[aliromman]

This package gives a local nonlinear centre reset on a declared physical dyadic input grid and a conditional cost theorem for every policy in a fixed pure-reset family. The local word returns (X-Y,Y), restores all 480 helper contents from those grid inputs, clears one owned private field, and adds 2,361 field passes with no added native child. It does not improve the complete kappa calculation. Read the proof, independent publication review, and source pins.

The version 2 local proof has a sufficient fine-accuracy profile of 124 rank-one and eight rank-four children, with mass 156. It keeps a separate absolute error budget for parent return rounding. A relative child error bound alone fails for a half-parent-grid output. Exact helper restoration applies after entry to the declared grid; it does not repair finer caller records.

The complete policy theorem covers all 2^36 effective policies through 12,288 exact group cases and all 255 nonempty query subsets. The all-query optimum has 121 rank-one and eight rank-four children, with mass 153. For every nonempty subset, the added rank moment exceeds the removed centre moment by at least 17 for 0 <= sigma <= 1; 17 is sharp. The class permits individual ideal pure-target resets after existing events before the fixed cut, with free reset cost. It keeps the chronology, basis, access edges, and independent relative-error width model. It is not a universal precision lower bound.

The publication base is b58ad9ef4d4e861b0eaa5dba41742bdc072dca51; its selected record identifies PR #4470. This package does not prove admission to that changed construction. The historical native lineage is pinned at bce54fb3f895cbe5cd4c0eb477c3afeed100190f; the historical precision source revision is 3ae9a6f3ac74ded022ab3b4872a7357a621af6fb. The notice preserves gauge44/t16 contributor credit and licenses. History-encoding supplied the cut response, circuit-interface supplied the matrix, schur-selection supplied the grid word, temporal-reuse supplied the policy theorem, and the separate reviews supplied the physical correction, return-grid correction, and sharp surplus. The retained source credits @icekylinx for the gauge and pipeline, @FanPUYen for t16, @jeremykothe2 for the retained local word, @chafreaky for the frame and owner constructions, and @rohanarun for the native recurrence correction. Substantial OpenAI Codex assistance was used; no independent human review is claimed.

Use CPython 3.10 or later, the standard library, one process, and at least 3 GiB of free space. From the repository root:

python3 -I -O -B research/finite-tagged-centre-reset/replay.py --out /tmp/finite-tagged-centre-reset-check

The output directory must not exist. The expected final status is:

PASS_FINITE_TAGGED_CENTRE_RESET_PACKAGE

Both unchanged independent checkers passed from a fresh relocated package, with complete receipt equality. Checks cover all 704 input columns, all 480 helpers, 11,025 decoder cases, 132 local native-error impulses, every effective policy, and 528 policy-error impulses. Additional independent controls reject zeroing nonzero helper baselines and verify the exact margins and parent return counterexample. A later input-grid clarification changed only the new proof; the final hash check binds the accepted package to that replay.

The full native phase and historical record/tape audits remain inherited and were not rerun. Finer-caller entry conversion, complete node endpoint and owner admission, and the finite bit and tape invoice remain open. No complete supplier or frontier update is claimed.