BUG: authority oracle rejects legal holder movement and attributes access to the birth holder #638

Closed
opened 2026-10-20 21:28:28 +02:00 by erikinkinen · 0 comments
Owner

Problem

P2-08-A's real apply_move_holder operation preserves issuance identity, grant, ancestry, object, rights and validity, but the previously accepted analytical oracle requires live.from == issuance.holder_id. A graph-valid move therefore turns a valid authority measurement into an invariant error. Removing that guard alone would still attribute effective access to the birth holder.

A standalone probe at signed source 0b9ac14f4ac6f1af808ff21ecc5d3a1373b6e4f2 creates Alice and Bob, issues one Read capability to each, records immutable issuance evidence, and moves Alice's capability to Bob. Before the move the oracle succeeds; afterward graph validation and actual UseCap succeed but the oracle reports live capability disagrees with issuance evidence (exit 1). Probe source and logs are preserved under /tmp/aes-phase2-execution/holder-oracle-probe for the final evidence archive.

Required correction

Keep policy source atoms identified by their immutable birth roles. Derive effective access from the current holder's fixed role binding. Permit explicit immutable holder-role binding for actors who have never received an issuance; fail closed for missing/conflicting bindings. Preserve rights, object, ancestry, deletion and validity checks, frozen denominators, and unmetered oracle isolation. The mechanism must never consume analytical history.

Acceptance

  • Reproduce the holder-movement failure and add regression coverage for current access versus immutable source atoms, empty destinations, conflicting/missing role bindings, validity and withdrawal cohorts.
  • Implement the correction in its own signed branch/PR, preserving legacy mechanism behavior and validating Debug, Release, ASan/UBSan and repository hooks.
  • Attach immutable before/after evidence, pass every final-head Forgejo check, then check this checklist once and merge.

Discovered during #97 integration review; blocks P2-08-A acceptance. Phase 2, milestone 3. No scientific data or protocol changes.

## Problem P2-08-A's real `apply_move_holder` operation preserves issuance identity, grant, ancestry, object, rights and validity, but the previously accepted analytical oracle requires `live.from == issuance.holder_id`. A graph-valid move therefore turns a valid authority measurement into an invariant error. Removing that guard alone would still attribute effective access to the birth holder. A standalone probe at signed source `0b9ac14f4ac6f1af808ff21ecc5d3a1373b6e4f2` creates Alice and Bob, issues one Read capability to each, records immutable issuance evidence, and moves Alice's capability to Bob. Before the move the oracle succeeds; afterward graph validation and actual UseCap succeed but the oracle reports `live capability disagrees with issuance evidence` (exit 1). Probe source and logs are preserved under `/tmp/aes-phase2-execution/holder-oracle-probe` for the final evidence archive. ## Required correction Keep policy source atoms identified by their immutable birth roles. Derive effective access from the current holder's fixed role binding. Permit explicit immutable holder-role binding for actors who have never received an issuance; fail closed for missing/conflicting bindings. Preserve rights, object, ancestry, deletion and validity checks, frozen denominators, and unmetered oracle isolation. The mechanism must never consume analytical history. ## Acceptance - [x] Reproduce the holder-movement failure and add regression coverage for current access versus immutable source atoms, empty destinations, conflicting/missing role bindings, validity and withdrawal cohorts. - [x] Implement the correction in its own signed branch/PR, preserving legacy mechanism behavior and validating Debug, Release, ASan/UBSan and repository hooks. - [x] Attach immutable before/after evidence, pass every final-head Forgejo check, then check this checklist once and merge. Discovered during #97 integration review; blocks P2-08-A acceptance. Phase 2, milestone 3. No scientific data or protocol changes.
erikinkinen added this to the Phase 2 milestone 2026-10-20 21:28:28 +02:00
Sign in to join this conversation.
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#638
No description provided.