Calendo Computer-Use Test Suite — Coverage Matrix

This document is the traceability matrix for the Calendo "Computer Use Test" suite: a set of browser-agent (computer-use) runbooks executed against production https://calendo.dev. It proves what is covered, at what verification tier, and lists every known gap honestly.

How to read this document

Verification tiers

Tier	Meaning
L1	UI-only — the agent observes the rendered page/state in the browser
L2	Calendo persistence — state survives reload / is read back from Calendo's own backend (D1, API)
L3	External reality — a real Google/Outlook calendar event and/or a real email, confirmed by opening Gmail/Calendar in the same browser

Suite index (21 authored suites)

Suite	Title	Pri	Accounts	Lane	Est (min)
CU-01	Core booking lifecycle (book → reschedule → cancel)	P0	P1 host + anon invitee	B	30
CU-02	Auth lifecycle (register, verify, login, reset, delete)	P0	fresh throwaway	D	22
CU-03	Google Calendar integration (conflict, buffers, two-way)	P0	P1 + Google Cal	D	40
CU-04	Microsoft / Outlook calendar integration	P2	Outlook acct	D	35
CU-05	Event-type config → public booking-page enforcement	P0	P1 host	B	85
CU-06	Availability engine (weekly, overrides, holidays, slot-debug)	P0	P1 host	X (exclusive)	35
CU-07	Host-side booking management	P1	P1 host	B	40
CU-08	AI booking chatbot (public page)	P1	anon invitee	B	30
CU-09	AI dashboard assistant (feature parity)	P1	P1 host	B	45
CU-10	Landing + marketing + static pages + mobile	P1	none (anon)	A	15
CU-11	Public booking page UX (timezones, nav, QR, mobile)	P1	anon	A/B	22
CU-12	Routing forms (build → submit → route → analytics)	P2	P1 + anon	B	18
CU-13	Meeting polls (create → vote → tally → finalize)	P2	P1 + anon	B	18
CU-14	Team / org scheduling (roles, round-robin, collective)	P2	P1 owner + P3	D	40
CU-15	Contacts, analytics dashboard, CSV export	P2	P1 host	A	25
CU-16	Settings & customization (profile, branding, BYOK, pixels)	P2	P1 host	B	30
CU-17	Slack notifications & outbound webhooks	P2	P1 host	B	30
CU-18	New-user onboarding wizard (4-step)	P2	fresh throwaway	D	22
CU-19	Embeddable booking widget (inline/popup/badge)	P3	P1 + external page	B	25
CU-20	Email sequences, reminders, reconfirmation (time-gated)	P3	P1 host	B	40
CU-22	Chrome extension for Gmail (manual-led)	P3	P1 (manual)	manual	20

Lanes / wave plan (see 00-setup-preconditions.md): Lane A = read-only (anytime); Lane B = host-writers using uniquely-named event types (parallel-safe); Lane D = account-isolated (fresh throwaway / teammate / Outlook). Wave 1 runs all A/B/D suites concurrently; Wave 2 runs CU-06 alone because it rewrites global host availability.

1. Feature-area → suite traceability matrix

Rows are capabilities aggregated from every suite's coverage_items, grouped logically. The Primary suite owns deep verification; Also touches suites exercise the area incidentally. The highest tier reached anywhere is shown.

1.1 Authentication, account & onboarding

Capability	Primary suite	Also touches	Top tier
Email/password registration (happy path, verification gate)	CU-02	CU-18	L2
Email verification email + link → account verified	CU-02	CU-18	L3
Resend verification email	CU-02	(CU-18 noted)	L2
Unverified-user dashboard gating → `/auth/verify.html`	CU-02	CU-18	L1
Login happy path / wrong-password failure	CU-02	—	L1/L2
Forgot-password request + reset-password email + token flow	CU-02	—	L3
Old-password-rejected / new-password-accepted after reset	CU-02	—	L2
Account deletion (Settings → Danger Zone, double confirm)	CU-02	CU-18	L2
Post-deletion login fails / dashboard inaccessible (401)	CU-02	CU-18	L2
OAuth button presence (Google/Microsoft) on auth pages	CU-02	CU-10	L1
Standalone `/onboarding/` calendar-connect page	CU-18	—	L1
In-dashboard 4-step wizard (Calendar/Schedule/Location/AI)	CU-18	—	L1
Onboarding completion persists server-side (`onboarding_completed=1`)	CU-18	—	L2
Empty-availability schedule-setup nudge + dismissal persistence	CU-18	—	L2

