Wiki · Hierarchical Active Inference
H-AIF Gate → Existing P-Ladder Map
How to read this page
A Plain and a Clear version of this page have not been written yet. What follows is the document itself.
Precise — the source document
This is the document. Rendered from the repository at the commit above, with nothing rewritten for the web. A gate re-renders it on every deploy and fails the build if a single byte differs.
This file is a MAPPING, not a new ladder.
The hierarchical-AIF gates do not replace the existing P0–P8 ladder. They produce new receipts that map onto the existing ladder definitions in
CLAUDE.md.
1. The existing ladder (authoritative, unchanged)
Defined in CLAUDE.md. Not redefined here. No v2. No parallel Q-ladder.
P0 = computational integrity
P1 = equation / implementation
P2 = observational
P3 = held-out predictive
P4 = transfer
P5 = intervention
P6 = structural / mechanistic
P7 = independent replication
P8 = full verdict
P8 is conjunctive: any single required FAIL, CONTRADICTED, NOT_ESTABLISHED,
BLOCKED_EXTERNAL, NOT_RUN, or INVALID_PROVENANCE makes FULL_PARITY = false.
2. Gate definitions (new work)
H-AIF-G1 = repo + frozen evidence integrity
H-AIF-G2 = correction notice + defect ledger
H-AIF-G3 = failing tests for verified defects
H-AIF-G4 = runner fixes and resource-bound reclassification
H-AIF-G5 = corrected full B4 reruns
H-AIF-G6 = full motor-stack AIF gap audit + isolated implementation
H-AIF-G7 = current design control + adversarial alternatives + corrected AIF scoring
H-AIF-G8 = raw archive / transfer protocol status
H-AIF-G9 = ledger/ladders/docs/handoff update
3. Mapping
| gate | maps to | status | receipt |
|---|---|---|---|
| H-AIF-G1 | P0 computational integrity |
ESTABLISHED | reports/H-AIF-G1-REPO-AND-FROZEN-EVIDENCE-INTEGRITY.md |
| H-AIF-G2 | P1; P3/P6 interpretation (via D1); P0/P1 provenance (via D2/D4) |
ISSUED | reports/CORRECTION-NOTICE-TO-REVIEWER.md, ledgers/HIERARCHICAL-AIF-DEFECT-LEDGER.md |
| H-AIF-G3 | P1 equation/implementation |
COMPLETE | results/motor_stack_aif/H-AIF-G3-FAILING-TESTS-BEFORE-FIX.txt |
| H-AIF-G4 | P1 equation/implementation; P0 provenance |
COMPLETE | reports/H-AIF-G4-RUNNER-FIX-REPORT.md, reports/RESOURCE-BOUND-RECLASSIFICATION.md |
| H-AIF-G5 | P3 held-out predictive; P6 structural/mechanistic (per cell) |
COMPLETE — all four corrected cells landed at full frozen N. B4C10, B4C02, B4C01, B4C11 (564a5b0f…, U4_OK, collapse 0.0055). Every cell's prediction record was committed ahead of its result except B4C10 (D9, structurally unattainable) B4C10 (959a00e9…), B4C02 (0633988d…), B4C01 (8256cb12…), B4C11 (564a5b0f…) — each + report. B4C11 prospectivity SATISFIED (prediction 897c8ab committed ~19.4 h before the observation) |
|
| H-AIF-G6 | P1; and P6 gap status |
AUDIT COMPLETE / IMPLEMENTATION NOT_STARTED | docs/GAP-AUDIT-FULL-HIERARCHICAL-MOTOR-MODEL.md |
| H-AIF-G7 | P3; P6 discriminator pressure |
EXECUTED — verdict NOT_ESTABLISHED against control and strongest adversaries; RESOLVED_ABOVE only vs M0/M5. Coverage extended 2026-07-22: M4/M6/M7 contrasted post-hoc, neither M4 nor M6 resolves; candidate ranks 5th of 10; M7 exposed D10 |
results/.../F_SIDE_MOTOR_STACK_SCORING_RESULT.json (b3b12720…), results/.../M4_M6_M7_PER_MOTOR_CONTRASTS_RESULT.json (751a59ef…), reports for both |
| H-AIF-G8 | P2 observational if raw archive rederived; P4 transfer only if an independent dataset exists |
NOT_LOCATED / NOT_ESTABLISHED | pending |
| H-AIF-G9 | no new evidence by itself; records mapped status changes only | IN PROGRESS | this file |
4. Ledger rule
A gate result may update a P-level only if all six hold:
- the existing P-level definition is named;
- the source artifact is named;
- the claim scope is named;
- the falsifier is carried;
- partial / not-run / negative states are preserved;
- the update does not redefine the P-level.
