Army Men actual-call observer: blocked before implementation
Historical source last updated, 2026-10-20. Runtime and implementation claims below describe that dated source; see current app notes and source for subsequent changes.
Source findings to preserve
Read the required agent instructions, CLAUDE.md, messageboard, Army Men notes,
conformance report and retained scroll-entry evidence:
scratch/runs/20261019-army-men-scroll-entry-remote-v1/
(review.json, controller.js, child.json, pins.json,
army-men-scroll-entry-plan.js, relevant output.log entries).
test/run.jstrace-at handling records registers/guest bytes at the breakpoint. The retained eight observations at original0x4074f1are pre-push observations, not return events. The plan limits hits to eight.src/09b-dispatch.watincrementsapi_callsand emits the API entry name before loading stack arguments. Resolved COM/ordinal calls emit0xC0DE0000 | api_idbefore that name. The ordinary dispatch path callshost_log_api_exitafter the generated handler and nonvolatile restoration, gated byapi_log_on. This is a candidate post-handler observation, not proof that the guest caller's next instruction executed.- Generated dispatch entry 1008 calls
handle_IDirectDrawSurface_BltFast. That handler reads argument six at ESP+24 and resolves both surface objects. It reads backing fromDxObject.misc1, dimensions, pitch and bpp from the object. Its early-return paths demonstrate why success cannot be inferred from an entry trace; EAX must be observed. tools/jig-owning-api-observer.jsalready wrapslog,log_i32andlog_api_exit, authenticates dispatch counters, keeps a nesting stack, and records entry ESP/return address and exit EAX/ESP. It has callbacks for entry/exit and explicit incomplete-pair states. This is reusable plumbing to evaluate, not a validated Army capture adapter.test/run.js:dxLookupThisresolves the object's slot from guest this+4 and usesRegionMap.BASE.DX_OBJECTS, rather than a copied movable base. Its existing decoder reports width/height/bpp/pitch. It does not capture bounded pre/post backing pixels for the same invocation.
Still missing: original instruction-byte/disassembly proof connecting
0x4074f1 through the six pushes and indirect call to its actual return
address; an authenticated owning-instance adapter; validated backing reads;
invocation-paired pre/post snapshots and initial corruption evidence. Actual pitch, backing, format and HRESULT remain unknown. The prior
downward overlap fix does not establish the cause of upward terrain stripes.
Validation for this documentation-only receipt: source inspection and
git diff --check. No test or capture-helper completion is claimed.