Add GovKM experimental authority-resolution fixtures and generator

experimental/authority_resolution/CONTRIBUTION_NOTES.md

@@ -0,0 +1,3 @@
+# Contribution review notes
+
+Candidate only; not submitted. GovKM is fixture author. Intended upstream licensing is ADEBench repository MIT, subject to separate approval. This archive contains public fictional fixtures only, zero holdout seed/material, no scoring or model-adapter integration. Avoid publishing until maintainer agrees exact directory/schema and oracle visibility rules.

experimental/authority_resolution/README.md

@@ -0,0 +1,28 @@
+# GovKM / ADEBench authority-resolution fixture generator v0.2.0
+
+**Scope:** Public, purely synthetic fixture-generation candidate for ADEBench's proposed experimental non-Core authority-resolution track. **Not** an ADEBench scorer, model-memory adapter, or completed benchmark. No scored tests, upstream submission, private holdout, or external communication.
+
+## Quick start (non-scored fixture self-validation)
+
+```sh
+python -m unittest discover -s tests -v
+python generate_authority_cases.py --seed 20261022 --count 2 --set public --output-dir examples/public_seed_20261022
+```
+
+`--count N` generates N variants **per** six-pattern family (6N total). The seed is a nonnegative integer <2^63. The output directory must not exist; no overwrite. Holdout export is explicitly rejected. Python's PRNG stream should be replayed using the Python runtime version recorded in metadata.
+
+## v0.2 protocol corrections
+
+- `cases.json` includes `source_inventory_text`, which is **only the input record collection** offered for indexing. It must never be counted as delivered-to-model evidence. The only scoring boundary accepted by Simone is ADEBench's independently captured `delivered_text_after_cut`. This implementation does not create, simulate or evaluate that field. `expected_outcomes.json` is a separated private-to-evaluator oracle; never provide it to a model or retrieval system.
+- Authorized supersession AR-01 requires source E1, superseding E2, and source grant G1 as **required** evidence rather than assuming supersession alone suffices. Revocation AR-04 likewise requires E1, E2 and G1.
+- AR-06 recognizes two different outcomes: `NO_AUTHORITY_IF_COVERAGE_PROVEN` only where grant-register coverage (SOURCE-REGISTRY) is independently supported in model-delivered evidence; otherwise `AUTHORITY_UNRESOLVED`. `source_coverage` is a fictional source assertion, **not proof that the memory delivered it**. External score thresholds and semantic paraphrase matching remain ADEBench decisions.
+- All scenarios preserve distinct `valid_at` (effective authority) versus `recorded_at` (ingestion/source chronology). AR-05 historical as-of t2 must not be superseded retroactively by t3. Exact boundary conditions remain subject to ADEBench's scoring contract.
+- Three separate evaluator dimensions are declared: operative value, authority outcome and source event delivery. This generator DOES NOT implement any evaluator, cutoff algorithm, or score.
+
+## Files and provenance
+
+`cases.json` public fictional events and coverage assumptions; `expected_outcomes.json` candidate oracle (never feed to memory); `metadata.json` seed/version/runtime; `sha256_manifest.json` SHA-256 for output files and generator/test source; `tests/test_generator.py` local non-scored consistency checks. Source and tests are included in manifest rather than embedding self-referential hashes into source.
+
+**Packaging boundary:** No partner records, GovKM doctrine, internal scoring methods, or proprietary AIL logic may be contributed. ADEBench controls its own scoring and experimental leaderboard; attribution of GovKM as fixture author and licensing under actual repo license must be verified at the separate upstream approval gate. ADEBench, not GovKM, retains any private holdout seeds.
+
+**Protocol note:** `source_coverage` currently declares a fictional complete registry as a source assertion. It does not prove actual retrieved or delivered coverage. This deliberately prevents false-positive claims that "no grant found" means "no authority." An independent evaluator must define a falsifiable model-visible coverage proof before scoring NO_AUTHORITY. Do not infer a passing result from this metadata.

