[Gonzalo-Mier]

Change and mathematical scope

Adds two source-bound negative-result packages. Neither changes the selected κ of 0.001672660197076918 or claims a universal obstruction.

• Full-frame outputs: the direct P111 word passes local scalar and bridged-column checks but fails the actual interstage address frames. The complete 5/9/11-stage bills exceed one at zero saving, even with free junction repairs. A separate E8 rotating-normal experiment exhausts its deficit on literal terminal closure. The package explicitly withdraws its earlier local-only optimistic ceilings and identifies the missing idle mass.
• Owned-zero scratch: three clean-input implementations replace borrowed-helper restoration with nonzero disposal, late source forks, or a common-frame source decoder. All reproduce the exact target map; the best still costs 5637/4736 at zero saving under an optimistic reverse-call allowance. Its full idle work is included. This rejects these implementations in the retained schedule, not all source-consuming programs.

Task and overlap

Canonical main was freshly fetched at 20ecf2891d80b5acf18d980b2164fb44ea7faa90. Task bit-helper-contract, changed scope under the explicit donor request to explore mechanisms crossing κ=0.002. Exact inputs, original card and bounded stop rules are pinned in each package's SOURCES.json and TASK.json.

PR #3721 tests P111 scalar plane matching; this PR tests physical endpoint compatibility and complete idle accounting. PR #1862 supplies the distinct complex source-code precedent. The local clean-ancilla packages require return to zero; the new scratch experiment permits nonzero disposal. Discussion #1587 contains related endpoint work. No novelty claim follows from the overlap search.

Validation

On CPython 3.12.15, with C++17 and Boost headers, from an isolated checkout of canonical main:

python3.12 -B scripts/research_package.py --package research/full-frame-bit-output-probe
python3.12 -B scripts/research_package.py --package research/owned-zero-native-dag

Both registered replays passed, including optimized-Python checks. Evidence includes all 111 exact rational port-boundary checks, 444/888/1110 scalar bridge columns, exact rotating-normal closure, three 222-column scratch experiments, source-code inversion and the complete pinned idle schedule. Controls detect changed sources, omitted bridges, omitted target additions and dirty-helper misuse; omitting idle work produces the retained false-positive control.

These are finite checks and written scoped arguments. Full selected-supplier cold replay, all-size/native admission, hosting, precision transfer and Lean verification were not run or established.

Credit

Preserves eumemic, DreamingOfClouds, the retained carrier/checker and compressed-geometry lineage, and the E8 Gaussian geometry attribution. Ali Romman's source-code precedent remains credited. Prepared by Gonzalo Mier with substantial OpenAI Codex assistance; Apache-2.0. No independent human review is claimed.

Open next questions

Each package records the smallest discriminating follow-up: a joint source/consumer word with matched physical endpoints and positive complete deficit, or a two-call encoded-borrowed-state program with a priced terminal decoder. Both remain unconstructed. The named failed schedules should not be expanded on scalar success alone.