⚠ Errors log — every defect found, fixed & re-shot (links to each scenario's full retry history)
| Scenario | Step | Failed-pixel observation | Fix (builder sub-agent) | Attempts | Final |
|---|---|---|---|---|---|
| S3 — booked on red-❗ → resume on URL save | 2 | ❗ cleared but no "🎨 Dizájn készül" chip; DOM showed no Studio project — URL-save did not re-fire create+kick | Wired URL-save → ensure_studio_project(); cleared studio_deferred in one transaction (TDD red→green); redeployed | 2 | ✓ green after fix |
Demo shows 1 logged error; the overnight runner appends one row per defect across all 55 scenarios, each linking down to the scenario toggle where the failed + passing frames are kept.
▸ S1 Enrichment guarantees a website URL; missing → red ❗ badge guardrail / upstream studio-S1bnf-r3-S8duo2-S14 LIVE ✅ PASS
| # | You do | You should see | Element that changes copy · look · where |
What changes underneath | Must NOT happen | 🕓 Touchpoint history | 🧪 Test |
|---|---|---|---|---|---|---|---|
| 1 | Enrichment runs on a new lead and finds + saves the prospect's website | The card shows the website (favicon/domain) and NO red ❗ — Studio readiness is implicitly met | Copy: domain shown (e.g. "beridoor.hu") · Look: normal card, optional 🌐 favicon · Where: card meta row | deal.website_url persisted; Studio-readiness = true |
Lead must NOT exist long-term with an empty website silently; readiness must NOT be assumed without a saved URL | SYSTEM 🔎 "Weboldal megtalálva" · "{domain} mentve" · by system | LIVE ✅PASS |
Step 1 — live test transcript① Before the shot — “What should I see?”
A saved website domain on this deal's Details panel (e.g. "beridoor.hu") and no red ❗ on the card face — enrichment should have persisted 📸 screenshot taken
frames/s1-01-details.png · 1280×800 · captured 21:18:04 ② What to look for (exactly)
On the Details panel, find the "Weboldal" row and confirm it carries a real domain. On the card header, confirm there is no red ❗ ribbon. Cross-check that the Legal-info fields (entity type, tax number) are populated — this row merges duo2-S14's ingestion fill. ③ What I see — Gemini · 5-sentence pixel report
The Details panel shows a populated contact block; the "Weboldal" row reads "beridoor.hu"; the legal-info fields below are filled; the card header carries no red or amber badge; there are no empty-field placeholders. ④ Verdict — pass or error?
Pixels (filled Weboldal row, no badge) and the DOM/state ( ⑤ Next action
PASS → Proceed to the no-URL branch (step 2) to confirm the red ❗ fires when enrichment finds nothing. | |||||||
| 2 | Enrichment finds NO usable website at all | A prominent red ❗ badge on the card face — the only thing that will later block the Studio auto-kick | Copy: tooltip "Nincs weboldal — add meg a Studio dizájnhoz" · Look: RED ❗ badge, pinned top-right · Where: card corner ribbon | deal.website_url empty → readiness = false; ❗ flag raised (after the bnf-r3 retry budget) |
The ❗ must NOT be hidden in a panel; a no-URL deal must NOT pass as Studio-ready (the duo2 silent path) | SYSTEM ❗ "Nincs weboldal" · "kézi megadás szükséges" · by system | LIVE ✅PASS |
Step 2 — live test transcript① Before the shot — “What should I see?”
On a lead where enrichment found no website, a prominent red ❗ badge pinned top-right of the card face — never hidden in a panel, never only an amber chip, never the duo2 silent path. There must be no "Dizájn készül" chip (the Studio kick is held). 📸 screenshot taken
frames/s1-02-noURL.png · 1280×800 · captured 21:18:41 ② What to look for (exactly)
Scan the card top-right corner for a red ❗ ribbon. Confirm there is no "🎨 Dizájn készül" chip. Hover to confirm the tooltip prompts the operator for a URL. ③ What I see — Gemini · 5-sentence pixel report
The board card carries a bright red ❗ ribbon top-right, above the fold; the card greys slightly to draw the eye; hovering shows the "add a URL" tooltip; no design chip is present. ④ Verdict — pass or error?
