{ "$schema": "https://json-schema.org/draft/2020-12/schema", "$id": "https://github.com/Imbad0202/academic-research-skills/shared/contracts/passport/experiment_provenance_entry.schema.json", "title": "Material Passport Experiment Provenance Entry", "description": "One entry in Material Passport Schema 9 experiment_provenance[] aggregate (#260). Scholar-entered at Stage 1 intake — NOT writer-emitted. Records the intake + disclosure layer for an experiment the scholar ran EXTERNALLY and brought results back from. ARS does not run experiments, does not auto-fill provenance, and does not judge whether an experiment was correctly designed/run/statistically-adequate/reproducible by ARS (POSITIONING non-goals, carried verbatim into the integrity gate wording). The entry backs manuscript claims that cite it via claim_intent_manifest.planned_experiment_ids[]. Cross-array invariants EP-INV-1..EP-INV-4 (in-passport experiment_id uniqueness, planned_experiment_ids resolution, empirical-kind gating, declaration symmetry) are lint-enforced in scripts/check_claim_audit_consistency.py — JSON Schema cannot express cross-array integrity. Shape-only validation also available standalone via scripts/check_experiment_provenance.py. The nested repro_lock sub-shape re-declares (does NOT $ref) the inline-object design from shared/artifact_reproducibility_pattern.md; the canonical field set is single-sourced in scripts/repro_lock_validation.py and a drift test asserts the two copies stay in sync. See docs/design/2026-06-08-260-experiment-provenance-intake-spec.md.", "type": "object", "additionalProperties": false, "required": [ "experiment_id", "title", "repro_lock", "planned_vs_executed", "negative_results", "known_limitations" ], "properties": { "experiment_id": { "type": "string", "minLength": 1, "pattern": "^[A-Za-z0-9][A-Za-z0-9._-]*$", "description": "Scholar-chosen identifier, flat within ONE Material Passport (a passport is one run). EP-INV-1 enforces in-passport uniqueness. Cross-passport consumers treat the durable key as (passport_hash, experiment_id). FROZEN at Stage 1 intake (D3 intake-freeze): once experiment_intake_declaration.status == experiments_declared is set, this is the stable key space writers reference via planned_experiment_ids. A post-intake rename is a re-intake event (re-run the manifest emitters), NOT a silent edit — EP-INV-2 FAILs a passport whose planned_experiment_ids dangle after a rename." }, "title": { "type": "string", "minLength": 1, "maxLength": 500, "description": "Human-readable label for the experiment (e.g. 'Ablation: remove attention head pruning')." }, "description": { "type": "string", "maxLength": 4000, "description": "Optional free-form description of what the experiment tested and why." }, "repro_lock": { "type": "object", "additionalProperties": false, "description": "Per-experiment reproducibility lock — the SAME inline-object shape as the passport-level repro_lock (shared/artifact_reproducibility_pattern.md), nested here so each experiment carries its own configuration trail. Re-declared (not $ref'd) because the source is an inline-prose object, not a schema file; the canonical required-field set is single-sourced in scripts/repro_lock_validation.py with a drift test. Documents configuration, NOT a deterministic replay guarantee.", "required": [ "schema_version", "stochasticity_declaration", "ars_version", "model", "prompts", "materials", "external_protocols", "cross_model" ], "properties": { "schema_version": { "const": "1.0" }, "stochasticity_declaration": { "type": "string", "minLength": 1 }, "ars_version": { "type": "string", "minLength": 1 }, "model": { "type": "object", "additionalProperties": false, "required": ["family", "id", "weight_stable"], "properties": { "family": { "type": "string", "minLength": 1 }, "id": { "type": "string", "minLength": 1 }, "weight_stable": { "type": "boolean" } } }, "prompts": { "type": "object", "additionalProperties": false, "required": ["hash_timing", "skill_md_hash", "agents_bundle_hash"], "properties": { "hash_timing": { "const": "skill-load" }, "skill_md_hash": { "type": "string", "minLength": 1 }, "agents_bundle_hash": { "type": "string", "minLength": 1 } } }, "materials": { "type": "object", "additionalProperties": false, "required": ["list_hash", "count"], "properties": { "list_hash": { "type": "string", "minLength": 1 }, "count": { "type": "integer", "minimum": 0 } } }, "external_protocols": { "type": "object", "additionalProperties": false, "required": ["s2_api_protocol_version", "s2_snapshot_available"], "properties": { "s2_api_protocol_version": { "type": "string", "minLength": 1 }, "s2_snapshot_available": { "type": "boolean" } } }, "cross_model": { "type": "object", "additionalProperties": false, "required": ["enabled", "secondary_model_id"], "properties": { "enabled": { "type": "boolean" }, "secondary_model_id": { "type": ["string", "null"] } } } } }, "planned_vs_executed": { "type": "array", "minItems": 1, "description": "Per-result-unit record of what the scholar PLANNED to measure vs what was actually EXECUTED. minItems 1 — an experiment_provenance entry with no planned_vs_executed unit is a degenerate declaration that backs no claim. The integrity gate (D6 check-2) reads executed:false + skip_reason and feeds the all-skipped derivation rule (every entry executed:false ⟹ a claim resting on this experiment is NOT_SUPPORTED_BY_PROVENANCE).", "items": { "type": "object", "additionalProperties": false, "required": ["planned", "executed"], "properties": { "planned": { "type": "string", "minLength": 1, "description": "What the scholar planned to measure (e.g. 'F1 on held-out test set with head-pruning removed')." }, "executed": { "type": "boolean", "description": "Whether this planned unit was actually run. executed:false MUST carry skip_reason (gate check, D6 check-2)." }, "skip_reason": { "type": "string", "minLength": 1, "description": "Required-by-gate when executed:false (D6 check-2, not a schema conditional so a malformed skip surfaces at the gate with full context). Why the planned unit was not run." }, "result_file": { "type": "string", "minLength": 1, "description": "Pointer to the scholar's reported result artifact (path/URI). The experiment_alignment_result.result_pointer points INTO this (e.g. result_file + metric) — experiment_id alone is too coarse for a claim like 'F1 improved 4.2%'." }, "metric": { "type": "string", "minLength": 1, "description": "Optional metric name the reported result reports (e.g. 'macro-F1')." }, "value": { "description": "Optional reported value for the metric (number or string)." } } } }, "negative_results": { "type": "array", "description": "Declared null/absent/contradicting findings. The KEY MUST be present (D6 check-0 absent-key rule, ported from #261 C3); an empty [] is well-formed and routes to the check-4 advisory. An ABSENT key is malformed → gate FAIL. A manuscript claim that contradicts a negative_results entry is NOT_SUPPORTED_BY_PROVENANCE (D4 derivation rule), not merely a disclosure advisory.", "items": { "type": "object", "additionalProperties": false, "required": ["description"], "properties": { "description": { "type": "string", "minLength": 1, "maxLength": 2000 }, "result_file": { "type": "string", "minLength": 1 } } } }, "known_limitations": { "type": "array", "description": "Declared material limitations of the experiment. The KEY MUST be present (D6 check-0 absent-key rule); empty [] is well-formed and routes to the check-4 advisory; an ABSENT key is malformed → gate FAIL. Surfaced in Limitations prose (D6 check-4 advisory).", "items": { "type": "object", "additionalProperties": false, "required": ["description"], "properties": { "description": { "type": "string", "minLength": 1, "maxLength": 2000 } } } } } }