1.2 Booking lifecycle (invitee-driven)

Capability	Primary suite	Also touches	Top tier
Anonymous invitee books a slot (name + email → Confirm)	CU-01	CU-05, CU-08, CU-11, CU-19, CU-20	L3
Booking Confirmed page (event/duration/date/time/tz/host)	CU-01	CU-11	L1
Confirmation page tokenized Reschedule/Cancel + add-to-calendar	CU-01	CU-11	L1
Booking persists in dashboard Upcoming, survives reload	CU-01	CU-07, CU-08, CU-20	L2
Invitee self-serve reschedule via token (`reschedule.html`)	CU-01	CU-07, CU-17	L3
Reschedule idempotency / already-rescheduled handling	CU-01	—	L1
Invitee self-serve cancel via token (`cancel.html` + reason)	CU-01	CU-16, CU-17	L3
Cancel idempotency / already-cancelled state	CU-01	—	L1
Confirmation / reschedule / cancellation emails (invitee + host)	CU-01	CU-03, CU-20	L3
Test event-type cleanup (no stray calendar events)	CU-01	—	L2

1.3 Calendar integration (Google & Outlook)

Capability	Primary suite	Also touches	Top tier
Google Calendar shows Connected; persistence	CU-03	—	L2
Google one-time OAuth connect (incl. "unverified app" warning)	CU-03	—	L1
Outbound conflict blocking: real GCal BUSY removes overlapping slot	CU-03	CU-06 (reason code)	L3
Non-overlapping slots remain (targeted conflict, not outage)	CU-03	—	L3
Per-event-type before/after buffers suppress adjacent slots	CU-03	CU-05, CU-06	L3
Booking creates real GCal event `<event> with <invitee>` + attendees	CU-03	CU-01, CU-07, CU-08, CU-14	L3
Reschedule moves the GCal event (old empty, new populated)	CU-03	CU-01, CU-07	L3
Cancel deletes the GCal event (`sendUpdates=all`)	CU-03	CU-01, CU-07	L3
Inbound two-way sync: GCal-created event appears in Calendo Calendar tab	CU-03	—	L2
Google Meet conferenceData link attached (spot-check)	CU-03	CU-05	L1
Outlook connection confirmed + persistence	CU-04	—	L2
Outlook OAuth connect (Microsoft consent, scopes, unverified warning)	CU-04	—	L1
Outlook BUSY event suppresses overlapping slot (Graph calendarView)	CU-04	—	L3
Outlook busy/tentative/oof = conflict; free/workingElsewhere = not	CU-04	—	L3
Deleting Outlook busy event restores slot (live, non-cached)	CU-04	—	L3
Booking writes real Outlook event + attendee	CU-04	—	L3
Cancel removes the Outlook event	CU-04	—	L3
Outlook primary-calendar dropdown lists the connection	CU-04	—	L1

1.4 Event-type configuration & enforcement

Capability	Primary suite	Also touches	Top tier
Event-type CRUD (overview + New flow, edit modal)	CU-05	CU-01, CU-07, CU-09	L2
Duration + slot interval/granularity (30-min spacing) enforced	CU-05	CU-11	L1
Description text rendered on public page	CU-05	—	L1
Event color (non-default swatch) reflected on public page	CU-05	—	L1
Location config (in-person address / Google Meet video)	CU-05	—	L1
Custom questions (text + select) render + required validation	CU-05	—	L1
Saved custom-question answers persisted, visible in booking detail	CU-05	—	L2
Conditional questions show/hide + conditional validation	CU-05	—	L1
Buffers before/after reduce offered slots vs zero-buffer twin	CU-05	CU-03, CU-06	L1
Per-day meeting cap (`max_per_day=1`) blocks further same-day slots	CU-05	—	L1
Minimum scheduling notice hides near-now slots (72h)	CU-05	CU-06	L1
Timezone lock disables selector, shows "(locked)"	CU-05	CU-11 (noted)	L1
Post-booking redirect URL with invitee/event params	CU-05	—	L1
Secret event (hidden from list, bookable via `?event=`, badge)	CU-05	—	L1/L2
Group event (max attendees, "X spots left", cap enforced)	CU-05	—	L1
Prefilled booking link `?name=&email=` auto-fills form	CU-05	—	L1
Allow-guests toggle (section shown/hidden, guest stored)	CU-05	CU-07	L2
Guest-notification email to P3 inbox	CU-05	CU-07	L3
Edit-panel setting persistence across reopen	CU-05	CU-09	L2

