[petalas]

Summary

PR3446 leaves 696 outgoing ADD effects on 93 omitted P104 banks outside its 4,027-event projection. This computes the complete normalized endpoint difference from the original PR2586 word, including every omitted event, and proves its rank is exactly eight.

F = full normalized word; S = terminal-history projection
F = (I + K) S
K has a 52-X-output / 104-Y-input nonzero block: R = C A / 2

for each of eight linear queries, using one arbitrary dirty helper h:
  X -= C[:,i] h
  h += A[i,:] Y / 2
  X += C[:,i] h
  h -= A[i,:] Y / 2  # restores the original dirty value

The explicit integer matrices and minor determinant -1/128 certify all 5,408 matrix entries and the sharp query dimension. The completion alone preserves a Gaussian module M for Y in 2M; the predecessor's sufficient 4M Y entrance meets it. Eight queries do not require eight simultaneous helpers.

All 93 omitted final frame changes remain unfilled, including the 41 zero normalized residual rows. Retained Y outputs have dimension eight; normalized readouts are not free same-frame physical ADDs. Native covariance is inherited. Physical adapters, owners/routes, scalar expansion, precision and the complete recursive bill remain open. No native supplier, improved kappa or selected-frontier change is claimed. The unchanged-profile stop remains.

Credit Hiroki Kumazaki/PR3446 for the projection and released handoff, Clear1nsight/PR3423/3382 for its open-port predecessors, ywz65/PR2586/2497 for the physical/logical word, Arham Tahir for retained source bytes, and the native/geometry/compiler lineage detailed in the package. Runtime uses only the two unchanged inputs already on canonical main. Base 6f100d4bb5083f8c35d96fd95a539811c62f245c, priority revision 6. Canonical investigation.

Evidence

• Before: a literal scalar fixture on the sufficient dyadic grid distinguishes the full word from the projection on an omitted X bank; dropping dirty compensation also changes the output.
  After: the one-helper completion matches all 358 endpoint payloads and restores the helper. Six full-map fixtures and every Y source column at three dirty values pass.
• Exact source replay reconstructs 675 ports, 4,027 events and 265 inputs, matching PR3446's event hash. Separate reverse covectors verify all 716 full/projection output rows. Every original private erasure and all native occurrence frame/rank bindings pass.
• Normal/optimized registered Python 3.12 replay, six behavioral test groups, source/certificate drift rejection through the actual optimized Git-free CLI, scoped Ruff and whitespace checks pass.
• Exact committed-byte and strict source-history checks pass. The committed source-only export passes three standalone commands with Git unavailable; every exported byte, mode and inode remains unchanged. Independent human/formal review and hosted acceptance are unclaimed.

python3.12 -B scripts/research_package.py --package research/p104-omitted-completion
python3.12 -B scripts/check_research_commit.py --package research/p104-omitted-completion --check-source-history

The package contains the proof, complete coefficient matrices, source pins, rejection controls and next caller/cost questions. Submission refresh found unchanged canonical guidance and predecessor head; 374 open PRs and 2,299 Discussions were fully paginated with no warnings. Sparse omitted notes, unpublished work and all historical diffs are outside the overlap coverage. The selected orientation command was attempted but has an omitted sparse dependency; no selected cold replay is claimed.

Merge Danger

Door: two-way

Blast Radius: research

Eight additions in one isolated package. Frozen inputs, selected frontier and shared execution code are unchanged. The main review boundary is the conditional normalized interpretation versus the explicitly unfilled physical frame/cost contract.