QA: qualify operator-first redesign for iOS beta

[jpapiez]

Scope

Run the independent end-to-end qualification gate for epic #705 after all feature implementation issues are complete. This issue is test-only and does not own product fixes.

Dependencies

• All F1–F10 work under F1 — 5-tab operator shell and navigation restructure #706–F10 — Offline tolerance and write queue #715 complete.
• All linked React gap issues complete where applicable.
• APNs topology decision in F3 — Native push with actionable categories (API + client) #708 implemented in the selected architecture.

Acceptance criteria

• Verify the primary operator loop on supported iPhone and iPad simulators/devices: attention resolution, farm glance, tasks, scan, inventory, swap, multi-slot, harvest, and offline replay.
• Verify notification tap/action routing for each supported category and failure behavior when the selected server is unreachable.
• Verify multi-server isolation: tokens, caches, queued writes, and deep links never cross server/user boundaries.
• Verify duplicate suppression for harvest, stock adjustment, task completion, and spool binding.
• Verify camelCase/string-enum decoding against the real API.
• Record pass/fail evidence and file focused defects; do not waive P0/P1 regressions.

Test expectations

• Run relevant .NET, React, and iOS automated suites once with captured logs.
• Execute a documented manual beta smoke matrix for lock-screen actions, camera/scan permissions, degraded Wi-Fi, and reconnect replay.

Dispatch

Owner: Kane. Blocked until implementation completion; apply go:yes only at the beta qualification handoff.

[github-actions]

⚠️ No squad member found matching label squad:kane. Check .squad/team.md (or .ai-team/team.md) for valid member names.

[github-actions]

📋 Assigned to 🧪 Kane (Tester)

Issue: #723 — QA: qualify operator-first redesign for iOS beta

🧪 Kane will pick this up in the next Copilot session.

For Copilot coding agent: If enabled, this issue will be worked automatically.
Otherwise, start a Copilot session and say:
🧪 Kane, work on issue #723

[jpapiez]

Beta Qualification Matrix v1 (Kane / QA)

Authoritative test matrix for epic #705. Every row maps to a source issue, a
layer, an automation classification, and objective pass/fail evidence. Rows
marked Proactive can be written now against contracts already published in
the feature issues — they do not need the implementation branch to land.

Legend:

• Layer: contract (JSON/enum shape), unit (backend xUnit), int (backend integration via CustomWebApplicationFactory), signalr, react (Vitest/RTL), e2e (Playwright), ios-unit (Swift XCTest under PrintFarmerTests/), ios-ui (PrintFarmerUITests/), manual (physical device / operator smoke), perf, a11y, migration.
• Auto: A = automatable now, A* = automatable once implementation lands, M = manual/physical only.
• Proactive: Y if I can write it before the feature PR (contract already frozen in issue), N otherwise.
• Evidence: the concrete artifact that decides pass/fail.

---

0. Cross-cutting contract & serialization gate (blocks all features)

ID	Source	Layer	Auto	Proactive	Case	Pass/fail evidence
CT-01	#705, #707, #709, #711, #713, #714	contract	A	Y	Every new API response uses camelCase property names in JSON	Golden-JSON test per DTO; JsonPropertyNamingPolicy assertion
CT-02	#705, #707, #708, #711	contract	A	Y	All new enums serialize as strings via JsonStringEnumConverter (attention kind/severity, task anchor, adjustment reason, APNs platform, toolheadType, fallback chain state)	Serialize sample → assert string, not int
CT-03	#705, #707	signalr	A	Y	New SignalR events are lowercase (attentionchanged / attentionitemcreated / attentionitemresolved) — no duplicate PascalCase listener registered	Grep + hub test that only lowercase name fires
CT-04	#708, #716	contract	A	Y	Notification preference DTO adds runout/harvest/maintenance/offline categories without breaking legacy job started/completed/failed/paused fields	Contract test: old client payload round-trips
CT-05	#715	contract	A*	N	Idempotency-Key header accepted on parts-adjust, harvest, task complete, spool bind; header echoes in 200/201/409 body	Integration test: same key twice → single mutation
CT-06	Epic invariant	react/ios	A	Y	React types/api.ts and iOS Models/ decoders reject int for enum, accept string	Vitest + XCTest decode failures

Gap flagged to Dallas (CT-G1): #707 acceptance criteria do not specify the SignalR event name(s). Confirm attentionchanged vs attentionitemcreated/resolved so CT-03 becomes deterministic.

