Index  |  Benchmarks  |  Mathematics  |  Graph  |  About

Schemas

Machine-readable public contracts for Assignee Research artifacts.

Schema and public data contract changes are tracked in the changelog.

Purpose

Schemas define the public JSON structure used by artifact manifests and benchmark evidence records. They are intended for indexing, citation tooling, reproducible audits, and external readers that need stable field names.

Available Schemas

1 Paper manifest v1. JSON Schema for per-paper artifact manifests, including work metadata, public artifacts, quality assessment, provenance, and limitations.
2 Benchmark evidence v1. JSON Schema for model-benchmark score evidence clusters, including reported ranges, spread, severity, source coverage, source profile, entries, interpretation, and limitations.
3 Quality dimensions v1. JSON Schema for multidimensional quality assessment records, including evidence strength, citation grounding, numerical consistency, benchmark validity, novelty, uncertainty disclosure, reproducibility status, and safety/impact.
4 Quality tiers v1. JSON Schema for descriptive public quality tier records, including tier, basis, inputs, thresholds, enforcement note, and limitations.
5 Review monitor v1. JSON Schema for public re-review trigger records, including monitor state, active triggers, monitored triggers, enforcement note, and limitations.
6 Verification levels v1. JSON Schema for paper-level public verification status, including level, basis, claim summary, unsupported upper levels, enforcement note, and limitations.
7 Math result v1. JSON Schema for public mathematical result manifests, including verification state, computational budget summary, proof-source availability, artifacts, interpretation, and limitations.
8 Source provenance v1. JSON Schema for public paper provenance records, including artifact inspectability, external archival record state, aggregate claim lineage, source-list boundary, interpretation, and limitations.
9 Correction record v1. JSON Schema for public correction and manifest-version records, including correction state, additive correction events, superseded manifest references, policy notes, interpretation, and limitations.
10 API envelope v1. JSON Schema for additive API envelope responses, including contract metadata, query echo, pagination, links, records, and compatibility notes.
11 Status record v1. JSON Schema for the public status surface, including public counts, contract versions, public routes, data change policy, verification policy, and limitations.
12 Public export v1. JSON Schema for public dataset export records, including dataset identifier, media type, download URL, fields, limits, update policy, privacy boundary, and limitations.

Quality Dimension Contract

Quality dimensions separate public assessment into inspectable signals instead of relying on a single aggregate score. Dimension fields are designed for triage and auditability; they do not replace source papers, independent replication, or domain expert review.

Paper manifests may include a quality_dimensions object derived conservatively from public manifest fields. Dimensions marked UNASSESSED mean no public dimension-specific scorer is attached; they are not negative findings.

Quality Tier Contract

Paper manifests include a quality_tier object that maps public aggregate signals into one of five descriptive inspection bands: FLAGSHIP_CANDIDATE, DOI_GRADE, PUBLIC_RECORD, WATCHLIST, or QUARANTINE_CANDIDATE.

Quality tiers are generated from public manifest fields such as review score, verified claim count, artifact types, and reproducibility level. They are descriptive public triage labels and do not alter current publication or DOI behavior.

Review Monitor Contract

Paper manifests include a review_monitor object that records public re-review trigger state. The current public states are NO_PUBLIC_TRIGGER and REVIEW_RECOMMENDED.

The monitor contract defines trigger names for benchmark contradictions, source availability, extraction-signal changes, corrected manifests, and low quality tiers. Triggers marked NOT_EVALUATED_PUBLICLY are not public findings; they indicate that no public trigger signal is attached to the manifest.

Verification Level Contract

Paper manifests include a verification_status object that reports the highest paper-level public evidence status currently supported by manifest fields. The contract uses levels L0 through L7.

The public manifest does not claim sandbox execution, independent reproduction, external review, or formal verification unless supporting public artifacts or event records are attached. Unsupported upper levels are listed explicitly so readers do not infer stronger evidence than the manifest provides.

The claim_summary field exposes aggregate public counts such as source_grounded_claims, aggregate_claim_records, and per_claim_public_record_count. Current public manifests do not expose individual claim text.

Source Provenance Contract

Paper manifests include a source_provenance object using the source-provenance-v1 contract. The contract separates public metadata, generated report artifacts, aggregate claim-lineage records, and external archival records into explicit provenance levels.

The current public paper provenance contract intentionally exposes aggregate source-grounding signals, artifact availability, and DOI or external-record state. It does not expose private extraction traces, local file paths, individual claim text, or a full cited-source list unless such a list is attached as a public artifact in a future contract.

Correction Record Contract

Paper manifests include a correction_record object using the correction-record-v1 contract. The public states are CURRENT, CORRECTED, and SUPERSEDED. The contract also exposes manifest contract_version and contract_updated fields so clients can detect public contract drift without scraping HTML.

Corrections are additive public records. A work marked CURRENT has no public correction event attached; this is not a claim that the synthesis is error-free. Private review notes, local storage paths, and operational logs are intentionally omitted from correction records.

API Envelope Contract

The api-envelope-v1 contract provides stable machine-readable metadata for newer public API surfaces. Envelope responses include contract_version, contract_updated, query echo, offset pagination, navigation links, and a records array.

Legacy list endpoints such as /api/v1/papers keep their existing list response shape for compatibility. New clients should prefer /api/v1/papers/envelope and can read /api/v1/api-contract for endpoint-level compatibility metadata.

Status Record Contract

The status-record-v1 contract powers /status and /api/v1/status. It reports public data counts, public contract versions, public routes, and change policy without exposing private infrastructure or operational logs.

Public Export Contract

The public-export-v1 contract powers /exports, /api/v1/exports, and public JSONL dataset downloads such as /api/v1/exports/papers.jsonl. Export records are public metadata records suitable for indexing, citation tooling, and reproducible audits.

Exports are additive public data surfaces. They do not expose private operational logs, local storage paths, credentials, or internal worker state.

Math Result Contract

Mathematical result manifests use the math-result-v1 contract. The contract separates FALSIFIED, COMPUTATIONAL_EVIDENCE, FORMAL_PROOF_ATTEMPTED, and FORMAL_PROOF_VERIFIED.

Computational evidence is not proof. Public Lean4 source is linked only when the manifest can support a formal verification claim. Python check code, local file paths, and private execution logs are intentionally omitted from public math manifests.

Benchmark Evidence Contract

Benchmark evidence records are audit surfaces, not verdicts. The source_coverage field reports how many extracted records and distinct public sources support the cluster. The source_profile field reports public source URL coverage, source domains, and available publication-year bounds. These fields describe source breadth and provenance; they do not assert correctness or independent replication.

The benchmark evidence index API at /api/v1/benchmark-evidence supports additive public filters: limit, min_spread, severity, and coverage or coverage_level. It also supports public source filters: source_domain, year_min, and year_max. Year filters match clusters whose public source-year range overlaps the requested bounds. Filters operate only on public evidence-cluster metadata and preserve the default unfiltered response shape.

Machine clients can read /api/v1/benchmark-evidence/meta for the current public filter contract, allowed enum values, response fields, and example query URLs. Benchmark evidence JSON and metadata expose contract_version and contract_updated so clients can detect public contract drift without scraping HTML.

Stability

Versioned schema URLs are stable public contracts. Additive fields may appear in future records, but incompatible structure changes should use a new versioned schema URL.

Privacy Boundary

Public schemas describe only public artifact fields. They do not expose private infrastructure, local storage paths, non-public operational metadata, or private credentials.