P08-03: 配置の通常操作受入

This commit is contained in:
Agent Control Runner
2026-10-14 07:18:17 +09:00
parent 0c5e0022f2
commit 1f0a2bcd53
14 changed files with 6962 additions and 24 deletions
@@ -246,3 +246,36 @@ and implementation.json. These are mocked Python controls, not native or OS
acceptance. Root native, OS, final full regression, independent review and
integration/current verification remain pending, including Windows device
acceptance. Existing driver2400, calibration and earlier repair work is retained.
## R07 actual-input observer repair (current submission)
Authority: ROOT_OS_R07_SELECTED_CELL_READBACK.json and its supplied raw writer
actions/observations and screenshot. These historical failures remain unchanged:
writer exit1, reader not started, and request64 is not current acceptance proof.
The five condition records and acceptance denominator remain unchanged.
The passive read-only observer captures left-press viewport coordinates in
_input before GUI side effects, inverse local coordinates, the actual converted
world cell, bounds, cell dimensions, map origin, pan/zoom and both affine
transforms. It also records the prior selection/occupant as separate dispatch
fields. The action records the post-input selected entity and cell occupant
separately. Later mouse-getter reads remain diagnostic only. The checker binds
the action to the actual pressed event, independently recomputes its cell and
transforms, and checks camera selection against complete-model occupancy.
Requested center pixels remain distinct from actual event pixels; a press
elsewhere within the same cell is valid.
All old tests remain; new whole-checker semantic controls cover missing dispatch,
wrong actual cell, mismatched event coordinates, transform drift, wrong selected
entity/occupancy and bypassed selection. A positive control demonstrates that
a later diagnostic getter cannot change acceptance. Final syntax, test counts
and source/input hashes are recorded in repair-r07-local-validation.json and
repair-r07-local-source-manifest.json, referenced by implementation.json.
No product changes, Godot/native/OS/full regression, git writes, Runner actions
or integration were performed by this worker. Driver receipt values remain
2400 seconds and all timeout cleanup guards remain. Root must collect actual
native and OS evidence on the final candidate, perform verify-only and final
full regression, then obtain a separate independent review of all five facets.
Any proposed historical reuse requires exact source/input applicability and
independent review; no changed-source PASS is inherited here.