the defeaters and the attack graph, the burden of proof, the evidence standard that changes with the claim type? Where is the declared causal graph? Where isd​o​\(x\)do\(x\), the identification condition that licenses an interventional claim, and the bounds that apply when it is not met?

Table[1](https://arxiv.org/html/2607.12480#S2.T1)enumerates the constructs; the answer to every question is the same, nowhere, and each row names the section of this paper where the TRACE schema gives the construct a field and a test\.

Table 1:The formal\-absence checklist\. Each construct is standard in its home discipline, absent as machinery in a language model, and supplied by TRACE as a field with a test\.The oldest layer of the tradition makes the same point procedurally\. The seven Socratic methods, definition, generalization, induction, elenchus, hypothesis elimination, maieutic, and dialectic, are not opinions about reasoning; they are testing procedures, and each maps onto the formal families: definition and dialectic exercise inductive and deductive reasoning together, generalization and induction are inductive, elenchus and hypothesis elimination are deductive refutation, and maieutic questioning is abductive \(Figure[1](https://arxiv.org/html/2607.12480#S4.F1)\)\. The mapping also reveals what the tradition lacks: it supplies testing procedures but no causal type system; that gap waits two millennia for the ladder of Section[4\.4](https://arxiv.org/html/2607.12480#S4.SS4)\. CRIT operationalizes the tradition for machine evaluation \(Section[4\.1](https://arxiv.org/html/2607.12480#S4.SS1)\), which is why it sits at the roof of the constructive sections\. The hammer question extends accordingly: where, in a language model, is elenchus? A refutation procedure requires a counter\-position held and pressed against the claim, and a solitary model sampling from adjacent conditionals holds none\.

One more absence completes the critique, and it is social rather than formal\. Mercier and Sperber argue that human reasoning evolved not mainly for solitary truth\-seeking, but for producing and evaluating arguments in social exchangeMercier and Sperber \([2011](https://arxiv.org/html/2607.12480#bib.bib54),[2017](https://arxiv.org/html/2607.12480#bib.bib55)\)\. The same lesson applies to machine reasoning\. A claim must be asked: What supports it? What would defeat it? What evidence is missing? What alternative explanation has not been considered? Evaluation is therefore a division of labor: one side proposes a claim, and another side tests it\. A solitary model, sampling from closely related conditionals, contains no second side\.

##### From critique to construction\.

The three layers agree on what is missing, and it is not more reasoning\-like text\. Every instrument surveyed above elicits reasoning; none of them writes reasoning down in a form that survives the episode\. A chain of thought is discarded when the answer is emitted; a debate transcript is prose, not data; a reward is a scalar that remembers nothing about which step earned it\. What the field lacks is a shared, typed, versioned schema for recording what was claimed, what kind of reasoning the claim required, what evidence and warrants were offered, what was tested and how, what failed, what is missing, and what verdict followed, with each field backed by the formal construct that Table[1](https://arxiv.org/html/2607.12480#S2.T1)names\. Procedures such as chain\-of\-thought, retrieval, self\-critique, discussion, refutation, and debate remain useful, but only insofar as they improve the typed support relation the record captures\. They are instruments; the reasoning relation is the object\. The next section defines the record\.

## 3The TRACE Schema

### 3\.1Terminology Definition

Table 2:The complete TRACE vocabulary\. Eight terms, no aliases, every term tied to TRACE\.
### 3\.2Schema Definition and Key Fields

The TRACE schema is a versioned JSON type definition, stated in the vocabulary of Table[2](https://arxiv.org/html/2607.12480#S3.T2)\. Version 1\.0 defines two types: the rootTraceRecordand its causal specializationTraceCausalRecord\. The complete type listings appear in Appendix[A](https://arxiv.org/html/2607.12480#A1); the fields that carry the paper’s arguments are:

- •*Typing fields*:claim\_text,claim\_type,reasoning\_family,argument\_scheme\. They fix the standard a claim is tested against \(Sections[4\.1](https://arxiv.org/html/2607.12480#S4.SS1)and[4\.3](https://arxiv.org/html/2607.12480#S4.SS3)\)\.
- •*Anatomy fields*:premises,warrants,role\_states,defeaters\. They make hidden premises visible, including reconstructed enthymemes and their charity delta \(Section[4\.3](https://arxiv.org/html/2607.12480#S4.SS3)\)\.
- •*Evidence and settlement fields*:evidence,source\_credibility,criticisms,attack\_graph\. They carry audited support and the accounting of objections \(Section[4\.3](https://arxiv.org/html/2607.12480#S4.SS3)\)\.
- •*Causal fields*, inTraceCausalRecord:causal\_rung,graph\_fragment,identifiability\_status,bounds,missing\_evidence\(Section[4\.4](https://arxiv.org/html/2607.12480#S4.SS4)\)\.
- •*Outcome fields*:final\_status,score\_vector,failed\_gates,missing,repair,cost\(Section[6](https://arxiv.org/html/2607.12480#S6)\)\.
- •*Infrastructure fields*:schema\_version,writer\_id,policy\_version,provenance,revision\_history,regret\_score,consumer\_actions\(Sections[6](https://arxiv.org/html/2607.12480#S6)and[7](https://arxiv.org/html/2607.12480#S7.SS0.SSS0.Px1)\)\.

Three rules govern the schema\. First, outcomes are fields: the verdict, score vector, failed gates, missing items, repair, and cost live inside the record, not beside it; a verdict separated from the record that justifies it cannot exist, and every consumer reads one artifact\. Second, versioning: fields are added by version and deprecated by version, never silently mutated; every consumer declares the schema versions it can read; freezing version 1\.0 is the first phase of the implementation plan\. Third, judgment: a schema cannot be optimal, only adequate, and adequacy is measured on three axes, expressiveness \(can the record represent rung collapse, an implicit but essential warrant, a defer that names its missing evidence\), cost \(reported as CostNorm in the metric vector\), and consumer value \(the admission, plan\-gating, and closed\-loop experiments of the next implementation plan\)\. None of these axes requires an objective scalar of reasoning quality\.

### 3\.3The TRACE Writer

The TRACE writer is an API\. Its signature:

> ``` write(record: TraceRecord) -> WriteResult WriteResult = { record_id, schema_version, policy_version, validation: pass | fail(violations), stored: true | false } ```

The input is a JSON document declaring conformance to a TRACE schema version\. The service validates the document against that version; on pass, the record is parsed and stored in the TRACE repository under the TRACE schema; on fail, the write is rejected with the violated constraints named, and nothing is stored\. Writes are atomic and the repository is append\-only: a record is never mutated in place, and corrections enter throughrevision\_history\.

Who constructs the record is deliberately outside the API\. Any process that emits a valid record is a conforming writer: a caller can construct its own, a single structured model call can emit one whole, and the reference writer, an eight\-stage procedure that builds a record from raw text under per\-stage contracts, is specified in Appendix[B](https://arxiv.org/html/2607.12480#A2)with its algorithms, stage table, modality contract, and cost model\. TRACE\-Bench compares the reference writer against the single\-pass writer as the staged decomposition’s null hypothesis \(Section[E](https://arxiv.org/html/2607.12480#A5)\)\.

One objection should be met head\-on rather than left implicit\. The reference writer’s model\-mediated stages use the very mechanism whose limits Section[2](https://arxiv.org/html/2607.12480#S2)argues, so the specification may appear circular: the associative generator audits itself\. The answer is architectural\. The schema converts an unreliable*generator*into a checkable*proposer*: every model\-mediated emission lands in a typed field that validation, gates, perturbations, and the human\-agreement boundary can check, and nothing a stage emits becomes durable except through those checks\. The specification does not assume the stages execute reliably; it requires their reliability to be measured at the typing boundary, and the falsification criteria of Section[7](https://arxiv.org/html/2607.12480#S7)already stake the pipeline’s warrant on that measurement\.

### 3\.4The Record\-Consumer Contract

The writer produces records; consumers, memory admission, plan gating, temporal regret, and verdict reuse, act on them\. Because a record is the only channel between reasoning and durable action, the relationship between a record and a consumer is itself a contract, and stating it is part of the specification\. ThewriteAPI above, and theconsumeAPI of the companion implementation, give the call signatures; this subsection states the invariants those calls must preserve, which a signature alone does not capture\. The contract has two sides\.

The record guarantees, to any conforming consumer, four properties\.*Immutability*: a stored record is never mutated in place, so a consumer that has read record versionvvmay rely onvvremaining exactly as read; corrections arrive as new revisions linked throughrevision\_history, never as silent edits\.*Version legibility*: every record names the schema and policy versions under which it was written, so a consumer can refuse a record whose versions it does not understand rather than misinterpret it\.*Self\-containment*: the verdict, failed gates, missing items, and repair live inside the record, so a consumer never needs the writer’s private state to act\.*Provenance*: every field is attributable to the stage and inputs that produced it, so a consumer can audit why a verdict holds\.

The consumer, in return, accepts four obligations\.*Version pinning*: a consumer declares which schema and policy versions it reads and rejects records outside that set rather than coercing them\.*Fail closed*: an incomplete, invalid, or unreadable record, or one whose decisive evidence is unavailable, must never be treated as acceptance; the safe default is to withhold the durable action\.*Authority separation*: the record’s verdict is advisory about*warrant*, not a grant of*authority*; the consumer owns the mapping from verdict to durable action under its own declared, versioned policy, and a consumer may hold a technically acceptable claim when its own stakes demand more, or escalate to a human\.*Action write\-back*: a consumer that acts on a record appends its decision toconsumer\_actions, so the record accumulates its own downstream consequences and the loop from reasoning to action to outcome stays auditable end to end\.

Two corollaries follow that the flood vignette \(Section[7\.1](https://arxiv.org/html/2607.12480#S7.SS1)\) makes concrete\. First, a TRACE verdict and a consumer decision are different objects:deferis a property of the record, whileHoldis a plan\-gate action the consumer writes back after reading it\. Second, the same record may license different actions for different consumers, because authority separation lets a high\-stakes consumer withhold what a low\-stakes consumer would commit, without any change to the record\. The contract is what makes the schema an operational interface rather than a passive document\.

## 4The Reasoning Hierarchy

The schema’s typing fields promise that every claim is tested at the standard its type demands\. This section builds the hierarchy that redeems the promise: the Socratic roof, the four evaluation layers, the reasoning families with their argumentation toolbox, and Pearl’s ladder as the causal type system\. The hierarchy has a cast, and Table[3](https://arxiv.org/html/2607.12480#S4.T3)introduces it at once: each figure, what he contributed, and which part of the TRACE contract his machinery became\. The scholars who contest the lists, Wellman, Govier, Pollock, Gordon, Freeman, and Hitchcock, appear in Appendix[C](https://arxiv.org/html/2607.12480#A3), where the contested extensions live\.

Table 3:The hierarchy’s cast: five bodies of theory and the TRACE machinery each became\. The contested extensions and their proponents are in Appendix[C](https://arxiv.org/html/2607.12480#A3)\.### 4\.1Reasoning Before Debate: The CRIT Roof

Section[1](https://arxiv.org/html/2607.12480#S1)separated reasoning from the facilities that may elicit or test it\. This section uses debate as the clearest stress case for that distinction\. Debate is the most explicit adversarial reasoning facility: it assigns roles, creates opposition, elicits counter\-reasons, and makes disagreement visible\. But the same point applies to chain\-of\-thought, retrieval, self\-critique, discussion, and refutation\. None of these facilities defines reasoning\. Each is useful only insofar as it improves the support relation linking a claim to its reasons, evidence, warrants, counter\-reasons, and context\.

Reasoning and debate therefore answer different questions\.*Reasoning*asks whether a conclusion is supported by premises under a valid or acceptable inference pattern\.*Debate*asks whether competing agents or roles can expose missing premises, alternative explanations, contradictions, and overclaims\. Debate can improve reasoning, but it does not define reasoning\. Before debate can be useful, the system must know what kind of claim is being made and what standard of support that claim requires\.

CRIT is the starting point for this volume’s architecture because it supplies a facility\-independent evaluator\. Given a documentdd, CRIT identifies the conclusionΩ\\Omega, extracts supporting reasonsRR, evaluates each reason\-to\-conclusion relationr⇒Ωr\\Rightarrow\\Omega, recursively audits reasons that are themselves claims, seeks rival reasonsR′R^\{\\prime\}, evaluates counter\-reasons, aggregates validity and credibility scores, and then asks for reflective synthesis in contextChang \([2023a](https://arxiv.org/html/2607.12480#bib.bib76)\)\. The public CRIT paper explicitly develops definition, elenchus, dialectic, maieutics, generalization, and counterfactual reasoning as prompt\-template methods connected to inductive, deductive, and abductive reasoningChang \([2023a](https://arxiv.org/html/2607.12480#bib.bib76)\)\.

This gives the paper its roof:

1. 1\.Claim first\.What is the conclusion, and what kind of conclusion is it?
2. 2\.Reason next\.What reasons are offered, and what inference relation connects them to the conclusion?
3. 3\.Evidence and credibility\.What evidence supports each reason, and how credible are the sources?
4. 4\.Rival reasons\.What counterarguments or alternative explanations have not been considered?
5. 5\.Context\.Would the support relation change under a different time, place, population, legal regime, institutional role, or value frame?
6. 6\.Status\.Should the claim be accepted, accepted with qualification, revised, deferred, or rejected?

Table 4:The CRIT roof in one table: each Socratic or legal\-evidentiary method, the evaluator role it plays, the reasoning families it invokes, and the question it asks\. Chain\-of\-thought, retrieval, self\-critique, refutation, discussion, or debate may instantiate some of these operations, but the operations themselves evaluate reasoning\.The legal flavor of CRIT is not accidental\. Elenchus is cross\-examination: it tests whether the reason actually supports the conclusion, whether the support rests on evidence or assertion, whether the source is credible, and whether a reason is itself an unsupported claim\. Dialectic introduces adverse positions and counter\-reasons\. Recursive validation resembles evidentiary tracing: when a premise is a quoted claim, the evaluator follows the source chain\. These are legal\-evidentiary techniques in computational form, even when the domain is not law\. Table[4](https://arxiv.org/html/2607.12480#S4.T4)assembles the full method set: for each CRIT method, the evaluator role it plays, the reasoning families it invokes, and the question it puts to the argument\.

Table[4](https://arxiv.org/html/2607.12480#S4.T4)also fixes a boundary\. Chain\-of\-thought, retrieval, self\-critique, discussion, refutation, SocraSynth, and EVINCE may instantiate some of these operations, and SocraSynth or EVINCE may organize or measure them in debate\. But CRIT defines what all of them are trying to improve: the reason\-to\-claim support relation\.

The third column of Table[4](https://arxiv.org/html/2607.12480#S4.T4)deserves emphasis\. CRIT’s Socratic methods are not another taxonomy competing with deduction, induction, and abduction; they are evaluator operations that invoke those families\. Definition fixes terms before any inference is scored\. Elenchus tests deductive and defeasible support\. Evidence typing and source audit carry the inductive and legal\-evidentiary load\. Dialectic and hypothesis elimination drive abductive comparison\. Generalization checks inductive strength, and counterfactual questioning probes causal and practical robustness\.

1\. definition2\. generalization3\. induction4\. elenchus5\. hypothesis elimination6\. maieutic7\. dialecticInductivereasoningDeductivereasoningAbductivereasoningFigure 1:The Socratic methods and the IDA mapping: each of the seven methods drawn to the classical family or families it exercises \(solid, inductive; dashed, deductive; dotted, abductive\)\. The view is deliberately coarse: Table[4](https://arxiv.org/html/2607.12480#S4.T4)refines it with the defeasible, legal\-evidentiary, causal, and practical families, and a method can invoke more than one family, as hypothesis elimination is deductive in form \(refutation\) and abductive in function \(narrowing the hypothesis set\)\. No method maps to a causal family: the tradition supplies testing procedures but no causal type system, the gap Pearl’s ladder fills in Section[4\.4](https://arxiv.org/html/2607.12480#S4.SS4)\.This mapping is the roof under the rest of the paper\. It also explains why causal reasoning cannot be treated as the whole of reasoning\. A causal claim may require deductive graph reasoning, inductive estimation, abductive diagnosis, and counterfactual evaluation\. Those operations play different roles and should not be collapsed into one score\.

### 4\.2Four Layers of Typed Evaluation

Table[5](https://arxiv.org/html/2607.12480#S4.T5)situates the related work by the layer it serves\.

The typed\-reasoning space divides into four layers\. Each layer answers a different evaluator question\.

1. 1\.Claim types:factual, definitional, causal, counterfactual, predictive, diagnostic, normative, legal, policy, preference, explanatory\.
2. 2\.Reasoning types:deductive, inductive, abductive, analogical, causal, probabilistic, defeasible, practical, deontic, dialectical, narrative, contextual\.
3. 3\.Argument schemes:expert opinion, witness testimony, sign, cause–effect, correlation–cause, analogy, precedent, rule, consequences, practical reasoning, values, ignorance, popular opinion, slippery slope, threat, commitment\.
4. 4\.Context and value evaluation:audience, stakes, vulnerability, culture, role, institutional norm, fairness, harm, autonomy, reversibility, proportionality\.

The second and third layers are easy to conflate, and the distinction matters: a reasoning family names the formal standard of inference, while an argument scheme names a recurring warrant pattern in discourse\. A causal claim may instantiate a correlation\-to\-cause scheme, but the causal family supplies the rung and identification tests\. The scheme tells the evaluator which critical questions to ask; the family tells it which formal standard the answers must meet\.

The literature base supports these layers but does not by itself provide an agentic evaluation architecture\. Argumentation schemes classify defeasible reasoning moves and associate each move with critical questionsWaltonet al\.\([2008](https://arxiv.org/html/2607.12480#bib.bib111)\); Walton \([2006](https://arxiv.org/html/2607.12480#bib.bib112),[2013](https://arxiv.org/html/2607.12480#bib.bib113)\)\. Computational argumentation formalizes attacks, defenses, and acceptability among competing argumentsDung \([1995](https://arxiv.org/html/2607.12480#bib.bib114)\); Prakken and Vreeswijk \([2002](https://arxiv.org/html/2607.12480#bib.bib117)\); Baroniet al\.\([2011](https://arxiv.org/html/2607.12480#bib.bib118)\); Bench\-Capon and Dunne \([2007](https://arxiv.org/html/2607.12480#bib.bib119)\), with structured frameworks supplying the internal anatomy of premises, rules, and attacksPrakken \([2010](https://arxiv.org/html/2607.12480#bib.bib115)\); Modgil and Prakken \([2014](https://arxiv.org/html/2607.12480#bib.bib116)\); Toni \([2014](https://arxiv.org/html/2607.12480#bib.bib120)\)\. Value\-based and practical\-reasoning accounts show why the same evidence can support different conclusions under different goals, audiences, and valuesBench\-Capon \([2003](https://arxiv.org/html/2607.12480#bib.bib121)\); Atkinson and Bench\-Capon \([2007](https://arxiv.org/html/2607.12480#bib.bib122)\)\. Argument mining and argument\-quality research provide computational tools for extracting and assessing claims, premises, support, attack, and quality dimensionsLippi and Torroni \([2016](https://arxiv.org/html/2607.12480#bib.bib91)\); Lawrence and Reed \([2019](https://arxiv.org/html/2607.12480#bib.bib123)\); Wachsmuthet al\.\([2017](https://arxiv.org/html/2607.12480#bib.bib124),[2024](https://arxiv.org/html/2607.12480#bib.bib125)\); Liet al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib126)\)\. Context\-aware reasoning motivates DIKE–ERIS: argument relevance depends on task, role, stakeholder, situation, and institutional settingDey \([2001](https://arxiv.org/html/2607.12480#bib.bib128)\); Pereraet al\.\([2014](https://arxiv.org/html/2607.12480#bib.bib129)\); Duet al\.\([2024a](https://arxiv.org/html/2607.12480#bib.bib130)\)\. Surveys of reasoning with language models map the modern elicitation and evaluation landscapeQiaoet al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib90)\); Sunet al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib132)\); Liuet al\.\([2025a](https://arxiv.org/html/2607.12480#bib.bib133)\); Huang and Chang \([2023](https://arxiv.org/html/2607.12480#bib.bib131)\), and causal inference disciplines embedded causal subclaimsPearl \([2009](https://arxiv.org/html/2607.12480#bib.bib92)\); Imbens and Rubin \([2015](https://arxiv.org/html/2607.12480#bib.bib95)\); Peterset al\.\([2017](https://arxiv.org/html/2607.12480#bib.bib96)\); Guoet al\.\([2020](https://arxiv.org/html/2607.12480#bib.bib97)\); Federet al\.\([2022](https://arxiv.org/html/2607.12480#bib.bib98)\)\.

