Research: observer-safe restoration with nonzero readers (no new kappa)

[adamgall]

Status: recursive-interface research construction/correction; no new kappa

PR1155's frozen restoration-coalescing note says read-freedom is necessary for deferred dirty-workspace restoration. Its bad-reader example establishes that unrestricted deferral can fail, not that every reader fails. This independent package supplies the exact criterion in its stated additive model:

sum_i B_i (G_{i-1}...G_1 - I) = 0.

The written proof covers arbitrary dimensions and fields, while the executable evidence is explicitly finite. Prefix observations require each prefix defect to vanish; terminal cancellation alone is not safe for adaptive observations.

• Nonzero invariant-reader shear batches permit genuine restoration elimination. The seed-selected five-child family pays 15N -> 11N scalar ADDs per node, including readers, forward drift and final restoration.
• A second example reads the changing coordinate and cancels its two observable defects only at the endpoint, separating terminal and prefix contracts.
• Complete recursion, scalar/memory-access invoices, child calls, descriptor/controller charges, integer magnitude bounds and storage limits are stated. An actual recursive event emitter is replayed forward and inverse on every dirty-state basis column in its bounded domain.
• Independent block-state multiplication checks the defect formula on all 46,656 three-child signed-reader/shear tuples. Noncommuting rational evidence and six mathematical controls are included, plus standalone relocation and a real source-corruption rejection.

Reproduce

python3 -I -O -B research/observer-safe-restoration/test_package.py

Expected final marker: PASS_OBSERVER_RESTORATION_PACKAGE relocation=1 source_corruption=1.

Base 1b59b624d21924f8e3537cdc1ac52d3be6eda2fe; OS assignment seed 52. PR1155 source pinned at 9821c6e9388f3fc097e94b72be9fa7b41a849527, with its complete README preserved and authenticated. The selected PR1094 construction, upstream/, and all existing supplier code are untouched. No selected cold replay, hosted supplier admission, compiler/tape realization, remote CI, full formal proof, or stronger multiplication exponent is claimed.

Credits Aaron Wang (@navinor), PR1155, and its retained restoration-interface predecessors. Distinct from PR889 common-factor query batching, PR1105 projected target-feedback cleanup, PR1012 hosted restoration mass, and PR1135 rounded borrowing. Fresh overlap metadata and the local claim are included; no pinned coordination issue was present. Prepared by Adam Gall (@adamgall), with substantial OpenAI Codex assistance. Apache-2.0; existing license and attribution retained.

[Mister-Meeseeks]

100-PR batch review at e2008d7ffcad464da7ec21c69e8d8a8c91de4930.

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

Reviewed scope: Observer-safe restoration criterion and nonzero-reader constructions with prefix versus terminal observation contracts; no outer recurrence admission.

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.