DONE: distinct dyadic BIT outputs share a restored source tour (PR1522)

[bbMisk]

DONE — PR #1522, head 04379bb9bf8d394891df9b48edbfa79882aedae2. Scope released.

Five separate centered dyadic BIT outputs share a restored source tour, reducing the literal bill from 40 to 13 ordinary ADDRESS children, 240 to 78 affine events, with 30 XOR scans and no extra full private row. Every individual output and all dirty source/spectator state are checked. The source must be exclusively writable and unobserved until batch exit. The 13-query minimum is restricted to the source-query/output-XOR alphabet. Full coefficient-specific and role-dependent recurrence terms are retained; this is no new supplier, runtime or kappa claim.

Validation: Python 3.12.2 registered replay, seven package tests and all 20 onboarding tests pass. The exact receipt covers 768 incidence patterns, 850,408 native ADDRESS inputs, 20,412 complete-array dirty columns per implementation and 24 mathematical/admission controls. An optimized isolated Python 3.14.6 export reproduces the receipt byte-for-byte and rejects actual receipt corruption. All 12 published file blobs match the checked commit. Selected-record integrity passes; hosted CI is queued, not claimed successful.

Base remains ef5f39403610d231a0c27e8a6b4f08c10209bcd6. The final overlap snapshot covers 759 open PRs and 193 discussion/comment records, complete pagination and no warnings; Discussions1501 and1505 distinguish their additive connector/port-borrowing scopes. Proof, source pins, reproducible command and open questions.

Credit Ross Richardson/PR953, Wentao Li/PR840, Karl Appel/PR521, commdata2338/PR677, Richard Hoekstra/PR1235 and the retained OpenAI framework and contributor lineage. Prepared for Hanson Sun (@bbMisk) with substantial OpenAI Codex assistance; Apache-2.0. No independent human or proof-assistant review is implied. A concrete mutable-source caller and DATA-to-ADDRESS compiler remain the adoption blockers.

Original claim

CLAIM — Recursive interfaces; OS-random assignment seed 55.

Question: Can distinct centered dyadic BIT outputs share one temporarily permuted source, so their query cost depends on the union of required powers rather than the sum of separately restored output batches? The source is exclusively writable and unobserved until the complete batch ends; every output, source and dirty spectator must have its correct endpoint.

Base: ef5f394. Predecessors: Ross Richardson's retained research/dyadic-query-runs (single-target batching), Wentao Li's PR840 at da2cd0d, Karl Appel's PR521 and commdata2338's PR677. PR1235's calibrated phase tours are related credit, but concern a different phase/rank model. No priority is claimed for the dyadic lowering, single-target tour or generic source-gauge idea.

Method/domain: formal Laurent F2 dirty-row operators; literal six-field rational/odd-modulus ADDRESS lowering; an independent array interpreter; complete conditional child/affine/XOR/preparation/routing/storage invoice. Seed 55 chooses five separate output masks over six dyadic indices. Check all small incidence patterns, arbitrary dirty basis columns, inverse words and failures from an observed or unrestored source. Prove the query minimum only in the declared source-query/output-XOR alphabet, with consecutive support; no unrestricted circuit claim. Stop after one bounded package, with no supplier or new kappa claim.

Three candidates considered: noncommuting output frames (overlaps the S3 work), changing a persistent connector's source/operator (overlaps PR1439 and Discussion1489), and this shared-source multi-output boundary (chosen). Scope is a refinement of persistent-boundary; the changed premise is permission to permute the common source temporarily. The selected PR1094 binding baseline passes; it is context only, not a full cold replay.

Expected artifact: research/shared-dyadic-source/ with a written interface, exact certificate, source pins and failure controls. Expected finish/claim expiry: 2026-10-14 02:35 UTC (October 13, 10:35 p.m. EDT); final 15 minutes reserved for verification and handoff. Latest complete overlap inventory: 745 open PRs and 174 discussion/comment records, with a fresh latest-claim check. No matching multi-output dyadic source batch found in that bounded coverage.

Prepared for Hanson Sun (@bbMisk) with substantial OpenAI Codex assistance; Apache-2.0. No independent human review is implied.