Four further traditions bear directly on the evaluator\. First, deductive reasoning has a mature verification stack of its own\. Automated theorem proving and proof assistants such as Lean give the deductive family a machine\-checkable standardde Moura and Ullrich \([2021](https://arxiv.org/html/2607.12480#bib.bib139)\); benchmarks now bridge informal and formal proofWellecket al\.\([2021](https://arxiv.org/html/2607.12480#bib.bib137)\); Zhenget al\.\([2022](https://arxiv.org/html/2607.12480#bib.bib138)\); Jianget al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib140)\); and trained verifiers with process supervision bring stepwise checking to natural\-language reasoningCobbeet al\.\([2021](https://arxiv.org/html/2607.12480#bib.bib136)\); Lightmanet al\.\([2024](https://arxiv.org/html/2607.12480#bib.bib135)\)\. When the family typing of Section[4\.3](https://arxiv.org/html/2607.12480#S4.SS3)says deductive, these are the plug\-in checkers\. Second, belief revision and truth maintenance supply the formal account of how a TRACE record must change when a premise falls: AGM belief revisionAlchourrónet al\.\([1985](https://arxiv.org/html/2607.12480#bib.bib141)\), truth\-maintenance systemsDoyle \([1979](https://arxiv.org/html/2607.12480#bib.bib20)\); de Kleer \([1986](https://arxiv.org/html/2607.12480#bib.bib21)\), and nonmonotonic logicsReiter \([1980](https://arxiv.org/html/2607.12480#bib.bib142)\); McCarthy \([1980](https://arxiv.org/html/2607.12480#bib.bib143)\)are the classical machinery behind the record’s revision history and RLER’s policy updates\. Third, decision theory prices the evaluator’s choices: subjective expected utilitySavage \([1954](https://arxiv.org/html/2607.12480#bib.bib146)\), applied statistical decision theoryRaiffa and Schlaifer \([1961](https://arxiv.org/html/2607.12480#bib.bib145)\), and the value of informationHoward \([1966](https://arxiv.org/html/2607.12480#bib.bib144)\)ground ERM’s regret weights, the stopping rules of Section[5\.4](https://arxiv.org/html/2607.12480#S5.SS4), and the defer verdict’s obligation to name the evidence that would most reduce uncertainty\. Fourth, the AI\-and\-law tradition formalizes the legal\-evidentiary techniques CRIT borrows: case\-based reasoning with precedentsAshley \([1990](https://arxiv.org/html/2607.12480#bib.bib147)\), dialectical assessment of conflicting legal argumentsPrakken and Sartor \([1996](https://arxiv.org/html/2607.12480#bib.bib149)\), case comparison incorporating theories and valuesBench\-Capon and Sartor \([2003](https://arxiv.org/html/2607.12480#bib.bib150)\), and computational models of burden of proofGordonet al\.\([2007](https://arxiv.org/html/2607.12480#bib.bib148)\)\. These give the roof’s cross\-examination, source audit, and burden allocation their disciplinary home\.

One prior artifact deserves direct comparison because it is the closest in kind\. The Argument Interchange Format \(AIF\) is the established schema for representing and exchanging arguments as typed graphs of information and scheme nodes, with an ecosystem of corpora and tools built on itChesñevaret al\.\([2006](https://arxiv.org/html/2607.12480#bib.bib191)\); Lawrenceet al\.\([2012](https://arxiv.org/html/2607.12480#bib.bib192)\)\. TRACE shares AIF’s premise, that arguments deserve a typed, machine\-readable representation, and departs from it on exactly the dimensions an operational consumer needs: AIF records an argument’s structure, while TRACE additionally records its*adjudication*\. The verdict, score vector, failed gates, missing items, and repair are fields of the record rather than annotations beside it; reasoning families route subclaims to formal plug\-ins with rung and identification status for causal claims; the schema is versioned with an append\-only revision history; and the record\-consumer contract of Section[3\.4](https://arxiv.org/html/2607.12480#S3.SS4)states what a downstream system may rely on\. AIF is an interchange format for arguments; TRACE is an interchange format for*evaluated*arguments and the commitments they license\.

The design gap is now precise\. Existing work gives taxonomies, dialogue semantics, extraction tools, causal methods, and context theories\. What the paper supplies is the evaluation\-side organization an agentic system needs: which quality standard applies to which reasoning family, how a mixed argument is decomposed and routed, how causal subclaims are mapped to Pearl’s rungs, and how final judgments are scored, audited, and revised\.

Table 5:How related work is integrated into the four\-layer evaluation architecture\.
### 4\.3Families of Reasoning

Once CRIT identifies the claim and reasons, the evaluator must classify the reasoning family\. A single passage may contain several families at once\. For example, a medical recommendation may contain a diagnostic inference, a causal treatment claim, a probabilistic risk estimate, a practical\-action argument, and a value judgment about patient goals\. Treating all of them as “causal” or all of them as “debate” loses the type\-specific standards\. Table[6](https://arxiv.org/html/2607.12480#S4.T6)defines the ten families this paper uses: the surface signature by which each is recognized in text, the standard it must meet, and its typical failure\.

Table 6:The ten reasoning families: the surface signature by which stage\-zero detection recognizes and routes each family, the quality standard it must meet, and the failure its tests most often catch\.#### 4\.3\.1The Argumentation\-Theory Toolbox: Anatomy, Schemes, Acceptability

Argumentation theory supplies a complementary layer, and it is worth being precise about what each of its three traditions contributes, because the pipeline of this paper uses them for three different jobs\. Toulmin supplies the*anatomy*of a single argument; Walton supplies the*tests*for a single inference pattern; abstract argumentation supplies the*accounting*when many arguments attack one another\. None of the three replaces CRIT; CRIT is the operator that runs all three\.

##### Toulmin: the anatomy of one argument\.

Toulmin’s model reconstructs any single\-step argument into six rolesToulmin \([1958](https://arxiv.org/html/2607.12480#bib.bib110)\): the*claim*\(what is being asserted\), the*grounds*\(the data offered in support\), the*warrant*\(the usually unstated bridge that licenses the step from grounds to claim\), the*backing*\(what supports the warrant itself\), the*qualifier*\(the claimed strength:*certainly*,*presumably*,*some*,*all*\), and the*rebuttal*\(the conditions under which the claim would fail\)\. Toulmin’s own example carries all six in one sentence:*Harry was born in Bermuda*\(grounds\),*so, presumably*\(qualifier\),*Harry is a British subject*\(claim\),*since a man born in Bermuda will generally be a British subject*\(warrant\),*on account of the relevant statutes*\(backing\),*unless both his parents were aliens*\(rebuttal\)\. Two properties of the layout matter as much as the roles themselves\. First, the reconstruction is not unique: the boundary between grounds and warrant shifts with the analyst, which is why the pipeline records the reconstruction it chose and why the charity delta measures that choice’s distance from the argument as stated\. Second, roles may be empty: a deductively valid argument has no substantive qualifier or rebuttal, and an evaluable record marks such roles absent rather than inventing content for them\. The anatomy matters operationally because most real arguments arrive with only the claim and the grounds visible; the warrant is hidden, the qualifier is inflated, and the rebuttal is missing\. Three of CRIT’s core moves are exactly the recovery of the missing roles: warrant explicitness surfaces the bridge, quantifier consistency compares the qualifier against the grounds, and rival\-reason search asks for the rebuttal the author omitted\. The Toulmin roles are also where theTraceRecordof Section[3](https://arxiv.org/html/2607.12480#S3)gets its fields; Table[13](https://arxiv.org/html/2607.12480#A3.T13)maps each role to the record field and pipeline component that handles it\.

##### Is the six\-role list complete, and does it matter?

No completeness theorem exists at the level of natural argument, and the literature contests the list in both directions, proposing additions such as the counter\-consideration, the undercutting defeater, and presumption with burden of proof, and proposing merges in the other directionWellman \([1971](https://arxiv.org/html/2607.12480#bib.bib154)\); Govier \([1987](https://arxiv.org/html/2607.12480#bib.bib155)\); Pollock \([1987](https://arxiv.org/html/2607.12480#bib.bib156)\); Gordonet al\.\([2007](https://arxiv.org/html/2607.12480#bib.bib148)\); Freeman \([1991](https://arxiv.org/html/2607.12480#bib.bib157)\); Hitchcock and Verheij \([2006](https://arxiv.org/html/2607.12480#bib.bib158)\)\. Two answers make the schema robust to this churn\. The first is formal: in the structured\-argumentation core used here, one argument step is premises, rule, and conclusion, so its attack surfaces are exactly three, premise, rule, and conclusion, and closure holds by constructionPrakken \([2010](https://arxiv.org/html/2607.12480#bib.bib115)\); Modgil and Prakken \([2014](https://arxiv.org/html/2607.12480#bib.bib116)\); Toulmin’s six roles are a readable projection of that closed core, and every proposed extension already has a home, counter\-considerations and undercutters in the attack graph, presumption and burden in external adjudication\. The second is architectural: the schema does not adjudicate taxonomy disputes, it versions them\. The role vocabulary, the scheme registry, the family taxonomy, and the gate metrics are all versioned under the TRACE policy, so adding a role, subtracting a scheme, or admitting a future theory’s family and its critical\-question battery is a version bump, not a refutation; the contract absorbs theoretical progress the way durable interchange standards do, by extension rather than by breakage\. Appendix[C](https://arxiv.org/html/2607.12480#A3)carries the full two\-tier development and the extended anatomy table \(Table[13](https://arxiv.org/html/2607.12480#A3.T13)\)\.

##### Role states and reconstruction\.

Number the rolesR1R\_\{1\}claim throughR6R\_\{6\}rebuttal, with extensionsR7R\_\{7\}counter\-considerations,R8R\_\{8\}undercutters, andR9R\_\{9\}presumption and burden\. For a given segment, each role is in exactly one of five states, the values of the record’srole\_statesfield:*stated*,*elicited*\(supplied by the source in answer to a query\),*reconstructed*\(supplied by the evaluator and marked as such\),*absent*, or*contradicted*\(the source’s own statements are mutually inconsistent on this role\)\. Reconstruction is a procedure, not an art: roles are filled in dependency order; the source is queried when one is available, which is CRIT’s elenchus in interactive form; a missing warrant is abduced as the minimal bridging proposition, and if no plausible bridge exists the step is rejected as a non sequitur, a hard gate\. The procedure emits four scores consumed by Section[6](https://arxiv.org/html/2607.12480#S6): warrant explicitness, critical\-question coverage, quantifier consistency, and the charity deltaΔ\\Delta, the load\-weighted share of reconstructed roles\. Elicited roles improve the argument; reconstructed roles improve only the evaluator’s reconstruction of it, andΔ\\Deltapreserves the difference\. The full algorithm, with its formulas and its human\-audit control on reconstructed warrants, is Appendix[B](https://arxiv.org/html/2607.12480#A2)\.

##### Walton: pre\-packaged cross\-examination\.

Where Toulmin gives the parts list, Walton, Reed, and Macagno give the catalog of*inference patterns*: some sixty presumptive argumentation schemes, each paired with the critical questions that probe precisely where that pattern failsWaltonet al\.\([2008](https://arxiv.org/html/2607.12480#bib.bib111)\); Walton \([2006](https://arxiv.org/html/2607.12480#bib.bib112),[2013](https://arxiv.org/html/2607.12480#bib.bib113)\)\. A scheme is a defeasible template: it confers provisional support, not proof, and each critical question names a way the presumption can be defeated\. The practical value for an automated evaluator is hard to overstate: once the scheme is identified, the cross\-examination comes pre\-written \(Table[7](https://arxiv.org/html/2607.12480#S4.T7)\)\. The evaluator does not need to invent objections; it needs to*route*to the right question list and record which questions were answered, which were dodged, and which defeat the argument outright\. That record is the scheme\-fit and critical\-question coverage that Section[6](https://arxiv.org/html/2607.12480#S6)scores\. In the reconstruction algorithm above, Walton schemes supply the question set used in the rebuttal, undercutter, and counter\-consideration steps; their answered\-over\-applicable ratio is the critical\-question coverage component ofSchemeFit\\mathrm\{SchemeFit\}\.

The correlation\-to\-cause questions of Table[7](https://arxiv.org/html/2607.12480#S4.T7)shadow the causal family’s own tests: the third\-factor question is the confounder audit, the reverse\-causation question is edge orientation, and the survives\-intervention question is identification itself\. The resemblance is the design, not a redundancy: a scheme’s critical questions are the cheap, discourse\-level interface to a family’s formal battery, and one rule prevents double charging\. Where a question has a formal counterpart, the plug\-in adjudicates and the record marks the question answered by it; where none exists, which holds for most of the catalog \(partial counterparts are emerging, such as source\-credibility models for expert opinion and decision\-theoretic checks for practical reasoning\), the critical\-question answer is itself the verdict, held to the defeasible standard\. In one sentence: schemes screen, families adjudicate\.

Table 7:Walton schemes ship with their own cross\-examination\. Identifying the scheme makes critique systematic rather than improvised: the critical\-question list is the intake test battery for that inference pattern\. Where a family plug\-in supplies a formal counterpart, the plug\-in adjudicates the answer; otherwise the critical\-question answers themselves carry the defeasible verdict\.
##### Dung and structured argumentation: the accounting\.

A real evaluation rarely ends with one argument\. The dialectic stage produces counterarguments; counterarguments attract rebuttals; rebuttals attract undercuts\. Abstract argumentationDung \([1995](https://arxiv.org/html/2607.12480#bib.bib114)\)answers the bookkeeping question this creates: given a set of arguments and an attack relation among them, which arguments are ultimately*acceptable*? The core intuition fits in a sentence: an argument stands if every argument attacking it is itself defeated by an argument that stands: innocent until attacked, reinstated when its attacker falls\. Different semantics make this circle precise in different waysBaroniet al\.\([2011](https://arxiv.org/html/2607.12480#bib.bib118)\), and structured frameworks such as ASPIC\+and assumption\-based argumentation extend the accounting inside arguments, distinguishing attacks on a premise, on the inference rule, and on the conclusion itselfPrakken \([2010](https://arxiv.org/html/2607.12480#bib.bib115)\); Modgil and Prakken \([2014](https://arxiv.org/html/2607.12480#bib.bib116)\); Toni \([2014](https://arxiv.org/html/2607.12480#bib.bib120)\); Prakken and Vreeswijk \([2002](https://arxiv.org/html/2607.12480#bib.bib117)\); Bench\-Capon and Dunne \([2007](https://arxiv.org/html/2607.12480#bib.bib119)\)\. In the pipeline, this layer is the formal backbone of external adjudication \(condition D4 of Section[5\.3](https://arxiv.org/html/2607.12480#S5.SS3)\): after debate has populated the attack graph and CRIT has audited each node, acceptability semantics, not the debaters’ consensus, determines which claims survive\.

##### Division of labor, in one sentence\.

Toulmin tells the evaluator what to extract from one argument; Walton tells it what to ask of one inference pattern; Dung tells it how to settle a field of competing arguments; and CRIT is the evaluator that performs the extraction, asks the questions, and reports the settlement\. These traditions do not replace CRIT; they give CRIT the scheme\-specific questions needed after the claim has been typed\. Section[5\.2](https://arxiv.org/html/2607.12480#S5.SS2)shows all three at work on a single sentence\.

### 4\.4Pearl’s Ladder Inside the Reasoning Roof

Table[8](https://arxiv.org/html/2607.12480#S4.T8)states the mapping this subsection develops: Pearl’s ladder as a causal type system inside the broader roof\.

Pearl’s hierarchy separates three causal question types: association, intervention, and counterfactualPearl \([2009](https://arxiv.org/html/2607.12480#bib.bib92)\); Pearl and Mackenzie \([2018](https://arxiv.org/html/2607.12480#bib.bib93)\)\. In this paper, the hierarchy is not the roof; it is the type system for causal claims inside the roof\.

1. 1\.Rung 1: Association\.What is observed together? This includes correlations, conditional probabilities, classification, prediction, and observational patterns\.
2. 2\.Rung 2: Intervention\.What would happen if an action were imposed? This requires a target such asd​o​\(X=x\)do\(X=x\)and assumptions that distinguish doing from seeing\.
3. 3\.Rung 3: Counterfactual\.What would have happened under an alternative condition, given what actually occurred? This requires an individual or case\-conditioned model, not merely a population association\.

The mapping to reasoning families is as follows\.

Table 8:Pearl’s ladder as a causal type system inside the broader reasoning roof\. Association is valuable formal reasoning, but it supports prediction and hypothesis formation rather than action by itself\.This answers three common confusions\.

##### Is deductive reasoning causal?

No\. Deduction is an inference relation: if the premises are true and the form is valid, the conclusion must be true\. Causality is semantic content about how variables, events, or actions affect one another\. A causal graph plus assumptions may allow some causal conclusions to be deduced, but that does not make deduction identical to causality\. Deduction tests validity; causal reasoning tests intervention, identification, counterfactual dependence, and assumptions\.

##### What is counterfactual reasoning?

Counterfactual reasoning asks about an alternative world conditioned on the actual world\. In Pearl’s structural account, it combines observed evidence with a model and an intervention\-like modification\. Informally, it often requires abduction to infer the case\-specific background, action to modify the antecedent, and prediction to compute the consequentPearl \([2009](https://arxiv.org/html/2607.12480#bib.bib92)\)\. In CRIT terms, counterfactual questioning tests robustness under context shift; in ERM/RLER terms, it is also how the system asks what mistake would have been avoided under a different reasoning policy\.

##### What is the value of association?

Association is not “mere” reasoning\. It is essential for detecting regularities, making predictions, selecting candidate hypotheses, estimating priors, calibrating confidence, and deciding what evidence to collect next\. Its limitation is precise: association cannot by itself support claims about what would happen under intervention or what would have happened otherwise\. A typed evaluator should preserve association’s value while blocking rung collapse\.

For interventional claims, once a causal graph and evidence set are declared, the target effect may be identifiable, partially identifiable, or underidentified\. If the available evidence cannot support the requested causal quantity, the correct output is not a confident answer\. It is a bounded statement: the claim is underidentified, the available bounds are reported when possible, and the missing evidence needed to tighten the claim is namedShpitser and Pearl \([2006](https://arxiv.org/html/2607.12480#bib.bib103)\); Manski \([1990](https://arxiv.org/html/2607.12480#bib.bib101)\); Balke and Pearl \([1997](https://arxiv.org/html/2607.12480#bib.bib102)\)\. This turns defer from a conversational refusal into a technical outcome\.