1.5 Availability engine

Capability	Primary suite	Also touches	Top tier
Weekly recurring schedule (Mon–Fri 09:00–17:00 = 40.0 hrs) + persistence	CU-06	—	L2
Public page offers only scheduled weekdays; weekends unavailable	CU-06	CU-11	L1
Date override that BLOCKS a weekday zeroes that day's slots	CU-06	—	L1
Date override that OPENS a normally-off Saturday surfaces slots	CU-06	—	L1
Holiday preset (US 2026) auto-blocks an upcoming holiday	CU-06	—	L1
Holiday renders blocked in override mini-calendar (tooltip)	CU-06	—	L1
Per-event buffers (30/30) + min-notice (48h) reduce slots vs default	CU-06	CU-05	L1
Multiple named schedules drive different event types independently	CU-06	CU-09	L2
Schedule-level daily meeting limit persists	CU-06	—	L2
`slot-debug` endpoint returns `available` + `reasons[]`	CU-06	—	L2
`slot-debug` reason codes (no_schedule, outside_hours, date_override, holiday, min_notice, max_future)	CU-06	—	L2
Create/select/delete non-default schedule; default protected	CU-06	CU-09	L2
Baseline capture + full restoration of global schedule	CU-06	—	L2

1.6 Host-side booking management

Capability	Primary suite	Also touches	Top tier
Book-on-behalf from dashboard (`/api/book-on-behalf`)	CU-07	CU-15 (seeding)	L3
Bookings filters (Upcoming/Past/Cancelled; no_show exclusion)	CU-07	CU-15	L1
Meeting search by invitee name	CU-07	—	L1
Booking detail sidebar (status/title/datetime/invitee/location/timeline)	CU-07	—	L1
Mark past booking as no-show	CU-07	CU-09, CU-15	L2
No-show reflected in Analytics No-Show Rate	CU-07	CU-09, CU-15	L2
Add internal note via sidebar, persists across reload	CU-07	CU-09	L2
Attach a guest at booking time; shown in host detail	CU-07	CU-05	L3
Cancel from dashboard sidebar → Cancelled filter + badge	CU-07	CU-01	L3
Reschedule via token link (no host-side reschedule button exists)	CU-07	CU-01	L3

1.7 AI features (the differentiator)

Capability	Primary suite	Also touches	Top tier
Public-page AI chat surface presence (finding: hidden in prod)	CU-08	CU-11 (noted)	L1
Conversational NL request → AI reply with real slots (`/api/ai/chat`)	CU-08	—	L1
AI latency vs ~2s target (measured, not certified)	CU-08	—	L1
AI proposes only real slots (no hallucinated times)	CU-08	—	L1
Conversational slot pick → inline booking form → real booking	CU-08	—	L3
AI-chat booking appears in dashboard with correct details	CU-08	—	L2
Ambiguity handling ("next Tuesday") + impossible-request decline	CU-08	—	L1
Scope enforcement (off-topic refused) in public chat	CU-08	CU-09	L1
Grid fallback completes booking independently; AI⊆grid slot set	CU-08	CU-11	L1
429/529 rate-limit static fallback recognized	CU-08	—	L1
Dashboard AI panel open (`#aiPanel`, mobile FAB)	CU-09	CU-16	L1
`get_event_types` / `get_bookings` / `get_analytics` (read tools)	CU-09	—	L1
`create_event_type` / `update_event_type` via AI + reload verify	CU-09	—	L2
`get_booking_link` + `copy_to_clipboard`	CU-09	—	L1
`add_booking_notes` / `mark_no_show` via AI	CU-09	CU-07	L2
`create_availability_schedule` / `delete_availability_schedule`	CU-09	CU-06	L2
`get_embed_code` (inline + popup snippets)	CU-09	CU-19	L1
`navigate_to` (limited tab enum)	CU-09	—	L1
`get_audit_log` (AI writes are audit-logged)	CU-09	CU-14	L2
`manage_tracking (get)` / `manage_slack (get)` read-only probes	CU-09	CU-16, CU-17	L1
`delete_event_type` with confirm guardrail	CU-09	—	L2
Off-topic guardrail refusal	CU-09	CU-08	L1
Feature-parity gap detection (polls have no AI tool)	CU-09	CU-13	L1
False-success detection (tool claims success w/o state change)	CU-09	—	L2
BYOK key applied to assistant (inferred via fake-key auth failure)	CU-16	CU-09	L1
Welcome message set via AI assistant → public page	CU-16	—	L1

