[rohanarun]

This 17-stage construction gives conditional κ = .00363797977, up .000000562650059 from #7112. The exact raw bracket is (.00363797978, .00363797979). The selected record is .00363005817.

• Exact Conditional κ = 0.003637417119941 (+9.6926221e-7 over #7110) #7112 native canonical SHA256 3454c4900245ed1814ceeff79a741a45160eb3a93cf8d15f891fe2a67ba036fc.
• Asymmetric signed caller emits 4,783 events and replays exactly on all 72 output fields.
• Localized grouped idles pass an independent integer-grid replay with charged caller scale 675, separate from native scale 1125. Physical ownership, phase, precision and all 58 outer checks pass locally.
• The unchanged native's 3,047 exact paid/helper chart maps are reused at the same bytes. The complete caller has 110,638 paid positions and 22,172 productive fields.

python3.12 -B scripts/research_package.py --package research/signed-caller-8x9

The bound uses the same five framework assumptions and completed BIT predecessor. The selected record is outside this package's scope. Credit and full evidence are in the package.

Verification

• Fresh Linux CPython 3.12.16 replay regenerated the package receipts. A separate fresh run checked all 3,047 native paid/helper maps and all 104 main boundary pairs, including absolute-phase rejection controls, and matched the frozen receipts.
• Registered signed-caller replay passed at exact head 07a76c5e2738882c0cb322278e56f6cae3c9885d with PASS_SIGNED_CALLER_8X9 363797977/100000000000.
• The CI guide update also selected the historical selected-replay-gap audit. Its preflight reports working source drift: Makefile. The unchanged base Makefile already differs from that historical package’s pin.

The replay covers the source-bound conditional application described in the package. Maintainer review and the five inherited assumptions remain.