`5x/REPORT.md`

38 lines
# /5x REPORT — MODEL WARS (http://localhost:9911/)

## Sweep ledger
| sweep | six-way core (M1–M3 + B4–B6) | click-through | caught | fixed | note |
|-------|------------------------------|---------------|--------|-------|------|
| 1 | **6/6 PASS** | 15 pass / 6 fail / 10 skip | 6 | 0 | all 6 fails = one hidden-tab control |

Stopped after 1 sweep: the six-way core is fully green and the only failures are a
proven test-harness artifact, not app defects — so there is nothing to *fix*, and the
no-progress guard forbids re-sweeping a non-defect.

## Final six-way state (the real verdict)
- **M1 HTTP contract** — PASS (200 text/html)
- **M2 headless render** — PASS (screenshot 1.25MB)
- **M3 automation E2E** — PASS (`#beginBtn` present, 0 JS errors)
- **B4 Google Chrome** — PASS
- **B5 Safari** — PASS
- **B6 Firefox** — PASS

## The 6 "failures" — classified, not fixed (correctly)
All six are `locator.selectOption` timeouts on **`#addProvider`**, the provider
dropdown inside the **King's Table** tab. That tab is `display:none` while the Arena
tab is active; the /3x click-through agent operates controls in DOM order and does
**not** switch tabs, so it cannot reach a hidden control → 5s timeout.

**Proven a non-defect:** with the King's Table tab active, `#addProvider` is visible
(`offsetParent !== null`) and `selectOption` succeeds (`value → 'anthropic'`), and the
full "Summon a champion" flow works end-to-end (verified earlier via `/api/champions`).

Why no code change was made: the only ways to flip this to PASS are (a) weaken/skip the
assertion — forbidden by /5x, or (b) surface the King's Table controls inside the Arena
tab — a real UX regression. Neither is a legitimate fix. The honest outcome is: core
six-way is clean; the click-through limitation is the harness, not the build.

## Open items
None that are app defects. The click-through simply doesn't tab-switch; if desired, a
future improvement is a tab-aware click-through pass (drive each tab, then its controls).