1.8 Public booking page UX & marketing/static pages

Capability	Primary suite	Also touches	Top tier
Timezone picker shifts displayed slot times + "Times in …" label	CU-11	CU-05 (lock)	L1
Locked-timezone behavior (disabled selector) noted	CU-11	CU-05	L1
Calendar month next/prev nav; prev disabled on current month	CU-11	CU-01	L1
Day selection reveals time-slots panel; unavailable greyed; today marker	CU-11	CU-01, CU-06	L1
Empty-availability handling ("No available times", not blank)	CU-11	—	L1
Event-list rendering (`.event-card`)	CU-11	CU-14 (team page)	L1
Brand color applied via `applyBranding()` (`--brand`)	CU-11	CU-16	L1
QR code render client-side (booking-link + per-event) + PNG download	CU-11	—	L1
Correction: confirmation page has NO QR (only calendar/manage links)	CU-11	—	L1
Mobile layout @375px (sidebar collapse, stacking, no overflow)	CU-11	CU-10	L1
Happy-path booking → "Booking Confirmed!" + add-to-calendar/manage	CU-11	CU-01	L1
Landing loads anon (no logged-out redirect to dashboard)	CU-10	—	L1
Hero AI messaging headline + subtitle + CTAs	CU-10	—	L1
Six feature cards with correct titles	CU-10	—	L1
How It Works (3 steps)	CU-10	—	L1
Pricing (Free $0 "Most Popular" + Pro $2/mo) cards + CTAs	CU-10	—	L1
"Why it's free" section + CTA	CU-10	—	L1
Header nav anchors smooth-scroll; Docs link → `/docs/`	CU-10	—	L1
Docs interactivity (expand/collapse, sidebar anchor, embed guide)	CU-10	CU-19	L1
Footer 5 links resolve (Docs/Login/Signup/Privacy/Terms)	CU-10	—	L1
Privacy + Terms pages render with expected headings/dates	CU-10	—	L1
Clean URLs and `.html` forms both resolve	CU-10	—	L1
Login + Register pages render forms + OAuth buttons	CU-10	CU-02	L1
Console free of JS/asset errors (desktop + mobile)	CU-10	—	L1
Mobile @375px: hamburger, drawer, no overflow, single-column stack	CU-10	—	L1

1.9 Routing forms & meeting polls

Capability	Primary suite	Also touches	Top tier
Create routing form (name, slug, fields, rules)	CU-12	—	L2
Conditional routing rules (first-match-wins ordering)	CU-12	—	L2
Form card shows field/rule counts + Active badge; persists	CU-12	CU-15 (surface)	L2
Public routing form loads anon; renders fields	CU-12	—	L1
Answer set A → route to event type A; set B → route to B	CU-12	—	L1
Required-field validation on public form	CU-12	—	L1
Routing analytics submission count increments	CU-12	CU-15	L2
Routing analytics conversion (0 because no booking) + per-ET breakdown	CU-12	CU-15	L2
Create meeting poll (title, desc, 3+ options); persists + "0 voters"	CU-13	CU-15 (surface)	L2
Resolve poll ID; correct public URL `/poll/?id=` (vs broken Copy Link)	CU-13	—	L1
Anonymous yes/maybe/no votes; multiple voter columns	CU-13	—	L1
Votes persist across reload; voter_count increments 0→2	CU-13	—	L2
Per-option tally + auto-computed "Best option"	CU-13	—	L1
Finalize poll → "finalized" badge + "Meeting confirmed!" + winner	CU-13	—	L2
`GET /api/polls/<ID>` confirms finalized status + winning option	CU-13	—	L2

1.10 Team / org scheduling

Capability	Primary suite	Also touches	Top tier
Create org as owner (auto slug); owner controls visible	CU-14	—	L1
Invite existing user (P3) → added immediately; persists	CU-14	—	L2
Role management (promote/demote); owner-only controls	CU-14	—	L1
Role constrains capability (admin vs member gating)	CU-14	—	L1
Round-robin team event type; union availability	CU-14	—	L1
Round-robin assignment rotates / least-booked (`assigned_user_id`)	CU-14	—	L2
Collective team event type; intersection slot logic	CU-14	—	L1
Public team page `/team/<slug>` lists org/members/event types	CU-14	CU-11	L1
Per-user audit/activity log loads + filters (event_type.created, booking.created)	CU-14	CU-09	L2
Team booking → invitee/host emails + GCal events	CU-14	CU-03	L3

1.11 Contacts, analytics & export

Capability	Primary suite	Also touches	Top tier
Contacts aggregate invitees (one row per unique email)	CU-15	—	L2
Contacts columns (Name/Email/Phone/Last/Next/Meetings)	CU-15	—	L1
Contacts search (name + email)	CU-15	—	L1
Contacts filters (All / New 30d / No upcoming)	CU-15	—	L1
Analytics headline stats (Total/ThisMonth/Cancel/No-Show/Popular)	CU-15	CU-07, CU-09	L2
30-day Booking Volume bar chart	CU-15	—	L1
Event Type Breakdown table (per-ET bookings/cancel/no-show)	CU-15	CU-09	L1
Quirk: `#exportRange` scopes CSV only, not on-screen numbers	CU-15	—	L1
CSV export `calendo-bookings-<date>.csv` — exactly 15 columns in order	CU-15	—	L2
CSV escapes commas/quotes; range param genuinely scopes export	CU-15	—	L2