The red ❗ is raised and unmissable, and the Studio kick is held → matches studio-S1, not the duo2 silent path. → PASS (the conflict resolution holds in pixels). ⑤ Next action
PASS → Carry into S3 — book a call on this red-❗ deal and confirm the kick is deferred, not fake-run. | |||||||
▸ S2 Sales call booked + URL present → auto-create project + auto-kick the Lab (Phase ①) happy path studio-S2 RENDERED ⚠️ PARTIAL
| # | You do | You should see | Element that changes copy · look · where |
What changes underneath | Must NOT happen | 🕓 Touchpoint history | 🧪 Test |
|---|---|---|---|---|---|---|---|
| 1 | The sales call is booked on a URL-present deal | A Studio project is auto-created, bound to this deal; the status chip "🎨 Dizájn készül" appears | Copy: chip "🎨 Dizájn készül" · Look: indigo chip · Where: card status-chip slot | Studio POST /api/projects {source_url, crm_ref=deal_id}; project id stored on the deal |
Must NOT create a 2nd project (idempotent); must NOT block the booking if Studio is slow (fail-open) | NEW 🎨 "Studio projekt létrehozva" · by system | LIVE ✅PASS |
Step 1 — live test transcript① Before the shot — “What should I see?”
After booking a URL-present deal, exactly one indigo chip "🎨 Dizájn készül" on the card, and a Studio project created and bound ( 📸 screenshot taken
frames/s2-01-chip.png · 1480×1000 · captured 21:22:10 · ZZ Studio Walk ② What to look for (exactly)
Confirm a single "🎨 Dizájn készül" chip (not a cluster of widgets). Read the DOM for ③ What I see — Gemini · 5-sentence pixel report
One indigo "🎨 Dizájn készül" chip on the card; DOM read confirms ④ Verdict — pass or error?
Pixels (single chip) ∧ state (project bound) ∧ idempotency (no duplicate) all hold. → PASS (LIVE ✅). ⑤ Next action
PASS → Watch the generation step — expect RENDERED, because the Lab spend/artifacts are back-end and not pixel-visible. | |||||||
| 2 | (Automatic) the Lab is kicked to generate the Phase-① draft | Studio begins generating a homepage design + SEO audit + suggested structure from the saved URL | Copy: chip stays "🎨 Dizájn készül" with a progress shimmer · Look: animated chip · Where: card status chip | Lab run started (scrape → SEO audit → structure → homepage); Gemini/DataForSEO spend on this paying-intent lead | Must NOT auto-send anything to the prospect; Lab kick is internal-only; no new Modal cron added | NEW ⚙️ "Dizájn generálás elindítva" · by system | RENDERED ⚠️UNVERIFIED |
Step 2 — live test transcript① Before the shot — “What should I see?”
The chip should stay "🎨 Dizájn készül" with a progress shimmer while the Lab runs (scrape → SEO → structure → homepage). Gemini/DataForSEO spend is incurred underneath, and nothing is sent to the prospect. 📸 screenshot taken
frames/s2-02-generating.png · RENDERED (not driven to completion) ② What to look for (exactly)
Confirm the chip still reads "készül" with a shimmer and there is no error state. Note in advance: the actual Lab run/spend is not something a screenshot can prove — flag it accordingly. ③ What I see — Gemini · 5-sentence pixel report
Gemini's call: not a clean pass — confidence medium, because the shimmer is present but completion can't be seen. The chip reads "🎨 Dizájn készül" with a subtle shimmer; the card is otherwise healthy; there is no visible error; completion is not yet shown. 🔬 Sonnet — second-judge adjudication (Gemini did not return a clean pass)
Re-examining the flagged frame against the spec: the shimmer is a legitimate loading / working state, not a product defect — so this is not a BUG. But it also cannot be GREEN: the asserted state-change (Lab run started + Gemini/DataForSEO spend) lives in the Studio back-end and is absent from these pixels. Adjudication: down-rank to RENDERED ⚠️ / UNVERIFIED rather than PASS or BUG, and require a Studio run-state assertion (API) to settle it. ④ Verdict — pass or error?
The surface renders correctly, but the asserted state-change (Lab run + spend) is back-end and not verified by pixels. Per Sonnet's adjudication → tagged RENDERED ⚠️, which cannot be GREEN → UNVERIFIED (not an error, but not a pass). ⑤ Next action
UNVERIFIED → Re-shoot once the Studio run reaches "ready" and assert run-state via the Studio API — the state assertion is what upgrades this frame to LIVE ✅. | |||||||
▸ S3 Booked on a red-❗ deal → NO auto-kick; saving a URL resumes the flow failure / deferred studio-S3 ✗→✓ 2 frames · PASS after fix
| # | You do | You should see | Element that changes copy · look · where |
What changes underneath | Must NOT happen | 🕓 Touchpoint history | 🧪 Test |
|---|---|---|---|---|---|---|---|
| 1 | The sales call is booked while the card still shows the red ❗ | NO Studio project, NO Lab kick; the card prompts: "Add meg a weboldalt a Studio dizájnhoz" | Copy: prompt "Weboldal szükséges a dizájnhoz" · Look: ❗ stays + inline "Weboldal megadása" field · Where: card body | Studio create/kick suppressed; deal flagged studio_deferred=true |
Must NOT auto-create a blank project; must NOT silently skip; must NOT fake-run a design on a missing URL | SYSTEM ⏸️ "Studio dizájn elhalasztva" · by system | LIVE ✅PASS |
Step 1 — live test transcript① Before the shot — “What should I see?”
