Research construction: paid nonzero reader-defect repair (no new kappa)

[bbMisk]

Fixed signed-permutation reader calls can defer restoration even when their reader defects do not cancel, provided the output correction is explicitly implemented and charged. This scoped recursive-interface construction reduces the complete seed-selected dirty echo from 380 to 244 ADD/NEG instructions. Both source-copy passes, all recursive children and inverses, correction and scratch restoration are included. No new kappa is claimed.

Only research/paid-reader-defect-seed58/ is added. It contains the written proof, a partition planner, an independent dense-state/cycle-count oracle, exact certificate, source pins and a standalone package replay. The planner minimizes over canonical repaired partitions, including required observation cuts; it does not optimize arbitrary linear circuits. At depth three the emitted count is 41,630 versus 26,534. Repair raises the exact intermediate magnitude envelope, which is recorded and discussed.

The fixed-child restriction is essential: opening these toy leaves permits direct evaluation in five ADDs. No real supplier with the required indivisible boundary, physical routing, fixed-precision admission or improved multiplication runtime is established.

Reproduce from the repository root with the Python standard library:

python3 -I -O -B research/paid-reader-defect-seed58/test_package.py

Expected: PASS_PAID_READER_DEFECT batches=5256 seeded=160 echo=380->244 controls=9 kappa=unchanged, followed by PASS_PAID_READER_PACKAGE relocation=1 corruptions=5.

The full relocated package passes on CPython 3.14.6; the exact verifier also passes on 3.11.5. Evidence covers 5,256 exhaustive small batches, all 10,440 partitions per policy, 160 seeded noncommuting cases, 120 generic recursive cases plus nine selected cases, literal inverses, arbitrary dirty basis columns, nine mathematical controls, eight invalid inputs and five on-disk corruptions. The selected source-binding baseline passes; no selected cold replay, hosted CI or full formal theorem is claimed.

Base: 49b3157d1d89adcef16df8be98445451672be5ba. Assignment seed: 58, recursive interfaces. Refreshed main dd6403f774a0891093a006ae9716c02e371b1fdb retains all five pinned selected-context files. The overlap record covers 843 open PRs and nine discussions. No pinned coordination issue existed; the claim is local.

Credit Aaron Wang/Navinor (#1155), Sagar Thakrar (#1199/#1227), mightyOil (#1203), Adam Gall (#1205), Ross Richardson/CubeJerry (#889), and the retained selected construction's contributors. The predecessor defect identity, cancelling-reader rules, sign frames and dirty echo are not claimed as new. This adds paid nonzero output repair and its checked partition recurrence. Three predecessor notes are retained unchanged with their notices and exact pins.

Hanson Sun (@bbMisk), with substantial OpenAI Codex assistance in algebra, implementation, verification and writing. Apache-2.0. The selected frontier, upstream inputs and existing packages are unchanged.

[Mister-Meeseeks]

100-PR batch review at 5ed1ce17a9df4c853a93f9758ba903d2620c3443.

Merged to main, preserving the original contributor commits and package attribution.

Reviewed scope: Paid output-defect repair and exact partition planner for indivisible drifting readers, with complete dirty echo, recursive instruction invoice and observation cuts; no supplier or tape-speedup claim.

The pinned scoped command passed; the batch ledger contains its command, input hashes, environment and log. This incorporates the research/tooling contribution and does not change the selected κ or discharge the inherited multiplication-theorem assumptions.

Batch review, attribution and evidence. Thank you for the contribution.