1.12 Settings, customization & integrations

Capability	Primary suite	Also touches	Top tier
Profile display name persists + renders on public page	CU-16	—	L2
Gap: no timezone field in profile (edited on schedule instead)	CU-16	CU-06	L1
Cancellation policy persists + shows on confirmation/cancel pages	CU-16	CU-01	L1/L2
Brand color (`#ff5500`) persists + restyles public page (`--brand`)	CU-16	CU-11	L1
Logo URL persists + renders as `img.host-avatar`	CU-16	—	L1
Email blocklist (exact + `@domain`) rejects at `/api/book` (403)	CU-16	—	L2
BYOK Anthropic key save/active/remove; never echoes secret; persists	CU-16	CU-09	L2
GA Measurement ID + Meta Pixel ID injected at runtime into public head	CU-16	—	L1/L2
Welcome message via AI → `#host-welcome` on public page	CU-16	—	L1
Register outbound webhook (`POST /api/webhooks`, 201 + secret)	CU-17	—	L2
List webhooks (persistence)	CU-17	—	L2
`booking.created` webhook received (headers + signed body)	CU-17	—	L3
`booking.cancelled` webhook (`cancelled_by:host`)	CU-17	—	L3
`booking.rescheduled` webhook (`previous_*` + changed time)	CU-17	—	L3
Slack incoming-webhook save validation + Connected + Test round-trip	CU-17	—	L3
Slack `booking.created` formatted message (when real Slack available)	CU-17	—	L3
Webhook deletion (`DELETE /api/webhooks/:id`)	CU-17	—	L2

1.13 Embeds, email sequences & extension

Capability	Primary suite	Also touches	Top tier
Dashboard generates 3 embed snippets (inline/popup/badge)	CU-19	CU-09 (`get_embed_code`)	L1
Two distinct embed scripts serve as JS (not 404/HTML)	CU-19	—	L2
Inline iframe render with `embed=1` chrome stripping (foreign origin)	CU-19	—	L1
Popup modal overlay iframe (close via x/Esc/backdrop)	CU-19	—	L1
Floating badge pill → modal iframe	CU-19	—	L1
Iframable from non-calendo origin (no `X-Frame-Options: DENY`)	CU-19	—	L1
Cross-origin script execution (CORS `*`)	CU-19	—	L1
Full booking completed inside embedded iframe	CU-19	—	L2
`calendo:booking-confirmed` postMessage re-dispatched to host window	CU-19	—	L1
In-embed booking cleanup (cancel)	CU-19	—	L2
Create email sequence (trigger/delay/subject/body) + persistence	CU-20	—	L2
Template variable substitution in delivered sequence emails	CU-20	—	L3
Immediate booking-confirmation email content (subject/links)	CU-20	CU-01	L3
Host notification email content	CU-20	CU-01	L3
Cron-driven sequence engine (after_booking delay-0 window)	CU-20	—	L3
Reminder config (24h/1h) + reconfirmation config (24/48/72) persist	CU-20	CU-05	L2
Chrome MV3 extension loads (name/version/icons)	CU-22	—	L1 (manual)
Popup link validation + save (chrome.storage.sync) + copy	CU-22	—	L1 (manual)
Gmail content-script button injection + link insertion at cursor	CU-22	—	L1 (manual)
Inserted link resolves to live public booking page	CU-22	—	L2

2. Automation status

2.1 Fully automated (agent completes end-to-end in-browser, including L3 where claimed)

These suites/areas are driven to completion by the computer-use agent with no required human step inside the run (transient retries excepted):

CU-01 — Core booking lifecycle, including real GCal create/move/delete and Gmail email checks.
CU-02 — Auth lifecycle, including clicking real verification + password-reset links from Gmail.
CU-03 — Google Calendar conflict/buffer/two-way sync, including L3 calendar + Gmail evidence.
CU-05 — Event-type config → public-page enforcement (L1/L2 across all settings + L3 guest email).
CU-06 — Availability engine (L1/L2), run alone in Wave 2; fully self-restoring.
CU-07 — Host-side booking management (book-on-behalf, no-show, notes, guest, cancel/reschedule) with L3.
CU-09 — AI dashboard assistant feature-parity actions (read + write tools, reload-verified).
CU-10 — Landing/marketing/static pages + mobile (anonymous, read-only, L1).
CU-11 — Public booking page UX (timezones, nav, empty-state, QR render, mobile, one booking).
CU-12 — Routing forms build → submit → route → analytics (L1/L2).
CU-13 — Meeting polls create → vote → tally → finalize (L1/L2).
CU-15 — Contacts, analytics, CSV export (L1/L2, including CSV column/escape verification).
CU-16 — Settings & customization (profile, policy, brand, logo, blocklist, BYOK, GA/Meta pixels).
CU-17 — Outbound webhooks via webhook.site receipt + Slack Test round-trip (L3 where Slack available).
CU-19 — Embeddable widget (inline/popup/badge) on a foreign-origin page, with in-embed booking.

