docs: P104 side-correction scope map (results by premise, shared 328/9 budget line; no new kappa)

[navinor]

What this adds

This adds a single file, docs/research/p104-side-correction-scope-map.md. It is a navigation map for the revision-4 p104-side-correction card, in the style of docs/research/complex-decoder-scope-map.md. No new mathematics, no review dispositions, no κ or selected-frontier change.

Fourteen PRs and threads now work on the same I-G/2 completion of the native P104 channel, many from the last two hours. The map pins each to its exact head and records three things: which premise it fixes (cut, transports kept, storage, frame paths, producer, precision), what it concludes, and what it explicitly leaves outside.

Sources covered:

• open PRs Prove sharp P104 temporal rank bound and correction-budget obstruction #2082, Prove a terminal-cut paid-rank floor for the P104 side correction #2083, Research exact local P104 I-G/2 tail correction #2089, Sharp eight-query temporal cut for the frozen P104 itinerary #2090, Prove the native P104 channel's all-cut precision interface #2096, Prove P104 rank-eight cut under arbitrary frame-path interleaving #2113 and Prove P104 transport floor with arbitrary added storage #2127;
• the channel in Draft construction: P104 dirty-restored gather and low-rank channel with COPY56 (no new kappa) #1972 and Native Clifford realization of the P104 low-rank channel (no new kappa) #1991;
• discussions RELEASED — P104 terminal correction overlaps #1167 #2080, FINDING: exact local P104 I-G/2 tail correction (Cycle 16) #2092, Cycle 17 P104 correction with paired dirty-query baseline cancellation #2107, FINDING / HANDOFF: delayed-query P104 side word and its cost limit #2121 and CLAIM: P104 side correction with single-room mixed carriers and changed frame paths #2131.

It follows revision 4's request to "summarize which common premise has been tested and what the next search would change".

The one derived observation

PR 2082 and PR 2083 price completions under the same conditional coverage contract (s = N = 9, m = 73, W = 832, unchanged idle requirements). Under it, a correction with no creditable progress keeps the zero-saving moment at or below one only if it adds at most 328/9 ≈ 36.4 rank units per call. PR 2082 states the same slack. Every route covered so far is either infeasible (the 8-array cuts) or needs at least 96 added transports. The constructions and candidates pay 208.

So within the shared premises the route is closed. Only two kinds of change remain open:

1. replacing or fusing original transports, or adding new frame paths (live claims in FINDING / HANDOFF: delayed-query P104 side word and its cost limit #2121 and CLAIM: P104 side correction with single-room mixed carriers and changed frame paths #2131);
2. a coverage interface that credits added transports.

The map also notes that closing the 96..208 gap doesn't matter under the retained contract.

Checks

• The doc's Python snippet reproduces the cited moments exactly with fractions: 7659/7592 at 96, 1917/1898 at 104, 7785/7592 at 208, 7551/7592 uncorrected, and the 328/9 slack. Constants are taken from the PR bodies, not from fresh package replays.
• Both relative links resolve. There are no bare PR references.
• python3.12 -B -m unittest discover -s tests -p 'test_research_*.py': 65 tests OK (3 skipped).
• All nine PR heads were re-checked unchanged at 20:25 UTC. Canonical main is still 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2 (priority revision 4).
• The overlap refresh (389 open PRs, 796 discussion records, complete pagination) found no other scope map. One P104 claim appeared at 20:17 UTC (CLAIM: P104 side correction with single-room mixed carriers and changed frame paths #2131) and is included.

Not checked: I did not replay any cited package, so the open-PR entries are their authors' claims as stated. The map is a dated snapshot and will go stale as these PRs change; each entry gives its head for that reason.

Task

• Drawn: complex-mechanisms, card p104-side-correction, chooser nonce babf7d7d03bf6b191d708ecb1d52d2aa, base 0fdd7d82.
• Scope changed to consolidation, because every concrete sub-question was already claimed or in an open PR.
• Claim: Handoffs and requests for independent checks #1166 (comment).

Credit

All results belong to their authors: Rohan Arun, Chafik Boukhalfa, Adam Gall, David Arias, Jeremy Diviney, Ross Richardson, 1337isnot1337, Jose Manuel and the credited lineage. The coverage contract sources are Dominik Scholz's paired-chain model, Sagar Thakrar (#1012) and Douglas Colkitt (#1240). Authors: corrections to how I summarized your scope are welcome.

Prepared by Aaron Wang / navinor, with help from Claude Code. Apache-2.0. Not requesting merge; selected frontier unchanged.

[Tairitsu-Aya]

One new scoped boundary for this map: PR2136, exact head 9b8689e0153e038b2c1179b06026adfaa7f7593b, takes #2113's intermediate-frame handoff. It allows every old transport occurrence to be subdivided through arbitrary nested intermediate binary charts, with arbitrary interleaving, same-chart linear resynthesis and added source-independent storage. Identity remains impossible: a new separator bounds the 56 outside-span source columns by eight arrays, hence total transfer rank by 56 arrays. This upper bound is not claimed sharp.

This narrows one part of “replacing transports”: nested subdivision inside the original intervals is excluded. Replacement routes outside those intervals, fusion, additional cross-chart transfers and the changed producer/paths in Discussions2121/2131 remain open. It is not a new native cost floor and does not invoke the shared 328/9 slack to turn step counts into excess cost. Two code-disjoint exact checks and a written transcript proof are in the package; the canonical investigation is this released claim on #2113. Substantial Codex assistance, no independent human review.