A live proof that deliberately does nothing physical

On 19 August the real core and HUD were booted on isolated loopback ports. Three unauthenticated gesture controls were refused with 403, an E-stop blocked the gesture pipeline, and the HUD rendered that truthfully. No camera opened and no window moved.

LIVE-PROVEN What this run supports. Recorded 19 August 2026.

What happened

Production entry points started a core and a HUD on isolated loopback ports with disposable stores. The initial gesture snapshot showed controller off, camera off and zero frames. Three mutation routes without the process token returned 403. An E-stop trip made the pipeline read blocked, then seven dark stages, and a headless browser confirmed the HUD panel showed exactly that. Both processes were restarted, re-verified and cleaned up.

What this proves

The real core and HUD boot on loopback, serve health, and refuse unauthenticated gesture mutation with 403.

An E-stop propagates to the gesture pipeline and the HUD renders the blocked state with accessible labels.

After a restart on the same stores, the controller is off, the E-stop is clear and effect recovery is clean.

What it does not prove

Physical hand perception: camera frame rate, calibration precision, target accuracy, follow latency and false activations were not probed.

Recorded on 2026-08-19 against a candidate that later received repairs; the project treats it as regression evidence, not the final commit-bound claim.

It does not show any gesture producing an effect.

Context

The Z11 order built governed spatial cognition: a camera pipeline, hand landmarks, a gesture state machine and eight gates before any gesture may cause an effect. A physical acceptance test needs the owner at the machine with a camera. Before that sitting, the project wanted a live proof of everything else that deliberately took no physical action.

The run

The production serve and serve_hud entry points started in hidden processes: the core and the HUD, each on an isolated loopback port. The run used a disposable local-data directory, disposable SQLite stores and fresh process-local tokens that were never printed.

Initial state. GET /gesture reported controller OFF, camera session OFF, zero hands, zero retained raw frames, effect authorisation false and recovery ready/clean.
Refusal. Posts to arm the controller, capture and confirm, all sent without the token, returned 403. Nothing further was attempted.
E-stop. A diagnostic E-stop trip made the pipeline read exactly blocked, dark, dark, dark, dark, dark, dark, dark, with the reason “E-stop engaged; gesture perception cancelled”.
HUD truth. A fresh headless Chromium opened the W5 view. The panel was bound, showed eight stages with the same states, labelled the first one “webcam frame: blocked. E-stop engaged; gesture perception cancelled”, showed zero landmark joints and logged zero console errors.
Restart. Both processes were stopped by exact PID and their ports were confirmed released. They were restarted on the same stores, where health passed, the controller was off, the E-stop was clear and recovery was clean. Everything was then stopped and cleaned up again.

The record’s tally: memory mutations 0, camera opens 0, raw frames 0, desktop effects 0, external effects 0, paid-provider calls 0.

The evidence

docs/proof/Z11_LIVE_EVIDENCE.md records the PIDs, ports, responses and rendered states. It notes that the candidate later received repairs, so this reading is regression evidence rather than the final commit-bound claim. No footage exists yet.

The limits

The record states its own boundary. This is “live server/browser evidence of absence, refusal, binding and truthful rendering. It is not physical hand-perception acceptance.” Camera frame rate, visual latency, calibrated precision, target accuracy and the one-hour false-activation target remain NOT_PROBED_OWNER_HARDWARE until the owner’s sitting. Journey J10, hand tracking a panel, is fixture-tested only.