experimental/authority_resolution/SOURCE_SHA256.json

@@ -0,0 +1,9 @@
+{
+  "README.md": "9db79f9da2668a144982acb21e1c9f1097c44c05517e2de2ba6f14f45188f4db",
+  "examples/public_seed_20261022/cases.json": "f6c2515f64ed6ddc2226cfef19c0e9db0426092893ee8d211195cbd119f1e0a5",
+  "examples/public_seed_20261022/expected_outcomes.json": "b8f64ef4f018958db56a4ed0a828dab006a56cfc09c26093e91cb4410232c192",
+  "examples/public_seed_20261022/metadata.json": "e169842f562b59cd3bad28e0ac8c710338f29df2712ee797cd761c9a02895908",
+  "examples/public_seed_20261022/sha256_manifest.json": "361ac71f6d0ccfb8cb5431b3290d0860a77dbbd5deab236874b5b8921c411a5b",
+  "generate_authority_cases.py": "a2321ffed53aaa7a9c42fb47293c8f4875581effe76bf9760b83e676fbfe4d58",
+  "tests/test_generator.py": "f9d4570b94002eac7ee51ccf4cf25994554b35ae9d207a73727a2d2e79e77dbc"
+}

experimental/authority_resolution/SUBMISSION.md

@@ -0,0 +1,11 @@
+# GovKM experimental authority-resolution contribution
+
+GovKM is fixture author. The synthetic contribution is offered under this repository's MIT license; see ../../LICENSE. The original v0.2 source, example files and manifests are preserved byte for byte. README.md and CONTRIBUTION_NOTES.md describe the original candidate preparation status; this submission note records the subsequent authorized proposal.
+
+This directory is a proposed experimental location outside Core, subject to maintainer review under issue #2. It adds no scorer, adapter, package dependency, CI configuration or leaderboard change. ADEBench retains scoring, metrics, experimental leaderboard and private holdout custody.
+
+Only source_inventory_text is offered as unscored source input. expected_outcomes.json is a public-example evaluation oracle and must never enter memory/model input. Only independently captured delivered_text_after_cut can support future ADEBench scores. No private holdout was generated or included.
+
+Validation: Python 3.12.14, 17/17 offline generator tests passed. Public seed 20261022, count 2 per family gives 12 cases. Cases and expected outcomes reproduce approved sample bytes. Original metadata records Python 3.13.5; current-runtime replay changes python_version and its manifest hash only. No scored benchmark or native AI-memory/model experiment was run.
+
+Maintainer review remains needed for directory/schema, public-oracle placement, delivered coverage proof for no-authority cases, temporal boundaries and matching/cut rules. No claim of adoption before merge.

experimental/authority_resolution/examples/public_seed_20261022/expected_outcomes.json