---

1. F1 — 5-tab shell (#706)

ID	Source	Layer	Auto	Proactive	Case	Evidence
F1-01	#706	ios-ui	A*	N	App cold-launches into Attention tab	UI test asserts selected tab identifier
F1-02	#706	ios-ui	A*	N	Exactly 5 tabs present: attention/farm/tasks/scan/inventory	UI test enumerates accessibility identifiers
F1-03	#706	ios-unit	A	Y	AppTab enum values match spec; snapshot of raw values	XCTest against AppTab
F1-04	#706	ios-unit	A*	N	DeepLinkHandler routes: printerReady → farm, spoolDetail → inventory, notification deep link → attention	Unit test with fake router
F1-05	#706	ios-ui	A*	N	Jog / Preheat / z-offset controls not reachable from any list/card; only from PrinterDetail → Advanced disclosure	UI test asserts absence in default hierarchy, presence after disclosure tap
F1-06	#706	ios-ui	A*	N	Settings + Server Switcher reachable via Attention overflow menu in ≤2 taps	UI test tap count
F1-07	#706	ios-ui	A*	N	Every retired destination (Notifications, Jobs, Maintenance, Settings) reachable in ≤2 taps from new shell or explicitly removed	UI journey matrix, one per destination
F1-08	#706	ios-ui	A*	N	Onboarding flow does not reference removed tabs	UI test walks onboarding, no missing-destination fatal error

---

2. F2 — Attention feed API + client (#707)

Backend

ID	Source	Layer	Auto	Proactive	Case	Evidence
F2-01	#707	unit	A	Y	AttentionService composes items from Failure, Maintenance, printer/job sources with correct kind and severity mapping	xUnit with fake source services
F2-02	#707	unit	A	Y	Severity ordering: failure > runout > maintenance-due > harvest-ready > offline; ties break by occurredAt descending	xUnit sorted assertion
F2-03	#707	int	A*	N	GET /api/attention returns 200 with expected DTO shape and camelCase; empty state returns [] not null	CustomWebApplicationFactory int test
F2-04	#707	int	A*	N	POST /api/attention/{id}/snooze persists per user; snoozed item excluded from feed until duration elapses	Int test + time provider
F2-05	#707	int	A*	N	Cross-user: user A's snooze does not hide item for user B	Int test with two auth contexts
F2-06	#707	signalr	A*	N	Attention change broadcast on state transition; verify lowercase event name and payload shape	SignalR hub int test
F2-07	#707	int	A*	N	Actions array only references whitelisted endpoints; server rejects unknown action ids	Int test asserting enum-bounded actions
F2-08	#707	unit	A	Y	Item without live F4 data still returns a well-formed record with runout omitted rather than false-positive	xUnit

iOS client

ID	Source	Layer	Auto	Proactive	Case	Evidence
F2-09	#707	ios-unit	A	Y	AttentionItem decodes from published DTO fixture; unknown kind produces safe unknown row, not crash	XCTest with JSON fixture
F2-10	#707	ios-unit	A*	N	AttentionViewModel calls one endpoint; number of fetches asserted at 1 per refresh	XCTest with fake APIClient
F2-11	#707	ios-ui	A*	N	Failure resolution reachable in ≤2 taps from Attention card open	UI test tap count
F2-12	#707	ios-ui	A*	N	Live SignalR update reflected within 5 s of server-side state change	UI test with test hub
F2-13	#707	ios-ui	A*	N	Snapshot inline on failure card via GET /api/printers/{id}/snapshot	UI test asserts image loaded
F2-14	#707	ios-ui	A*	N	Healthy printers collapsed to one row; expand reveals list	UI test

Gap flagged to Dallas (F2-G1): #707 acceptance says "updates within 5 s" but does not specify measurement anchor (server broadcast timestamp vs UI paint). I will measure broadcastAt → onReceive for automation; ask if operator-perceived paint time is the release gate. Gap F2-G2: No explicit criterion for what happens when the feed exceeds N items (pagination? cap?). Ask Dallas.

---

3. F3 — Native push with actionable categories (#708)

Backend

ID	Source	Layer	Auto	Proactive	Case	Evidence
F3-01	#708	migration	A*	N	DeviceToken EF migration exists for both PostgreSQL and SQL Server providers	dotnet ef migrations has-pending-model-changes clean on both
F3-02	#708	int	A*	N	POST /api/notifications/device-tokens registers token, enforces one-per-(user, platform, environment) — duplicate updates in place	Int test
F3-03	#708	int	A*	N	DELETE unregisters token on logout / server switch	Int test
F3-04	#708	unit	A	Y	Attention kind → APNs category → action identifier map matches spec (PAUSE/RESUME/CANCEL/SNOOZE_15/OPEN_SWAP)	xUnit table-driven
F3-05	#708	unit	A	Y	Category dedupe/rate limit: same printer+kind within window → single delivery	xUnit with in-memory dedupe store
F3-06	#708	int	A*	N	Preference opt-out for a category suppresses APNs delivery for that user only	Int test
F3-07	#708	unit	A*	N	APNs sender uses token auth; p8 key from config; falsely configured relay does not throw at cold-start	xUnit with fake HTTP client
F3-08	#708	int	A*	N	Rollback path: master feature flag / config disable stops delivery within one refresh cycle without dropping tokens	Int test toggling config

