# Concourse Divide — MiniMax M3.1 Flash validation

- Artifact: `index.html` (43,012 bytes)
- Source: complete document from a fresh MiniMax M3.1 Flash authoring agent on the DeepSeek Harness; packaged without operator code changes
- Runtime: Chrome `155.0.8059.84`, Node `v22.23.3`, offline `file://` CDP loads
- Review date: 2026-10-27

These are post-delivery operator checks, not authoring-agent claims. No score, rank or rubric passage is claimed. One required behaviour failed and is reported as a failure.

| Check | Result | Evidence |
| --- | --- | --- |
| Complete standalone document | Pass | Doctype and closing `</html>` present; one inline script parsed with `new Function`; no external asset reference; no `fetch`, `XMLHttpRequest` or `WebSocket` |
| Offline desktop load | Pass | 1280×720 `file://` load produced a 914×633 diorama canvas, a trail canvas, three passage buttons and Pause/Reset; no console error, no page exception, no external request |
| Offline mobile load | Pass | 390×844 produced a 370×300 canvas with no horizontal overflow and the gate buttons still present |
| Fixed population of 48 | Pass | The panel reported `POPULATION 48`, `ACTIVE 48`, `ARRIVED 0`, `TOTAL 48` at load, and `Active 48 + Arrived 0 = 48`. Inspecting all 48 travelers returned 48 distinct IDs and 48 distinct positions |
| Stable IDs and inspectability (A5) | Pass | The inspector showed `AMBER MOVING A-23`, `Destination West arrivals · bay 23`, the planned route, waypoints left, speed and position. Cycling with `,` and `.` reached all 48 travelers |
| Waiting explanation (A5) | Pass | The inspector reported `Waiting because Yielding to C-23 crossing the passage (higher ID yields)` and `Waiting because Queued behind C-20 — no clear space ahead` |
| Truthful counts (A5) | Pass | `Active + Arrived = 48` held at every reading of the run: 48+0, 46+2, 44+4, 42+6 |
| Closed network waits safely (A3) | Pass | With all three passages closed, 32 of 48 travelers reported `No route` and waited; the remaining 16 were already across or reported another reason. Nobody was removed from the population |
| Gate drain, amber then red (A4 setup) | Mixed | North closed on its own and reached `NORTH · CLOSED` after about three seconds. Center stayed at `CENTER · CLOSING (12 inside)` for the whole run: the travelers trapped inside the doorway never drained it, so the passage never became solid |
| Pause freezes the simulation | Pass | With the network sealed, the counters held at 48/0/22 for 2.5 s of paused time and resumed unchanged afterwards |
| Seeded reset | Pass | `RESET` restored `ACTIVE 48`, `ARRIVED 0`, `WAITING 0` and all three passages to `OPEN`, with the same starting layout |
| Keyboard-accessible gate controls | Pass | Pressing `N` reopened North and the button text followed |
| South-only recovery (A4) | **Fail** | With North and Center closed and South reopened, 6 of 48 travelers had arrived after 95.4 s of simulation. The brief requires all 48 within 90 s |
| South-only recovery, Center closed first | **Fail** | With Center closed before anyone entered it, then North and South closed, then South reopened, 0 of 48 travelers had arrived after 95.3 s of simulation and all 48 were still waiting |
| Route diversion on closure (A1) | Not confirmed | The inspector's planned-route line for the sampled traveler read `East staging → direct corridor → West arrivals` both before and after Center closed. The crowd's behaviour did change, and 32 travelers replanned to `No route`, but a rerouted route line was not observed |
| Bodies respect walls, gates and each other (A2) | Not confirmed | No traveler was seen crossing a wall or a solid shutter, and no agent disappeared from the population, but no pair-distance or wall-overlap measurement was taken |
| Runtime during interactions | Pass | No uncaught exception, no console error, no failed or external request across two full runs |

## Reading the result

The simulation runs at real time: the page holds 60 frames per second in headless Chrome and the fixed-step loop advances at 1/60 s per frame, so the wall-clock waits above are simulation seconds. The failure is in the artifact, not in the measurement.

The jam is visible in the published screenshots. After the passages were sealed, the travelers bunch against the divider, and the divider wall in the fixed camera hides most of that crowd, so the rendered station understates the reported population. In the first run 12 travelers stayed inside the Center doorway for the whole session, which kept that gate in its amber `Closing` state; the gate never became solid and the queued crowd never got a third route. In the second run the gates did close, but the crowd still did not clear through South within the deadline.

## Limits and unchecked items

- No numeric score, rank or pass badge is stored, and no complete rubric passage is claimed. A2 and A1 stay not-observed or not confirmed.
- Only gate toggles were driven. The operator did not click travelers on the diorama; selection was exercised through the `,` and `.` keys.
- The trail canvas was confirmed to exist and to paint, but the two-second trail was not reviewed frame by frame.
- The authoring agent rewrote the file once inside the same authoring turn after a mental trace of its own code. The published file is that final complete document; no operator repair was made and no repair turn followed delivery.
- Screenshots were captured but only the recovery frame was visually reviewed.