###### Proposition 1\(Rung\-boundedness of debate\)\.

Relative to a declared structural\-model class and assumption set, if every unit in the evidence ledger is rung\-1 \(observational\) and the claim under debate is rung\-2 or rung\-3, then no debate protocol, of any length and among any number of agents, can warrant a verdict tighter than the partial\-identification bounds that ledger licensesManski \([1990](https://arxiv.org/html/2607.12480#bib.bib101)\); Balke and Pearl \([1997](https://arxiv.org/html/2607.12480#bib.bib102)\); Shpitser and Pearl \([2006](https://arxiv.org/html/2607.12480#bib.bib103)\)\. The optimal reachable verdict is the bound, or defer\. Debate can add evidence, and it can surface assumptions which, once granted, change the ledger’s rung composition; but exchange alone cannot climb the ladder\.

Proposition[1](https://arxiv.org/html/2607.12480#Thmproposition1)is the point at which the paper’s two directions meet:*Pearl’s hierarchy bounds what debate can achieve; debate determines whether the bound is achieved\.*The semantic axis sets the ceiling; the procedural axis determines the distance to it\.

## 5The Writer Stages

This section depicts the reference writer’s stages in operation: the formulation gate, one claim traced from sentence to verdict, the gated debate of stage 5, and the rules that stop an evaluation\. Figure[2](https://arxiv.org/html/2607.12480#S5.F2)shows the writer whole: the formulation gate decides whether the text formulates an argument at all, and stages 1 to 7 then construct, test, settle, and score the record\. The stage table, the two algorithms, the detail tables, and the cost model are in Appendix[B](https://arxiv.org/html/2607.12480#A2)\.

inputtextformulationgate1intake2type, route3question4test5elicit6settle7verdict, scoreTraceRecordwrittenargumentFigure 2:The reference writer\. The formulation gate admits text that formulates an argument and rejects the rest at negligible cost; stages 1 to 7 construct, test, settle, and score the accepted argument’s TRACE record\. Stage 5 is gated by flags, stakes, and budget; contracts, algorithms, and cost are in Appendix[B](https://arxiv.org/html/2607.12480#A2)\.### 5\.1The Formulation Gate: Detecting and Typing Reasoning in Narrative

The pipeline of this paper begins with a typed claim\. But narratives do not arrive typed; most text is narration, description, or reporting, and only some segments*conduct reasoning*\. The evaluator therefore needs a stage\-zero operator,

ℛ:text⟶\{\(segment,claim,premises,markers,family,scheme,confidence\)\},\\mathcal\{R\}:\\text\{text\}\\longrightarrow\\\{\(\\text\{segment\},\\text\{claim\},\\text\{premises\},\\text\{markers\},\\text\{family\},\\text\{scheme\},\\text\{confidence\}\)\\\},which segments the narrative, decides which segments reason, and routes each reasoning segment to its family\-specific tests\. Detection rests on three signal classes, in decreasing order of reliability\.

1. 1\.Structural dependence \(the criterion\)\.A segment conducts reasoning if and only if it advances a claim whose acceptance is presented as depending on other assertions in the segment; that is, Toulmin roles \(claim, grounds, warrant\) are assignableToulmin \([1958](https://arxiv.org/html/2607.12480#bib.bib110)\)\. Narration and description fail this test: they assert without staking acceptance on support\.
2. 2\.Inferential markers \(the flag\)\.*Therefore*,*because*,*it follows*,*suggests that*,*must have*,*if…then*, and concessives \(*although…nevertheless*\)\. Markers are sufficient to flag but not necessary: enthymemes carry no markers, which is why the charity delta belongs partly to detection: the reconstruction cost of an unmarked argument is itself a detection output\.
3. 3\.Family signatures \(the router\)\.Each reasoning family has a lexical–modal profile that both identifies it and routes it; the surface signature column of Table[6](https://arxiv.org/html/2607.12480#S4.T6)lists these profiles\.

Two properties make this stage more than preprocessing\. First, it is*measurable*: segmentation and typing can be scored against annotated argument corpora with standard extraction metricsLippi and Torroni \([2016](https://arxiv.org/html/2607.12480#bib.bib91)\); Lawrence and Reed \([2019](https://arxiv.org/html/2607.12480#bib.bib123)\), giving the success\-criteria table of Section[6](https://arxiv.org/html/2607.12480#S6)a row it otherwise lacks: claim\-type accuracy presupposes detection accuracy\. Second, the causal rung markers of Table[6](https://arxiv.org/html/2607.12480#S4.T6)makerung collapse detectable at the surface: a segment whose premises carry rung\-1 signatures and whose conclusion carries rung\-2 modality \(“is associated with…therefore we should administer”\) is flagged before any graph is drawn\. This cheap test catches a large share of causal overclaims at intake, reserving the full identification machinery of Section[4\.4](https://arxiv.org/html/2607.12480#S4.SS4)for claims that survive it\.

### 5\.2From Sentence to Verdict: One Claim Through the Whole Pipeline

The preceding sections introduced many instruments: a detection operator, a claim anatomy, reasoning families, argument schemes, causal rungs, debate\-ness conditions, and convergence metrics\. A reader meeting them for the first time can reasonably ask how they fit together on an actual sentence\. This section answers by tracing one claim end to end, the way a compiler text traces one small program from source code to machine code, so that every operator is seen doing its one job on the same artifact\. The input is deliberately ordinary:

> *“Students who take music lessons score higher on math tests, so schools should fund music programs to raise math scores\.”*

The sentence is a good specimen precisely because it is so agreeable\. It compresses an association, a causal leap, and a policy demand into one fluent breath\. As we will see at Stage 5, its conclusion sits squarely in the modal basin: it is what an unconditioned language model is most likely to say\. The pipeline’s job is to take it apart\.

##### The formulation gate \(Section[5\.1](https://arxiv.org/html/2607.12480#S5.SS1)\)\.

The gate admits the segment and, in admitting it, already finds the story: the premise carries a rung\-1 signature \(*score higher*, an observed association\) while the conclusion carries rung\-2 modality \(*fund…to raise*\) wrapped in practical\-deontic*should*, so the surface rung\-collapse flag fires before any deeper machinery runs, and the segment is routed to three batteries at once: inductive, causal, and practical\.

##### Stage 1: CRIT intake and Toulmin anatomy \(Sections[4\.1](https://arxiv.org/html/2607.12480#S4.SS1),[4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1)\)\.

CRIT fills the TRACE record\. ClaimΩ\\Omega: schools should fund music programs*in order to raise math scores*\. The italicized purpose clause matters, because it makes the policy conditional on a causal mechanism\. Grounds: the observed score gap between students who do and do not take lessons\. Warrant \(hidden, now surfaced\):*the association exists because music training causes mathematical improvement, and what causes improvement should be funded*\. Backing: none offered\. Qualifier mismatch: the grounds are a population\-level tendency; the conclusion is an unqualified prescription\. Rebuttal: absent; the sentence considers no alternative explanation and no downside\. Already the anatomy shows the argument is an enthymeme with a critical hidden warrant; the charity delta will be large\.

##### Stage 2: Family typing \(Section[4\.3](https://arxiv.org/html/2607.12480#S4.SS3)\)\.

The single sentence decomposes into three typed subclaims, each owed a different standard:S1S\_\{1\}\(inductive/statistical\): music students score higher, a claim checkable against data; standard: sample adequacy, reference class\.S2S\_\{2\}\(causal, rung 2\): funding music lessons would*raise*scores; standard: identification\.S3S\_\{3\}\(practical/normative\): therefore schools should fund them; standard: means–end fit, alternatives, side effects, plus a value premise about how school resources ought to be allocated\. Collapsing the three into one verdict is precisely the error the paper is built to prevent:S1S\_\{1\}may be true,S2S\_\{2\}unidentified, andS3S\_\{3\}unjustified, all at once\.

##### Stage 3: Scheme and critical questions \(Section[4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1)\)\.

The argument instantiates two Walton schemes in series:*correlation to cause*\(fromS1S\_\{1\}toS2S\_\{2\}\) and*practical reasoning from positive consequences*\(fromS2S\_\{2\}toS3S\_\{3\}\)\. The critical questions arrive pre\-written\. For correlation to cause: could a third factor produce both variables? \(Family income and parental involvement predict both music lessons and math scores\.\) Could causation run in reverse? \(Mathematically stronger students may be steered toward music\.\) Would the relation survive intervention? \(Unknown; that is exactlyS2S\_\{2\}\.\) For practical reasoning: are there alternative actions that achieve the goal at lower cost? \(Direct math tutoring is an obvious rival\.\) What are the side effects and what is displaced by the funding? The scheme layer has converted a vague unease into an explicit, finite question list, and the record now shows which questions the original argument answered: none\.

##### Stage 4: Pearl typing and identification \(Section[4\.4](https://arxiv.org/html/2607.12480#S4.SS4)\)\.

The causal subclaimS2S\_\{2\}is typed formally: the grounds reportP​\(math∣music\)P\(\\text\{math\}\\mid\\text\{music\}\); the conclusion needsP​\(math∣d​o​\(music\)\)P\(\\text\{math\}\\mid do\(\\text\{music\}\)\)\. With plausible confounders \(household income, parental involvement, school quality\) unmeasured in the ledger, the interventional quantity is not identified from the observational association; the honest output is a bounded statement, not a confident effectManski \([1990](https://arxiv.org/html/2607.12480#bib.bib101)\); Balke and Pearl \([1997](https://arxiv.org/html/2607.12480#bib.bib102)\)\. The evaluator names what would identify it: randomized assignment of music instruction, a lottery\-based natural experiment, or within\-family comparisons\. It also records that controlled studies of music training have generally failed to find the far\-transfer effect on academic outcomes that the argument assumesSala and Gobet \([2020](https://arxiv.org/html/2607.12480#bib.bib153)\)\. Note what just happened: the pipeline did not declare the claim false; it declared it*unwarranted at the requested rung given the present ledger*, which is a different and more useful thing\.

##### Stage 5: The basin, and the debate that escapes it\.

Why is a procedure needed at all? Would a single model not notice all this if simply asked? Often, no, and the maximum\-likelihood trap explains why\.*Music makes children better at math*is a culturally beloved claim; text asserting it vastly outnumbers text dissecting it, so the mode ofPθ\(⋅∣sentence\)P\_\{\\theta\}\(\\cdot\\mid\\text\{sentence\}\)is fluent elaboration, the Mozart\-effect genre, rather than critique\. A single unconditioned pass tends to stay in that basin, andkkunconditioned passes agree with one another because they sample from near\-identical distributions, a point Section[5\.3](https://arxiv.org/html/2607.12480#S5.SS3)makes precise: agreement among strongly correlated belief generators is cheap\. Role conditioning changes the distribution being sampled\. A proponent, a skeptic instructed that its success criterion is finding faults, and a moderator holding contentiousness high force low\-probability, coherent continuations into the exchange: the confounder objection, the reverse\-causation objection, the tutoring alternative, the null experimental literature\. EVINCE’s dials show the signature of a healthy episode: cross\-agent divergence*rises*while the hypothesis set grows, then falls as audited evidence units, not concessions, enter the ledger; the warranted convergence ratio, a metric defined in Appendix[F](https://arxiv.org/html/2607.12480#A6), stays high because each belief movement co\-occurs with a new ledger entry\. The debate need not create new knowledge; often it moves probability mass toward stored or retrievable knowledge that an unconditioned pass would not surface\.

##### Stage 6: Attack graph and acceptability \(Section[4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1)\)\.

The exchange leaves a small argument graph\.AA: the original argument\.BB: the confounding objection, attackingAA’s warrant\.CC: the reverse\-causation objection, also attacking the warrant\.DD: the experimental\-evidence argument \(null far\-transfer results\), attackingS2S\_\{2\}directly\.EE: a reply toBB\(“some studies control for income”\), attackingBB; but CRIT’s audit findsEE’s grounds weak \(the controls are partial\), soEEdoes not defeatBB\. Under grounded semantics,BB,CC, andDDare acceptable;AAis not: every path to reinstatingAArequires defeating all three attackers, and none is defeated\. The verdict is computed from the graph and the audited ledger, not voted by the debaters; that is external adjudication \(D4\) in action\.

##### Stage 7: Verdict, repair, and the metric readout \(Sections[5\.4](https://arxiv.org/html/2607.12480#S5.SS4)–[6](https://arxiv.org/html/2607.12480#S6)\)\.

The typed verdict separates the three subclaims\.S1S\_\{1\}: provisionally accepted if the reported association is supported by the supplied data\.S2S\_\{2\}: defer, underidentified from the ledger; the named missing evidence is randomized or quasi\-experimental assignment; the prior experimental record points against far transfer\.S3S\_\{3\}: rejected as stated, because a practical conclusion cannot stand on an unidentified causal premise, and its critical questions \(alternatives, side effects, value premise\) went unanswered\. The repaired argument a proponent could honestly make is much more modest:*music programs may be worth funding for their intrinsic and demonstrated musical benefits, but not on the promise of raising math scores; if raising math scores is the goal, directly targeted interventions have stronger evidence*\. The distance between the original and this repair is the charity delta, and it is recorded\. The episode also emits its convergence diagnostics: the stop occurred when the log\-odds onS2S\_\{2\}crossed the defer threshold \(Proposition[2](https://arxiv.org/html/2607.12480#Thmproposition2)\), procedural\-invariance reruns with permuted roles reproduce the verdict, and the ledger alone, without the transcript, suffices for a cold judge to re\-derive it\.

Readers who want the same pipeline at clinical stakes will find the illustration in Appendix[D](https://arxiv.org/html/2607.12480#A4)\.

##### What the trace shows\.

The pipeline’s value in this trace did not come from creating new knowledge\. It came from redistributing probability mass so that stored, retrievable, or otherwise available knowledge that an unconditioned pass would not surface gets said, audited, and counted\. That is the paper’s argument in miniature: reasoning quality is not a property of fluent output; it is a property of a record that survives inspection\. The trace is Algorithms[1](https://arxiv.org/html/2607.12480#alg1)and[2](https://arxiv.org/html/2607.12480#alg2)executing line by line; the record states above are their intermediate values\.

##### One architecture, interchangeable plug\-ins\.

The trace ran the causal plug\-in because Pearl’s hierarchy gives causal reasoning the most mature formal semantics, but the architecture is family\-neutral: every family passes the same gate and stages and contributes its own test battery and outputs\. The causal instance is representative, not privileged\. The plug\-in inventory and the causal module mapping are Tables[11](https://arxiv.org/html/2607.12480#A2.T11)and[12](https://arxiv.org/html/2607.12480#A2.T12)in Appendix[B](https://arxiv.org/html/2607.12480#A2)\.

### 5\.3Stage 5: Gated Debate

Stage 5 is the writer’s only unbounded\-cost stage, and it is therefore gated: debate runs only when a hard flag, the stakes, and the budget jointly warrant it\. When it runs, its purpose is elicitation under opposition, surfacing the evidence, alternatives, and defeaters that an unconditioned pass leaves inside the modal basin\. Exchange carries evaluative force only when four properties hold:

1. D1Role separation\.Proponent and critic occupy distinct contexts, not merely distinct sentences in one context window\.
2. D2Objective opposition\.The critic’s success criterion is finding faults, not reaching agreement\. Sycophancy induced by preference training attacks D2 directlySharmaet al\.\([2024](https://arxiv.org/html/2607.12480#bib.bib49)\)\.
3. D3Epistemic diversity\.The parties’ priors and likelihoods are not identical; otherwise agreement is preordained and uninformative\.
4. D4External adjudication\.A standard outside the disputants \(a judge, an evidence ledger, a formal checker, a CRIT audit\) settles the verdict; the disputants’ own consensus does not\.

These four properties define a debate\-ness profile on which every reasoning\-as\-debate method sits, and the empirical record orders itself along it \(Table[18](https://arxiv.org/html/2607.12480#A6.T18), Appendix[F](https://arxiv.org/html/2607.12480#A6)\): methods that realize none of the four fail to self\-correct, same\-model panels that realize only D1 degenerate toward premature agreement, and protocols that enforce D2 and D4 earn formal guarantees\. The SocraSynth, EVINCE, and CRIT line of this research program is an engineering program for restoring D2, D3, and D4 to exchanges that would otherwise possess only D1Chang \([2023b](https://arxiv.org/html/2607.12480#bib.bib77),[2024](https://arxiv.org/html/2607.12480#bib.bib78),[a](https://arxiv.org/html/2607.12480#bib.bib76)\)\.

Classical agreement results issue the right warning first\. Agents with a common prior cannot agree to disagreeAumann \([1976](https://arxiv.org/html/2607.12480#bib.bib81)\); iterated communication reaches consensus that need not aggregate all private informationGeanakoplos and Polemarchakis \([1982](https://arxiv.org/html/2607.12480#bib.bib82)\); and mechanical opinion averaging converges with no truth content at allDeGroot \([1974](https://arxiv.org/html/2607.12480#bib.bib83)\)\. Multi\-LLM panels, whether instantiated from one base model or from several vendors trained on overlapping corpora, are strongly correlated belief generators, so convergence among them should be treated as carrying little evidential weight unless it is coupled to new audited evidence\. Agreement is cheap\. The consequence shapes every metric below:the object that carries evidential weight is not the consensus but the ledger, the set of typed, CRIT\-audited evidence units surfaced during the exchange, together with the trajectory by which beliefs moved\. Every metric that follows is a ledger or trajectory metric, not an agreement metric\. The validity metrics that make these properties measurable, procedural invariance, warranted convergence, evidence\-counterfactual sensitivity, the dual\-phase signature, effective panel size, and the external\-judge gap, are defined in Appendix[F](https://arxiv.org/html/2607.12480#A6), together with the imported agreement and optimality results; the companion volume computes them\.

One boundary closes the stage\. Debate is a search procedure: it can expose assumptions, improve evidence coverage, and reduce overclaiming, but it cannot by itself certify truth, and a converged verdict remains a proposal that must pass the admission control appropriate to its domain before any durable state changes\. The pipeline improves proposals; it does not make them self\-authorizing\.

### 5\.4Stopping Conditions

A reasoning controller should not run until agents merely agree\. The four conditions below are not ad hoc: they are the qualitative face of a sequential test\. Treating each turn’s audited evidence as an increment to the log\-odds of the claim, the controller stops when the accumulated log\-odds cross thresholds set by the costs of the two errors, thresholds that ERM’s regret asymmetries supply directly \(Proposition[2](https://arxiv.org/html/2607.12480#Thmproposition2)\)\. What follows is the operational statement of that rule; the controller stops when one of the following conditions holds:

###### Proposition 2\(Optimal stopping\)\.

Under the idealization that turns yield conditionally independent evidence increments for a binary claim, the sequential probability ratio test with thresholdsA=log⁡1−βαA=\\log\\frac\{1\-\\beta\}\{\\alpha\}andB=log⁡β1−αB=\\log\\frac\{\\beta\}\{1\-\\alpha\}, with error ratesα,β\\alpha,\\betachosen from the regret asymmetries priced by ERM, minimizes the expected number of turns among all stopping rules with equal error probabilitiesWald \([1945](https://arxiv.org/html/2607.12480#bib.bib85)\); Wald and Wolfowitz \([1948](https://arxiv.org/html/2607.12480#bib.bib86)\)\. The four stopping conditions of Section[5\.4](https://arxiv.org/html/2607.12480#S5.SS4)are operational counterparts of this sequential\-testing view: qualified proposal and rejection correspond to threshold crossings, diminishing returns to vanishing increments, and defer to a value\-of\-information judgment that further exchange is not worth its cost\.

1. 1\.Qualified proposal:the claim is typed, grounded, audited, calibrated, and has no unresolved high\-severity objection\.
2. 2\.Rejected claim:the claim fails claim typing, scheme fit, premise support, source credibility, causal rung discipline, identifiability, or contradiction checks\.
3. 3\.Defer:the evaluation exposes missing evidence, underidentification, unresolved defeaters, or value\-context ambiguity that cannot be resolved with available information\.
4. 4\.Diminishing returns:additional exchange produces low evidence gain or repeated arguments without new information\.

