← Testing HQ
Brain seeded journey · Round 3 of 3

Ambre Studio

Instagram-only fork (@ambrestudio). Account: peytonworkprojects@gmail.com, wiped clean beforehand (all 11 Brain tables verified 0 rows). First round with a real founder name present in the source material — the first genuine live test of the [PERSON_NAME] redaction bug.

Status: Complete
Cost: $27.01 → $25.80 (net −$1.21, elevated by a discarded-and-redone first attempt)
Run date: 2026-07-30
Interceptor: installed before first request — 0 hits on /rest/v1/* confirmed this round
1 · Raw image extraction (verbatim)

Ground truth established by reading all 4 source files directly BEFORE looking at any product output.

Image 1 — "Medicine Drawing" post (no tiling, normal aspect)
POST brain-generate
REQUEST: {"template_id":"TMPL-BRVX01","section":"voice_sample","mode":"vision","config":{"image":{"base64":"<2,549,392 chars, omitted>","mediaType":"image/png"}}}
RESPONSE:
{
  "success": true, "mode": "vision",
  "extracted_text": "What needs to be seen\nCiò che deve emergere\n\n—ENG\nA session dedicated to your soul.👁\n\nMedicine Drawing is a profound journey to quiet the mind and receive what emerges:\n\n⋱ The Silence | I guide you through a shamanic meditation—a deep journey to still the mind and open the heart.\n\n👁 The Channel | While you rest in your inner space, I channel images, symbols, and frequencies for you, creating your personal work of art\n\n↭ Your Mark | Following your instinct, you express what you feel through your own artistic process (no experience needed, only your intuition).\n\n🧎 The Guide | We conclude together, giving voice to what emerged. We integrate and compare the drawings in a moment of contemplation. I will guide your reflection and understanding.\n\n1:1 Sessions available Online and in person.\n📩 DM\n\nWith love\nAmbre\n\n#drawmedicine #arteterapia #disegnomedicina #channeling #ubud",
  "structured": {
    "voice_notes": "Spiritual, gentle, introspective, and professional. The tone is inviting and focused on healing and self-discovery.",
    "visual_style": "Earthy and artistic. The image features a hand-drawn, abstract eye symbol in bold colours (red, blue, black) on white paper, held by a hand with gold rings. The background is a warm, wooden surface.",
    "offers_or_events": ["1:1 Sessions available Online and in person"],
    "writing_style": "Structured with clear headings using symbols (⋱, 👁, ↭, 🧎). Uses a mix of English and Italian. Professional yet warm, with minimal emoji usage."
  },
  "run_id": "5dbd3eb1-edaf-482f-8a3f-01c759752648", "model": "google/gemini-3.1-flash-lite"
}

organize-content after this alone: business_facts.business_name = "emergere" (a fragment of "Ciò che deve emergere" — wrong, not a business name at all), founder_facts.founder_name = "Ambre" (correct, first appearance).

Image 2 — TikTok EU Summit post (no tiling)
RESPONSE:
{
  "extracted_text": "ambrestudio During the Tik Tok Eu summit I was invited by @tiktok_it to lead a healing and art experience with women from all over Europe\n\n@francescatittarelli @morningritual.club organized the \"The empower(her) Dinner\": a night to celebrate the women of TikTok — creators, visionaries, and leaders shaping the future.\n\n👁️ During the dinner, each woman received a personalized canvas I had created for\n\n👁️During the dinner, I performed a live painting — a celebration of femininity, ancestral wisdom, flowers, goddesses, and raw emotion. Together, we co-created a final artwork by adding messages, drawings, or signatures on the canvas\n\n✳️ A SHARED MOMENT. A SHARED CREATION. A MEMORY THAT STAYS ✳️\n\nSo thankful to have been part of this powerful feminine circle.🙏\n\n#femininecircle #tiktok #companysummit #womeninspiringwomen #milanevent #livepainting #artistlive #artista #divineartworks #customizedevent #girlspower💪❤️",
  "structured": {
    "voice_notes": "Professional, reflective, and appreciative. The tone is elevated and community-focused, emphasizing empowerment and shared experiences.",
    "offers_or_events": ["The empower(her) Dinner"],
    "writing_style": "Uses specific emojis (👁️, ✳️, 🙏, 💪, ❤️) for emphasis and structure..."
  },
  "run_id": "ff69bff5-b39c-4088-a0fb-92106c5a7769", "model": "google/gemini-3.1-flash-lite"
}

organize-content after images 1+2: business_facts.business_name = "Ambre" (now the founder's personal name used as the business name — also wrong).

