research: price full P104 end-cut correction

[1337isnot1337]

Summary

Adds one exact full-frame end-cut I-G/2 correction for the 104 source and 104
target endpoints represented by PR #2981's 208 static flags. At the cut after
the pinned 4,058-event channel, every source is in F9, and every target is in
its final eight-dimensional output frame. For each target, the word raises it
to F9, applies its 48 exact coefficients, and returns it to the original
frame.

The correction has 4,992 scalar additions: 3,376 coefficients -1/2 and 1,616
coefficients +1/2. Exact rational composition gives Y_t = y_t + x_t for all
104 targets; all sources and 72 arbitrary dirty helpers are restored. The 208
new target transports are individually compiled as rank-one native recipes
and checked for signed Pauli action and global phase.

Finite invoice

• Base channel: 4,058 events, 2,624 scalar additions, 1,410 transports, rank
  mass 2,368.
• Correction suffix: 4,992 scalar additions and 208 H1 transports, with no
  new dirty or private register.
• Combined word: 9,258 events, 7,616 scalar additions, 1,618 transports, rank
  mass 2,576. The paid profile changes by +208 H1.
• Under the documented endpoint frame-coordinate convention, peak width is
  1,769 and the 5,200-event suffix has 9,198,800 frame-coordinate event-holds.

The last quantity is a finite scheduling count; it is not presented as the
selected recursive owner/idle tariff. No supplier admission or new κ is
claimed.

Scope relative to existing P104 work

PR #3618 already supplies a 48-add correction for flag 9, row 23, at its
earlier event-3489 cut. This contribution extends the direct word to all 104
target rows at the later post-channel cut. It uses paid target-frame lifts and
descents, so it changes the monotone equal-frame contract used by the fixed
cover support cut reviewed in PR #3810. The 208 internal bases and 1,872-step
static screen from PR #2981 are not treated as an executable program.

The exact selected recursive owner/idle charge, consumer composition,
precision, and all-size lifting remain open. No selected frontier or κ changes.

Reproduction

python3 -B scripts/research_package.py --package research/p104-fullframe-endcut-correction --check-only
python3 scripts/research_package.py --package research/p104-fullframe-endcut-correction

Both commands passed locally on the pinned base. The package pins all live
inputs and retains the exact 4,992-entry correction and 208 transport recipes.

Credit and review limits

Credits: Rohan Arun / PR #1972 for the scalar/frame channel; Chafik Boukhalfa /
PR #1991 for its concrete native realization; Nick Petalas / PR #2981 and the
PR #2920 endpoint geometry; and the related flag-specific work in PRs #3614,
#3618, #3601, and #3610. This package was produced with Codex assistance. No
independent human or formal-kernel review is claimed.

Current overlap

Open PR #3882 packages a source-consuming correction tail with eight residual replacements. This package contributes the explicit 104-target full-frame I-G/2 endpoint word and its 4,992-addition, 208-transport invoice. These are distinct finite methods; neither establishes physical supplier admission.

Publication validation

python3.12 scripts/research_package.py --package research/p104-fullframe-endcut-correction passed on CPython 3.12.3, and python3.12 scripts/check_research_commit.py --package research/p104-fullframe-endcut-correction --check-source-history passed. The registered replay left the worktree clean. This supersedes any earlier pending note above for the registered package replay; separate full selected cold-replay limits remain as stated.