The defer outcome is essential\. A reasoning system must be able to say that the evidence is insufficient, the causal quantity is underidentified, the legal standard is unclear, the value frame is unresolved, or the intervention is not defined well enough to support action\. A useful defer decision reports more than*unknown*: it reports the unresolved claim type, missing variables or assumptions, relevant bounds if available, and the next evidence that would most reduce uncertainty\.

## 6Measuring the Writer and the Record

The pipeline is evaluable because each stage produces observable outputs\. The metrics below \(Table[9](https://arxiv.org/html/2607.12480#S6.T9)\) are not measurements of objective reasoning quality; they measure record completeness, writer behavior, gate behavior, and consumer usefulness\. The objective is not agreement, but improvement of the TRACE record\.

Table 9:Writer and record\-quality criteria\. Debate can help improve these dimensions, but the dimensions evaluate the record and its writer, not debate performance\.Success criteria become measurements only when each dimension has a unit, a label source, a scoring rule, and a declared role in the verdict\. Table[15](https://arxiv.org/html/2607.12480#A5.T15)\(Appendix[E](https://arxiv.org/html/2607.12480#A5)\) supplies these operational definitions; the gate column marks the dimensions that participate in the hard gates of the final status rather than in the optimized residual\.

Four boundary notes on the gate column\. First, DetectionF1 is marked as a precondition rather than a gate: TypeAcc is meaningful only over the segments detection found, so a weak detector invalidates the rest of the vector before any gate applies\. Second, claim typing gates operationally even though TypeAcc does not gate benchmarks: in high\-stakes settings, when type confidence falls below a declared threshold the correct action is to defer or route to an external adjudicator, because a mistyped claim makes every downstream verdict malformed\. Third, signed quantities such as CalibrationGain and RegretReduction are normalized before aggregation; negative values are reported explicitly and clipped or rescaled according to the task’s declared scoring convention\. Fourth, defects are counted once: when a critical question is formally subsumed by a family test, as the correlation\-to\-cause questions are by the causal battery, the failure is charged to the family metric \(here RungCollapseRate or BoundViolationRate\), and CQCoverage records the question as answered by the plug\-in rather than charging SchemeFit a second time\.

Several metrics can be estimated with existing or constructible benchmarks\. Rung accuracy and rung\-collapse rates can be measured on tasks organized by association, intervention, and counterfactual reasoning; CausalT5K, CLadder, and correlation\-to\-causation benchmarks provide examples of this style of evaluationGenget al\.\([2026](https://arxiv.org/html/2607.12480#bib.bib108)\); Jinet al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib106),[2024](https://arxiv.org/html/2607.12480#bib.bib107)\)\. Calibration improvement can be measured with proper scoring rules, Brier score, or expected calibration errorGuoet al\.\([2017](https://arxiv.org/html/2607.12480#bib.bib109)\)\. Clinical examples require expert annotation, but the same structure applies: expert labels define relevant alternatives, missing tests, overclaims, and high\-regret errors\.

The benchmark and ablation protocol that operationalizes these metrics and gates, TRACE\-Bench, is specified in Appendix[E](https://arxiv.org/html/2607.12480#A5); its baselines include the single\-pass writer as the staged decomposition’s null hypothesis and ungated debate, this research program’s earlier evaluation mechanism, as the within\-family comparison\.

## 7Claims, Non\-Claims, and Falsification

The paper’s claims should be stated at their exact strength, because both overclaiming and underclaiming misdescribe the design\.

What is claimed\. TRACE is not defended as an optimal eight\-stage reasoning evaluator or as a universal reasoning\-quality metric\. It is defended as a typed, versioned reasoning\-log schema and admission\-outcome explanation contract for agentic systems: the schema and the stage contracts are the contribution, the pipeline is one implementation for writing the log, and the value is measured by whether downstream consumers, Trivium, MACI, ALAS, SagaLLMChang and Geng \([2025](https://arxiv.org/html/2607.12480#bib.bib187)\), and MnemosyneChanget al\.\([2027](https://arxiv.org/html/2607.12480#bib.bib190)\)make better, more auditable, more repairable decisions when they consume the records\. TRACE still performs evaluation\-like functions; what is disclaimed is only the pretension that its verdicts measure reasoning quality against an objective standard\. The scores measure compliance with the declared, versioned TRACE policy, not objective reasoning quality\. Within that policy, some conditional verdicts are largely model\-independent once their representation is declared: identification against a declared graph is decided by do\-calculus, deductive validity by a proof or entailment check, and acceptability by a deterministic semantics over a declared attack graph\. The representation boundary, the graph, subclaims, warrants, and attack edges themselves, remains model\- or human\-mediated and is therefore measured separately by typing, extraction, and human\-audit agreement\. In addition, the perturbation metrics of Section[F\.1](https://arxiv.org/html/2607.12480#A6.SS1), procedural invariance and evidence\-counterfactual sensitivity, require neither gold labels nor policy weights\. Policy dependence is confined to the aggregate scalar of Equation \([1](https://arxiv.org/html/2607.12480#A5.E1)\), which is a reporting device and is never reported without its vector and failed gates\.

##### The consumer claims\.

Two consumer\-facing claims complete the set, and both are central to the falsification criteria below\. Across all four consumers, the constant contribution is not that the accept or reject decision beats a baseline gate; on many items it may tie\. It is that every outcome, admit, reject, and defer alike, arrives with a typed account of why, on what evidence, and what would change it\. This claim is uniform, it survives wherever gating accuracy ties, and it is operational in three measurable senses\. Faithful: the verdict is reconstructable from the record alone by a cold judge, and it moves under evidence ablation, so the stated reasons are decisive rather than decorative\. Actionable: repairs, when applied, flip verdicts at a measurable rate, and defers name evidence that resolves them when supplied\. Consequential: systems that learn from records outperform systems that learn from scalar outcomes on downstream task error and error recurrence\. The closed\-loop experiment of the implementation plan, two identical agents on a multi\-week task stream, one writing memory through the gate and one writing freely, measures the third sense directly, and it is the headline test of the schema’s consumer value\.

Here a division of labor must be stated to avoid an overclaim\. Structural admission, whether the workflow is well formed, its dependencies acyclic, its resource and temporal constraints stated and satisfiable, its preconditions met, belongs to the plan validator and needs no reasoning evaluator; much of it is checkable by solvers and consistency checks\. TRACE gates the justificatory residue: the claims a plan rests on that constraint checking cannot reach\. Why this branch rather than that one; whether the causal premise behind an action supports the expected effect at the rung the plan requires; whether an assumed precondition is evidence\-backed or merely asserted\. TRACE audits the plan’s warrants, not its wiring\. The measured quantity is accordingly the prevented\-error rate on plans that pass structural validation yet fail justification, together with the added latency priced against the cost model of Section[3\.3](https://arxiv.org/html/2607.12480#S3.SS3)\.

### 7\.1Illustrative Consumer Vignette: Flood Search\-and\-Rescue

![Refer to caption](https://arxiv.org/html/2607.12480v1/flood_rescue_graphical_abstract.png)Figure 3:The flood search\-and\-rescue setting for the vignette\. A single Mission Controller runs the mission state, planner, predictive world model, TRACE gate, and action dispatcher on the command computer; a survey drone supplies observations; a rescue boat executes an authorized dispatch; and a human Incident Commander approves high\-stakes actions\. TRACE neither flies the drone nor drives the boat: it audits, at the gate, whether a prediction may authorize an irreversible dispatch\. The scene is illustrative and the consequences are simulated\. The scene is an AI\-generated illustration\.The music\-lessons trace shows how the writer types and evaluates an argument\. A second illustration is needed to show why the resulting record is useful to a consumer\. Consider a simulated flood\-response mission \(Figure[3](https://arxiv.org/html/2607.12480#S7.F3)\) with one central*Mission Controller*, one survey drone, one rescue boat, and one human Incident Commander\. The Mission Controller contains the mission state, planner, predictive world model, TRACE gate, and action dispatcher\. The drone observes; the boat executes an authorized rescue; and the Flood Environment supplies observations and consequences\. Simulation ground truth may contain facts that the Mission Controller has not observed\. In particular, the environment contains a debris obstruction in the North Channel, while the Mission Controller legitimately knows only that the route is unverified\. Hidden simulation state is not a premise available to the writer\.

Four residents are waiting at Riverside Apartments\. The planner proposes three structurally valid candidates: dispatch the boat through the short North Channel, send the drone to inspect that channel, or dispatch the boat through a longer observed\-open South Detour\. A prediction service reports a plan success value of0\.920\.92for the northern dispatch\. The same evidence object records model support of0\.280\.28and an out\-of\-distribution score of0\.820\.82\. These values are deterministic fixtures chosen to expose the record logic; they are not experimental measurements\. The distinction they make concrete is decisive: a high reported success value does not license action when the forecast was produced outside its declared support\. Two record mechanisms act together on the northern claim, and the vignette keeps them distinct: a failed hard support gate*forbids*clearance now, while thedeferstatus with a namedmissingfield and a boundedrepairmakes the hold*resolvable*rather than terminal\. A support failure is thus not a bare refusal; it is a refusal that carries the evidence request that could lift it\.

Table[10](https://arxiv.org/html/2607.12480#S7.T10)follows the records rather than the agents\. The predictive route claim, the bounded evidence\-seeking action, the revised route claim, and the supported alternative each receive their own TRACE record because they are different claims with different warrants\. The consumer decisions are not TRACE verdicts:HoldandClearare plan\-gate actions written toconsumer\_actionsafter the consumer reads the record\.

Table 10:An illustrative TRACE history for flood search\-and\-rescue\. The numbers are pedagogical fixtures, not results\. The example demonstrates how one schema represents defer, evidence\-seeking repair, append\-only revision, and a later consumer action without treating prediction as permission\.RN\(1\)R\_\{N\}^\{\(1\)\}predictive claimdefermissing: route viewplan consumerHoldboatRV\(1\)R\_\{V\}^\{\(1\)\}practical claimverificationaccept/qualifydispatchsurvey dronenew observationNorth Channel blockedRN\(2\)R\_\{N\}^\{\(2\)\}revisionrejectpreserveRN\(1\)R\_\{N\}^\{\(1\)\}RS\(1\)R\_\{S\}^\{\(1\)\}South Detour claimaccept/qualifyhuman approval, thenClearboatrepairlocal replanFigure 4:From prediction to attributable action\. A defer names an information\-gathering repair; the repair has its own record; new evidence creates an append\-only revision; and a later record licenses a different route\. TRACE neither supplies the planner nor controls the drone or boat\. It records the justificatory boundary through which their outputs become durable commitments\.The vignette complements the music\-lessons trace of Section[5\.2](https://arxiv.org/html/2607.12480#S5.SS2)\. The music example runs an argument through the writer and shows why association, intervention, and practical prescription must be typed separately\. The flood example starts after candidate plans exist and shows why the resulting record needsfailed\_gates,missing,repair,revision\_history, andconsumer\_actions\. It also makes the structural–justificatory division concrete: the planner and validator decide whether a route and asset assignment are well formed; TRACE audits whether the claims licensing that commitment are sufficiently warranted\. Appendix[H](https://arxiv.org/html/2607.12480#A8)gives the same vignette as a plan\-gating usage scenario\.

This example is a specification illustration, not evidence that TRACE improves rescue performance or that any predictive model is accurate\. A future closed\-loop study must hold the flood environment, observations, candidate plans, predictive outputs, policies, and scenario seeds fixed while varying the audit substrate\. Pre\-registered outcomes should include unsupported\-commitment rate, false\-hold rate, decision latency, record reconstructability, repair success, and localization of invalidated plan branches\.

The vignette also clarifies how the paper’s measurements should be read, because the same record fields it exercises, verdicts, gates, evidence, warrants, carry very different evidential weight\. The metric vector should accordingly be read in epistemic tiers, from strongest to weakest warrant\. Tier one is formally checkable: rung discipline, identification status, bound violations, and deductive validity have ground truth that is a theorem or an oracle model, not an annotator’s opinion\. Tier two is checkable without any ground truth: the perturbation metrics ask whether a verdict survives role, order, and seed permutation and whether it moves when decisive evidence is ablated and stays when only rhetoric is paraphrased; a memorized or sycophantic verdict is unlikely to survive a sufficiently rich perturbation suite, because the expected verdict transitions are specified at the level of decisive evidence, warrants, and gates rather than surface labels, which makes this tier the contamination\-resistant core of TRACE\-Bench, resistant by test design rather than by secrecy\. Tier three is moderate\-agreement annotation: detection, anatomy, and scheme fit, bounded by inter\-annotator agreement and reported with it\. Tier four is the contested holistic scalar, which inherits the disagreement of its annotators and is quarantined by the label\-source declarations of Table[15](https://arxiv.org/html/2607.12480#A5.T15)\(Appendix[E](https://arxiv.org/html/2607.12480#A5)\)\. The paper’s warrant lives in tiers one and two\.

What is not claimed\. The following are deliberately left open, and stating them as open is part of the contract:

1. 1\.The eight\-stage decomposition is not claimed to be optimal; stages may merge or split as the ablations direct\.
2. 2\.The family taxonomy is not claimed to be complete\.
3. 3\.The score vector is not claimed to be uniquely correct, and its weights are not claimed to be universal; they are governance decisions, versioned and owned by the governance layer \(DIKE and ERIS\) of the companion volume\.
4. 4\.The model\-mediated stages are not claimed to execute reliably; their reliability is measured, not assumed\.
5. 5\.The convergence metrics are not claimed to correlate with truth; they measure stability and responsiveness\.
6. 6\.The benchmark is not claimed to represent all reasoning\.

Falsification\. A decomposed framework has a characteristic failure mode: every wrong verdict can be blamed on an upstream mis\-typing, so the framework never appears to fail\. The defense is to name the judges in advance\. The following criteria are pre\-registered as the results that would count against the design, and no post hoc reattribution to upstream stages is admissible against them:

1. 1\.If the single\-pass writer matches the staged writer on record quality and error attribution at comparable cost, the decomposition is unjustified\.
2. 2\.If the no\-gates ablation shows no separation in WrongAcceptRate on fluent but unsupported or formally invalid items, the gates are decorative\.
3. 3\.If procedural invariance and evidence\-counterfactual sensitivity fail to separate systems that differ in evaluation discipline, the consistency metrics are uninformative\.
4. 4\.If evaluator\-versus\-human agreement at the typing boundary falls below its declared threshold, the downstream formal checks are checking the wrong objects and the pipeline’s warrant collapses with it\.
5. 5\.If the record\-gated agent of the closed\-loop experiment does not outperform its ungated twin on downstream task error and error recurrence, the record earns no consumer value and the schema is an archive, not an instrument\.

One baseline in this program deserves its own sentence\. Ungated debate scored holistically is this research program’s earlier evaluation mechanism, CRIT over SocraSynth and EVINCE exchange; it is retained inside stages 5 and 6 as elicitation and scoring machinery, and retained in TRACE\-Bench as a baseline\. The benchmark is therefore also the volume auditing its predecessor with its own methods, and the comparison is within\-family: what is tested is whether typing and gating improve the instrument that proposed them\.

### 7\.2Limits

No TRACE record, writer, or debate system can guarantee complete causal truth or wise action\. Several limits are unavoidable:

1. 1\.Hidden confounders may remain undiscovered\.
2. 2\.Evidence may be absent, corrupted, or unobservable\.
3. 3\.A causal quantity may be mathematically underidentified\.
4. 4\.Counterfactuals may remain unknowable under the available assumptions\.
5. 5\.Agents may share priors and therefore share errors\.
6. 6\.Correct causal reasoning does not by itself determine wise action\.
7. 7\.Gates verify field consistency, not field truth\. A deterministic gate reads what the record says; a writer that fabricates evidence entries or misstates support can populate fields that pass every gate\. The schema’s countermeasures are provenance and source\-credibility fields that make each entry attributable, and the perturbation suite, which tests that verdicts track decisive evidence; but record faithfulness is not world truth, and the schema does not claim otherwise\.

The aim is therefore not certainty\. The aim is disciplined improvement: claims should become more explicit, better grounded, more robust to critique, less overconfident, easier to revise, and less likely to repeat the same failure modes over time\. The flood vignette adds two further boundaries\. Its simulation truth is not an input to the Mission Controller, and technical TRACE clearance does not manufacture emergency\-command authority\. Both are external constraints on any future implementation\.

## 8Summary

The paper’s product is an artifact and a discipline\. The artifact is the TRACE record: a typed, versioned schema for logging what was claimed, what kind of reasoning the claim required, what evidence and warrants were offered, what was tested, what failed, what is missing, and what verdict followed, with the TRACE causal record as its specialization for the family with the most mature semantics\. The discipline is closure: no durable state change without a record\.

Around the artifact the paper arranged four things\. A type system gives the fields their semantics: CRIT supplies the Socratic roof; the reasoning families and the Toulmin, Walton, and Dung toolbox supply claim types, roles, schemes, and critical questions; Pearl’s ladder supplies rung, identification, and bounds, with association, intervention, and counterfactuals positioned as one disciplined family inside reasoning rather than reasoning itself\. A reference writer, TRACE, fills the fields under stage contracts, with cost controlled by a cascade and escalation gated by flags, stakes, and budget; the writer is one conforming implementation, and the schema outlives it\. A measurement regime holds writer and record to account: hard gates first, scalar optimization on the residual, the vector never reported without its failed gates, and TRACE\-Bench, specified in Appendix[E](https://arxiv.org/html/2607.12480#A5), whose baselines include the single\-pass writer as the decomposition’s null hypothesis and this research program’s earlier evaluation mechanism, ungated debate scored holistically, as the within\-family comparison\. And a set of consumers gives the record its value: memory admission, plan gating on a plan’s warrants rather than its wiring, temporal regret over attributed steps, and verdict reuse\. The music\-lessons trace shows the writer separating an association, an intervention, and a practical conclusion; the flood\-rescue vignette shows a plan consumer usingmissing,repair,revision\_history, andconsumer\_actionsto hold one branch, request evidence, preserve a contradiction, and authorize another\.

The final output for any claim remains a qualified proposal, a rejection, or a defer that names its missing evidence\. A fluent consensus can be evidence that reasoning has improved; it is not truth and it is not authority\. What the record buys is not correctness but attributability: when the system is wrong, the record says where\. Whether that is worth its cost is the falsifiable question the companion volume’s experiments are built to answer\.

## Appendix ATRACE Schema v1\.0: Full Type Listing

TRACE schema v1\.0 defines two types\. The root type:

> ``` TraceRecord = { schema_version, writer_id, policy_version, claim_id, claim_text, claim_type, reasoning_family, argument_scheme, premises, warrants, role_states, evidence, source_credibility, counter_reasons, defeaters, context_frame, value_frame, embedded_subclaims, routing_decisions, criticisms, stage_results, attack_graph, audit_result, score_vector, failed_gates, missing, repair, cost, provenance, regret_score, revision_history, consumer_actions, final_status } ```

For causal claims, the record specializes into the TRACE causal record:

> ``` TraceCausalRecord = TraceRecord + { causal_rung, variables, graph_fragment, assumptions, missing_evidence, alternative_explanations, intervention, counterfactual_query, identifiability_status, bounds } ```

## Appendix BThe Reference Writer: Stages, Algorithms, Cost, and Detail Tables

The reference writer constructs aTraceRecordfrom raw text in eight stages, each holding a declared contract\. The decomposition is a reference architecture induced by the information dependencies among the record’s fields: a claim cannot be tested before it is typed, a scheme’s critical questions cannot be loaded before the scheme is identified, and an attack graph cannot be adjudicated before its edges are audited\. Whether the staging earns its cost is settled by the ablations of Section[E](https://arxiv.org/html/2607.12480#A5)\. This appendix specifies the pipeline at a glance, the two algorithms split at the record boundary, the modality contract, and the cost model\.

