- five-component status table (1/2/3/5-contract closed both signatures,
4 seeded, liris readback in progress)
- documents the bilateral method: two machines, two different AI agents,
GitHub as the only channel, operator-driven copy-paste rehydration,
seed/attack alternation, sha256 sidecar lattice, parity baselines
re-proven on every test run, 11 cross-vantage catches in 28 commits
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- tools/behcs/nnest-watcher-gate.mjs: CHILD_MAY_ACT requires BOTH
reported==recomputed at every level AND a complete, contiguous, unique,
ordered chain from depth 0 to max_depth (operator amendment 2026-06-11)
- truncation, duplicate, skipped, and reordered depths all return
CHILD_HELD gate=chain-topology-invalid with first_bad_depth - matching
hashes are necessary but NOT sufficient; topology checks run BEFORE
hash comparison by design
- divergence caught at the exact level (tested at every k in 0..5, the
planted-confab N-Nest-Prime behavior); depth bound 16 = the proven depth
- consent anchors at apex-T0 only: act-on-world/spawn-child/modify-self
DEFER_TO_OPERATOR even on a perfect chain
- HELD rows leak only first_bad_depth + chain_sha16; level hashes never
echoed; executable=0 every row; no spawn, no recursion into real
agents, no harness edits
- sealed 16-case time-free parity baseline; pyramid 103/103 on acer
- the LAST open vision component is now seeded - liri…
- tools/behcs/token-cube-catalog-binder.mjs: deterministic time-free
contract binding crypto/hash token REFERENCES (token_id, kind,
digest_sha16, scope, catalog) to vantage-qualified cube BH addresses
- HARD RULES: tokens are references, never material - material-shaped
rejects (32+ hex, key-block markers, bearer shapes, base64 runs) are
REDACTED from rows, not echoed; digest exactly 16 lowercase hex kills
length spoofing; every row executable=0
- BH addresses MUST be vantage-qualified (BH-ACER-942, never BH-942) per
hilbert-collision canon; disputed band 930-1229 DEFERS to operator
regardless of vantage or scope (bilateral-ack pending)
- mint/write scopes and mode=live cap at DEFER_TO_OPERATOR; 12-rung ladder
- registries (6 kinds, 6 catalogs) declared DRAFT_ASSUMPTION; exact-match
case-sensitive - hilbert-omni-47D keeps its canonical capital D
- self-correction in receipt: acer's own test wrongly claimed all-lowercase
registries; fixed the test to exact-match semant…
- tools/behcs/watcher-supervisor-suggestion-emitter.mjs: deterministic
WATCHSUGGEST row contract, repo-side draft ONLY - never touches the live
/hbp/supervisors surface, office-705 feed, fabric bus, or any runtime
watcher; live wiring is a separate gated follow-up
- HARD INVARIANT: executable=0 on every row, all verdicts (closed set:
DRAFT_SUGGESTION_READY / DEFER_TO_OPERATOR / DRAFT_SUGGESTION_BLOCKED)
- 10-rung ladder: dirty/unknown-watcher/supervisor/action/reason/missing-
evidence/invalid-confidence/malformed-future-stale-ts all BLOCK;
live-fabric actions (dispatch-agent, restart-daemon, request-cosign-mint)
cap at DEFER_TO_OPERATOR regardless of input quality
- closed-set registries: raw input never reaches a row (spoof-proof);
evidence referenced by sha16, never inlined; confidence integer 0-100,
bands tile 0-33/34-66/67-100 with no gap or overlap
- route via component-1 resolver as pure import: tightest when proven,
broadest on conflict, none without tuple
- …