Note on CU-08 (AI booking chatbot): the suite is fully runnable, but its headline finding is that the public AI chat is intentionally hidden in production (inline #ai-chat-prompt is display:none and the floating chat-widget.js bubble is not imported). The agent verifies /api/ai/chat behavior directly and books via the conversational form, but the differentiator is "dark" in the live UI — a product decision, not a test failure.

2.2 Partially automated — has manual residue

These suites run but leave specific items a human must confirm out-of-band. Source: each suite's manual_residue.

Suite	Manual residue (must be confirmed by a human)
CU-01	Email deliverability/spam placement & non-Gmail rendering; `.ics` DTSTART/DTEND contents (agent can't parse); attendee-side RSVP delivery is ambiguous (invitee is a plus-alias of the host's own Gmail); whether GCal is truly two-way vs one-way write.
CU-02	In-dashboard verification banner unreachable in normal flow (dead code?); second delete dialog says "type your email" but offers no input; anti-enumeration forgot-password path only unit-tested; token expiry/reuse only unit-tested; can't confirm D1 rows purged on delete; OAuth flows not exercised.
CU-03	Webhook-driven inbound auto-cancel (delete GCal event → Calendo auto-cancel) non-deterministic in-browser; freeBusy/push latency may need retries; whether the "unverified app" warning still shows depends on Google status; watch-channel renewal cron unobservable.
CU-04	Whole suite is residue if the Microsoft/outlook.com session needs a mid-run password (no cold login); publisher-verification warning documented only; token-refresh/offline_access lifecycle can't be force-triggered; reminder/organizer semantics not asserted; email L3 residue if P1 Gmail not logged in.
CU-05	Per-event WEEKLY cap not settable in edit panel; SMS reminder delivery unverifiable; reconfirmation emails outside run window; primary-invitee email deliverability owned by notifications suite; deep GCal attendee/Meet verification owned by CU-03; buffered slot count verified qualitatively only.
CU-06	`slot-debug` reasons `google_calendar` / `outlook_calendar` need a connected conflicting calendar (deferred); `booking_conflict` / event-type `daily_cap` may be unexercised in a clean run (unit-tested); full meeting-limit emptying is unit-tested; DST-boundary correctness is unit-tested.
CU-07	No host-side dashboard Reschedule button (only token link); no API to add a guest to an existing booking; dashboard shows no-show rate not raw count; verifying the P3 guest email needs that inbox open; email timing depends on Resend; past-start book-on-behalf behavior needs human confirm.
CU-08	Product decision to actually expose AI chat (currently hidden); sub-2s SLA over time; 429/529 + Workers-AI llama-3.3 fallback quality (external); cross-timezone label correctness; chat session isolation across invitees (localStorage leakage).
CU-09	Embed snippet rendering on a real external page; no `create_poll` AI tool (parity gap to fix); no un-mark for no-show (human must restore real bookings); high-risk tools (delete_account, upgrade_to_pro, manage_paypal/slack set, trigger_password_reset, resend_verification, submit_feedback, update_settings) skipped on prod; can't independently verify the model self-ID claim.
CU-10	Subjective visual/aesthetic quality; SEO/meta/OG/favicon correctness; real OAuth consent flows; cross-browser + real-device rendering; whether `/docs` (no slash) 404s while `/docs/` works.
CU-11	Empty event-list state can't be triggered on P1 without deleting all event types; whether the downloaded QR PNG actually scans to the right URL; the exact intended P1 brand color; pixel-perfect mobile ergonomics; L3 reality for the Section G booking owned by CU-01/02.
CU-12	Conversion attribution (booked count / %) not verified (no confirmed booking); "No matching event type" 404 path not exercised; `contains` operator + multi-condition AND rules not covered; inactive-form 404 not covered; OS clipboard read not relied upon.
CU-13	No in-UI poll delete (D1 cleanup if needed); finalize produces no external artifact (no event/email) — confirm intended; defect: dashboard Copy Link copies broken `/booking/?poll=`; defect: dashboard polls-list Finalize button POSTs with no body and is rejected; deadline expiry untestable (no deadline field).
CU-14	Whether the round-robin assigned member actually got the GCal event/email can't be confirmed in-browser (side-effects route to booking-page owner P1); server-side 403 role enforcement only UI-verified; no delete control for org/team event types (D1 cleanup); org-level actions not written to audit log (possible gap); round-robin rotation non-deterministic.
CU-15	Contacts have no in-UI delete (RUNID rows persist); no-show-rate correctness needs real no-show data; routing/poll math not exercised end-to-end here; precise 90d/1y CSV cutoffs need bookings older than the windows.
CU-16	Can't prove the BYOK key is actually used for a billed call (only inferred from fake-key failure); profile timezone change propagation owned elsewhere; welcome-message editing has no dashboard input (AI/API only); encryption-at-rest of the key can't be observed in-browser; real GA/Meta ingestion out of scope.
CU-17	HMAC `X-Calendo-Signature` validity can't be recomputed in-browser (verify one out-of-band); real Slack channel rendering needs a real workspace; whether clearing a test Slack URL is safe depends on P1 usage; supporting Gmail emails are the email path, not the webhook path.
CU-18	Full click-through of wizard steps 2→4 needs a human with a pre-consented disposable Google account (Step 1 has no skip); the "OK" branch of the schedule-setup prompt only code-inspected; verification-email deliverability depends on Resend; resend-verification flow not exercised; wizard location → event-type persistence needs the blocked hand-driven path.
CU-19	Creating a true foreign-origin host page depends on agent browser capability (else human pastes snippets on a real site); inspecting the postMessage/CustomEvent needs JS injection in host context; exact response headers (CORS, no `X-Frame-Options`) need network-panel/curl; optional downstream email/GCal owned by other suites.
CU-20	Reminder timing (24h/1h) can't be confirmed in a short run (note quirk subject "in 1 hours"); reconfirmation delivery + Confirm/Cancel-Attendance clicks; after_booking delay-0 email may need a later Gmail re-check (cron lag); before/after_meeting trigger types configured-but-not-delivery-verified; SMS out of scope; Resend dashboard not openable.
CU-22	Loading the unpacked MV3 extension via `chrome://extensions` (agent can't); driving the real Gmail compose DOM (obfuscated selectors); cross-device `chrome.storage.sync` propagation (needs a 2nd signed-in Chrome); cloud-sync latency/quota; manual extension removal + storage cleanup.

