Research: optimize observable connector repair across source changes (no new kappa)

[bbMisk]

A common-B connector spanning a restored source-mutation tour can repair only the component visible to each current reader. The fixed seed-50 instance needs four connector corrections instead of fourteen for full coherence, or three corrections when one paid output repair is allowed. Including all source updates, both B calls, both reads per output and terminal restoration, the complete word falls 72 to 61 XORs.

Status: scoped recursive-interface construction, with no new kappa or selected-supplier adoption.

The package adds a written observation contract and complete cost recurrence, exact mask planner, separate shortest-path oracle with locally checkable lower-bound potentials, literal words and tape accounting. The three-correction minimum is sharp in the declared correction alphabet. The same path minimizes the stated correction tariff for every positive block length. Opening and fusing the toy matrices costs only 26 XORs; this explicit control excludes an unrestricted circuit-optimality claim.

Reproduce with Python 3.12 and standard library only:

python3.12 scripts/research_package.py --package research/observable-connector-repair

Expected: PASS_OBSERVABLE_CONNECTOR strict=4 mixed=3 coherent=14 complete_xors=72_to_61 small_cases=6220 controls=14 no_new_kappa, five passing package tests, then PASS_REGISTERED_RESEARCH_PACKAGE.

Local validation passed on Python 3.12.13, with optimized direct replay on 3.14.6. Coverage includes 6,220 complete small demand/pricing cases, 648 literal forward/inverse basis-column cases, all 256 dirty inputs of a smaller batch, 5,696 observation checks, fourteen semantic/domain controls, source corruption rejection and read-only relocation. The selected-record integrity check passes unchanged. Physical hosting, instruction-tape admission, a contracting input-size recurrence, selected cold replay, formalization and independent human review remain separate.

Base and refreshed main: ef5f39403610d231a0c27e8a6b4f08c10209bcd6. Assignment seed 50; recursive interfaces. Claim. Final overlap inventory covers 762 open PRs and 196 discussion records; limits and exact heads are retained in OVERLAP.json.

Credit woklerx #1439 for full-coherence source repair, mightyOil #1203 for reader kernels, Hanson Sun #1246 for paid output defects, Sagar Thakrar #1199/#1227, and Ross Richardson's universal-connector lineage. The three closest predecessor notes are pinned and retained unchanged. Concurrent #1506 changes B and source over characteristic zero while retaining full coherence; this package keeps B fixed and optimizes weaker observable intermediate states. No priority is claimed for the predecessor identities or dynamic programming.

Hanson Sun (@bbMisk), with substantial OpenAI Codex assistance in derivation, both implementations, verification and prose. Apache-2.0. Separate implementations were authored in the same AI-assisted session.

[Mister-Meeseeks]

Batch review at 6fe9310331e942a9879135ea8db07f7f75fce23b.

Merged to main, preserving the original contributor commits and package attribution.

Reviewed scope: Observable connector corrections have an exact finite-state tariff and fully restored dirty pilot, with paid output-defect correction and observation cuts. Independent shortest-path potentials and full columns pass; optimum is only within the fixed-query interface and a direct circuit is explicitly cheaper.

The pinned scoped command passed; the batch ledger contains its command, input hashes, environment and log. This incorporates the research/tooling contribution and does not change the selected κ or discharge the inherited multiplication-theorem assumptions.

Batch review, attribution and evidence. Thank you for the contribution.