R03.3: finite-state tracking and planning reference audit #15

New issue

Closed

opened 2026-09-21 02:01:22 +00:00 by pyrrha · 0 comments

pyrrha commented

2026-09-21 02:01:22 +00:00

Member

Scope

Implement R03 Family B after the accepted records-family foundation: deterministic finite-state tracking and bounded planning tasks, an independent transition reference and BFS planning reference, and small exhaustive audits.

Boundary

Model-free only. No model, tokenizer, data download, GPU, or experiment run.
Tracking and planning references must be separate from generator generation logic.
Planning scorer semantics accept any shortest goal-reaching plan, not a generator tie-breaker.
Reject unreachable/ambiguous/shortcut instances before candidate assignment.
Vesper reviews the final semantic/audit boundary.

Acceptance

Deterministic typed state/action generation and canonical episode records.
Tracking reference and separate bounded BFS planning reference.
Exhaustive small audit for two binary variables, up to three actions, horizon four.
Rejection reasons/counts and stable fixtures.
Fresh-venv tests, documentation/ledger/diff checks; no unearned results.

## Scope Implement R03 Family B after the accepted records-family foundation: deterministic finite-state tracking and bounded planning tasks, an independent transition reference and BFS planning reference, and small exhaustive audits. ## Boundary - Model-free only. No model, tokenizer, data download, GPU, or experiment run. - Tracking and planning references must be separate from generator generation logic. - Planning scorer semantics accept any shortest goal-reaching plan, not a generator tie-breaker. - Reject unreachable/ambiguous/shortcut instances before candidate assignment. - Vesper reviews the final semantic/audit boundary. ## Acceptance - Deterministic typed state/action generation and canonical episode records. - Tracking reference and separate bounded BFS planning reference. - Exhaustive small audit for two binary variables, up to three actions, horizon four. - Rejection reasons/counts and stable fixtures. - Fresh-venv tests, documentation/ledger/diff checks; no unearned results.

pyrrha referenced this issue from a pull request that will close it,

2026-09-21 02:23:59 +00:00

feat(R03): add finite-state tracking and planning audit #16

pyrrha closed this issue

2026-09-21 02:31:04 +00:00