iOS

ID	Source	Layer	Auto	Proactive	Case	Evidence
F3-09	#708	ios-unit	A*	N	PushNotificationManager posts token to endpoint on auth success and on token refresh; token cleared on logout / server switch	XCTest with fake APIClient
F3-10	#708	ios-unit	A	Y	UNNotificationCategory set registered matches server category ids	XCTest inspecting registered categories
F3-11	#708	ios-ui	M	N	Lock-screen APNs arrives ≤10 s after simulated failure detection; long-press exposes PAUSE / CANCEL	Physical device, screen recording
F3-12	#708	ios-ui	M	N	PAUSE / CANCEL / SNOOZE_15 execute without foregrounding the app; verify with API side-effect log	Physical device + server log
F3-13	#708	ios-ui	A*	N	Tap deep-link: notification → correct Attention card / printer detail / swap flow	UI test with mocked userNotifications

APNs action paths (dedicated)

ID	Source	Layer	Auto	Proactive	Case	Evidence
F3-14	#708	manual	M	N	Every advertised action id (PAUSE, RESUME, CANCEL, SNOOZE_15, OPEN_SWAP) fires the expected backend endpoint with correct printer/job ids from lock screen	Physical device + backend log capture
F3-15	#708	manual	M	N	Notification action while offline: queued locally OR fails clearly with user-visible reason (never silently swallowed)	Airplane-mode device test
F3-16	#708	manual	M	N	Duplicate delivery suppression: same category+printer within rate-limit window produces one banner	Physical device

Gap flagged to Dallas (F3-G1): #708 says "attention item kinds → APNs categories" but does not explicitly list the deep-link URL scheme/host. Confirm printfarmer://attention/{id} etc. so DeepLinkHandler tests are deterministic. F3-G2: No acceptance criterion for rollback / kill-switch, but F3 is release-critical and epic gate #705 requires a "tested disable/rollback path". I will treat F3-08 as a gate; ask Dallas to add the explicit criterion.

---

4. F4 — Spool coverage and runout prediction (#709)

ID	Source	Layer	Auto	Proactive	Case	Evidence
F4-01	#709	unit	A	Y	Coverage math: single-tool, current-job progress prorate → correct currentJobRemainingGrams	xUnit with fixture gcode metadata
F4-02	#709	unit	A	Y	Multi-toolhead job with per-tool grams → per-toolhead coverage rows	xUnit
F4-03	#709	unit	A	Y	Unknown gcode usage → covers = null and no predictedRunoutAt (no false positive)	xUnit
F4-04	#709	unit	A	Y	Queued required grams added to demand only for jobs explicitly assigned to the printer/toolhead	xUnit — cites epic decision "coverage counts active + explicitly assigned queued"
F4-05	#709	int	A*	N	GET /api/printers/{id}/filament-coverage shape + camelCase	Int test
F4-06	#709	int	A*	N	Batch GET /api/printers/filament-coverage returns per-printer rows for all authorized printers	Int test
F4-07	#709	int	A*	N	Runout attention item appears in F2 feed ≥30 min before predictedRunoutAt when data allows	Int test with time provider
F4-08	#709	int	A*	N	Consumption reconciliation on job complete: verify Spoolman remaining is decremented via actual filament-used stat, not estimate (per epic decision)	Int test asserting single decrement path
F4-09	#709	signalr	A*	N	Coverage refresh emitted on progress tick and on spool binding change	SignalR int test

React (#717)

ID	Source	Layer	Auto	Proactive	Case	Evidence
F4-10	#717	react	A	Y	Compact/detailed printer cards render covers / runoutAt / unknown states	Vitest + RTL snapshot per state
F4-11	#717	react	A*	N	Query invalidation: spool assignment mutation invalidates fleet coverage query	Vitest with queryClient.getQueryState
F4-12	#717	react	A	Y	No PascalCase duplicate SignalR listener registered for coverage events	Grep + component test
F4-13	#717	e2e	A*	N	Playwright: assign spool with low remaining → runout appears on card without full reload	Playwright
F4-14	#717	react	A	Y	Never displays runout when API returns null — displays "unknown" chip	Vitest

iOS

ID	Source	Layer	Auto	Proactive	Case	Evidence
F4-15	#709	ios-ui	A*	N	Farm grid card shows coverage badge, not raw percent-only	UI test
F4-16	#709	ios-ui	A*	N	Slot rows in printer detail show remaining grams + coverage line	UI test

