Sources we considered, with review status and reasons.
3070 recorded entries · 652 awaiting detailed review · 584 excluded · 672 on hold · 3 publisher withdrawals · 1159 related to curated records.
Suggest a missing contribution or correction.
See omissions and decisions reviewed against public discussion.
Entries without a recorded subject appear only under All subjects. Coverage totals describe all subjects.
1911 matching entries · Page 37 of 39
Collected arXiv metadata screened for scope. This is discovery, not a completed primary-paper review or proof check. Supplied-target benchmarks, applied prediction and human technical work are not automatically curated.
Screening recorded: 2026-10-21 · Decision reviewed: 2026-10-21T07:53:19.130Z
Collected arXiv metadata screened for scope. This is discovery, not a completed primary-paper review or proof check. Supplied-target benchmarks, applied prediction and human technical work are not automatically curated.
Screening recorded: 2026-10-21 · Decision reviewed: 2026-10-21T07:53:19.130Z
Thread and comment leads read. Larsson repository explicitly describes an extraction weaker than an existing numerical cutoff; model-only audit is not human checking. LemmaLab repository/API404, with original QRH dependencies and uncompiled-check caveats; proposed class-number/idoneal extensions remain unreviewed. No forum claim promoted from anonymous or model checking alone.
Screening recorded: 2026-10-21 · Decision reviewed: 2026-10-21T07:53:19.130Z
AI-greement generated protocol candidates, but the final A-greement needed human removal of benchmark exploitation and an inherited Bodega flaw. Before inclusion isolate the surviving AI-origin protocol mechanism and any essential mathematical repair. Empirical latency gains and Lean safety models remain separate; no universal performance or fault-tolerance claim inferred.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
Formal addendum builds on a 2018 human planning thesis. Determine whether any new central theorem is AI-origin rather than a formal reconstruction of supplied game/planning definitions and proofs.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
PerMIT’s security theorems concern specified memory/provenance assumptions and implementation faithfulness. Contribution-specific AI mathematical origin is not established by machine-checked security models.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
The reviewed disclosure credits Claude/GPT with the Lean formalization of a human depth-separation argument. A new central AI mathematical contribution is not established by this formalization.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
The Eshelby manuscript separates analytical theorem, finite-shape enclosure, FEM and AI-process evidence. General candidate-formulation/exploration assistance does not yet allocate a specific central retained stability result to AI. Physical assumptions and numerical limits must be preserved.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
SpecGuard concerns generated specifications and supplied-target benchmark conflicts. Its measured certification rates are not newly discovered mathematical theorems.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
Interactive specification clarification is evaluated on supplied tasks; no distinct central AI mathematical research result is established by this scope.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
Autoformalizes a previously known core theorem behind MIP*=RE. Agent audits and repairs of statement correspondence are formalization work, not a new AI-origin solution to the original theorem.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
Authors used GPT discussion and generated the Lean formalization while guiding and rewriting the paper. Isolate the central half-density uncertainty-principle idea from the prior human strategy before inclusion.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
Authors report extensive Claude assistance with constructions/proofs and their own complete proof checking, but the specific central locator/new-domain contribution needs allocation between human prior work and AI. Preserve unconditional k=2 versus conditional higher-k restrictions before inclusion.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
The revised v2 is titled Electron quasicrystal in semiconductor quantum wells. Its neural variational calculation reports metallic/crystalline/bilayer states; finite-cell competing configurations and energy ordering need a fuller v1/v2 and size/error comparison before a new phase assessment. Numerical evidence is not exact ground-state proof or experimental confirmation; keep the candidate pending with this concrete scope gap.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
Both the original and revised v2 entropy-functional paper retrieved; specific central AI origin of its analytical result is not documented in the reviewed passages. Do not infer eligibility from neural-functional applications or the query match.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
Hydrodynamics disclosure describes Codex manuscript revision, literature cross-checking and review. This does not document central AI origin of the stable/causal-frame construction.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
Selected primary paper retrieved. Specific central AI origin, exact theorem and prior-work allocation remain unresolved; general AI assistance is insufficient.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
Hecke-zero-free-region follow-up retrieved. Compare exact families and bounds with existing OpenAI/Cureton–Thorner records, and separately attribute any human strengthening. No duplicate zero-free discovery from a changed title.
