Phase 2: Measure usable authority with an independent oracle #620

Merged
erikinkinen merged 2 commits from phase2/p2-04-usable-authority into main 2026-10-20 17:19:23 +02:00
Owner

Task

Closes #87 (P2-04).

Summary

Add an independent Read-authority evaluator and withdrawal measurement interface. Epoch-revoked tokens can now be reported as physically present but unusable, while a benign parallel grant preserves legitimate effective access without becoming target leakage.

Scope

Included

  • Reference validity evaluation in issuance order, independent of the real per-use parent walk and all strategy resolvers.
  • Exact target structural/usable residue, pre-query benign loss, nullable rates and deduplicated current effective-access triples.
  • Isolated actual-UseCap comparisons, hand-derived strategy expectations, negative fixtures and nonmutation checks.
  • API documentation and source-pinned acceptance/evidence records.

Explicitly excluded

  • Scheduled exposure/censoring aggregation, new validity profiles and Phase 5 token-copy semantics, owned by later issues.

Design intent

Only the oracle may read the complete analytical ledger. It accepts const state, advances no clock or identity cursor, and contributes no mechanism work. Frozen target/benign denominators are retrieved from the private policy ledger. Empty denominators produce null with an explicit reason. The documentation distinguishes later-epoch benign cohort loss from causal attribution requiring the declared paired comparison.

Phase discipline

The evaluator matches legacy_phase1 semantics, including root-only lazy markers and missing-parent epoch boundaries. Legacy metrics, logs and snapshots remain unchanged. Validation fixtures are not production research samples.

Verification

All 47 CTests pass at the source commit in Debug, Release and Clang 22 ASan/UBSan. The new suite includes 21 hand-derived strategy cases, seven collateral cases, 56 branching cases, rights/empty-denominator/new-grant cases and malformed evidence. Every canonical state's oracle answers match actual UseCap on separate engine copies; snapshots and the research GrantId cursor remain unchanged. All hooks pass and archived legacy bytes still match.

Commands, logs and JUnit evidence are pinned in artifacts/backlog/2.2.1/P2-04/acceptance.json. All eight exact-head Forgejo checks must pass before checklist update and merge.

# Task Closes #87 (P2-04). ## Summary Add an independent Read-authority evaluator and withdrawal measurement interface. Epoch-revoked tokens can now be reported as physically present but unusable, while a benign parallel grant preserves legitimate effective access without becoming target leakage. ## Scope ### Included - Reference validity evaluation in issuance order, independent of the real per-use parent walk and all strategy resolvers. - Exact target structural/usable residue, pre-query benign loss, nullable rates and deduplicated current effective-access triples. - Isolated actual-UseCap comparisons, hand-derived strategy expectations, negative fixtures and nonmutation checks. - API documentation and source-pinned acceptance/evidence records. ### Explicitly excluded - Scheduled exposure/censoring aggregation, new validity profiles and Phase 5 token-copy semantics, owned by later issues. ## Design intent Only the oracle may read the complete analytical ledger. It accepts const state, advances no clock or identity cursor, and contributes no mechanism work. Frozen target/benign denominators are retrieved from the private policy ledger. Empty denominators produce null with an explicit reason. The documentation distinguishes later-epoch benign cohort loss from causal attribution requiring the declared paired comparison. ## Phase discipline The evaluator matches `legacy_phase1` semantics, including root-only lazy markers and missing-parent epoch boundaries. Legacy metrics, logs and snapshots remain unchanged. Validation fixtures are not production research samples. ## Verification All 47 CTests pass at the source commit in Debug, Release and Clang 22 ASan/UBSan. The new suite includes 21 hand-derived strategy cases, seven collateral cases, 56 branching cases, rights/empty-denominator/new-grant cases and malformed evidence. Every canonical state's oracle answers match actual UseCap on separate engine copies; snapshots and the research GrantId cursor remain unchanged. All hooks pass and archived legacy bytes still match. [Commands, logs and JUnit evidence](https://git.erikinkinen.fi/attachments/99f157d4-b9e3-4455-bd6f-c39c1512af1c) are pinned in `artifacts/backlog/2.2.1/P2-04/acceptance.json`. All eight exact-head Forgejo checks must pass before checklist update and merge.
erikinkinen added this to the Phase 2 milestone 2026-10-20 17:14:15 +02:00
Phase 2: Record P2-04 usable authority acceptance (#87)
All checks were successful
clang-format / check-format (pull_request) Successful in 21s
markdownlint / markdown-lint (pull_request) Successful in 22s
ci / sanitizers (pull_request) Successful in 4m2s
ci / smoke (pull_request) Successful in 4m6s
markdownlint / markdown-lint (push) Successful in 24s
clang-format / check-format (push) Successful in 25s
ci / smoke (push) Successful in 3m16s
ci / sanitizers (push) Successful in 3m21s
48d72bcd58
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
erikinkinen/AES!620
No description provided.