##### The pipeline at a glance\.

Everything that follows elaborates one eight\-stage pipeline, shown in Figure[2](https://arxiv.org/html/2607.12480#S5.F2)of the main text\. Stages read and write one shared artifact, the TRACE record; the two algorithms state the executable contract, split at the record boundary\.

The pipeline admits an implementation\-level statement\. TRACE is intended not as a metaphor for evaluation, but as an executable methodology: each stage has a declared input, output, state update, failure mode, and score contribution\. Given an input passage, the evaluator detects whether the passage contains a reasoning relation, constructs aTraceRecord, decomposes the passage into typed subclaims, routes each subclaim to the appropriate reasoning\-family standard, applies scheme\-specific critical questions and formal plug\-ins where available, elicits and audits counter\-reasons when the escalation gate warrants, and emits a typed verdict with a score vector\. This turns the paper’s survey of reasoning theories into an operational pipeline for writing and auditing TRACE records; the sections that follow develop each line, and Section[5\.2](https://arxiv.org/html/2607.12480#S5.SS2)runs the whole algorithm on one sentence\. The statement is split at its natural seam: Algorithm[1](https://arxiv.org/html/2607.12480#alg1)turns raw text into a typed record, and Algorithm[2](https://arxiv.org/html/2607.12480#alg2)turns the typed record into a verdict; the record is the only artifact that crosses the boundary\.

We call the executable formTRACE:*Typed Reasoning Argument Critique and Evaluation*\. TRACE is not a truth oracle and not a replacement for domain authority\. Its output is a structured audit: what was claimed, what kind of reasoning the claim requires, which evidence and warrants support it, which defeaters remain unresolved, what verdict follows, and which error and cost terms were incurred\. The contribution of TRACE is to make reasoning evaluation executable without pretending that all reasoning standards reduce to one mechanism: symbolic checks, causal identification, argument\-scheme questions, model\-mediated critique, retrieval, and human adjudication all become plug\-ins behind a shared record contract\.

Algorithm 1TRACE, Part I: Detect, Intake, and Type \(stages 0–2\)0:Input passage

xx, context

cc
0:Typed TRACE record

RRwith routed subclaims, orNo Argument

1:

D←ℛ​\(x\)D\\leftarrow\\mathcal\{R\}\(x\)\{segment text; detect claim–support relations\}

2:if

DDcontains no support relation above threshold

τR\\tau\_\{R\}then

3:returnNo Argument

4:endif

5:Initialize

RRwith claim text, candidate premises, evidence links, context frame, value frame, and empty critique fields\.

6:Apply CRIT–Toulmin intake to fill claim, grounds, warrants, backing, qualifiers, rebuttals, undercutters, counter\-considerations, and burden fields\.

7:Mark each role in

RRasStated,Elicited,Reconstructed,Absent, orContradicted\.

8:Decompose

RRinto typed subclaims

\{S1,…,Sk\}\\\{S\_\{1\},\\ldots,S\_\{k\}\\\}\.

9:for allsubclaims

SiS\_\{i\}do

10:Assign claim type, reasoning family, argument scheme, and confidence\.

11:Route

SiS\_\{i\}to the corresponding family plug\-in\.

12:endfor

13:return

RR

Algorithm 2TRACE, Part II: Question, Test, Elicit, Settle, and Score \(stages 3–7\)0:Typed record

RRfrom Algorithm[1](https://arxiv.org/html/2607.12480#alg1), evidence ledger

ℰ\\mathcal\{E\}, stakes and budget

0:Completed record

RRcarrying verdict, score vector, failed gates, repair proposal, and cost

1:for allsubclaims

SiS\_\{i\}in

RRdo

2:Attach Walton\-style critical questions and compute critical\-question coverage\.

3:Route any critical question with a formal family counterpart to that plug\-in; record the defect once, at the strongest applicable level\.

4:Apply the family\-specific tests required by

SiS\_\{i\}: validity and soundness; sample and reference class; alternative explanations; Pearl rung, identification, and bounds; means–end fit; rule fit; or value/context fit\.

5:ifa hard flag is raised and declared stakes and remaining budget warrant escalationthen

6:Convene conditioned debate under the stopping rule to elicit counter\-reasons, missing evidence, alternative explanations, undercutters, and rebuttals\.

7:endif

8:Audit each proposed support or attack against the evidence ledger

ℰ\\mathcal\{E\}and write pass, fail, bound, or defer status into

RR\.

9:endfor

10:Build the attack graph

ARA\_\{R\}from surviving supports, rebuttals, and undercutters\.

11:Adjudicate

ARA\_\{R\}with the applicable acceptability rule or external judge and write the verdict

V∈\{Accept,Qualify,Revise,Defer,Reject\}V\\in\\\{\\textsc\{Accept\},\\textsc\{Qualify\},\\textsc\{Revise\},\\textsc\{Defer\},\\textsc\{Reject\}\\\}into

RR\.

12:Write the score vector, failed gates, and cost into

RR: detection, type accuracy, scheme fit, evidence support, warrant explicitness, value transparency, alternative recall, overclaim rate, rung\-collapse rate, bound violations, calibration gain, regret reduction, recurrence, and cost\.

13:Write the missing and repair fields of

RRfrom absent roles, failed critical questions, unresolved defeaters, missing evidence, and failed gates\.

14:return

RR

TRACE treats implementation modality as part of the contract\. A plug\-in may be symbolic, statistical, model\-mediated, retrieval\-mediated, human\-mediated, or hybrid\. The requirement is not that every plug\-in be purely algorithmic; the requirement is that every plug\-in declare its input fields, output fields, failure modes, confidence, cost, and audit trail\. What makes the pipeline automatic is the contract at each step: a detector must emit segments and confidence; a router must emit claim types, families, and schemes; each plug\-in must emit pass, fail, bound, or defer status; and the adjudicator must write a final status and metric vector back to the record\. This is why the record, not the transcript, is the central artifact\.

##### Cost anatomy\.

TRACE is affordable because it is a cascade: each step is a filter, and the expensive machinery runs only on the fraction of inputs that survives the filters ahead of it\. Writingpap\_\{a\}for the fraction of segments that are arguments andpep\_\{e\}for the fraction of arguments whose flags, stakes, and budget trigger the debate escalation in Algorithm[2](https://arxiv.org/html/2607.12480#alg2), the expected cost per segment is

𝔼​\[cost\]≈c0\+pa​\(cintake\+ctests\)\+pa​pe​cdebate,\\mathbb\{E\}\[\\text\{cost\}\]\\;\\approx\\;c\_\{0\}\+p\_\{a\}\\,\(c\_\{\\mathrm\{intake\}\}\+c\_\{\\mathrm\{tests\}\}\)\+p\_\{a\}\\,p\_\{e\}\\,c\_\{\\mathrm\{debate\}\},where the debate term dominates the constants\. Detection is a small classifier over every segment; intake, typing, and scheme selection are one structured call per argument; and the gate keepspep\_\{e\}small and stakes\-directed\. Some family tests, such as proof checking, causal identification, bounds, and graph acceptability, can be implemented algorithmically\. Others, such as practical\-reasoning assessment, source\-credibility judgment, value\-context fit, and alternative generation, may be model\-mediated, retrieval\-mediated, human\-mediated, or hybrid\. Three economies compound over episodes: stakes\-proportional gating ties spend to the expected regret of acting unevaluated; adjudicated claim patterns cache through RLER; and Trivium compiles recurring error classes into cheap formulation\-gate flags and stage\-3 questions\. TRACE therefore does not merely score an isolated argument; it accumulates reusable reasoning\-control policy\. The cascade controls cost, not model dependence: intake, typing, value\-sensitive judgment, and debate remain model\-mediated or human\-mediated where the contract requires them, and the auditing caveats of the reconstruction algorithm apply to them unchanged\.

The immediate design consequence is routing\. A causal\-sounding claim must be routed to causal standards; a legal\-sounding claim to legal and deontic standards; a policy recommendation to practical reasoning, embedded empirical claims, causal subclaims, and value warrants\. A model that saysXXcausedYYhas not yet made an evaluable causal claim unless the evaluator can tell whether the claim is associational, interventional, or counterfactual\. A model that says a policy is justified has not yet made an evaluable practical argument unless the evaluator can separate facts, causal effects, alternatives, costs, institutional constraints, and value assumptions\.

##### Detail tables\.

The family plug\-ins and the causal module mapping:

Table 11:Family plug\-ins in the typed reasoning pipeline\. Every plug\-in runs through the same shared stages; causal reasoning is the worked instance, not the whole architecture\.Table 12:Causal reasoning as the worked plug\-in inside the typed reasoning pipeline\. CRIT is placed before debate because it defines the reasoning standard; SocraSynth and EVINCE are elicitation and measurement procedures\.
##### The enthymeme\-reconstruction algorithm \(stage 1\)\.

Once the roles are numbered, reconstruction stops being an art and becomes a procedure\. Number the rolesR1R\_\{1\}claim,R2R\_\{2\}grounds,R3R\_\{3\}warrant,R4R\_\{4\}backing,R5R\_\{5\}qualifier,R6R\_\{6\}rebuttal, and the extensionsR7R\_\{7\}counter\-considerations,R8R\_\{8\}undercutters, andR9R\_\{9\}presumption and burden\. For a given segment, each role is in exactly one of five states:*stated*\(present in the text\),*elicited*\(supplied by the source in answer to a query\),*reconstructed*\(supplied by the evaluator and marked as such\),*absent*\(marked empty after elicitation and reconstruction both fail\), or*contradicted*\(the source’s own statements are mutually inconsistent on this role\)\. The procedure fills the roles in dependency order, queries the source when one is available, and emits the scores that the metric definitions of Section[6](https://arxiv.org/html/2607.12480#S6)already provide\.

1. 1\.Claim\.ExtractR1R\_\{1\}\. If no claim is found, the segment is not an argument; return it to detection\.
2. 2\.Grounds\.ExtractR2R\_\{2\}\. If absent, query the source:*what supports this claim?*If elicitation fails, mark absent; an ungrounded claim is scored as assertion, not argument\.
3. 3\.Warrant\.Test whetherR2R\_\{2\}defeasibly licensesR1R\_\{1\}without additions\. If not, generate the minimal bridging propositionWWby abduction and present it:*your argument assumesWW; is that your assumption?*Confirmed: elicited\. Unconfirmed but plausible: reconstructed\. No plausibleWW: hard gate; reject the step as a non sequitur\.
4. 4\.Backing\.For each warrant, ask what supports it, and recurse: backing is itself an argument, and the recursive source audit applies to it\.
5. 5\.Qualifier\.Read the stated modality; compute the strongest qualifier the grounds license; if the stated strength exceeds the licensed strength, raise the quantifier\-mismatch flag, an overclaim signal\.
6. 6\.Rebuttal\.Load the identified scheme’s critical questions; every unanswered exception\-type question becomes a candidate rebuttal condition, queried if a source is available\.
7. 7\.Extensions\.Scan for concessive constructions \(R7R\_\{7\}\), applicability challenges \(R8R\_\{8\}\), and default or burden context \(R9R\_\{9\}\); routeR7R\_\{7\}andR8R\_\{8\}to the attack graph andR9R\_\{9\}to external adjudication\.
8. 8\.Score\.Emit four reconstruction scores:WarrantExplicitness\\mathrm\{WarrantExplicitness\}, the support\-weighted fraction of essential warrant roles that are stated or elicited;CQCoverage\\mathrm\{CQCoverage\}, the fraction of applicable critical questions answered; quantifier consistency, a flag comparing the stated qualifier against the strongest the grounds license; and the charity deltaΔ\\Delta, the support\-weighted fraction of essential roles that had to be reconstructed\. Weights reflect each role’s support load\. The reasonableness of the argument*as stated*is reported as this vector under the gate\-then\-optimize rule; elicited roles improve the*argument*, while reconstructed roles improve only the evaluator’s*reconstruction*, andΔ\\Deltapreserves the difference\.

Two remarks close the loop\. First, the queries of steps 2, 3, and 6 are CRIT’s elenchus and dialectic in interactive form: the algorithm is the Socratic method run against a schema, which is why its outputs land directly in the TRACE record rather than in a side channel\. Second, steps 3 and 5 are themselves evaluator judgments; when the evaluator is a language model, they inherit the reliability questions of any model judgment, and sampled human audit of reconstructed warrants is the corresponding control\.

## Appendix CThe Extended Anatomy and the Contested Lists

##### The contested list\.

No completeness theorem exists at the level of natural argument, and the literature contests the list in both directions: proposed additions include the*counter\-consideration*of conductive argumentWellman \([1971](https://arxiv.org/html/2607.12480#bib.bib154)\); Govier \([1987](https://arxiv.org/html/2607.12480#bib.bib155)\), the*undercutting defeater*that attacks the warrant’s applicability rather than the claimPollock \([1987](https://arxiv.org/html/2607.12480#bib.bib156)\), and*presumption with burden of proof*Gordonet al\.\([2007](https://arxiv.org/html/2607.12480#bib.bib148)\), while Freeman would merge the warrant into the grounds altogetherFreeman \([1991](https://arxiv.org/html/2607.12480#bib.bib157)\); Hitchcock and Verheij \([2006](https://arxiv.org/html/2607.12480#bib.bib158)\)\. A taxonomy that serious scholars want to both extend and shrink is a useful schema, not a closed enumeration\.

Table[13](https://arxiv.org/html/2607.12480#A3.T13)assembles the resulting extended anatomy: the six canonical roles, the three best\-supported extensions, their reading in the formal core, and the pipeline component that handles each\. Nothing on the extended list is homeless, and nothing requires the six\-role list to be closed: counter\-considerations and undercutters are resolved in the attack graph of the acceptability layer, and burden of proof is a parameter of external adjudication\.

Table 13:The extended anatomy under the two\-tier view\. Closure holds at the formal tier, where one step has three components and therefore three attack surfaces; the working tier is an open checklist whose known extensions are all handled by existing pipeline components\.

## Appendix DIllustrations at Clinical and Policy Stakes

This section illustrates the record where mistakes are expensive: stylized clinical records, a worked policy argument, and the typed evaluation that replaces holistic scoring\.

### D\.1Stylized Clinical TRACE Records

The pipeline is easiest to understand in clinical reasoning, where causal mistakes have different consequences at different stages\. Diagnosis begins with symptoms and uncertain causal hypotheses\. Test selection asks which observation would reduce uncertainty\. Treatment selection becomes an interventional and often counterfactual question\. Follow\-up asks whether the chosen intervention caused benefit, harm, tolerance, resistance, or recurrence\. The following examples are stylized records used only to illustrate causal reasoning\. They are not clinical advice, and they are not intended to reflect full clinical guideline pathways; they illustrate reasoning\-record transitions\.

##### Example 1: lung cancer treatment\.

Consider a stylized case in which a patient presents with persistent cough, weight loss, and blood\-tinged sputum, and imaging reveals a lung mass\. The initial claim is not yet “this patient has lung cancer\.” It is a diagnostic causal claim: the observed symptoms and image may be produced by infection, inflammation, metastatic disease, or primary lung cancer\. Pearl typing therefore places the early question near association and causal explanation, not yet treatment intervention\. SocraSynth generates the differential diagnosis\. CRIT rejects invalid jumps such as “a lung mass implies cancer” or “symptom improvement after antibiotics proves infection\.” EVINCE then asks what evidence would distinguish the hypotheses: prior imaging, infection markers, biopsy, histology, stage, molecular profile, comorbidities, functional status, and patient preference\.

Only after diagnosis and staging does the reasoning problem become primarily interventional: which treatment is expected to improve outcome under this patient’s disease state and constraints? At this point the TRACE causal record changes from a diagnostic record to a treatment record\. Candidate interventions may include surgery, radiation, systemic therapy, targeted therapy, immunotherapy, palliative symptom control, or clinical\-trial referral\. CRIT rejects treatment claims that ignore stage, histology, molecular evidence, toxicity, or the difference between tumor response and cure\. EVINCE compares evidence for each intervention\. ERM penalizes high\-regret errors such as delayed biopsy, omitted molecular testing, unnecessary toxicity, or overconfident recommendation before sufficient staging evidence\. If treatment fails or recurrence occurs, RCA localizes the failure: wrong diagnosis, incomplete staging, missing biomarker, toxicity underestimated, resistance not monitored, or goals of care mis\-specified\. RLER converts the failure into a revised policy, for example: do not recommend disease\-control treatment before diagnosis, staging, and treatment\-relevant evidence are explicit\. Trivium then evaluates whether future cases show fewer missing\-test, premature\-treatment, or recurrence\-monitoring errors\.

The example shows why the pipeline must distinguish a valid causal claim from an actionable causal proposal; a second stylized record, early cognitive decline with treatment selection under asymmetric regret, follows the same transitions and is left to the record repository\. A claim can be well supported and still fail to justify action if the intervention is infeasible, too risky, poorly aligned with patient goals, or insufficiently evidenced for the decision context\. Operational causal reasoning therefore requires more than a correct explanation\. It requires knowing which rung the system is on, what evidence is missing, when additional tests are justified, when treatment should be deferred, and how later outcomes should revise future reasoning\.

### D\.2Evaluating Practical Arguments: A Worked Policy Example

Consider the statement:*ads can be deceiving, so ads should not appear in children’s programs*\. This is not mainly a causal argument\. It is a practical\-policy argument with embedded empirical, causal, and normative subclaims\. A proper decomposition is:

> P1\.Some ads can deceive\.*empirical claim*P2\.Children are vulnerable to deception\.*empirical, developmental claim*P3\.Exposure to ads in children’s programs can harm children\.*causal, interventional claim*P4\.Removing ads from children’s programs would reduce that harm\.*causal, policy claim*P5\.Protecting children from deception outweighs the costs of removing ads\.*normative, value claim*C\.Therefore, ads should not appear in children’s programs\.*normative policy conclusion*

The primary scheme is the argument from negative consequences: if actionAApermits harmful consequenceHH, andHHshould be avoided, thenAAshould be restricted or replaced\. But practical reasoning is not validated by empirical evidence alone; it also requires a value premise\. That is Hume’s gap: facts do not by themselves yield*should*\. The original argument is therefore incomplete, not necessarily false, and the evaluation must say which\. Seven dimensions make the evaluation systematic\.

##### Claim\-type classification\.

The evaluator first identifies the conclusion type \(normative policy claim\), the embedded subclaims \(empirical, causal, value\-based\), and the primary scheme \(negative consequences, practical reasoning\)\. This prevents applying the wrong evaluator: a causal evaluator asks whetherXXcausesYY; a practical\-policy evaluator asks whether, given evidence, consequences, values, alternatives, and affected stakeholders, the action is justified\.

##### Premise support\.

Each premise gets its own evidence question: are ads sometimes deceptive \(P1\); are children less able than adults to recognize persuasive intent \(P2\); does exposure cause measurable harm \(P3\); would removal reduce the harm \(P4\); what costs and competing values would a ban impose \(P5\)? P2 is well supported: industry self\-regulation itself treats children as having limited knowledge, experience, sophistication, and maturity, noting that younger children may not understand persuasive intent or even that they are being advertised toChildren’s Advertising Review Unit \(CARU\) \([2021](https://arxiv.org/html/2607.12480#bib.bib134)\)\.

##### Warrant explicitness\.

The original sentence hides its warrant:*because children are vulnerable, society has a duty to protect them from deceptive persuasion\.*That is not a data claim; it is a normative bridge from evidence to policy, and an argument that hides it cannot be evaluated until it is surfaced\. Warrant explicitness is low in the original and high after repair, and the score records the difference\.

