[flyingviper]

An active injection into the retained L240 paid-query cut can fit a desired main coefficient while losing an independent helper output. This package tracks both obligations and supplies exact rational certificates for the resulting observation-space obstruction.

The source-bound replay checks all 632 original input columns, including 408 arbitrary helpers. For the granted injection immediately before the rank-one climb at word event 17978, all 4,868 chronological full-frame observations still have rank 482 after adjoining the injection-response column. Direct witnesses separate both the missing mode and original helper240. A scoped proof extends the observation limitation to fixed-mode changes that preserve the verified full-frame sink interface.

Direct correction remains possible in principle: a second exact replay restores the helpers and repairs one mode if a specified payload can be supplied at frame 280. Its currently proposed ground-to-frame-280 ADD fails literal coframing. This is an explicit unresolved physical transport requirement, not a legal replacement or new bound. The package does not rule out general active feedback, change the selected frontier, or claim compiler/caller admission.

The contribution retains the original archive unchanged, pins canonical main df0137efd35e11b4445734bf612bdd4c7ebc330c, and distinguishes this changed-history restoration test from the retained passive observer and source-DAG work. README and TASK include concrete next questions. Prepared for flyingviper with substantial OpenAI Codex assistance and separate agent review; predecessor attribution is retained.

Validation:

• Registered source-bound replay passes normally and with -O, including meaningful failure controls and an independent direct witness check.
• Committed-byte and source-history preflight passes.
• Committed export integrity and Git-free, read-only replay pass with empty PATH; all exported files remain unchanged.
• Independent agent review confirmed source custody, the fixed-interface proof scope, and the exact main/helper impulse identity. This is not human peer review or a formal proof of the multiplication theorem.

GitHub CI limitation: the planning job fails the same two unchanged upstream test_research_current_status assertions seen on #5717 and #5718: the current-status navigation lacks the selected review and source-manifest links. It ran 449 tests with two failures and 21 skips. Scientific replay was skipped by CI; the package's registered and cold-export replays passed locally as recorded above.