Image 3 — "THE FIRST HOUSE" post (no tiling)
RESPONSE:
{
  "extracted_text": "ambrestudio Some codes only needs to be remembered, as we already know. ☂️🗝️A\n\nHere some details from my last work soon unveiled.\n\nGolden codes on raw canvas\n\"THE FIRST HOUSE\"\n120x113cm\n✉️",
  "structured": {
    "voice_notes": "Mystical, artistic, and slightly cryptic. The tone is elevated and intentional, focusing on the spiritual or symbolic nature of the artwork.",
    "offers_or_events": [],
    "writing_style": "Short, poetic sentences. Uses specific emojis (umbrella, key, envelope) as punctuation..."
  },
  "run_id": "8614b8bd-7986-4d5a-8cbe-e5ac31f62d51", "model": "google/gemini-3.1-flash-lite"
}

organize-content after images 1+2+3: business_facts.business_name = "ambrestudio" (now the IG handle, lowercase).

Image 4 — profile grid (2560×6776, aspect 2.65), "reading it in 3 parts" — TILED
Tile 1/3: extracted_text: "ambrestudio\nAmbre | Artist & Art Therapist\n343 posts 7,184 followers 528 following\nArt as expression of the soul ☉ Bali\n✳ Disegno medicina ✳ Channeling\n✳ Art Therapy 1:1 online ✳ live\n✳ Medicine drawing... more\nlinktr.ee/AmbreStudio\nMy soul My ...\nFormazione\n1:1 live\nMy Artworks\nWearable Art\nWorkshops\nTik Tok Collab\n1:1 MEDICINE DRAWING\n1:1 DISEGNO MEDICINA\nQuesto è il tuo segno\ndi venire a Bali con me"

Tile 2/3: extracted_text: "Questo è il tuo segno di venire a Bali con me\nTHE SECRET\nART HEALS YOUR INNER CHILD\nNOT ALL ART IS MEANT TO BE \"NICE\"\nSOMETIMES IT'S MEANT TO HEAL\nNON TUTTA L'ARTE NASCE PER ESSERE BELLA\nA VOLTE NASCE PER GUARIRE\nYour soul speaks in symbols\nI just translate them\nLa tua anima parla per simboli, io li traduco\nART FOR THE SOUL\nThank you\nCASA MAGA\nWHEN YOU FEEL HEAVY INSIDE\n28TH MARCH\n2:30 - 4:30 PM\nMASTERCLASS\nI CONFINI\nCon Ambre\nTO MY POWER"

Tile 3/3: extracted_text: "WHEN YOU FEEL HEAVY INSIDE\nI CONFINI\nCon Ambre\nCASA MAGA\nTO MY POWER\nARTE TERAPIA\nONLINE | 1:1"

organize-content after all 4 images: business_facts = {"business_name":"AmbreStudio","description":"Art as expression of the soul, offering Medicine Drawing, channeling, and art therapy sessions."}, founder_facts = {"founder_name":"Ambre","founder_role":"Artist & Art Therapist"}.

2 · Analysis — §8.5 hard pass bar

Graded against the actual image files, read directly, before looking at any product output.

MarkerGround truthProduct outputVerdict
Glyph bullets⋱ 👁 ↭ 🧎 as line markersPreserved verbatim; writing_style names them explicitlyPASS
All-caps refrain"✳️ A SHARED MOMENT. A SHARED CREATION. A MEMORY THAT STAYS ✳️"Preserved verbatim, caps intactPASS
Quoted title + dimensions"THE FIRST HOUSE", 120x113cmPreserved verbatim, quotes intactPASS
Zero exclamation marksNone in any of the 4 captionsZero "!" anywhere in AI-generated content or the rendered playbookPASS
—ENG bilingual separator"—ENG" before the English half of the Medicine Drawing captionPreserved verbatimPASS
Non-native grammar slip"Some codes only needs to be remembered..."Preserved verbatim, un-corrected ("needs" not "need")PASS
This round clears the hard bar. No "warm, creative, spiritual" generic mush anywhere — every voice_notes/writing_style field names something specific and checkable. No generic-and-nice description found this round.
3 · Deliberate experiment — "My Business" root cause
The planned single-field "Use" click test couldn't run: this fork's Instagram-only image path only exposes ONE commit action ("Confirm all") — no per-field "Use" button exists for image-derived drafts (that control is specific to the website-research pathway from Round 1). So the batch-vs-per-field distinction wasn't directly testable here.