##### Quantifier and modality consistency\.

The premise is weak and existential, some ads*can*deceive; the conclusion is strong and universal, all ads banned from the context\. Conclusion strength exceeding premise strength is the practical\-argument form of overclaim\. The repaired argument needs stronger premises: children cannot reliably distinguish persuasive intent; child\-directed advertising predictably exploits this vulnerability; screening only deceptive ads is insufficient or infeasible; therefore broad restriction is justified\.

##### Embedded causal typing\.

Inside the practical argument sits an interventional claim, P4:d​o​\(remove ads\)→do\(\\text\{remove ads\}\)\\rightarrowlower deception and harm\. This subclaim is extracted and routed to the causal plug\-in of this paper: Pearl typing marks it interventional; SocraSynth generates the alternatives \(labeling, parental controls, advertising literacy, platform rules\); CRIT rejects the unsupported causal jump; EVINCE gathers exposure and harm evidence; ERM penalizes overconfident policy when the causal evidence is weak; RLER and Trivium track whether future policy arguments distinguish exposure, deception, harm, and remedy\.

##### Counter\-consideration recall\.

A practical argument is evaluated partly by what it considered and rejected\. Missing here: could ads be regulated rather than banned; could educational and public\-service advertising remain; would removing ads reduce access to free programming; would children migrate to less regulated platforms; would parental controls be more targeted? Pragma\-dialectics makes this a general standard: argument quality is judged not only by internal consistency but by whether the discussion reasonably addresses the points at issue, the implicit premises, and the scheme\-appropriate criticismsvan Eemeren and Grootendorst \([2004](https://arxiv.org/html/2607.12480#bib.bib127)\)\.

##### Value\-context fit: DIKE and ERIS\.

The value premise P5 is not resolvable by evidence, and its evaluation is reasoning with declared criteria\. DIKE asks which fairness, protection, harm\-prevention, rights, and regulatory norms apply: a declared, versioned criteria set, the normative analog of𝒞\\mathcal\{C\}\. ERIS asks whose context is missing: advertisers, parents, children of different ages, low\-income families relying on free content, educational broadcasters, different cultures and legal systems\. The argument is not fully evaluable without this context\. In one jurisdiction a full ban may be proportionate; in another, restricting manipulative formats, requiring disclosure, or banning categories may fit better\. An evaluation is complete only when it names the criteria set and version it applied; the governance of those criteria is developed in the volume’s later chapters\.

### D\.3From Holistic Scoring to Typed Argument Evaluation

The evaluation practice this implies is a typed extension of CRIT\. The original CRIT question, does this argument support the claim, remains necessary, but it is now refined by claim type, scheme, context, and value frame\. The upgraded question is:

> Given the claim type, argument scheme, context, evidence standard, and value frame, does this argument provide sufficient support for this conclusion?

Table[14](https://arxiv.org/html/2607.12480#A4.T14)gives the scoring dimensions that operationalize the upgraded question\. Each is checkable separately, which is what makes the evaluation stronger than a single holistic score\.

Table 14:Typed argument evaluation: the upgraded CRIT scoring dimensions\. Each dimension is separately checkable; the final status extends the paper’s three verdicts with the graded outcomes practical arguments require\.Applied to the running example, the judgment as stated is: type, practical\-policy argument with embedded empirical and causal subclaims; scheme, argument from negative consequences; quality, weak to moderate; reason, the conclusion may be defensible but the argument is underdeveloped, with a hidden normative warrant, a quantifier jump, a missing child\-vulnerability premise, missing intervention evidence, and missing counter\-considerations\. The repaired argument reads:

> Because young children have limited ability to recognize persuasive intent, advertising embedded in children’s programs can exploit a vulnerable audience\. If such exposure predictably increases deception, unhealthy persuasion, or parent–child pressure, and if less restrictive controls are insufficient, then restricting or banning ads in children’s programs is justified to protect children from unfair persuasion\.

The post\-repair judgment: much stronger, because claim type, empirical support, causal mechanism, value warrant, and policy scope are now explicit; still open, the age threshold, the definition of an ad, the evidence of harm, the effectiveness of a ban, exceptions for educational and public\-service content, and migration to other platforms\. The distance between the two judgments is the charity delta, and it is itself a quality signal about the source of the original argument\.

The TRACE record of Section[3](https://arxiv.org/html/2607.12480#S3)generalizes accordingly, with four added fields: the scheme, the scheme’s critical questions with their answers, the warrants, and the embedded subclaims with their routing\. The paper’s summary statement for this section is one sentence:

> CRIT does not measure whether a sentence sounds reasonable; it types the claim, identifies the argument scheme, reconstructs hidden warrants, evaluates embedded subclaims under the right reasoning discipline, and scores whether the conclusion is justified in context\.

## Appendix ETRACE\-Bench: Benchmark and Ablation Protocol

The operational metric definitions and their label sources:

Table 15:Operational definitions for the writer and record\-quality dimensions\. Gated metrics participate in the hard gates of the final status; the rest are optimized on the residual and reported in the vector\. Provenance \(SourceTraceCompleteness\) is objective and separated from credibility \(SourceReliabilityScore\), which is partly subjective\.An implementation\-level methodology needs an implementation\-level benchmark\. TRACE\-Bench is the benchmark protocol implied by the pipeline: it does not ask only whether the final answer is correct\. It asks whether the evaluator detected the argument, reconstructed its roles, typed the claim, routed the subclaims to the right standards, asked the right critical questions, invoked the right plug\-ins, surfaced the right defeaters, and assigned the right verdict and repair\. The benchmark is designed to evaluate both the full TRACE implementation and each stage\-level contract separately\.

Each TRACE\-Bench item is a short passage, usually one to five sentences, paired with a context and optional evidence ledger\. Items include non\-arguments, single\-family arguments, and mixed arguments\. The gold record contains: the reasoning segment; claim and subclaims; premises, warrants, backing, qualifier, rebuttal, counter\-considerations, undercutters, and burden fields; claim type; reasoning family; argument scheme; applicable critical questions; family\-specific checks; evidence status; unresolved defeaters; final status; repair; and score vector\. The same item can therefore evaluate both local operators and end\-to\-end judgment\.

Table 16:TRACE\-Bench evaluates the whole typed\-reasoning methodology and its operators\. Unlike answer\-only benchmarks, it scores the intermediate record: detection, anatomy, routing, scheme questions, family tests, defeaters, verdict, repair, and cost\.The benchmark should be deliberately mixed\. Deductive items test validity and soundness; inductive items test sample and reference\-class errors; abductive items test alternative\-explanation recall; causal items test association, intervention, counterfactuals, confounding, identification, and bounds; practical\-policy items test means–end fit, side effects, alternatives, and value warrants; legal or deontic items test rule fit, authority, precedent, burden, and institutional context\. Non\-argument passages are necessary because a detector that treats every fluent statement as an argument will look strong on downstream tasks while failing at intake\.

A minimal useful TRACE\-Bench can be built in three tiers\. Tier 1 uses synthetic and templated examples whose gold decomposition is controlled\. Tier 2 adapts existing annotated argument\-mining and causal\-reasoning datasets into the TraceRecord schema\. Tier 3 uses expert\-written mixed arguments in medicine, policy, law, science, and everyday explanation, because the central challenge is mixed reasoning: a single paragraph may contain empirical, causal, practical, and value claims at once\. Annotation should be redundant: at least two annotators fill the record independently, disagreements are adjudicated, and the adjudicated record becomes the gold ledger\.

The ablation suite tests whether the architecture matters\. The full system is TRACE\. Baselines include direct answer, chain\-of\-thought only, self\-consistency, LLM\-as\-judge holistic scoring, single\-pass record emission in which one structured call writes the entire TraceRecord, debate without typed routing, CRIT without family plug\-ins, and family plug\-ins without Walton critical questions\. Two baselines carry special weight\. Single\-pass record emission is the null hypothesis for the staged writer: it conforms to the schema, so if it matches TRACE on record quality and error attribution at comparable cost, the decomposition is unjustified \(Section[7](https://arxiv.org/html/2607.12480#S7)\)\. Debate without typed routing, scored holistically, is the volume’s own earlier evaluation mechanism, so this row is the volume auditing its predecessor with its own methods\. The important ablations remove one control surface at a time\.

Table 17:Ablation plan for TRACE\-Bench\. Each ablation removes one design commitment of the paper and predicts the failure mode it should expose\. A useful benchmark must show not only that TRACE improves final verdicts, but also which stage is responsible for the improvement\.The primary end\-to\-end measures are*WrongAcceptRate*and*DeferQuality*\. WrongAcceptRate is the fraction of unsupported, overclaiming, or formally invalid arguments that receiveAcceptorQualify\. DeferQuality asks whether a defer verdict names the missing evidence or assumption that would actually tighten, flip, or justify the claim when supplied\. These two measures capture the practical value of the pipeline: it should not merely sound critical; it should prevent wrong acceptance and make uncertainty actionable\.

A useful reporting scalar is record utility:

RecordUtility=\\displaystyle\\mathrm\{RecordUtility\}=\{\}α1​TypeAcc\+α2​SchemeFit\+α3​EvidenceSupport\+α4​AltRecall\\displaystyle\\alpha\_\{1\}\\mathrm\{TypeAcc\}\+\\alpha\_\{2\}\\mathrm\{SchemeFit\}\+\\alpha\_\{3\}\\mathrm\{EvidenceSupport\}\+\\alpha\_\{4\}\\mathrm\{AltRecall\}\(1\)\+α5​CalibrationGain\+α6​WarrantExplicitness\\displaystyle\+\\alpha\_\{5\}\\mathrm\{CalibrationGain\}\+\\alpha\_\{6\}\\mathrm\{WarrantExplicitness\}\+α7​ValueTransparency\+α8​RegretReduction\\displaystyle\+\\alpha\_\{7\}\\mathrm\{ValueTransparency\}\+\\alpha\_\{8\}\\mathrm\{RegretReduction\}−β1​UnsupportedRate−β2​OverclaimRate\\displaystyle\-\\beta\_\{1\}\\mathrm\{UnsupportedRate\}\-\\beta\_\{2\}\\mathrm\{OverclaimRate\}−β3​RungCollapseRate−β4​RepeatedErrorRate\\displaystyle\-\\beta\_\{3\}\\mathrm\{RungCollapseRate\}\-\\beta\_\{4\}\\mathrm\{RepeatedErrorRate\}−β5​CostNorm\.\\displaystyle\-\\beta\_\{5\}\\mathrm\{CostNorm\}\.
The weights should be task dependent\. In low\-risk explanatory tasks, evidence gain and alternative recall may dominate\. In high\-stakes action settings, overclaim reduction, regret reduction, and recurrence control should receive higher weight\.

The Cost term is not incidental\. Reasoning models routinely produce redundant and overlong traces, and surveys of efficient reasoning document substantial token expenditure for little accuracy gainChenet al\.\([2025b](https://arxiv.org/html/2607.12480#bib.bib152)\); Suiet al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib151)\)\. Cost\-aware quality metrics and the stopping rules of Section[5\.4](https://arxiv.org/html/2607.12480#S5.SS4)are the evaluator’s countermeasures to overthinking\.

##### Gate, then optimize\.

Equation \([1](https://arxiv.org/html/2607.12480#A5.E1)\) should be read as a reporting device, not a training target\. Optimized directly, a scalarized score invites Goodhart behavior: alternative recall can be inflated with junk alternatives to offset overclaim penalties, and cost terms can be gamed by early stopping\. The operational form is lexicographic: hard gates on the non\-negotiables \(rung discipline, the unsupported\-premise rate, and bound violations, per Proposition[1](https://arxiv.org/html/2607.12480#Thmproposition1)\), followed by scalar optimization on the residual\. As a reporting rule, the scalar is never reported without the vector and the list of failed gates: the scalar is a dashboard summary, not the scientific result\. The weightsαi,βj\\alpha\_\{i\},\\beta\_\{j\}are governance decisions: versioned, audited, and owned by the governance layer \(DIKE and ERIS\) of the companion volume\. All component terms should be normalized to\[0,1\]\[0,1\]before aggregation\.

## Appendix FConvergence Validity: Metrics, Imported Propositions, and an Open Problem

The six validity metrics below make convergence validity measurable; the companion volume computes them\. The propositions that follow are the formal spine: imported classical results stated as theorems, their transfer to multi\-LLM systems stated as design principles, and the open problem the experiments are designed to inform\.

### F\.1The Validity Metrics

Let an episode consist of turnst=1,…,Tt=1,\\dots,T; letbtb\_\{t\}denote the consensus\-relevant belief state after turntt\(for a binary claim, a probability\); letEt⊆ℰE\_\{t\}\\subseteq\\mathcal\{E\}be the evidence units newly admitted to the ledger at turnttafter CRIT audit at thresholdθ\\thetaChang \([2023a](https://arxiv.org/html/2607.12480#bib.bib76)\); and letVVbe the final verdict\.

1. M1Procedural invariance \(convergence robustness\)\.Re\-run the episode under a perturbation setΠ\\Pithat ought not to matter: role reassignment, speaking order, persona relabeling, sampling seed\. DefineCR=1−Prπ∈Π⁡\[Vπ≠V\]\\mathrm\{CR\}=1\-\\Pr\_\{\\pi\\in\\Pi\}\[\\,V^\{\\pi\}\\neq V\\,\]\. A consensus that flips with seating order is rhetorical, not epistemic\.
2. M2Warranted convergence ratio\.Decompose belief movement by turn and call a movement*warranted*when it co\-occurs with newly admitted evidence: WCR=∑t\|bt−bt−1\|​1​\[Et≠∅\]∑t\|bt−bt−1\|\.\\mathrm\{WCR\}\\;=\\;\\frac\{\\sum\_\{t\}\\,\\lvert b\_\{t\}\-b\_\{t\-1\}\\rvert\\;\\mathbf\{1\}\[\\,E\_\{t\}\\neq\\varnothing\\,\]\}\{\\sum\_\{t\}\\,\\lvert b\_\{t\}\-b\_\{t\-1\}\\rvert\}\.The complement1−WCR1\-\\mathrm\{WCR\}is a per\-episode*sycophancy index*: concession without evidence, the failure mode documented empirically in multi\-agent debateWynnet al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib72)\); Yaoet al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib73)\)\.
3. M3Evidence\-counterfactual sensitivity\.Ablate ledger items and re\-adjudicate: the verdict should move when decisive evidence is removed and stay fixed when rhetoric\-only turns are removed or paraphrased\. This is the debate analogue of chain\-of\-thought faithfulness tests by corruption and truncationLanhamet al\.\([2023](https://arxiv.org/html/2607.12480#bib.bib42)\)\. A verdict insensitive to its own evidence is fluent consensus, now detectable\.
4. M4Dual\-phase divergence signature\.LetDtD\_\{t\}be a cross\-agent divergence \(Jensen–Shannon or Wasserstein\) over predictive distributions, as tracked by EVINCEChang \([2024](https://arxiv.org/html/2607.12480#bib.bib78)\)\. A healthy episode shows an*exploration*phase in whichDtD\_\{t\}rises or holds while the hypothesis set grows, then a*contraction*phase in which the fall ofDtD\_\{t\}correlates with evidence arrivalsEtE\_\{t\}\. Monotone contraction from the first turn with a static hypothesis set is premature collapse: degeneration of thoughtLianget al\.\([2024](https://arxiv.org/html/2607.12480#bib.bib66)\)as a measurable waveform rather than an anecdote\.
5. M5Effective panel size\.With mean pairwise error correlationρ¯\\bar\{\\rho\}amongnnagents, estimated on a calibration set of labeled claims, defineneff=n/\(1\+\(n−1\)​ρ¯\)n\_\{\\mathrm\{eff\}\}=n/\\bigl\(1\+\(n\-1\)\\bar\{\\rho\}\\bigr\)\. Jury\-style aggregation benefits requireneff≫1n\_\{\\mathrm\{eff\}\}\\gg 1Ladha \([1992](https://arxiv.org/html/2607.12480#bib.bib84)\); same\-model panels haveρ¯→1\\bar\{\\rho\}\\to 1and henceneff→1n\_\{\\mathrm\{eff\}\}\\to 1, a one\-line account of why homogeneous debate underperforms and heterogeneous panels are the remedyZhanget al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib70)\)\. Reportneffn\_\{\\mathrm\{eff\}\}, notnn\.
6. M6External\-judge gap\.Give the ledger, and only the ledger, to an adjudicator that did not participate: a different\-family model, a formal checker, or CRIT run cold\. Measure the gap between its verdict and the participants’ consensus\. A real convergence survives being re\-derived from its own evidence by a stranger\.

The empirical record of reasoning\-as\-debate methods, ordered by the four debate\-ness conditions of the main text:

Table 18:The debate\-ness profile D1–D4 orders reasoning\-as\-debate methods and explains the mixed empirical record: exchange helps where D2–D4 are enforced and disappoints where they are simulated\.
### F\.2Imported Propositions and the Open Problem

###### Proposition 3\(Aumann agreement\)\.

If agents share a common prior and truthfully exchange posteriors that become common knowledge, they cannot agree to disagree, and iterated communication converges in finitely many rounds regardless of the truth of the claimAumann \([1976](https://arxiv.org/html/2607.12480#bib.bib81)\); Geanakoplos and Polemarchakis \([1982](https://arxiv.org/html/2607.12480#bib.bib82)\)\.

##### Design principle: consensus is weak evidence under correlated generators\.

The transfer of Proposition[3](https://arxiv.org/html/2607.12480#Thmproposition3)to multi\-LLM debate is a modeling assumption, not a theorem\. Agents instantiated from a common base model are strongly correlated belief generators that only approximate the common\-prior setting\. Under that assumption, the likelihood ratio of the event “the debate converged” with respect to the truth of the claim should be treated as close to uninformative unless the convergence is accompanied by ledger movement, external adjudication, and procedural invariance: consensus alone is not evidence\. Evidential weight resides in the ledger and the belief trajectory, not in the agreement\.

###### Proposition 4\(Diversity, not head\-count\)\.

Under the standard equal\-variance, equal\-correlation approximation, majority aggregation overnnagents with mean pairwise error correlationρ¯\\bar\{\\rho\}behaves as an independent panel of sizeneff=n/\(1\+\(n−1\)​ρ¯\)n\_\{\\mathrm\{eff\}\}=n/\(1\+\(n\-1\)\\bar\{\\rho\}\), and jury\-theorem guarantees weaken accordinglyLadha \([1992](https://arxiv.org/html/2607.12480#bib.bib84)\)\. The reduction to a single mean correlation is a design approximation, not an exact law of arbitrary error dependence, but the direction of the effect is general\. Corollary: a homogeneous multi\-agent debate approximates a single agent, and the marginal value of an added agent is the marginal decorrelation it contributes\.

###### Proposition 5\(Verification asymmetry\)\.

In the prover–verifier formalization, debate with optimal play permits a polynomial\-time judge to decide any problem in PSPACEIrvinget al\.\([2018](https://arxiv.org/html/2607.12480#bib.bib56)\); doubly\-efficient debate makes the honest strategy itself efficientBrown\-Cohenet al\.\([2024](https://arxiv.org/html/2607.12480#bib.bib58)\), and prover–estimator debate addresses obfuscated argumentsBrown\-Cohenet al\.\([2025](https://arxiv.org/html/2607.12480#bib.bib59)\)\. Debate is therefore not a brainstorming heuristic but a verification protocol: it extends the reach of a bounded judge beyond anything solitary deliberation at the judge’s capacity can attain\. These guarantees apply to formal debate protocols with optimal play and specified judge models, not to ordinary multi\-agent LLM discussion\.