Screening recorded: 2026-10-22 · Decision reviewed: 2026-10-22T07:39:59.456Z
The total–total randomized composition counterexample is substantial and the source reports a Lean check. The AI statement documents failed direct attempts, human decomposition and proposed construction families, but does not isolate which retained detector/certificate mechanism was supplied centrally by AI. Keep the logarithmic saving and exponential size coupling separate from polylogarithmic composition.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Authors report important human and LLM ideas, with initial ideas supplied by humans, and that they checked and understand all results. Isolate a specific central AI balancing/path modification idea before inclusion. Lean formalization retains Asimow–Roth equivalence as an explicit hypothesis.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Recovered PDF explicitly attributes original concept, framework and primary results to the human authors. Aristotle assisted error verification and simplification; this scope fails the central-AI eligibility requirement.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
The CM-period result uses the OpenAI separated-weight interpolation program and specified uniformization/CM/theta inputs. Generic ChatGPT exploration, exposition and Lean assistance does not yet identify the central AI-origin determinant/selected-minor result. Do not infer irrationality exponent two from formal conditional wrappers.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Retrieval/alignment dataset and LLM reranking benchmarks connect known Stacks/Mathlib statements, without a distinct AI-origin mathematical theorem.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
The new1/4 bounds and exact Subspace-Theorem dependencies were read. Formalization and a manuscript’s authorship do not by themselves document central AI origin of the strengthened argument.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Mathlib retrieval and query-rewriting benchmark. AI is the evaluated system; retrieval gains and generated training queries are not a scoped new AI mathematical discovery.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-26T08:59:58.123Z
AI-led detailed-balance proof development is explicitly documented. Determine the substantive new proof increment beyond the already retained OpenAI weak-reversibility persistence argument and classical entropy convergence. Seventy-one upstream Lean modules and known global-attractor results are credited; no duplicate theorem count or automatic formalization-only inclusion.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Certificate synthesis on81 supplied control instances is a proof/repair evaluation, not a scoped new mathematical or theoretical-physics result. Verification percentages are benchmark outcomes.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
The Lean/ZFC arithmetic metatheory statement and formalization are research claims, but a contribution-specific central AI role is not established in the retrieved source.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Changed v3 preserves a forward/inverse NTT model with characteristic constraints. Allocate a specific AI-origin rank result and compare revisions before publishing.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-28T08:47:35.674Z
Execution-grounded formal framework and nine formalized known quantum algorithms. Statement fidelity and resource contracts matter, but no new retained AI-origin quantum algorithm is identified by the source.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-26T08:59:58.123Z
Agentic verified optimization measures program performance against supplied reference semantics. No specific new theoretical-computer-science discovery is established by benchmark speedups.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
The algebra classification’s exact polynomial-identity checking is documented, but central AI origin of the mathematical classification remains unestablished.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Reviewed actual v2. Numerical/vacuity/semantic verification and labeled DLMF statement datasets are benchmark and formalization-method work; no specific new central-AI mathematics result is established.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Reviewed actual v3 PDF after HTML404. It strengthens positive SoS coverage through order50, with Lean only through20, and large-order non-SoS checking. Claude/Codex/Qwen candidate/code/proof assistance is generic; identify the central retained construction or argument supplied by AI before inclusion.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Reviewed actual v2. Langlands’s known lemma is formalized; ChatGPT transferred formulas/citations and helped wording. That disclosure does not establish central AI mathematical work.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Compressible-Euler cutoff implosion paper discloses Claude editing and numerical-method reproducibility checking. These roles are not central AI discovery; simulations plus asymptotics are not an exact general blow-up theorem.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Flux-vacua scan discloses AI coding and language polishing for ML analysis. SLSQP candidate minima and embedding patterns do not establish a central AI theoretical contribution or a complete physical-vacuum classification.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Actual v2 gives AI exploration/centering/corrections disclosures. Isolate the new central aggregation construction from human design and prior Coppersmith–Winograd/Pan identities before inclusion; no automatic matrix-exponent upgrade.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Actual v2 reports ChatGPT candidate counterexamples and proof-strategy exploration. Exact central retained AI lower-bound or certificate idea remains unresolved.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
The source calls this a proof draft and identifies further checking points. No contribution-specific central AI role established; do not relabel the proposed axisymmetric regularity threshold as checked.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Black-hole-scattering thesis uses Claude for language and figures. Fifth-order post-Minkowskian/self-force scope does not make that editing central AI theory discovery.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Exact few-magnon Bethe-scar v2 read in its alternating-chain model. No documented central AI role established; numerical or analytical quantum work alone is not eligible.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
Functional-analytic Hardy–Littlewood-equivalence paper has no identified central AI contribution in the retrieved version; no inclusion inferred from a collector match.
Screening recorded: 2026-10-23 · Decision reviewed: 2026-10-23T08:28:30.727Z
The reciprocal-RIS result documents Claude-assisted Lean verification. This does not identify central AI origin of the electromagnetic design theorem; no inclusion inferred from machine verification.
Screening recorded: 2026-10-24 · Decision reviewed: 2026-10-24T07:28:43.626Z
Proof-carrying contracts concern agent commitments and checking infrastructure. Examples of LLM-generated proofs and known formal-method protocols are not a scoped new central-AI mathematics or theory discovery.
Screening recorded: 2026-10-24 · Decision reviewed: 2026-10-24T07:28:43.626Z
The AI disclosure limits assistance to symbolic-experiment code and formal verification of the two main theorems. The new irrationality arguments are not centrally attributed to AI.
Screening recorded: 2026-10-24 · Decision reviewed: 2026-10-24T07:28:43.626Z
Actual v2 algebraic/geometric classification read. A new central AI contribution is not documented in the retrieved source; changed formalization metadata does not establish mathematical origin.
Screening recorded: 2026-10-24 · Decision reviewed: 2026-10-24T07:28:43.626Z
Mechanized communication-safety metatheory needs a specific central AI-origin theorem, rather than verification of a human-designed type system alone.
Screening recorded: 2026-10-24 · Decision reviewed: 2026-10-28T08:47:35.674Z
Exact compilation and resource laws are formal-method results. No contribution-specific central AI origin established from the manuscript; do not promote a proof-assistant listing automatically.
Screening recorded: 2026-10-24 · Decision reviewed: 2026-10-24T07:28:43.626Z
HTML unavailable; versioned PDF recovered. The power-of-two cycle statement in projective cubic graphs has precise edge-width assumptions, but its central AI origin remains unestablished.
Screening recorded: 2026-10-24 · Decision reviewed: 2026-10-24T07:28:43.626Z
Discovery metadata attribution: What AI Found, CC BY 4.0 · VibeMathed contributors, editorial layer CC BY 4.0. AIProofWatch supplies the screening decisions.
Allow optional Google Analytics to help improve AIProofWatch? Privacy details.