Booking a call while the red ❗ is present must not create a project or kick the Lab. Instead the card should show an inline "Weboldal megadása" field, and the deal should be flagged 📸 screenshot taken
frames/s3-01-deferred.png · 1280×800 · captured 21:26:33 ② What to look for (exactly)
Confirm there is no "🎨 Dizájn készül" chip; confirm the red ❗ persists; confirm an inline URL field rendered; read state for ③ What I see — Gemini · 5-sentence pixel report
The red ❗ persists; an inline "Weboldal megadása" field is rendered in the card body; no "Dizájn készül" chip exists; the kick was suppressed. ④ Verdict — pass or error?
The kick is correctly deferred (no blank project, no fake run) and the rep is prompted for the URL. → PASS (LIVE ✅). ⑤ Next action
PASS → Paste a valid URL into the inline field and confirm the S2 create+kick resumes on save alone (step 2). | |||||||
| 2 | Mátyás pastes a valid website URL into the inline field and saves | The ❗ clears and the normal S2 flow resumes automatically: project created + Lab kicked | Copy: ❗ gone; chip flips to "🎨 Dizájn készül" · Look: red badge → indigo chip · Where: card | website_url saved; studio_deferred cleared; same create+kick as S2 fires |
Must NOT require re-booking; the save alone must resume it | NEW 🎨 "Studio projekt létrehozva" · "Weboldal megadva" · by operator | LIVE ✅PASS |
Step 2 — live test transcript · 2 attempts (✗ Attempt 1 → ✓ Attempt 2)① Before the shot — “What should I see?”
After pasting a valid URL and saving, the red ❗ should clear and the normal S2 flow should resume automatically — project created + Lab kicked — without any re-booking; Attempt 1 LIVE ✅BUG · 21:27:40
📸 screenshot taken — attempt 1
frames/s3-02-retry1-FAIL.png · 1280×800 · captured 21:27:40 ② What to look for (exactly)
After save: confirm the ❗ cleared, an indigo "🎨 Dizájn készül" chip appeared, and the DOM shows exactly one new project bound to the deal — with no re-booking. ③ What I see — Gemini · 5-sentence pixel report
Gemini's call: not a pass. The red ❗ has cleared, but the card shows no design chip at all; the status-chip slot is empty; nothing indicates a Studio project spun up; the card otherwise looks idle. 🔬 Sonnet — second-judge adjudication (Gemini did not return a clean pass)
Adjudicating against pixels + DOM: the ❗ did clear, but there is no chip and the DOM shows no Studio project → a genuine BUG (the URL-save did not re-fire create+kick), not a Gemini misread or a transient loading state. ④ Verdict — pass or error?
Resume did not fire on save alone → BUG (LIVE ✅ — the failure is real, not a render artifact). ⑤ Next action
ERROR → Back up (git bundle), spawn a builder sub-agent (own worktree, TDD) to wire the URL-save to Attempt 2 · after fix LIVE ✅PASS · 21:28:12
📸 screenshot taken — attempt 2 (after the builder fix + redeploy)
frames/s3-02-retry2-PASS.png · 1280×800 · captured 21:28:12 ② What to look for (exactly)
Same checklist: ❗ gone, indigo "🎨 Dizájn készül" chip present, exactly one new project in the DOM, ③ What I see — Gemini · 5-sentence pixel report
Gemini's call: clean pass (conf high). The badge is gone; an indigo "🎨 Dizájn készül" chip now renders in the status slot; DOM confirms a single new project bound to the deal; ④ Verdict — pass or error?
On attempt 2 the resume fires on save alone (one project, chip flipped, no re-booking). Gemini is a clean pass → no Sonnet adjudication needed. → PASS. ⑤ Next action
PASS → Scenario complete; advance to S4 (open the Phase-① draft from the card). error-arc — full history kept (both frames above stay in this toggle)
ERRORAttempt 1: URL-save cleared the ❗ but did NOT re-fire create+kick — no chip, no project. Violated "save alone resumes it".
FIXBuilder sub-agent wired the URL-save handler to
ensure_studio_project(); cleared studio_deferred in one transaction (TDD red→green). Redeployed via snap_deploy.PASSAttempt 2: save alone resumed; one project, chip flipped, no re-booking. Verified LIVE.
| |||||||