###### Open Problem 1\(Convergence to the ideal aggregator under engineered diversity\)\.

Characterize the conditions on role conditioning \(contentiousness schedules, persona assignment, corpus separation\) under which the consensus of a conditioned multi\-LLM debate converges to the ideal aggregatorb∗​\(ℰ\)b^\{\*\}\(\\mathcal\{E\}\); that is, when does*simulated*prior diversity purchase the aggregation benefits of*real*prior diversity? A stylized sufficient condition would require the conditioned agents’ likelihood contributions to be conditionally independent given the claim; whether and when role conditioning achieves this is an empirical question that the validity metrics of Section[F\.1](https://arxiv.org/html/2607.12480#A6.SS1)are designed to test\.

## Appendix GUsage Scenario: Mnemosyne Memory Admission

Mnemosyne enforces the TRACE discipline at the memory boundary: nothing enters durable memory without a TRACE record carrying a verdict\. The gate is one function,admit\(record\), mapping the record’sfinal\_statusandfailed\_gatesto one of four outcomes: commit, commit with qualifier, quarantine, or reject\. Quarantined records persist with their verdicts, retrievable but marked, so a later consumer can distinguish an unproven claim from an absent one\.

The music\-lessons record of the worked trace \(Section[5\.2](https://arxiv.org/html/2607.12480#S5.SS2)\) walks the gate as follows\. SubclaimS1S\_\{1\}, the observed association, arrives with statusAcceptunder an inductive qualifier and commits with that qualifier attached\. SubclaimS2S\_\{2\}, the causal claim, arrives asDeferwithmissingnaming randomized or quasi\-experimental assignment; it is quarantined, and the named missing evidence is registered as a trigger\. SubclaimS3S\_\{3\}, the policy conclusion, arrives asRejectwith its repair proposal; the rejection and the repair are logged, not discarded, because Trivium’s regret accounting consumes them\.

Two behaviors then depend on fields no scalar carries\. Defer is*prospective memory*: when evidence matching the trigger arrives, a lottery\-based school assignment study, say, the quarantined record is re\-adjudicated; if the verdict flips toAccept, the record commits and the trigger retires, and if it hardens toReject, the quarantine converts, in either case exercising the actionable sense of the explanation claim \(Section[7](https://arxiv.org/html/2607.12480#S7)\)\. Reconsolidation is*re\-audit*: if a committed record’s premise is later defeated, belief revision propagates throughrevision\_historyrather than silently overwriting, and every downstream record whoseprovenancecites the defeated premise is queued for re\-audit\. The measured quantities are admission precision and recall against expert judgment, stale\-memory correction rate, and defer quality, the fraction of defers whose named evidence, when supplied, in fact changes the verdict\.

## Appendix HUsage Scenario: MACI Plan Gating

MACI enforces the TRACE discipline at the commitment boundary of a multi\-agent workflow, under the division of labor stated in Section[7](https://arxiv.org/html/2607.12480#S7): structural admission, whether the workflow is well formed, its dependencies acyclic, and its resource and temporal constraints satisfiable, belongs to the plan validators\. TRACE gates the justificatory residue, the claims a plan rests on that constraint checking cannot reach\.

The flood\-rescue vignette of Section[7\.1](https://arxiv.org/html/2607.12480#S7.SS1)makes the division concrete\. The structural validator admits all three candidates: a northern boat dispatch, a bounded drone verification, and a southern boat dispatch\. The boat exists, its capacity is sufficient, the destination and deadline are declared, and each route is syntactically usable by the planner\. The validator therefore does not reject the northern branch\. TRACE reads the record licensing that branch and finds a different defect: the reported success value is high, but model support is below policy and the OOD score is above policy\. MACI writesHoldtoconsumer\_actions, citesRN\(1\)R\_\{N\}^\{\(1\)\}, and leaves the alternative branches available\.

Therepairfield ofRN\(1\)R\_\{N\}^\{\(1\)\}names a current route observation\. The planner proposes the survey\-drone action, and a separate practical recordRV\(1\)R\_\{V\}^\{\(1\)\}clears or qualifies that bounded information\-gathering step\. When the drone reports the debris obstruction, the repository appendsRN\(2\)R\_\{N\}^\{\(2\)\}rather than editing the earlier forecast\. Therevision\_historylink identifies which commitments depended on the failed route premise, so MACI invalidates the northern branch only\. The South Detour remains structurally valid; after its own predictive and practical record passes policy and the Incident Commander supplies external approval, MACI writesClearfor the southern dispatch with the exact authorizing record version\.

The measured quantities are the prevented\-error rate on plans that pass structural validation yet fail justification, false\-hold rate, added latency, repair success, and branch\-localization accuracy\. The claim defended is still the explanation and control\-interface claim, not a rescue\-performance claim: every held, qualified, or cleared commitment arrives with why, on what evidence, under which policy, and what later observation changed it\. A controlled gated\-versus\-ungated implementation is future work\.

## Acknowledgments

The photorealistic flood search\-and\-rescue scene \(Figure[3](https://arxiv.org/html/2607.12480#S7.F3)\) was generated with OpenAI’s GPT\-4o image generation\. All schematic and lifecycle diagrams are the authors’ own vector figures\. The generated image is an illustration only; it depicts no real person, place, or deployment\.

