Aranna native hover journal, 2026-11-07

The run-local harness repair passed both-module controls. The single registered original advanced through Main Menu → Single Player → Start New Game → Choose Hero → Difficulty. Player-controlled world movement remains untested: the global observer transport reached its unchanged 8,192-record limit at Difficulty, and native point certification refused. No production emulator defect or repair is established. Evidence and reproducible helper patches are retained in scratch/runs/20261107-aranna-hover-journal/.

The previous run had no original target click: lock expired before dispatch, producing zero effects. Its generated post-arm acquisition moved the cursor to (396,-22); measured movement reached (399,156), but DRAW and MANAGER shared an exhausted eight-store journal. This investigation repaired that demonstrated observation gap before launching another original.

Run-local repair and controls

observer-transform.cjs separates bounded DRAW/input stores (eight) from MANAGER stores (sixteen). Stores no longer consume hover transaction admissions. Hover deduplication also requires an unchanged manager-store revision: returning to an identical coordinate after acquisition produces fresh evidence. The observer still executes each guest store once. Arming remains one-way; the 32-transaction hover limit and global 8,192-record transport remain bounded. Saturated manager coverage, stale evidence and incomplete transactions refuse certification.

The adapter preserves actual Worker/session/module, app, scene/provenance, receiver/vtable/native return, focus, geometry, movement and protocol checks. Ordinary mouse operations retain separately timed down/up attempt and completion receipts. Unknown or partial outcomes are terminal. A stale lock transition remains a typed refusal before input, with zero effects; no held operation is replayed.

Capture now records the adapter's initial browser pointer bookkeeping, including the ordinary launcher click. Node-only adapter reload reconstructs from that captured baseline and completed receipts, verifies it against live bookkeeping, and resumes without browser evaluation or guest-state writes. Reconstruction rejects unknown/partial receipts. Actual reload evidence is in control-target-standard2/input-module-reload.json and its compat counterpart. An earlier generated reload exposed the missing launcher baseline and terminated without further input; its failed evidence remains retained.

route-qualification.json passes the actual natural expiry → stale zero-effects refusal → trusted canvas acquisition → measured retarget → fresh native hover → target activation route in both modules. Reviewed standard2/response-08 and compat3/response-10 captures show ACTIVATED TARGET A, hit=1 and down/up=3/3. Acquisition itself had hit=0 and down/up=2/2. The fresh final point (398,155) equals the pre-acquisition point, explicitly exercising identical-coordinate revisits. Browser traces establish trusted requestGuestPointerLock acquisition and natural releaseExpiredAutomaticLock expiry.

In compat3, the first completed return movement instead reached (400,332). Strict proof refused before a target command was queued. A fresh reviewed capture and measured correction returned to (398,155), after which a new complete proof admitted activation. No uncertain input was retried. Compat2's activation passed but did not exercise identical coordinates; it is not the final qualifying case. The evidence retains these distinctions.

The two full-shell routes produced no post-arm DRAW writes. Separate generated controls therefore establish the actual DRAW-before-later-MANAGER ordering: draw-manager-ordering.json records eight DRAW stores through serial 1341 and sixteen MANAGER stores through serial 4810 in both modules, via ordinary API and owning Worker execution. The generated saturation fixture itself exhausts hover admissions and is not accepted as a live target proof. Fresh hover certification is established by the full-shell routes above.

Both twelve-cell baseline/disabled/armed passivity matrices pass: actual module and owner identity, CPU/lazy flags/live FPU, guest memory, input/public API state, COM/draw behavior, DLL initialization, reserve and SEH results match. The changed observer remains inactive when disabled. Results are in controls/ and controls-draw-churn/. The first matrix's initial offline validator referenced predecessor metadata by mistake; the corrected review passes on its retained results. The second matrix passed directly. Incorrect initial full-shell DRAW coverage assertions were replaced with the separate actual ordering evidence.

Additional passing controls cover 22 mutations of actual route receipts, including wrong owner/module/native target, stale protocol/point/movement, later stores and exhausted/missing journal coverage; 32 contract boundaries; 12 retained-proof negatives; pointer reconstruction; and five dispatch mocks including partial/unknown terminal outcomes. The retained prior exhausted journal still rejects. Unchanged earlier controls are referenced rather than rerun.

Original and concrete remaining blocker

The sole original ran from 00:32:31.271Z to normal exit at 00:42:53.967Z, within the authorized forty minutes including review and cleanup. The actual owning standard observer module was 56f8521938f3fcc8e0b2806926b436575a64cbde270ff30e5136aade1027b860. original/action-10.json records fresh native Single Player proof at (398,155), serials 2548–2553, and completed ordinary down/up. Reviewed response-11 shows Single Player; response-15 shows Choose Hero; response-19 visibly contains the ordinary typed name Aranna; response-22 shows Difficulty after Next.

At Difficulty, action-23 completed balanced ordinary reacquisition. The transport first reached 8192 at 00:42:36.000Z, event 89, on the actual owning Worker. The subsequent read-only measure-point.cjs original response-23 refused on the global saturation assertion. The exact first helper invocation time is unavailable; the refusal is retained in original/difficulty-measure-refusal.log. No Normal click was queued. Reviewed response-23 is the final Difficulty screenshot.

original-assessment.json records fifteen completed input actions, nine balanced clicks, and 545 complete observer transactions with no incomplete packets. There was no safety flag, unknown/partial input outcome, forced lock, injected guest state, second original, or guest/media/config edit. The original did not require the generated control's post-arm first-target recovery: its first target click completed before expiry. Later ordinary reacquisitions were reviewed separately.

The next work is a bounded fresh observation window or reduced unrelated trace traffic that preserves complete owner/scene/native-target proof throughout later setup. Qualify its changed observer controls before any separately authorized original. Do not reuse this saturated journal as proof. World loading, movement, combat, audio and sustained performance are not established by these captures.

Reproduction and handoff

source-identity-v7.json pins Git revision f4473ab2936e22ac5fdd02267eec3504bd425aa2, relevant binary diff and individual run-local helper patches. Observer builds derive from production revision e5c075a59ea84bcb3abc0334c6e3986ebc96a96b; baseline module bytes match shipped standard/compat builds. Source and HTTP identity checks passed before the original, and source identity still matched after closure. Media remains at its existing path with SHA-256 references. No source tree or media archive was made. This publication changes documentation only; no production build is claimed. Aranna has no candidate-corpus manifest entry; candidateId: null retains this as investigation evidence and is an explicit dashboard-schema exception.

runtime-cleanup.json verifies all owned runners, Chrome processes and recorded descendants absent, ports refused and temporary profiles absent. All browser sessions closed. runtime-release.json records terminated holder 1110100 and all four local runtime/build locks independently free. Runtime/source/build release was posted explicitly for Lords2 immediately after cleanup. No boxes were allocated or changed. Free /home space remained above 2 GB.