2.3 Out of scope / TBD

A. Global out-of-scope (no test mode / not feasible for a browser agent) — never silently dropped

These are explicitly excluded across the whole suite and require separate manual/human handling:

Area	Why excluded
Stripe payments (paid bookings, holds, refunds)	Live mode only, no Stripe test mode in production.
PayPal	Live integration, no safe test path.
Coupons at checkout	Tied to the paid-booking flow (above).
Pro-upgrade flow	Real billing; not exercised on the production account.
All Pro-gated features	Includes SMS/Twilio reminders (also need a real phone), custom domains, and remove-branding. Setting persistence is sometimes checked, but delivery/effect is not.
Admin panel	Gated to a different hardcoded email; not accessible to P1.

These are intentionally listed, not dropped. If/when Stripe test mode, a phone inbox, or an admin account become available, dedicated suites should be authored: paid-booking lifecycle, coupon enforcement, Pro-upgrade, SMS reminders, custom-domain, and admin-panel.

B. Adjacent capabilities with no authored suite

No suite currently covers (treat as TBD / future work):

Native mobile apps — none exist; only responsive web (@375px) is tested (CU-10/11).
CRM integrations / SSO (SAML) / Zapier / generic third-party automation — no suites; only the generic outbound-webhook + Slack paths (CU-17) and the Chrome extension (CU-22) exist as integration coverage.