## References

- C\. E\. Alchourrón, P\. Gärdenfors, and D\. Makinson \(1985\)On the logic of theory change: partial meet contraction and revision functions\.The Journal of Symbolic Logic50\(2\),pp\. 510–530\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- K\. D\. Ashley \(1990\)Modeling legal argument: reasoning with cases and hypotheticals\.MIT Press,Cambridge, MA\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- K\. Atkinson and T\. J\. M\. Bench\-Capon \(2007\)Practical reasoning as presumptive argumentation using action based alternating transition systems\.Artificial Intelligence171\(10–15\),pp\. 855–874\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- R\. J\. Aumann \(1976\)Agreeing to disagree\.The Annals of Statistics4\(6\),pp\. 1236–1239\.Cited by:[§5\.3](https://arxiv.org/html/2607.12480#S5.SS3.p4.1),[Proposition 3](https://arxiv.org/html/2607.12480#Thmproposition3.p1.1.1)\.
- A\. Balke and J\. Pearl \(1997\)Bounds on treatment effects from studies with imperfect compliance\.Journal of the American Statistical Association92\(439\),pp\. 1171–1176\.Cited by:[§4\.4](https://arxiv.org/html/2607.12480#S4.SS4.SSS0.Px3.p2.1),[§5\.2](https://arxiv.org/html/2607.12480#S5.SS2.SSS0.Px5.p1.3),[Proposition 1](https://arxiv.org/html/2607.12480#Thmproposition1.p1.1.1)\.
- P\. Baroni, M\. Caminada, and M\. Giacomin \(2011\)An introduction to argumentation semantics\.The Knowledge Engineering Review26\(4\),pp\. 365–410\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px5.p1.1)\.
- T\. J\. M\. Bench\-Capon and P\. E\. Dunne \(2007\)Argumentation in artificial intelligence\.Artificial Intelligence171\(10–15\),pp\. 619–641\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px5.p1.1)\.
- T\. J\. M\. Bench\-Capon and G\. Sartor \(2003\)A model of legal reasoning with cases incorporating theories and values\.Artificial Intelligence150\(1–2\),pp\. 97–143\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- T\. J\. M\. Bench\-Capon \(2003\)Persuasion in practical argument using value\-based argumentation frameworks\.Journal of Logic and Computation13\(3\),pp\. 429–448\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- J\. Brown\-Cohen, G\. Irving, and G\. Piliouras \(2024\)Scalable AI safety via doubly\-efficient debate\.InICML,Note:arXiv:2311\.14125Cited by:[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.7.4.6.1.1),[Proposition 5](https://arxiv.org/html/2607.12480#Thmproposition5.p1.1.1)\.
- J\. Brown\-Cohen, G\. Irving, and G\. Piliouras \(2025\)Avoiding obfuscation with prover\-estimator debate\.Note:arXiv:2506\.13609Cited by:[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.7.4.6.1.1),[Proposition 5](https://arxiv.org/html/2607.12480#Thmproposition5.p1.1.1)\.
- E\. Y\. Chang, L\. Geng, and E\. J\. Chang \(2027\)Mnemosyne: agentic transaction processing for validating and repairing AI\-generated workflows\.Note:Under submission, Proceedings of the VLDB EndowmentCited by:[item 3](https://arxiv.org/html/2607.12480#S1.I1.i3.p1.1),[§7](https://arxiv.org/html/2607.12480#S7.p2.1)\.
- E\. Y\. Chang and L\. Geng \(2025\)SagaLLM: context management, validation, and transaction guarantees for multi\-agent LLM planning\.Proceedings of the VLDB Endowment18\(12\),pp\. 4874–4886\.External Links:[Document](https://dx.doi.org/10.14778/3750601.3750611)Cited by:[§7](https://arxiv.org/html/2607.12480#S7.p2.1)\.
- E\. Y\. Chang and L\. Geng \(2026\)Epistemic regret minimization: label\-free causal critique beyond outcome reward\.External Links:2602\.11675,[Link](https://arxiv.org/abs/2602.11675)Cited by:[item 3](https://arxiv.org/html/2607.12480#S1.I1.i3.p1.1),[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px2.p1.1)\.
- E\. Y\. Chang \(2023a\)Prompting large language models with the socratic method\.InIEEE 13th Annual Computing and Communication Workshop and Conference \(CCWC\),Note:arXiv:2303\.08769Cited by:[§F\.1](https://arxiv.org/html/2607.12480#A6.SS1.p1.7),[§2\.1](https://arxiv.org/html/2607.12480#S2.SS1.SSS0.Px1.p2.1),[§2](https://arxiv.org/html/2607.12480#S2.p1.1),[§4\.1](https://arxiv.org/html/2607.12480#S4.SS1.p3.5),[§5\.3](https://arxiv.org/html/2607.12480#S5.SS3.p3.1)\.
- E\. Y\. Chang \(2023b\)SocraSynth: multi\-LLM reasoning with conditional statistics\.InProceedings of the 10th International Conference on Computational Science and Computational Intelligence \(CSCI\),Las Vegas, NV\.Note:Extended version: arXiv:2402\.06634Cited by:[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.6.3.6.1.1),[item 3](https://arxiv.org/html/2607.12480#S1.I1.i3.p1.1),[§2\.1](https://arxiv.org/html/2607.12480#S2.SS1.SSS0.Px1.p2.1),[§5\.3](https://arxiv.org/html/2607.12480#S5.SS3.p3.1)\.
- E\. Y\. Chang \(2024\)EVINCE: optimizing multi\-LLM dialogues using conditional statistics and information theory\.Note:arXiv:2408\.14575Cited by:[itemM4](https://arxiv.org/html/2607.12480#A6.I1.ix4.p1.4),[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.6.3.6.1.1),[item 3](https://arxiv.org/html/2607.12480#S1.I1.i3.p1.1),[§2\.1](https://arxiv.org/html/2607.12480#S2.SS1.SSS0.Px1.p2.1),[§5\.3](https://arxiv.org/html/2607.12480#S5.SS3.p3.1)\.
- E\. Y\. Chang \(2025\)Multi\-LLM agent collaborative intelligence; the path to artificial general intelligence, volume i\.ACM Books\.Note:\(First Amazon edition published in March 2024\)External Links:[Document](https://dx.doi.org/10.1145/3749421),ISBN 979\-8\-4007\-3197\-6Cited by:[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.6.3.6.1.1),[item 3](https://arxiv.org/html/2607.12480#S1.I1.i3.p1.1)\.
- E\. Y\. Chang \(2026a\)Diagnosing and mitigating sycophancy and skepticism in LLM causal judgment\.InProceedings of the 64th Annual Meeting of the Association for Computational Linguistics \(ACL\),Cited by:[item 3](https://arxiv.org/html/2607.12480#S1.I1.i3.p1.1),[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px2.p1.1)\.
- E\. Y\. Chang \(2026b\)System\-2 reasoning: from semantic anchoring to causal intelligence: the path to artificial general intelligence, volume ii\.ACM Books\.External Links:[Document](https://dx.doi.org/10.1145/3822380),ISBN 979\-8\-4007\-2804\-4Cited by:[item 3](https://arxiv.org/html/2607.12480#S1.I1.i3.p1.1)\.
- E\. Y\. Chang \(2026c\)Trivium: temporal regret as a first\-class objective for causal\-memory controllers\.External Links:2606\.04421,[Link](https://arxiv.org/abs/2606.04421)Cited by:[item 3](https://arxiv.org/html/2607.12480#S1.I1.i3.p1.1),[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px2.p1.1)\.
- Q\. Chen, L\. Qin, J\. Liu,et al\.\(2025a\)Towards reasoning era: a survey of long chain\-of\-thought for reasoning large language models\.Note:arXiv:2503\.09567Cited by:[§2](https://arxiv.org/html/2607.12480#S2.p1.1)\.
- X\. Chen, J\. Xu, T\. Liang,et al\.\(2025b\)Do NOT think that much for 2\+3=? On the overthinking of long reasoning models\.InICML,Note:arXiv:2412\.21187Cited by:[Appendix E](https://arxiv.org/html/2607.12480#A5.p11.1)\.
- Y\. Chen, J\. Benton, A\. Radhakrishnan,et al\.\(2025c\)Reasoning models don’t always say what they think\.Note:arXiv:2505\.05410Cited by:[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px1.p1.1)\.
- C\. Chesñevar, J\. McGinnis, S\. Modgil, I\. Rahwan, C\. Reed, G\. Simari, M\. South, G\. Vreeswijk, and S\. Willmott \(2006\)Towards an argument interchange format\.The Knowledge Engineering Review21\(4\),pp\. 293–316\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p7.1)\.
- Children’s Advertising Review Unit \(CARU\) \(2021\)Self\-regulatory guidelines for children’s advertising\.Technical reportBBB National Programs\.Note:Revised editionCited by:[§D\.2](https://arxiv.org/html/2607.12480#A4.SS2.SSS0.Px2.p1.1)\.
- H\. K\. Choi, X\. Zhu, and Y\. Li \(2025\)Debate or vote: which yields better decisions in multi\-agent large language models?\.InNeurIPS,Note:arXiv:2508\.17536Cited by:[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.2.6.1.1)\.
- K\. Cobbe, V\. Kosaraju, M\. Bavarian,et al\.\(2021\)Training verifiers to solve math word problems\.Note:arXiv:2110\.14168Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- J\. de Kleer \(1986\)An assumption\-based TMS\.Artificial Intelligence\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- L\. de Moura and S\. Ullrich \(2021\)The Lean 4 theorem prover and programming language\.InCADE\-28,LNCS, Vol\.12699,pp\. 625–635\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- M\. H\. DeGroot \(1974\)Reaching a consensus\.Journal of the American Statistical Association69\(345\),pp\. 118–121\.Cited by:[§5\.3](https://arxiv.org/html/2607.12480#S5.SS3.p4.1)\.
- A\. K\. Dey \(2001\)Understanding and using context\.Personal and Ubiquitous Computing5\(1\),pp\. 4–7\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- J\. Doyle \(1979\)A truth maintenance system\.Artificial Intelligence\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- H\. Du, S\. Thudumu, R\. Vasa, and K\. Mouzakis \(2024a\)A survey on context\-aware multi\-agent systems: techniques, challenges and future directions\.Note:arXiv:2402\.01968Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- Y\. Du, S\. Li, A\. Torralba, J\. B\. Tenenbaum, and I\. Mordatch \(2024b\)Improving factuality and reasoning in language models through multiagent debate\.InICML,Note:arXiv:2305\.14325Cited by:[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.2.6.1.1)\.
- P\. M\. Dung \(1995\)On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n\-person games\.Artificial Intelligence77\(2\),pp\. 321–357\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px5.p1.1),[Table 3](https://arxiv.org/html/2607.12480#S4.T3.1.5.4.1.1.1)\.
- A\. Feder, K\. A\. Keith, E\. Manzoor,et al\.\(2022\)Causal inference in natural language processing: estimation, prediction, interpretation and beyond\.Transactions of the Association for Computational Linguistics10,pp\. 1138–1158\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- J\. B\. Freeman \(1991\)Dialectics and the macrostructure of arguments: a theory of argument structure\.Foris Publications,Berlin\.Cited by:[Appendix C](https://arxiv.org/html/2607.12480#A3.SS0.SSS0.Px1.p1.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px2.p1.1)\.
- J\. D\. Geanakoplos and H\. M\. Polemarchakis \(1982\)We can’t disagree forever\.Journal of Economic Theory28\(1\),pp\. 192–200\.Cited by:[§5\.3](https://arxiv.org/html/2607.12480#S5.SS3.p4.1),[Proposition 3](https://arxiv.org/html/2607.12480#Thmproposition3.p1.1.1)\.
- L\. Geng, A\. Ouyang, T\. Wu, D\. Barretto, M\. J\. Hayes, R\. Cooper, Y\. Zeng, S\. Vijay, G\. Ancone, A\. Rai, M\. Wolfman, P\. Flanagan, and E\. Y\. Chang \(2026\)CausalT5k: diagnosing refusal and failure modes in trustworthy causal reasoning across causal rungs\.Note:arXiv:2602\.08939; verify KDD 2026 acceptance before citing as KDDCited by:[§6](https://arxiv.org/html/2607.12480#S6.p4.1)\.
- T\. F\. Gordon, H\. Prakken, and D\. Walton \(2007\)The Carneades model of argument and burden of proof\.Artificial Intelligence171\(10–15\),pp\. 875–896\.Cited by:[Appendix C](https://arxiv.org/html/2607.12480#A3.SS0.SSS0.Px1.p1.1),[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px2.p1.1)\.
- T\. Govier \(1987\)Problems in argument analysis and evaluation\.Foris Publications,Dordrecht\.Cited by:[Appendix C](https://arxiv.org/html/2607.12480#A3.SS0.SSS0.Px1.p1.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px2.p1.1)\.
- C\. Guo, G\. Pleiss, Y\. Sun, and K\. Q\. Weinberger \(2017\)On calibration of modern neural networks\.InICML,Cited by:[§6](https://arxiv.org/html/2607.12480#S6.p4.1)\.
- D\. Guo, D\. Yang, H\. Zhang,et al\.\(2025\)DeepSeek\-R1 incentivizes reasoning in LLMs through reinforcement learning\.Nature645,pp\. 633–638\.Note:arXiv:2501\.12948Cited by:[§2](https://arxiv.org/html/2607.12480#S2.p1.1)\.
- R\. Guo, L\. Cheng, J\. Li, P\. R\. Hahn, and H\. Liu \(2020\)A survey of learning causality with data: problems and methods\.ACM Computing Surveys53\(4\),pp\. 75:1–75:37\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- D\. Hitchcock and B\. Verheij \(Eds\.\) \(2006\)Arguing on the toulmin model: new essays in argument analysis and evaluation\.Springer,Dordrecht\.Cited by:[Appendix C](https://arxiv.org/html/2607.12480#A3.SS0.SSS0.Px1.p1.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px2.p1.1)\.
- R\. A\. Howard \(1966\)Information value theory\.IEEE Transactions on Systems Science and Cybernetics2\(1\),pp\. 22–26\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- J\. Huang and K\. C\. Chang \(2023\)Towards reasoning in large language models: a survey\.InFindings of ACL,Note:arXiv:2212\.10403Cited by:[§2](https://arxiv.org/html/2607.12480#S2.p1.1),[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- J\. Huang, X\. Chen, S\. Mishra,et al\.\(2024\)Large language models cannot self\-correct reasoning yet\.InICLR,Note:arXiv:2310\.01798Cited by:[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.5.2.6.1.1)\.
- G\. W\. Imbens and D\. B\. Rubin \(2015\)Causal inference for statistics, social, and biomedical sciences: an introduction\.Cambridge University Press\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- G\. Irving, P\. Christiano, and D\. Amodei \(2018\)AI safety via debate\.Note:arXiv:1805\.00899Cited by:[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.7.4.6.1.1),[Proposition 5](https://arxiv.org/html/2607.12480#Thmproposition5.p1.1.1)\.
- A\. Q\. Jiang, S\. Welleck, J\. P\. Zhou,et al\.\(2023\)Draft, sketch, and prove: guiding formal theorem provers with informal proofs\.InICLR,Note:arXiv:2210\.12283Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- Z\. Jin, Y\. Chen, F\. Leeb,et al\.\(2023\)CLadder: assessing causal reasoning in language models\.InNeurIPS,Note:arXiv:2312\.04350Cited by:[§6](https://arxiv.org/html/2607.12480#S6.p4.1)\.
- Z\. Jin, J\. Liu, Z\. Lyu,et al\.\(2024\)Can large language models infer causation from correlation?\.InICLR,Note:arXiv:2306\.05836Cited by:[§6](https://arxiv.org/html/2607.12480#S6.p4.1)\.
- E\. Kıcıman, R\. Ness, A\. Sharma, and C\. Tan \(2024\)Causal reasoning and large language models: opening a new frontier for causality\.Transactions on Machine Learning Research\.Note:arXiv:2305\.00050Cited by:[§2\.1](https://arxiv.org/html/2607.12480#S2.SS1.p1.1)\.
- T\. Kojima, S\. S\. Gu, M\. Reid, Y\. Matsuo, and Y\. Iwasawa \(2022\)Large language models are zero\-shot reasoners\.InNeurIPS,Note:arXiv:2205\.11916Cited by:[§2](https://arxiv.org/html/2607.12480#S2.p1.1)\.
- K\. K\. Ladha \(1992\)The Condorcet jury theorem, free speech, and correlated votes\.American Journal of Political Science36\(3\),pp\. 617–634\.Cited by:[itemM5](https://arxiv.org/html/2607.12480#A6.I1.ix5.p1.8),[Proposition 4](https://arxiv.org/html/2607.12480#Thmproposition4.p1.3.3)\.
- T\. Lanham, A\. Chen, A\. Radhakrishnan,et al\.\(2023\)Measuring faithfulness in chain\-of\-thought reasoning\.Note:arXiv:2307\.13702Cited by:[itemM3](https://arxiv.org/html/2607.12480#A6.I1.ix3.p1.1),[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.4.1.6.1.1),[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px1.p1.1)\.
- J\. Lawrence, F\. Bex, C\. Reed, and M\. Snaith \(2012\)AIFdb: infrastructure for the argument web\.InComputational Models of Argument \(COMMA\),pp\. 515–516\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p7.1)\.
- J\. Lawrence and C\. Reed \(2019\)Argument mining: a survey\.Computational Linguistics45\(4\),pp\. 765–818\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§5\.1](https://arxiv.org/html/2607.12480#S5.SS1.p3.1)\.
- H\. Li, V\. Schlegel, Y\. Sun, R\. Batista\-Navarro, and G\. Nenadic \(2025\)Large language models in argument mining: a survey\.Note:arXiv:2506\.16383Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- K\. Li, A\. K\. Hopkins, D\. Bau, F\. Viégas, H\. Pfister, and M\. Wattenberg \(2023\)Emergent world representations: exploring a sequence model trained on a synthetic task\.InInternational Conference on Learning Representations \(ICLR\),Cited by:[§2\.1](https://arxiv.org/html/2607.12480#S2.SS1.p2.1)\.
- T\. Liang, Z\. He, W\. Jiao,et al\.\(2024\)Encouraging divergent thinking in large language models through multi\-agent debate\.InEMNLP,Note:arXiv:2305\.19118Cited by:[itemM4](https://arxiv.org/html/2607.12480#A6.I1.ix4.p1.4),[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.2.6.1.1)\.
- H\. Lightman, V\. Kosaraju, Y\. Burda,et al\.\(2024\)Let’s verify step by step\.InICLR,Note:arXiv:2305\.20050Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- J\. Lindsey, W\. Gurnee, E\. Ameisen, B\. Chen,et al\.\(2025\)On the biology of a large language model\.Note:Transformer Circuits Thread, AnthropicExternal Links:[Link](https://transformer-circuits.pub/2025/attribution-graphs/biology.html)Cited by:[§2\.1](https://arxiv.org/html/2607.12480#S2.SS1.p2.1)\.
- M\. Lippi and P\. Torroni \(2016\)Argumentation mining: state of the art and emerging trends\.ACM Transactions on Internet Technology16\(2\),pp\. 10:1–10:25\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§5\.1](https://arxiv.org/html/2607.12480#S5.SS1.p3.1)\.
- H\. Liu, Z\. Fu, M\. Ding,et al\.\(2025a\)Logical reasoning in large language models: a survey\.Note:arXiv:2502\.09100Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- M\. Liu, S\. Diao, X\. Lu,et al\.\(2025b\)ProRL: prolonged reinforcement learning expands reasoning boundaries in large language models\.InNeurIPS,Note:arXiv:2505\.24864Cited by:[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px2.p1.1)\.
- C\. F\. Manski \(1990\)Nonparametric bounds on treatment effects\.American Economic Review80\(2\),pp\. 319–323\.Cited by:[§4\.4](https://arxiv.org/html/2607.12480#S4.SS4.SSS0.Px3.p2.1),[§5\.2](https://arxiv.org/html/2607.12480#S5.SS2.SSS0.Px5.p1.3),[Proposition 1](https://arxiv.org/html/2607.12480#Thmproposition1.p1.1.1)\.
- J\. McCarthy \(1980\)Circumscription—a form of non\-monotonic reasoning\.Artificial Intelligence13\(1–2\),pp\. 27–39\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- H\. Mercier and D\. Sperber \(2011\)Why do humans reason? Arguments for an argumentative theory\.Behavioral and Brain Sciences34\(2\),pp\. 57–74\.Cited by:[§2\.3](https://arxiv.org/html/2607.12480#S2.SS3.p4.1)\.
- H\. Mercier and D\. Sperber \(2017\)The enigma of reason\.Harvard University Press\.Cited by:[§2\.3](https://arxiv.org/html/2607.12480#S2.SS3.p4.1)\.
- I\. Mirzadeh, K\. Alizadeh, H\. Shahrokhi,et al\.\(2025\)GSM\-Symbolic: understanding the limitations of mathematical reasoning in large language models\.InICLR,Note:arXiv:2410\.05229Cited by:[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px1.p1.1)\.
- S\. Modgil and H\. Prakken \(2014\)The ASPIC\+ framework for structured argumentation: a tutorial\.Argument & Computation5\(1\),pp\. 31–62\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px2.p1.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px5.p1.1)\.
- M\. Nye, A\. J\. Andreassen, G\. Gur\-Ari,et al\.\(2021\)Show your work: scratchpads for intermediate computation with language models\.Note:arXiv:2112\.00114Cited by:[§2](https://arxiv.org/html/2607.12480#S2.p1.1)\.
- OpenAI \(2024\)OpenAI o1 system card\.Technical reportNote:arXiv:2412\.16720Cited by:[§2](https://arxiv.org/html/2607.12480#S2.p1.1)\.
- C\. Opus and A\. Lawsen \(2025\)The illusion of the illusion of thinking: a comment on Shojaee et al\. \(2025\)\.Note:arXiv:2506\.09250Cited by:[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px1.p1.1)\.
- J\. Pearl and D\. Mackenzie \(2018\)The book of why: the new science of cause and effect\.Basic Books\.Cited by:[§4\.4](https://arxiv.org/html/2607.12480#S4.SS4.p2.1)\.
- J\. Pearl \(2009\)Causality: models, reasoning, and inference\.2nd edition,Cambridge University Press\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§4\.4](https://arxiv.org/html/2607.12480#S4.SS4.SSS0.Px2.p1.1),[§4\.4](https://arxiv.org/html/2607.12480#S4.SS4.p2.1),[Table 3](https://arxiv.org/html/2607.12480#S4.T3.1.6.5.1.1.1)\.
- C\. Perera, A\. Zaslavsky, P\. Christen, and D\. Georgakopoulos \(2014\)Context aware computing for the internet of things: a survey\.IEEE Communications Surveys & Tutorials16\(1\),pp\. 414–454\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- E\. Perez, S\. Ringer, K\. Lukošiūtė,et al\.\(2023\)Discovering language model behaviors with model\-written evaluations\.InFindings of ACL,Note:arXiv:2212\.09251Cited by:[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px2.p1.1)\.
- J\. Peters, D\. Janzing, and B\. Schölkopf \(2017\)Elements of causal inference: foundations and learning algorithms\.MIT Press\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- J\. L\. Pollock \(1987\)Defeasible reasoning\.Cognitive Science11\(4\),pp\. 481–518\.Cited by:[Appendix C](https://arxiv.org/html/2607.12480#A3.SS0.SSS0.Px1.p1.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px2.p1.1)\.
- H\. Prakken and G\. Sartor \(1996\)A dialectical model of assessing conflicting arguments in legal reasoning\.Artificial Intelligence and Law4\(3–4\),pp\. 331–368\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- H\. Prakken and G\. Vreeswijk \(2002\)Logics for defeasible argumentation\.InHandbook of Philosophical Logic,Vol\.4,pp\. 219–318\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px5.p1.1)\.
- H\. Prakken \(2010\)An abstract framework for argumentation with structured arguments\.Argument & Computation1\(2\),pp\. 93–124\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px2.p1.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px5.p1.1)\.
- S\. Qiao, Y\. Ou, N\. Zhang,et al\.\(2023\)Reasoning with language model prompting: a survey\.InACL,Note:arXiv:2212\.09597Cited by:[§2](https://arxiv.org/html/2607.12480#S2.p1.1),[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- H\. Raiffa and R\. Schlaifer \(1961\)Applied statistical decision theory\.Division of Research, Graduate School of Business Administration, Harvard University,Boston\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- R\. Reiter \(1980\)A logic for default reasoning\.Artificial Intelligence13\(1–2\),pp\. 81–132\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- G\. Sala and F\. Gobet \(2020\)Cognitive and academic benefits of music training with children: a multilevel meta\-analysis\.Memory & Cognition48,pp\. 1429–1441\.Cited by:[§5\.2](https://arxiv.org/html/2607.12480#S5.SS2.SSS0.Px5.p1.3)\.
- L\. J\. Savage \(1954\)The foundations of statistics\.John Wiley & Sons,New York\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- R\. Shao, S\. S\. Li, R\. Xin,et al\.\(2025\)Spurious rewards: rethinking training signals in RLVR\.Note:arXiv:2506\.10947Cited by:[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px2.p1.1)\.
- M\. Sharma, M\. Tong, T\. Korbak,et al\.\(2024\)Towards understanding sycophancy in language models\.InICLR,Note:arXiv:2310\.13548Cited by:[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px2.p1.1),[itemD2](https://arxiv.org/html/2607.12480#S5.I2.ix2.p1.1)\.
- P\. Shojaee, I\. Mirzadeh, K\. Alizadeh,et al\.\(2025\)The illusion of thinking: understanding the strengths and limitations of reasoning models via the lens of problem complexity\.Note:arXiv:2506\.06941Cited by:[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px1.p1.1)\.
- I\. Shpitser and J\. Pearl \(2006\)Identification of joint interventional distributions in recursive semi\-Markovian causal models\.InAAAI,Cited by:[§4\.4](https://arxiv.org/html/2607.12480#S4.SS4.SSS0.Px3.p2.1),[Proposition 1](https://arxiv.org/html/2607.12480#Thmproposition1.p1.1.1)\.
- C\. Snell, J\. Lee, K\. Xu, and A\. Kumar \(2025\)Scaling LLM test\-time compute optimally can be more effective than scaling parameters for reasoning\.InICLR,Note:arXiv:2408\.03314Cited by:[§2](https://arxiv.org/html/2607.12480#S2.p1.1)\.
- Y\. Sui, Y\. Chuang, G\. Wang,et al\.\(2025\)Stop overthinking: a survey on efficient reasoning for large language models\.Transactions on Machine Learning Research\.Note:arXiv:2503\.16419Cited by:[Appendix E](https://arxiv.org/html/2607.12480#A5.p11.1)\.
- J\. Sun, C\. Zheng, E\. Xie,et al\.\(2023\)A survey of reasoning with foundation models\.Note:arXiv:2312\.11562; expanded version in ACM Computing Surveys, 2025Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- F\. Toni \(2014\)A tutorial on assumption\-based argumentation\.Argument & Computation5\(1\),pp\. 89–117\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px5.p1.1)\.
- S\. E\. Toulmin \(1958\)The uses of argument\.Cambridge University Press\.Cited by:[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px1.p1.1),[Table 3](https://arxiv.org/html/2607.12480#S4.T3.1.3.2.1.1.1),[item 1](https://arxiv.org/html/2607.12480#S5.I1.i1.p1.1)\.
- M\. Turpin, J\. Michael, E\. Perez, and S\. R\. Bowman \(2023\)Language models don’t always say what they think: unfaithful explanations in chain\-of\-thought prompting\.InNeurIPS,Note:arXiv:2305\.04388Cited by:[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.4.1.6.1.1),[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px1.p1.1)\.
- F\. H\. van Eemeren and R\. Grootendorst \(2004\)A systematic theory of argumentation: the pragma\-dialectical approach\.Cambridge University Press\.Cited by:[§D\.2](https://arxiv.org/html/2607.12480#A4.SS2.SSS0.Px6.p1.1)\.
- H\. Wachsmuth, G\. Lapesa, E\. Cabrio, A\. Lauscher, J\. Park, E\. M\. Vecchi, S\. Villata, and T\. Ziegenbein \(2024\)Argument quality assessment in the age of instruction\-following large language models\.InLREC\-COLING,pp\. 1519–1538\.Note:arXiv:2403\.16084Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- H\. Wachsmuth, N\. Naderi, Y\. Hou,et al\.\(2017\)Computational argumentation quality assessment in natural language\.InEACL,Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1)\.
- A\. Wald and J\. Wolfowitz \(1948\)Optimum character of the sequential probability ratio test\.The Annals of Mathematical Statistics19\(3\),pp\. 326–339\.Cited by:[Proposition 2](https://arxiv.org/html/2607.12480#Thmproposition2.p1.3.3)\.
- A\. Wald \(1945\)Sequential tests of statistical hypotheses\.The Annals of Mathematical Statistics16\(2\),pp\. 117–186\.Cited by:[Proposition 2](https://arxiv.org/html/2607.12480#Thmproposition2.p1.3.3)\.
- D\. Walton, C\. Reed, and F\. Macagno \(2008\)Argumentation schemes\.Cambridge University Press\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px4.p1.1),[Table 3](https://arxiv.org/html/2607.12480#S4.T3.1.4.3.1.1.1)\.
- D\. Walton \(2006\)Fundamentals of critical argumentation\.Cambridge University Press\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px4.p1.1)\.
- D\. Walton \(2013\)Methods of argumentation\.Cambridge University Press\.Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p5.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px4.p1.1)\.
- X\. Wang, J\. Wei, D\. Schuurmans,et al\.\(2023\)Self\-consistency improves chain of thought reasoning in language models\.InICLR,Note:arXiv:2203\.11171Cited by:[Table 18](https://arxiv.org/html/2607.12480#A6.T18.1.1.6.1.1),[§2](https://arxiv.org/html/2607.12480#S2.p1.1)\.
- J\. Wei, X\. Wang, D\. Schuurmans,et al\.\(2022\)Chain\-of\-thought prompting elicits reasoning in large language models\.InNeurIPS,Note:arXiv:2201\.11903Cited by:[§2](https://arxiv.org/html/2607.12480#S2.p1.1)\.
- S\. Welleck, J\. Liu, R\. Le Bras, H\. Hajishirzi, Y\. Choi, and K\. Cho \(2021\)NaturalProofs: mathematical theorem proving in natural language\.InNeurIPS Datasets and Benchmarks Track,Note:arXiv:2104\.01112Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- C\. Wellman \(1971\)Challenge and response: justification in ethics\.Southern Illinois University Press,Carbondale, IL\.Cited by:[Appendix C](https://arxiv.org/html/2607.12480#A3.SS0.SSS0.Px1.p1.1),[§4\.3\.1](https://arxiv.org/html/2607.12480#S4.SS3.SSS1.Px2.p1.1)\.
- Z\. Weng, G\. Chen, and W\. Wang \(2025\)Do as we do, not as you think: the conformity of large language models\.InICLR,Note:arXiv:2501\.13381Cited by:[Table 18](https://arxiv.org/html/2607.12480#A6.T18.2.2.6.1.1)\.
- Z\. Wu, L\. Qiu, A\. Ross,et al\.\(2024\)Reasoning or reciting? exploring the capabilities and limitations of language models through counterfactual tasks\.InNAACL,Note:arXiv:2307\.02477Cited by:[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px1.p1.1)\.
- A\. Wynn, H\. Satija, and G\. K\. Hadfield \(2025\)Talk isn’t always cheap: understanding failure modes in multi\-agent debate\.Note:arXiv:2509\.05396Cited by:[itemM2](https://arxiv.org/html/2607.12480#A6.I1.ix2.p1.1)\.
- B\. Yao, C\. Shang, W\. Du,et al\.\(2025\)Peacemaker or troublemaker: how sycophancy shapes multi\-agent debate\.Note:arXiv:2509\.23055Cited by:[itemM2](https://arxiv.org/html/2607.12480#A6.I1.ix2.p1.1)\.
- S\. Yao, D\. Yu, J\. Zhao,et al\.\(2023a\)Tree of thoughts: deliberate problem solving with large language models\.InNeurIPS,Note:arXiv:2305\.10601Cited by:[§2](https://arxiv.org/html/2607.12480#S2.p1.1)\.
- S\. Yao, J\. Zhao, D\. Yu,et al\.\(2023b\)ReAct: synergizing reasoning and acting in language models\.InICLR,Note:arXiv:2210\.03629Cited by:[§2](https://arxiv.org/html/2607.12480#S2.p1.1)\.
- Y\. Yue, Z\. Chen, R\. Lu,et al\.\(2025\)Does reinforcement learning really incentivize reasoning capacity in LLMs beyond the base model?\.Note:arXiv:2504\.13837Cited by:[§2\.2](https://arxiv.org/html/2607.12480#S2.SS2.SSS0.Px2.p1.1)\.
- M\. Zečević, M\. Willig, D\. S\. Dhami, and K\. Kersting \(2023\)Causal parrots: large language models may talk causality but are not causal\.Transactions on Machine Learning Research\.Note:arXiv:2308\.13067Cited by:[§2\.1](https://arxiv.org/html/2607.12480#S2.SS1.p1.1)\.
- H\. Zhang, Z\. Cui, X\. Wang,et al\.\(2025\)Stop overvaluing multi\-agent debate—we must rethink evaluation and embrace model heterogeneity\.Note:arXiv:2502\.08788Cited by:[itemM5](https://arxiv.org/html/2607.12480#A6.I1.ix5.p1.8)\.
- K\. Zheng, J\. M\. Han, and S\. Polu \(2022\)miniF2F: a cross\-system benchmark for formal olympiad\-level mathematics\.InICLR,Note:arXiv:2109\.00110Cited by:[§4\.2](https://arxiv.org/html/2607.12480#S4.SS2.p6.1)\.
- D\. Zhou, N\. Schärli, L\. Hou,et al\.\(2023\)Least\-to\-most prompting enables complex reasoning in large language models\.InICLR,Note:arXiv:2205\.10625Cited by:[§2](https://arxiv.org/html/2607.12480#S2.p1.1)\.