Original Win32 Pegged: released bounded drag

Run: scratch/runs/20261114-pegged-released-drag/.

The exact original is test/binaries/entertainment-pack/pegged.exe, 43,792 bytes, SHA-256 bb5a4e90b71363cb7df5f818fc9a2de9841dbdfcb8a74498756b2c11438fe4c1. Its registry and media are unchanged. The earlier six-peg original remains closed; its unchanged source-click image proves neither selection nor a jump.

Static original-handler evidence establishes left-button drag semantics: WM_LBUTTONDOWN at 0x0100403b accepts an occupied source, sets the active flag at 0x01004268, and calls SetCapture. WM_MOUSEMOVE at 0x010044ad handles that active drag. WM_LBUTTONUP at 0x01004b28 uses its destination, releases capture, and validates an orthogonal distance-two jump, empty destination and occupied midpoint through 0x01006680. The active flag is cleared at 0x010056df. Board WM_KEYDOWN/WM_CHAR are not handled by this wndproc. The referenced matching pegged.hlp is absent; no visible Help contents are claimed. Focused static excerpts are retained in the run.

The adopted tools/pegged-native-controller.py and its two tests implement one personally chosen fixed drag through existing public guarded pointer edges. No public input or renderer change is needed. One fresh complete scene must identify the occupied source, occupied neighbor, empty destination, legal reason, exact centers and actual count before DOWN. An independent child acknowledges its immutable 1750 ms fallback before DOWN; destination UP requires identity-matched consumed DOWN and MOVE receipts. The complete normal gesture is bounded by 1500 ms including movement. EOF, cancellation, late ACK or uncertainty releases at the original source and closes input. Physical cleanup precedes durable audit. No capture or model call occurs while held. Every completed gesture requires a separate personal visual removal review before another gameplay gesture. UNKNOWN remains terminal, without retry or rebinding.

The coordinator's explicit prerequisite amendment permits this successor after the predecessor's actual native, watchdog and supervisor-child waits with both EOFs, supervisor heavy release and exact process absence. Those records were personally checked. StarCraft's ancillary outer-wrapper wait remains UNKNOWN; its independently proved native Quit is preserved. The previous report 82d9d0ea is read-only and retains its historical gate.

Three personally verified removals, native Game Over with three pegs, then native Game → Exit and actual exit code 0. No win is claimed. The fresh original opened directly on the actual six-peg default board. Binding truthfully recorded the board phase, despite the first capture's historical filename bootstrap1.png. No Start, pattern selection, reset, undo or second puzzle was sent. There was no numeric remaining-count text; counts below were personally established from complete native pixels.

Gesture Personally chosen jump Count after visual review
1 r3c4 over r2c4 into r1c4 6 → 5
2 r5c4 over r4c4 into r3c4 5 → 4
3 r3c4 over r3c5 into r3c6 4 → 3

Each review explicitly identifies the source and jumped neighbor now empty and destination now occupied. The third drag produced an actual native Pegged dialog with message Game Over. Because it obscured the source, the third removal was withheld until a reviewed native OK dismissal exposed the complete board. Its three remaining pegs at r1c4, r3c3 and r3c6 have no occupied orthogonal neighbor, consistent with the actual native result. Independent destination-UP receipts establish DOWN-to-UP intervals of 519, 514 and 512 ms; cleanup acknowledged all three gestures.

Two five-second admission attempts expired before any input. Subsequent admission and drag reviews used a fresh complete capture, requiring exact byte equality with the personally viewed complete scene before acting. The review names both images and records their complete-image SHA-256. Any mismatch would refuse the operation. This compares image identity only; it reads no board or guest state and generates no coordinates or moves. All chosen coordinates came from this fresh original's personally viewed scene. No timestamp, visible selection or count was fabricated.

The actual result dialog also exposed a derivative-helper assumption: the original title is Pegged, whereas Game Over is its message. Before OK, the owned dismissal policy was corrected to check title and message separately. The old policy failed the actual-title fixture; the correction passed that fixture and refused incorrect titles/messages. Fifteen policy tests and another full locked build passed before the correction was published as 25a76f2a. Public input and resident runtime code were untouched. The two changed owned files are explicitly recorded as a second source phase, with old/new hashes and dialog-owned-change.patch; they are not reported as zero launch-pin drift.

The initial coherent controller was published as af185f28 before original launch: 14 policy, 9 HTTP and 8 actual CLI controls passed, along with the released 23 policy, 12 HTTP and 10 actual CLI controls, and flock /tmp/wine-build.lock bash tools/build.sh. Actual CLI fixtures test the real public endpoint/renderer, cancellation, identity replacement, lost ACK, parent death, late MOVE, independent fallback and persistence ordering; they are not Pegged gameplay proof. No public endpoint extension or callback change was needed.

The original recipe was bounded at 900 seconds, default batch size 1000 and real ticks, with an independently prearmed 920-second watchdog and irreversible final 180-second Quit reserve. Reviewed Game → Exit was consumed. The guest logged [Exit] code=0 at 367.790 CLI seconds; actual native wait followed 368.021 seconds after launch, before the reserve boundary. Native, watchdog, supervisor-child and outer-wrapper actual waits all returned 0 with both EOFs. Their exact PIDs were absent, port 8615 was closed and the heavy lock was released. The watchdog exited unfired on actual native wait.

The final independent release found OWNER_EXITED: physical UP was not acknowledged. That is retained separately from the reviewed Exit command, consumed receipt and actual normal guest/process exit. One held main-owner critical section was logged at exit; no all-thread-clean claim is made.

All 423 launch pins were checked immediately before releasing runtime ownership: two explained owned title/message-policy changes, zero unexpected drift. All ten read-only pins and registered media matched. Evidence contains Git SHA plus relevant binary patches, compact logs and selected complete PNGs named by result.json; no source tree, media or WASM copies or frame series. No private board access, seed/clock/config writes, automated reader/solver, browser, full-27, FPS, audio or new-title compatibility claim is involved. No box or additional worker was created: box ID/expiry null, state NOT_CREATED, stop receipt NOT_APPLICABLE.

This bounded task is complete. No further input or original launch is planned.