Qa Test Fast

---
name: qa-test-fast
description: "[Testing] Methodology behind the /qa-test-fast command — the grounded quick ticket test between /qa-test FAST and FULL. Use when running or changing /qa-test-fast, or when deciding whether a ticket fits it."
argument-hint: "(read by /qa-test-fast — not invoked directly)"
disable-model-invocation: true
---

# /qa-test-fast — methodology

**The promise:** a ticket verdict in one sitting that is still **grounded**. The checklist is traced to
a fault model, and the fault model is traced to the diff, the ticket and the domain. `/qa-test` FAST
buys speed by dropping the model. FULL buys certainty with a verifier per step and suite authoring.
This flow keeps the model, drops the verifiers and the suite writes, and runs discovery **alongside**
execution instead of before it.

## Stage flow

```
Step 0  route · env · prior artifacts · slug · kb ask
Stage 1 Wave 1  A ticket ‖ B PRs ‖ C domain map (only if ABSENT)   ── join → bundle (+ schema:refresh)
        Wave 2  /qa-test-model ‖ /qa-test-mind-map (update|build, no stamps)
        Wave 3  /qa-checklist --from-model  ──  stage gate   [--dry-run stops here]
Stage 2         runner(chrome) ‖ runner(edge) ‖ exploratory(firefox), data on the fly ── settings restored
                + visual(devtools, ui-ux-expert → Skill qa-design) when visual_surface; ≤3 at once
Stage 3         qa-triage-results ticket --verify (→ vc-fix:qa-investigate per real bug) → /vc-fix:qa-bug per
                confirmed bug, with its package ── teardown → verdict → files → HTML page → ask → comment → kb
```

**Why three waves, not six parallel agents:** the dependencies are real. The mind map is derived
from the domain map. The model needs the ticket and the diff. The checklist needs the model and the
mind map. `/qa-exploratory ticket` stops without a model. Parallelism lives *inside* each wave.

## Fit

| Use `/qa-test-fast` | Use something else |
|---|---|
| a `feature-test` Story/Task that crosses 1–3 layers and needs a verdict today | a bug fix → `/qa-verify-fix` · a hotfix → `/qa-hotfix-check` · a refactor → `/qa-test` (technical-change) |
| no suite authoring is wanted from this run | new regression cases are wanted → `/qa-test` FULL |
| | a release gate that needs independent verifiers → `/qa-test` FULL |

## What each stage may write

| Stage | Writes | Never writes |
|---|---|---|
| 1 | `reports/ba/test-models/<TICKET>-<date>.md` · the domain map (only via `/qa-domain-map`) · the mind map (only via `/qa-test-mind-map`) · `testing-checklist.md` | suite CSVs, the tracker |
| 2 | evidence under `reports/tickets/{SPRINT}/<TICKET>/screenshots/` · `design-report.md` (visual lane) · `AGENT-TEST-` entities (then deletes them) · the SBTM session file · a setting or fixture a Data cell names (`flip+restore` / `mutate` + its seeder part, re-read after) · the Results in `testing-checklist.md` (you, at the join) | any other shared fixture, other lanes' settings |
| 3 | `summary.json`, `verdict.md`, `triage-report.md` + the investigation packages (via `qa-triage-results`), bug reports **via `qa-bug`** · the Artifact page (shared with Anyone at Virto Commerce) + its `Page:` line · one tracker comment after a yes · kb entries | a status transition, a hand-written bug |

## Rationalization table (each row was produced by a baseline run without this skill)

| Excuse | Reality |
|---|---|
| "The mind map doesn't exist and nothing requires it — skip it." | Rule 4: build it (domain map present) or record the observable reason. The checklist's node tracing is where it pays off. |
| "That node is already DRIFT in the map, so there's nothing to check." | A DRIFT is a question. Re-observe it: `HOLDS` or `RESOLVED` with evidence is the finding the next map update needs ([`context-wave.md`](context-wave.md) §Wave 3). |
| "Pre-seed with the domain seeder first, then test." | Rule 3: data is made per item. A seeder runs only when an item's Data cell names it. A pre-seed wave is FULL's `3a`, not this flow. |
| "The runner already saw it fail — call `qa-bug` straight away." | A runner's FAIL is a candidate, not a bug: it can be a wrong checklist item, stale data or an env blip. Triage classifies it, and `vc-fix:qa-investigate` reproduces it and finds the owning layer. Only then does `qa-bug` file it, from that package. |
| "Triage is quick — I'll classify inline." | `qa-triage-results ticket` runs the same classifier as a regression run. Inline classification is the second copy of that taxonomy that drifts. |
| "I'll write the bug reports into `reports/bugs/open/` myself." | Rule 1: `/vc-fix:qa-bug` does the 4-layer validation, the owning-repo resolution and the `/qa-fix` handoff block. A hand-written report has none of them. |
| "The opening status hop is automatic." | Rule 2: this flow makes no transition, and that includes the opening one. |
| "Contract refresh isn't needed for a verdict." | Without it, every `{DOC}` oracle on a GraphQL field is `{HYPOTHESIS}`. `schema:refresh` takes about 9 s. |
| "A 120-line report is within the per-ticket cap." | The deliverable is the HTML page + a ≤60-line `verdict.md`. The cap is a ceiling, not a target. |
| "The page is optional — offer it." | Rule 6: it is the deliverable. |
| "FAST allows one agent, so ask whether to go FULL." | This command *is* the answer to that question. Do not re-route. |
| "This morning's run already has results — reuse them." | On the same build, show that verdict and ask (Step 0.3). Never silently reuse another run's results, and never silently overwrite them. |
| "Sweep leftovers with `seed:*:teardown`." | Persistent fixtures share the `AGENT-TEST-` prefix, and a sweep deletes them. Teardown is by ledger id only. |
| "The checklist runner already looked at the page — skip the visual lane." | A functional runner cannot see a contrast failure, a token collision or design drift. `visual_surface: true` ⇒ the lane runs, or `--no-visual` is recorded ([`../qa-test/visual-axis.md`](../qa-test/visual-axis.md) §5). |
| "I'll invoke `/qa-design` myself, inline." | The command is not model-invocable, and the skill is ~30K chars. The `ui-ux-expert` brief makes the **agent** invoke the `qa-design` skill ([`../qa-test/visual-axis.md`](../qa-test/visual-axis.md) §2). |
| "25 minutes is enough exploring." | The exploratory box floor is 30 minutes ([`../qa-test/exploratory-lane.md`](../qa-test/exploratory-lane.md) §5). |

## Red flags — stop and re-read the six rules

- You are about to `Write` a file under `reports/bugs/`.
- You are about to call `/vc-fix:qa-bug` for a finding that has no `vc-fix:qa-investigate` package.
- You are about to call a tracker transition.
- The Stage-2 message contains a `seed:*` call that no checklist item's Data cell names.
- `verdict.md` lists a Stage-1 artifact as neither produced nor SKIPPED-with-reason.
- You wrote "optional" next to the HTML page.
- A `seed:*:teardown` appears anywhere in your teardown.