@@ -0,0 +1 @@
+[{"case_id":"AR-01-S20261022-V0001","conflicting_event_ids":[],"delivery_boundary":"ADEBench measured delivered_text_after_cut only; source_inventory_text never scores","expected_resolution":"RESOLVED","historical_event_ids":["E1"],"independent_dimensions":["operative_value","authority_outcome","source_event_delivery"],"interpretation_note":"Expected event and authority evidence pertains to delivered text after cut, not retrieval pool or final model answer.","operative_values":["Fern"],"required_event_ids":["E1","E2"],"required_grant_ids":["G1"],"supersession_event_ids":["E2"],"temporal_basis":"valid_at; recorded_at provenance only"},{"case_id":"AR-01-S20261022-V0002","conflicting_event_ids":[],"delivery_boundary":"ADEBench measured delivered_text_after_cut only; source_inventory_text never scores","expected_resolution":"RESOLVED","historical_event_ids":["E1"],"independent_dimensions":["operative_value","authority_outcome","source_event_delivery"],"interpretation_note":"Expected event and authority evidence pertains to delivered text after cut, not retrieval pool or final model answer.","operative_values":["Amber"],"required_event_ids":["E1","E2"],"required_grant_ids":["G1"],"supersession_event_ids":["E2"],"temporal_basis":"valid_at; recorded_at provenance only"},{"case_id":"AR-02-S20261022-V0001","conflicting_event_ids":[],"delivery_boundary":"ADEBench measured delivered_text_after_cut only; source_inventory_text never scores","disallowed_event_ids":["E2"],"expected_resolution":"RESOLVED","historical_event_ids":[],"independent_dimensions":["operative_value","authority_outcome","source_event_delivery"],"interpretation_note":"Expected event and authority evidence pertains to delivered text after cut, not retrieval pool or final model answer.","operative_values":["Ember"],"required_event_ids":["E1"],"required_grant_ids":["G1"],"supersession_event_ids":[],"temporal_basis":"valid_at; recorded_at provenance only"},{"case_id":"AR-02-S20261022-V0002","conflicting_event_ids":[],"delivery_boundary":"ADEBench measured delivered_text_after_cut only; source_inventory_text never scores","disallowed_event_ids":["E2"],"expected_resolution":"RESOLVED","historical_event_ids":[],"independent_dimensions":["operative_value","authority_outcome","source_event_delivery"],"interpretation_note":"Expected event and authority evidence pertains to delivered text after cut, not retrieval pool or final model answer.","operative_values":["Jade"],"required_event_ids":["E1"],"required_grant_ids":["G1"],"supersession_event_ids":[],"temporal_basis":"valid_at; recorded_at provenance only"},{"case_id":"AR-03-S20261022-V0001","conflicting_event_ids":["E1","E2"],"delivery_boundary":"ADEBench measured delivered_text_after_cut only; source_inventory_text never scores","expected_resolution":"CONFLICT_UNRESOLVED","historical_event_ids":[],"independent_dimensions":["operative_value","authority_outcome","source_event_delivery"],"interpretation_note":"Expected event and authority evidence pertains to delivered text after cut, not retrieval pool or final model answer.","operative_values":[],"required_event_ids":["E1","E2"],"required_grant_ids":["G1","G2"],"supersession_event_ids":[],"temporal_basis":"valid_at; recorded_at provenance only"},{"case_id":"AR-03-S20261022-V0002","conflicting_event_ids":["E1","E2"],"delivery_boundary":"ADEBench measured delivered_text_after_cut only; source_inventory_text never scores","expected_resolution":"CONFLICT_UNRESOLVED","historical_event_ids":[],"independent_dimensions":["operative_value","authority_outcome","source_event_delivery"],"interpretation_note":"Expected event and authority evidence pertains to delivered text after cut, not retrieval pool or final model answer.","operative_values":[],"required_event_ids":["E1","E2"],"required_grant_ids":["G1","G2"],"supersession_event_ids":[],"temporal_basis":"valid_at; recorded_at provenance only"},{"case_id":"AR-04-S20261022-V0001","conflicting_event_ids":[],"delivery_boundary":"ADEBench measured delivered_text_after_cut only; source_inventory_text never scores","expected_resolution":"NO_AUTHORITY","historical_event_ids":["E1"],"independent_dimensions":["operative_value","authority_outcome","source_event_delivery"],"interpretation_note":"Expected event and authority evidence pertains to delivered text after cut, not retrieval pool or final model answer.","operative_values":[],"required_event_ids":["E1","E2"],"required_grant_ids":["G1"],"supersession_event_ids":["E2"],"temporal_basis":"valid_at; recorded_at provenance only"},{"case_id":"AR-04-S20261022-V0002","conflicting_event_ids":[],"delivery_boundary":"ADEBench measured delivered_text_after_cut only; source_inventory_text never scores","expected_resolution":"NO_AUTHORITY","historical_event_ids":["E1"],"independent_dimensions":["operative_value","authority_outcome","source_event_delivery"],"interpretation_note":"Expected event and authority evidence pertains to delivered text after cut, not retrieval pool or final model answer.","operative_values":[],"required_event_ids":["E1","E2"],"required_grant_ids":["G1"],"supersession_event_ids":["E2"],"temporal_basis":"valid_at; recorded_at provenance only"},{"case_id":"AR-05-S20261022-V0001","conflicting_event_ids":[],"delivery_boundary":"ADEBench measured delivered_text_after_cut only; source_inventory_text never scores","expected_resolution":"RESOLVED","future_event_ids":["E2"],"historical_event_ids":[],"independent_dimensions":["operative_value","authority_outcome","source_event_delivery"],"interpretation_note":"Expected event and authority evidence pertains to delivered text after cut, not retrieval pool or final model answer.","operative_values":["Fern"],"required_event_ids":["E1"],"required_grant_ids":["G1"],"supersession_event_ids":[],"temporal_basis":"valid_at; recorded_at provenance only"},{"case_id":"AR-05-S20261022-V0002","conflicting_event_ids":[],"delivery_boundary":"ADEBench measured delivered_text_after_cut only; source_inventory_text never scores","expected_resolution":"RESOLVED","future_event_ids":["E2"],"historical_event_ids":[],"independent_dimensions":["operative_value","authority_outcome","source_event_delivery"],"interpretation_note":"Expected event and authority evidence pertains to delivered text after cut, not retrieval pool or final model answer.","operative_values":["Granite"],"required_event_ids":["E1"],"required_grant_ids":["G1"],"supersession_event_ids":[],"temporal_basis":"valid_at; recorded_at provenance only"},{"case_id":"AR-06-S20261022-V0001","conflicting_event_ids":[],"delivery_boundary":"ADEBench measured delivered_text_after_cut only; source_inventory_text never scores","expected_resolution":"NO_AUTHORITY_IF_COVERAGE_PROVEN","historical_event_ids":[],"incomplete_coverage_resolution":"AUTHORITY_UNRESOLVED","independent_dimensions":["operative_value","authority_outcome","source_event_delivery"],"interpretation_note":"Expected event and authority evidence pertains to delivered text after cut, not retrieval pool or final model answer.","operative_values":[],"required_coverage_refs":["SOURCE-REGISTRY"],"required_event_ids":["E1"],"required_grant_ids":[],"supersession_event_ids":[],"temporal_basis":"valid_at; recorded_at provenance only"},{"case_id":"AR-06-S20261022-V0002","conflicting_event_ids":[],"delivery_boundary":"ADEBench measured delivered_text_after_cut only; source_inventory_text never scores","expected_resolution":"NO_AUTHORITY_IF_COVERAGE_PROVEN","historical_event_ids":[],"incomplete_coverage_resolution":"AUTHORITY_UNRESOLVED","independent_dimensions":["operative_value","authority_outcome","source_event_delivery"],"interpretation_note":"Expected event and authority evidence pertains to delivered text after cut, not retrieval pool or final model answer.","operative_values":[],"required_coverage_refs":["SOURCE-REGISTRY"],"required_event_ids":["E1"],"required_grant_ids":[],"supersession_event_ids":[],"temporal_basis":"valid_at; recorded_at provenance only"}]