Gap F4-G1: #709 does not name the SignalR event for coverage refresh. Ask Dallas to specify (reuse printerupdated? new coverageupdated?). F4-G2: "Thresholds configurable" — no config surface named. Ask where the ≥30 min lead time lives (appsettings? per-user prefs?). F4-G3: Epic invariant says "Coverage counts the active job and explicitly assigned queued jobs. Unassigned shared-queue jobs remain dispatch-candidate demand" — F4-04 will enforce, but acceptance criteria on #709 do not restate this; strengthen.

---

5. F5 — Guided filament swap flow (#710)

Backend

ID	Source	Layer	Auto	Proactive	Case	Evidence
F5-01	#710	migration	A*	N	Per-tool requirements columns/table migrated on PostgreSQL + SQL Server; legacy RequiredMaterialType still readable	Migration check on both providers
F5-02	#710	unit	A	Y	Slicer output → per-tool [{tool, materialType, colorHint?, estimatedGrams}] extraction	xUnit with sample gcode
F5-03	#710	int	A*	N	GET /api/printers/{id}/toolheads/{i}/swap-validation?spoolId= returns ok for matched material, mismatch with expected/scanned/affectedJobs[] otherwise	Int test
F5-04	#710	int	A*	N	Unload response includes residual spool weight	Int test asserting field
F5-05	#710	int	A*	N	Backward compatibility: job with only RequiredMaterialType still validates single-tool swap	Int test

iOS

ID	Source	Layer	Auto	Proactive	Case	Evidence
F5-06	#710	ios-unit	A*	N	SwapFlow state machine transitions Unload→Scan→Validate→Load→Confirm; mismatch state reachable and requires explicit override	XCTest against VM
F5-07	#710	ios-ui	A*	N	Wrong-material scan blocks with override; override logged (verify server log call)	UI test
F5-08	#710	ios-ui	A*	N	Full swap completes with zero keyboard input (no .keyboard() gesture in trace)	UI test asserts no keyboard shown
F5-09	#710	ios-ui	A*	N	Unloaded spool's residual weight returned to inventory (verify API call)	UI test
F5-10	#710	ios-ui	A*	N	Coverage badge updates immediately after swap (F4 refresh)	UI test
F5-11	#710	ios-ui	A*	N	Mid-print variant: pause → swap → resume; progress preserved (job.progress unchanged pre/post)	UI test with fake printer
F5-12	#710	manual	M	N	NFC scan path on physical device with bound tag	Physical device
F5-13	#710	manual	M	N	Camera permission denied path: user sees fallback (barcode entry) and can complete swap without camera	Physical device

---

6. F6 — Multi-toolhead / fallback chains (#711)

ID	Source	Layer	Auto	Proactive	Case	Evidence
F6-01	#711	migration	A*	N	Slot metadata columns and fallback-group tables migrated on PostgreSQL + SQL Server	Migration check
F6-02	#711	int	A*	N	GET /api/printers/{id}/details exposes slot kind (toolhead/feeder) and sharedHotend flag	Int test
F6-03	#711	int	A*	N	Fallback group CRUD under api/printers/{id}/fallback-groups; ordering preserved on create/update	Int test
F6-04	#711	int	A*	N	Auto-switch event downgrades runout severity to info in F2 feed when backup loaded and same material	Int test
F6-05	#711	int	A*	N	Uncovered runout with no fallback → timed task in F8	Int test
F6-06	#711	int	A*	N	Per-toolhead maintenance components: nozzle hours accrue per active tool, not per printer	Int test with tool-switch
F6-07	#711	int	A*	N	Queue candidates score printers by slot loadout vs per-tool requirements from F5	Int test
F6-08	#718	react	A*	N	Fallback group CRUD UI: create/reorder/delete with mixed-material warning; backend rejection surfaced	Vitest/RTL
F6-09	#718	react	A	Y	Toolhead vs feeder distinguished via toolheadType string enum, not integer	Vitest
F6-10	#719	react	A*	N	Maintenance plan deployment can target whole printer or a specific toolhead	Vitest
F6-11	#719	react	A*	N	AMS/MMU feeder gates not shown as hotend maintenance targets unless API marks eligible	Vitest
F6-12	#711	ios-ui	A*	N	U1 renders 4 independent slots with per-tool maintenance; AMS renders bays + fallback chain	UI snapshot