5. Current P-level status after this correction round
| level | status | note |
|---|---|---|
P0 computational integrity |
holds — strengthened | H-AIF-G1 receipts; D2/D4 are provenance defects in reason text, corrected in new reports without editing frozen artifacts |
P1 equation/implementation |
defect found and fixed in the new namespace | D1 bootstrap collapse + D3 seeding; corrected implementations under hierarchical-aif/src/; the committed runner retains its defect and its results are withdrawn, not silently patched |
P2 observational |
unchanged | no new observation; raw archive not located |
P3 held-out predictive |
unchanged; B3 stands — and its INTERPRETATION is now measured | D1 does not touch the B3 leaderboard. The adverse M2-over-M3 result remains the retained headline. B4C01 (8256cb12…) measured what this design can resolve: under correct specification the two structurally distinctive generators self-win at 0.885 and 0.935, but M0_EXPONENTIAL — nested inside M1 and M5 and a degenerate limit of M3 — self-wins only 0.290, against a >0.5 criterion. The assay is nesting-blind at 19 holdout motors, not uniformly underpowered. An INCONCLUSIVE contrast between nested near-equivalent models is therefore the EXPECTED output of the design and is not evidence about the models. This constrains interpretation and raises no level. H-AIF-G7 added a candidate without moving the level: the F-side motor stack scored NOT_ESTABLISHED against CONTROL_CURRENT and every serious adversary, and reproduces the frozen M7 to 2.5e-7 nats. A NOT_ESTABLISHED verdict is not equivalence and does not raise P3 |
P4 transfer |
NOT_ESTABLISHED |
single study; no independent dataset |
P5 intervention |
NOT_ESTABLISHED |
unchanged |
P6 structural/mechanistic |
scoped — see the four statements below; there is no single P6 verdict |
An unscoped "P6 is weakened" is forbidden by CLAUDE.md and was replaced by the closure ledger. P6 is carried per scope, never in aggregate |
P7 independent replication |
NOT_ESTABLISHED |
external review in progress; not complete |
P8 full verdict |
FULL_PARITY = false |
unchanged. First unsatisfied level remains P4 transfer |
5a. P6 carried per scope (the required form)
An unscoped weakening statement is a contract violation: a local failure must never be reported as
a global retreat. P6 is therefore carried as four separate statements, each with its own receipt
and its own falsifier.
P6 scope |
status | receipt / blocker |
|---|---|---|
P6 for C11 U4 (M7 dispersion stability) |
RE-ESTABLISHED on the corrected full-N run only (U4_OK, collapse 0.0055, N=2000). The withdrawn 30-replicate artifact stays withdrawn |
B4C11_CORRECTED_FULL_RESULT.json (564a5b0f…), reports/B4C11-CORRECTED-FULL-REPORT.md; D1 CLOSED_BY_CORRECTED_RERUN |
P6 for Wadhwa mark-process mechanism |
RETROSPECTIVE-ONLY, TRANSFER-REQUIRED | D5 (holdout mark channel burned), D6 (nextStateN unchecked); needs an independent dataset |
P6 for duration-only B3/B4 |
UNCHANGED | D1 does not touch the B3 leaderboard; the adverse M2-over-M3 result stands |
P6 for full motor-stack AIF |
PENDING constrained implementation + scoring | H-AIF-G6 built / H-AIF-G7 scoring |
P6 for B4C10 / M4 identifiability is a separate scope again: B4C10 at full frozen N=2000
supports M4 identifiability on the derived_eligible_1_to_8 cohort under the frozen U2/U3/U4
criteria. Identifiability is not correctness and is not mechanism, and this result may not be
transferred to C11 (different model, different likelihood structure).
No P-level was raised by this round. P6 for C11 U4 was lowered (withdrawing defective
evidence) and then re-established on the corrected run only — the gate working end to end:
withdraw on defect, restore on a valid rerun, and never restore the defective artifact itself.
B4C10/B4C02/B4C01/B4C11 landing at full N removed four RESOURCE_BOUND/withdrawn states without
raising any level. Identifiability is not correctness and is not mechanism.
sha256 ef2daedc050e735c — of the original file, so what was ingested stays checkable.