experimental/authority_resolution/examples/public_seed_20261022/metadata.json

@@ -0,0 +1 @@
+{"case_count":12,"classification":"PUBLIC_SYNTHETIC","count_per_pattern":2,"generator_version":"0.2.0","patterns":["AR-01","AR-02","AR-03","AR-04","AR-05","AR-06"],"prng":"python.random.Random","python_version":"3.13.5","schema_version":"2.0","seed":20261022,"source_inventory_is_not_measured_delivery":true}

experimental/authority_resolution/examples/public_seed_20261022/sha256_manifest.json

@@ -0,0 +1 @@
+{"algorithm":"SHA-256","files":{"cases.json":"f6c2515f64ed6ddc2226cfef19c0e9db0426092893ee8d211195cbd119f1e0a5","expected_outcomes.json":"b8f64ef4f018958db56a4ed0a828dab006a56cfc09c26093e91cb4410232c192","generate_authority_cases.py":"a2321ffed53aaa7a9c42fb47293c8f4875581effe76bf9760b83e676fbfe4d58","metadata.json":"e169842f562b59cd3bad28e0ac8c710338f29df2712ee797cd761c9a02895908","tests/test_generator.py":"f9d4570b94002eac7ee51ccf4cf25994554b35ae9d207a73727a2d2e79e77dbc"},"generator_version":"0.2.0","scope":"public fixture files and local generator/test sources only"}