[adamgall]

Result and scope

Independent exact finite audit of the complex k8 R480 scalar word selected at base 0b1cccf86e2f5132ed286a26ef0a681a495740ea. No current acceptance bug, new supplier, or new kappa. All changes are confined to research/selected-k8-omission-observability/; frozen sources and the selected frontier remain unchanged.

All 19,669 single ADD/SCALE omissions in each orientation change the full 704-register map. Forward target-only observation nevertheless misses 7,184 faults, observing both data banks misses 6,489, and clean-helper tests miss 8,977 even when every output is observed. 188 faults evade both latter shortcuts, in each orientation.

Two representative semantic mutants are executed against the unchanged selected standalone scalar auditor:

• Omit A[3]: only forward defect r311 += initial_r498; rejected at all-column endpoint failure (False, [311]).
• Omit cleanup[541]: only forward defect r680 += initial_r686; rejected at all-column endpoint failure (False, [680]).
• Original source passes all 704 formal columns.

Thus arbitrary dirty inputs and helper-output observation are material defenses, not redundant checks. The README traces this predicate through the slot driver and frontier1042 into selected frontier1094, including source-bound baseline reuse. This source trace is not an executed full cold caller replay.

Method, sources and novelty boundary

Sparse exact rational prefix rows and reverse suffix columns classify each rank-one omission defect. Compensation and early/full cleanup order are reconstructed from the pinned raw word. Six focused tests include all 125 three-event words over a rational operation alphabet in both orientations, compared with literal all-column mutant maps.

TASK.json, SOURCES.json, VERIFY.json and EXPECTED.json retain the assigned supplier-boundary-audit scope, Git/SHA256 source pins, selected-word binding, complete deterministic receipt, registered commands and open next question. Coordination: #1884.

PR1017/1102/1433 are explicitly credited for historical Boolean omission diagnostics; PR1583 audits selected helper endpoints while preserving scalar maps. This result instead concerns the selected rational ADD/SCALE program, both orientations, and the intersection of output-observation and clean-helper blind spots. Their full bodies were read and exact heads rechecked live. Cached overlap searches covered 543 PRs and 529 discussions with complete pagination but may miss later changes; they do not establish exhaustive novelty.

Observed verification

macOS arm64, CPython 3.12.12, standard library only:

uv run --python 3.12 --no-project scripts/research_package.py --package research/selected-k8-omission-observability
uv run --python 3.12 --no-project python -I -O -B research/selected-k8-omission-observability/verify.py

Final pre-publication registered replay and separate optimized smoke both passed, 5.37 seconds combined. Registered replay includes source/provenance metadata, all six tests, exact receipt comparison, the positive source control and both semantic rejection controls.

All 39,338 omission positions are classified by exact factors; representative mutants are literally executed. This does not claim a separate complete 704-column run for every mutant, syndrome uniqueness, multi-fault detection, native phase/frame or owner admission, hosted-CI success, full selected cold replay, compiler/tape/analytic obligations, or a complete multiplication theorem. No speedup claim.

Credit and next question

Credit Rohan Arun's PR1094 placement, Jeremy Kothe's k8/leaver/slot lineage and PR1087, Hao Yan's T16 controller, icekylinx's scalar auditor, Josie Elliston's PR1091 word pin, Sagar Thakrar's PR1049 and PR1433 work, woklerx's omission diagnostics, and retained supplier lineage. Prepared for Adam Gall with substantial OpenAI Codex assistance in scoping, implementation, experiments and writing; no independent human-review claim. Apache-2.0.

Open next question: find a small, bounded-height rational probe set detecting every single omission, then literally replay every mutant; no optimal probe set or practical advantage is claimed here, and such diagnostics must not replace complete admission.

[Mister-Meeseeks]

Batch review at 7e2aebdf2ae24cc3613731a95d71d2c46ac8173b.

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

Reviewed scope: Independent rational prefix/suffix census covers39338 forward/dual scalar omissions;188 defects evade both data-only and clean-helper shortcuts. Two semantic mutants reject in the unchanged complete704-column auditor. Six tests and caller source trace pass; no current acceptance defect or native/cold theorem replay claimed.

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.