Testing artifacts · 2026-08-22 · brain-brand-voice CT-BVOICE-13/14/15 · post-fix re-verify (PRs #3840, #3842, #3845)

Brand Voice — three post-fix scenarios, all PASS

Real browser against https://platform.mystflo.com/brain and /assistants, signed in as Peyton WP (peytonworkprojects@gmail.com). Blocked earlier in the same day's session on Chrome browser-tool unavailability (CLI/native-host version skew); resumed once reconnected in a fresh session. Ran in the order CT-BVOICE-15 → CT-BVOICE-14 → CT-BVOICE-13, deliberately — CT-BVOICE-15 needs a stale voice as its precondition, and a successful CT-BVOICE-13 regenerate clears staleness.

Verdict

Scenario	PR	Result	Why
CT-BVOICE-15	#3845	PASS	Stale-voice nudge live on both surfaces (Branding row, Brain Check), non-blocking, no charge on open.
CT-BVOICE-14	#3842	PASS	All 3 composers (Tweak, Share-what-to-change note, pending-preview Regenerates note) restore text on reopen, clear cleanly on cancel.
CT-BVOICE-13	#3840	PASS	"You, in one line" now renders a specific voice sentence, not the stale "friendly" fallback.

Precondition drift, noted, did not block: the plan's authoring snapshot named the QA account's business "The Offline Club® | Bali" at 88% playbook. This session found it as "QA Fixture Business" at 63% — most likely another parallel session's activity on this shared QA account. The staleness precondition CT-BVOICE-15 depends on held true regardless of the specific business name/percentage, so all three scenarios ran as specified.

1. CT-BVOICE-15 — stale-voice nudge, live on both surfaces

The Branding row's "How you sound" sub-line on /brain renders a muted, non-blocking nudge:

"Your playbook changed since this voice was written — worth a regenerate."

Branding row showing the stale-voice nudge line — Branding row, /brain — the nudge line renders live under "How you sound."

The same signal appears in Assistants → On-Brand Support → Connect tab → Brain Check module, as a ⚠ non-blocking attention row under BRANDING — styled identically to the other non-blocking items (Hours, profile touched today, Questions). The "Complete Setup" CTA stayed enabled the whole time, confirming this groups under attention, not blockers.

Brain Check module showing the How you sound attention row — Brain Check module, Connect tab — "How you sound" renders as a non-blocking attention row, same visual treatment as Hours/Questions/profile-touched.

Clicking either surface's nudge routes to the identical "How you sound" sheet / Regenerate composer — confirmed single entry point, not a duplicate. Checked read_network_requests immediately after opening on both paths: only a Clarity analytics beacon fired, zero calls to brain-generate or any billing endpoint — opening the composer alone fires no charge. Closed without submitting on both checks.

2. CT-BVOICE-14 — three composers, no lost text on close

Each composer was probed the same way: type distinctive marker text → close without saving/submitting → soft reopen → confirm restored → explicit cancel/clear → reopen → confirm no residue.

Composer	Marker text	Restore on reopen	Clears on cancel
Tweak textarea	`\|\|CTBVOICE14-TWEAK-PROBE-2026-08-22\|\|`	Auto-opened, text present	Yes, no residue
Share-what-to-change note	`CTBVOICE14-REGENNOTE-PROBE-2026-08-22`	Auto-opened (with $5.00 button), text present	Yes, no residue
Pending-preview "Regenerates" note (step 7, opportunistic)	`CTBVOICE14-STEP7-PENDINGNOTE-PROBE`	Text present on reopen	Yes, no residue

Step 7 was exercised live during CT-BVOICE-13's pending preview (that scenario's real regenerate was already fired) — reusing it avoided firing a second, unauthorized paid generate just to reach this field. Clicked Cancel, deliberately never "Write new preview" (that button fires a second real charge).

No ai_usage row and no brain_profile write observed from any open/type/close/reopen/clear activity across all three composers — only network traffic seen anywhere in this scenario was a Clarity beacon.

3. CT-BVOICE-13 — the real regenerate

Real ~$5 charge, fired this sitting with Peyton's explicit per-sitting sign-off (obtained before firing).

Pre-run, "You, in one line" still read the pre-fix stale value:

How you sound sheet before regenerate, one-liner reads friendly — Pre-run: "You, in one line" = "friendly" — confirms the fix does not retroactively rewrite old saved data, only future generations.

Opened "Share what to change → Regenerate," price shown $5.00 before submit, submitted with an empty/neutral note. Landed on the pending-preview three-door review sheet. Wallet debited $231.10 → $226.02 ($5.08 — real cost × markup).

New "You, in one line": "You write warm and plain — short answers, contractions, a quick reassurance like 'happy to help', and you tell people exactly how things work with no small print."

Pending preview sheet showing the new specific one-liner — Post-regenerate: a specific, grounded voice-summary sentence — not "friendly," not a verbatim `audience_one_liner` reuse, not identical to the pre-run value. Closure bar met.

Disposition: Discarded (Peyton's explicit call this sitting, matching the plan's stated default of preserving the account for future re-tests). The sheet reverted byte-identical to the pre-run state.

Branding row after discard, nudge line still present — Post-discard: nudge line still present (voice was never re-saved, so staleness holds) — confirms Discard preserved the account's pre-run state.

Closure

PR #3840 — "real paid Regenerate on the funded QA account must show a new specific voice sentence in 'You, in one line' (not 'friendly')" — closed.
PR #3842 — "the implementing agent had no browser tool available... flagging for a follow-up Chrome pass" — closed.
PR #3845 — "Branding row shows the nudge line, Brain Check shows brand_voice as attention not blocking, no charge fires until Regenerate is explicitly clicked" — closed.

Full write-up: docs/testing/click-through/brain-brand-voice-findings-2026-08-22-ct13-15.md. Full scenario specs: docs/testing/click-through/brain-brand-voice.md (CT-BVOICE-13/14/15).