What happened instead was sharper: closing the composer without confirming silently discarded all 4 already-paid reads with no warning — required a full re-upload and re-read, paying twice. On the redo, the final organize-content call returned no business_name key at all in business_facts. Confirmed via the raw payload before clicking Confirm all; clicking it then produced the literal "My Business". Round 2's confirm (also batch-only) DID include a business_name key in its payload and saved correctly. Same confirm mechanism in both cases — the only variable that changed was whether business_name existed in the confirmed payload. Root cause later confirmed and fixed in #2988/PR #2997: the bootstrap default fires on every first-ever profile write regardless of the AI's output, and (before the fix) the required-field check read the placeholder as satisfying the requirement.

Side effect worth flagging on its own: discarding a closed-without-confirming session doesn't refund or warn about wallet spend already incurred on the reads that already ran — a founder who closes early loses both the money and the results silently. (Fixed as a warning-before-discard in #2988/PR #2997; refund-on-discard remains a separate Finance policy call, not yet decided.)
4 · Defect watch
D3 (Rounds 1-2)
[PERSON_NAME] / [ADDRESS] — did NOT appear this round
Checked every raw brain-generate/organize-content response and the full confirmed playbook page text — "Ambre" appears correctly, unredacted, everywhere it's real (founder_name, "With love / Ambre", "Con Ambre" in a confirmed offering). Genuinely surprising given Round 1's reproduction — worth reconciling: Round 1's occurrence came out of the website-research/reputation-read pipeline, not the Instagram-vision pipeline used here and in Round 2. Likely pipeline-scoped, not universal. (Guard now wired into the missing call sites in #2988/PR #2997 regardless.)
D6 (Rounds 1-2)
Masthead readiness — same bare format, reproduced again
"63% of required" — same bare format as Round 2, no denominator (AC-BRAIN-60 pattern holds).
NEW
business_name/about flapping — worse than Round 2
"emergere" (nonsense fragment) → "Ambre" (founder name as business name) → "ambrestudio" (handle) → "AmbreStudio" (final, reasonable). A founder stopping after image 1 would see their business name listed as "emergere."
organize-content call count — confirms Round 2's pattern
One call per uploaded image (not per tile) — 4 images → 4 organize-content calls, each cumulative. The profile image's 3 tiles share ONE organize-content call after all 3 finish, not 3 separate ones.
Wallet + billing line
$27.01 → $25.80 (−$1.21, elevated because the discarded first attempt's reads were paid for and not refunded, then redone). Billing → Recent Activity again shows only $0.00 "AI usage" rows. (Root-caused as a sub-cent display bug, fixed in #2988/PR #2997.)
NEW
FAQ mis-categorization
The entire glyph-bulleted Medicine Drawing service description (a non-question, structured offering explanation) got filed under "Customer questions & policies" as a single FAQ answer, not under About/What-you-sell.
NEW
D7 (concatenated voice text) at its worst so far
Branding's "how you sound" field is 10 separate voice_notes descriptions run together in one paragraph blob (one per image/tile read) — same pattern as Rounds 1-2, scaled up with more images. (Interim fix — de-duplication before join — shipped in #2988/PR #2997; full synthesis pass remains a follow-up.)
NEW
Duplicate offerings, still not deduped
"1:1 MEDICINE DRAWING" appears at least twice plus a lowercase "Medicine drawing" variant among the 10 "what you sell" items.
5 · Post-fix live verification (2026-07-31)

After #2988's fixes deployed to production (brain-profile v50→v51, brain-ops v92→v93), this exact account's existing data was re-checked live — no new AI reads, zero additional cost.

A1 fix confirmed live, for free: the same stored "My Business" value that used to satisfy the readiness check no longer does. Masthead dropped from 63% → 50%, and "About your business" now shows "1 to answer" where it previously showed nothing — the required-field guard correctly re-flagging already-corrupted data the moment the fix deployed.