[DreamingOfClouds]

Adds research/p104-two-handoffs/: two successor hand-offs on PR5551's P104 word, conditional complete kappa .0030428242 (+1.29e-6 over #5551's .0030415312).

Conditional on (a) the five framework assumptions retained by #5544/#5545/#5551; (b) #5551's validity (unreproduced at the time of writing); (c) admissibility of a fresh helper starting at a non-canonical mid frame. For (c), "Variant A" (helpers start at frames 608/613) passes the unmodified PR5545 local.py; the canonical-start rule exists only in the #4610 compact-width verifier. No independent human or proof-kernel acceptance is claimed; the selected record is unchanged.

Mechanism. Helpers 473 (event 13725) and 531 (event 13941) end their first rank-1 climb at a dim-2 frame and then idle, with no other register holding their symbol. Following the successor hand-off of #5472, the rest of each helper's life moves to a fresh helper starting at that frame (8 -> 1+7, 7 -> 1+6). Pure rename: same 25,310 events, no new ADD, identical paid profile; productive fields 12,814 -> 12,850. By subadditivity the concave moment drops at all 18 calls.

Verify: research/p104-two-handoffs/verify.sh (stdlib, Python 3.12). It regenerates the word from #5551's pinned word, checks local.py/pricing.py are literal-only edits (735->737, 743->745, 12814->12850, kappa constants; see LITERAL-CHANGES.diff), replays scalar/native/physical caller, and checks all 54 strict + 4 functional outer checks and the exact crossing interval (76070606559/25000000000000, 304282426237/100000000000000), for both the hand-off word and Variant A.

Independent audit: an internal AI audit (separate read-only Claude agent, own builder and pipeline) reproduced the word byte-for-byte and the crossing; it rated mid-frame-start legality as unresolved, hence the conditional claim.

Builds on #5551 (thu-nmrc), the successor hand-off pattern of #5472 (GamingPuzzled), and the P104 word line of #5544/#5545 (aliromman; lineage PR2305/2152/2219).

Prepared by DreamingOfClouds with Claude Opus 5.5 (Anthropic), via Claude Code.

🤖 Generated with Claude Code

[DreamingOfClouds]

CI note: the plan failure was not this package. Current main's research/wordram-ntt-kernel/VERIFY.json declares the non-repo input "MANIFEST.json" (from #5466), which breaks test_research_matrix_batches for every new PR. Fixed separately in #5555; the same one-line fix is cherry-picked here so this PR's CI can run now. It will drop out cleanly once #5555 merges. No change to the claim or the package.

[DreamingOfClouds]

CI: plan and replay (research/p104-two-handoffs) pass. The remaining red job, replay (research/wordram-ntt-kernel), is selected only because this branch carries the one-line main fix from #5555; that package's own replay fails independently of this PR. It drops out once #5555 (or an equivalent fix) lands on main and the cherry-pick is removed.