GCTS sits at the intersection of several mature research streams. The safest academic posture is straightforward: the components are crowded, and the research contribution is the specific architecture-level composition.
Closest Neighboring Areas
| Area | Representative work | What it already covers | GCTS boundary |
|---|---|---|---|
| Automated fact verification | FEVER, SciFact, FEVEROUS, AVeriTeC | Claim/evidence retrieval, support/refute labels, insufficient-evidence labels | GCTS ranks claims across access-aware possible worlds and record contingencies |
| Attribution-grounded generation | ALCE, FActScore | Citation quality, atomic factuality, supported generation | GCTS treats citation/provenance as inference input and audit output |
| Truth discovery | Truth discovery surveys, Knowledge-Based Trust | Source reliability, multi-source conflict, web-scale fact probability | GCTS adds record control, generation duty, access state, and strategic non-production |
| Provenance systems | W3C PROV-O, C2PA, provenance semirings, ProvSQL | Derivation, content authenticity, provenance-aware data | GCTS uses provenance in claim ranking and status assignment |
| Probabilistic logic | MLNs, PSL, ProbLog, WFOMC, AMC | Weighted rules, relational probability, possible-world inference | GCTS worlds include access models and institutional-incentive hypotheses |
| Argumentation and legal evidence | Dung frameworks, Carneades, Wigmore charts, ATMS, BARD, Co-Arg | Attack/support graphs, proof standards, assumption contexts, competing hypotheses | GCTS combines evidential argument with record-access and oracle-boundary constraints |
| Missingness and omission | Rubin missing-data theory, open-world databases, Rule 37(e), selective disclosure, TRACER | Missing-data mechanisms, non-production, omission-aware verification | GCTS makes typed absence states runtime inference objects |
| Evaluation leakage | benchmark contamination, hidden tests, leakage surveys | Separation of evaluation artifacts from model behavior | GCTS formalizes runtime exclusion of gold labels and oracle answers |
Core Distinction
GCTS should not be framed as inventing fact checking, source scoring, provenance, probabilistic logic, possible worlds, contradiction detection, or missing-data analysis.
The defensible boundary is narrower:
- Evidence atoms include source, span, time, quality, access path, and provenance.
- Expected records are represented as typed record-access states.
- Missingness is conditioned on generation duty, expected observability, access path, control, production response, and incentives.
- Possible worlds branch over facts, rules, assumptions, access models, and institutional-incentive hypotheses.
- Claim ranking uses posterior mass across those worlds.
- Strict proof support is emitted separately from likely-truth posterior mass.
- Contradiction and chirality residuals remain visible in reports.
- Runtime scoring is barred from gold labels, hidden benchmark answers, or LLM truth votes.
Feature-to-Prior-Art Chart
| GCTS feature | Prior-art coverage | Distinguishing requirement |
|---|---|---|
| Evidence atoms | Atomic factuality, citation-grounded generation, claim decomposition | Access path and record-contingency metadata are part of the atom model |
| Typed record-access states | Missing-data theory, open-world databases, legal spoliation, omission detection | Absence state directly affects world ranking and claim status |
| Generation duty | Legal and compliance reasoning | Duty becomes a computational precondition for absence penalties |
| Contradiction-preserving graph | Argumentation, ATMS, legal evidence models | Contradiction is preserved as residual structure tied to evidence and access |
| Possible-world ranking | Probabilistic databases, MLNs, PSL, WFOMC | Worlds vary over record-access hypotheses and fact assignments |
| Strict proof separation | Proof theory, legal proof standards, hard/soft rule systems | `P0(c |
| Oracle boundary | Leakage and hidden-test practice | Runtime truth mass cannot come from labels, expert answers, or LLM judgments |
| Audit report | Fact-check explanations, provenance reports, legal charts | Report links status to evidence, missing records, proof traces, worlds, and next records |
Academic Claim Discipline
A strong paper should say:
GCTS proposes an evidence-first architecture for likely-truth ranking where typed record-access states and generation-duty-aware missingness participate directly in possible-world scoring, claim-status assignment, and audit output.
A weak paper would say:
GCTS is a new truth discovery system.
The second version is too broad. It collides with fact verification, truth discovery, probabilistic logic, provenance, and legal argumentation work.
Sources To Cite First
Start with the primary or official sources listed in References, then expand into a full BibTeX bibliography before arXiv submission.