C. Per-suite TBD items (deferred to future or separate suites)

Suite	Deferred (`tbd_items`)
CU-01	Reminder emails (time-based); cross-timezone slot correctness; recurring booking lifecycle; paid/Stripe-held lifecycle; Outlook parity for the lifecycle; host-initiated dashboard cancel/reschedule.
CU-02	Full OAuth register/login (Google + Microsoft); browser-verified session-invalidation-on-reset; deep server-side data-deletion audit; rate-limiting on repeated attempts.
CU-03	Outlook conflict + two-way sync (→ CU-04); multiple/non-primary Google calendars; Google Meet conferenceData depth; behavior when `GOOGLE_CLIENT_ID` unset / refresh fails (unit-tested).
CU-04	Teams `OnlineMeetings` join-link verification; multi-Outlook-account aggregation; buffer × Outlook conflict interaction; Outlook events in dashboard Calendar tab.
CU-05	Per-event weekly/daily meeting-limit caps; SMS reminder E2E; reconfirmation scheduled-send; paid/coupon enforcement; recurrence options; round-robin/collective (→ CU-14).
CU-06	L3 external reality (out of scope for availability-only); team/collective slot computation; onboarding-wizard availability path; API validation error paths (unit-tested).
CU-07	Recurring-booking cancel scopes; reconfirmation status surfacing; merged external calendar events in list; paid-booking hold/cancel + webhook delivery; AI-assistant parity for host ops (→ CU-09).
CU-08	Dedicated timezone-conversion suite for AI slots; floating `chat-widget.js` bubble (once wired); formal `/api/ai/chat` latency benchmarking; Outlook L3 for AI-chat booking.
CU-09	No `create_poll`/meeting-poll AI tool (parity gap); `navigate_to` can't reach calendar/contacts/event-types tabs; high-risk billing/auth/integration tools deferred to throwaway account; embed render on external site.
CU-10	Register form submission; login/OAuth completion; Pro "Upgrade" destination (auth-gated); unlinked docs sub-pages; server-side HTTP status assertion.
CU-11	Whether a reserved always-empty event exists in prod; locked-timezone event coverage (needs fixture); AI inline chat on booking page (→ AI suites).
CU-12	E2E conversion tracking; editing an existing routing form (PUT); slug-collision 409; `contains` + multi-condition AND; no-match 404 path.
CU-13	Voting-deadline enforcement (no UI field); whether finalize should create a booking/event/email; poll edit/delete UI; poll votes in main Analytics.
CU-14	Email-based pending-invitation acceptance; booking via P3's own slug for member-scoped L3; org/team-event-type deletion (API/D1); per-member round-robin tuning; collective multi-attendee invites.
CU-15	E2E routing conversion (→ CU-12); E2E poll voting/finalize (→ CU-13); no-show lifecycle (→ CU-07); JSON export variant of analytics export.
CU-16	Real GA/Meta ingestion; E2E BYOK with a billed key; profile timezone propagation; Outlook-specific settings.
CU-17	Webhook retry/backoff + auto-disable after 5 failures; UTM propagation in payload; `PUT /api/webhooks/:id` update flows.
CU-18	Whether standalone `/onboarding/` and the 4-step wizard should unify (product Q); OAuth-signup onboarding (auto-verifies); mobile wizard rendering; mid-wizard calendar-connect resume logic.
CU-19	`data-calendo-event`/`-color`/`-text` customization (documented but not emitted by generator); badge color/text E2E; mobile/responsive embed rendering; http-vs-https mixed-content.
CU-20	SMS reminders/reconfirmation (Twilio); before/after_meeting sequence delivery; Resend delivery-log inspection; calendar event assertions (→ CU-03/04).
CU-22	Chrome Web Store packed distribution; true cloud-sync latency/quota; Gmail DOM/class drift resilience; Firefox/Edge ports; plain-text compose insertion.

3. Known blind spots (read this)

A few things no suite verifies, and a human should keep them in mind:

Money paths are entirely untested. Stripe, PayPal, coupons, the Pro upgrade, and every Pro-gated feature (SMS reminders, custom domains, remove-branding) have no automated coverage at any tier because production has no test mode and the agent has no phone inbox or billing-safe account. Where these settings exist in the UI, only persistence is checked — never the effect. If a paid feature regresses, this suite will not catch it.
The agent cannot inspect bytes it can't render. It does not parse .ics files, recompute HMAC signatures, read raw HTTP response headers/status codes, or read encrypted values at rest in D1. So ICS DTSTART/DTEND correctness, webhook signature validity, exact CORS/X-Frame-Options headers, soft-404 vs true-404, and BYOK key encryption are all asserted indirectly or handed to a human.
Time-gated delivery is structurally unprovable in a single run. 24h/1h reminders, reconfirmation windows, before/after-meeting sequences, and webhook retry/backoff/auto-disable all depend on cron crossing a future boundary; the suites verify configuration and persistence, not eventual delivery. Resend's actual delivery status (bounces, spam placement, non-Gmail rendering) is never visible to the agent.
The "differentiator" is dark. The public AI booking chat is intentionally hidden in production (display:none, floating widget not imported). CU-08 documents this honestly, but it means the marquee invitee-facing AI experience is effectively unshipped right now.
Self/cross-account and external-side-effect attribution is fuzzy. Invitees are plus-aliases of the host's own Gmail, so true separate-mailbox RSVP/invite delivery is ambiguous; round-robin side-effects route to the booking-page owner rather than the assigned member; and OAuth/Outlook flows collapse to manual residue the moment a cold login is required. These are correctly flagged, but they mean "the invitee really got X" is often an inference, not a proof.
Subjective quality and real-device behavior are out of frame. Visual polish, typography, logo sharpness, SEO/OG cards, cross-browser rendering, and physical-device touch ergonomics are left to human eyeballing of screenshots — the agent only checks one emulated 375px viewport and the absence of console/overflow errors.