---

7. F7 — Printer Detail v2 (#712)

ID	Source	Layer	Auto	Proactive	Case	Evidence
F7-01	#712	ios-ui	A*	N	Section order: header → camera → current job (with coverage) → slots → queue → maintenance odometer → history tail → Mainsail link	UI snapshot
F7-02	#712	ios-ui	A*	N	Temperature graph and jog controls not present in default layout; only under Advanced disclosure	UI test asserts absence
F7-03	#712	ios-ui	A*	N	Camera snapshot tap opens MJPEG live via MJPEGStreamView	UI test
F7-04	#712	ios-ui	A*	N	All six operator questions answered without scrolling past two viewport heights on iPhone 15 baseline	UI test measures scroll offset
F7-05	#712	ios-ui	A*	N	Queue rows show per-tool match state from F6 candidates	UI test
F7-06	#712	ios-ui	A*	N	Maintenance-due row deep-links to POST /api/maintenance/logs completion flow	UI test

---

8. F8 — Shift plan compiler / Tasks tab (#713)

Backend

ID	Source	Layer	Auto	Proactive	Case	Evidence
F8-01	#713	unit	A	Y	Compiler emits task with anchor = at(predictedRunoutAt - leadTime) for runout source	xUnit
F8-02	#713	unit	A	Y	Maintenance task lands in genuine idle window computed from queue ETAs	xUnit
F8-03	#713	unit	A	Y	Spool burn-rate reorder threshold triggers restock task at correct on-hand level	xUnit
F8-04	#713	int	A*	N	Dedupe: existing open task for same sourceRef not duplicated on next compile tick	Int test
F8-05	#713	int	A*	N	Source resolution auto-completes task (swap done via F5 → runout task auto-completed)	Int test
F8-06	#713	int	A*	N	GET /api/tasks?view=shift returns anchor-grouped list with now/at/window/today	Int test
F8-07	#713	int	A*	N	Config: lead times + reorder thresholds honored from settings	Int test

React (#720) & iOS

ID	Source	Layer	Auto	Proactive	Case	Evidence
F8-08	#720	react	A*	N	TasksWidget renders Now/timed/windowed/anytime groups from server anchors	Vitest
F8-09	#720	react	A*	N	Unknown future task kinds render safe generic row (no crash)	Vitest with future-kind fixture
F8-10	#720	react	A*	N	Manual + profile-import legacy tasks still supported	Vitest
F8-11	#720	react	A*	N	Complete/skip/dismiss invalidate task query and refresh badge count	Vitest
F8-12	#713	ios-ui	A*	N	Checking harvest task in TasksView triggers F9 inventory increment path	UI test

---

9. F9 — Scan station and printed-parts inventory (#714)

Backend

ID	Source	Layer	Auto	Proactive	Case	Evidence
F9-01	#714	migration	A*	N	PartInventory, Bin, adjustment ledger migrations on PostgreSQL + SQL Server	Migration check
F9-02	#714	int	A*	N	POST /api/parts-inventory/{sku}/adjust records ledger row with delta + reason (harvest/qc-reject/manual); on-hand never blindly overwritten	Int test — cites epic decision "immutable adjustment ledger"
F9-03	#714	int	A*	N	POST /api/job-queue/{id}/harvest is atomic and duplicate-safe: re-post returns 200 with prior result, no second increment	Int test — cites epic decision "harvest atomic and duplicate-safe"
F9-04	#714	int	A*	N	Harvested job remains Completed with harvest metadata (per epic decision)	Int test
F9-05	#714	int	A*	N	Multi-SKU plate: single harvest call produces N adjustment rows in one transaction	Int test
F9-06	#714	int	A*	N	Harvest resolves the associated attention item (F2) and completes the associated task (F8)	Int cross-service test
F9-07	#714	int	A*	N	Reorder-point evaluation feeds F8 restock task at threshold	Int test
F9-08	#714	unit	A	Y	Job → SKU mapping: project-file mapping, gcode metadata, and manual fallback resolved in that precedence	xUnit

React (#721 config, #722 harvest)

ID	Source	Layer	Auto	Proactive	Case	Evidence
F9-09	#721	react	A*	N	SKU CRUD + bin CRUD + G-code → SKU output mapping, including multi-SKU	Vitest
F9-10	#721	react	A	Y	Printed-parts inventory visually and semantically separate from maintenance-components (verify distinct route + component tree)	Vitest + route test
F9-11	#721	react	A*	N	Stock displayed from ledger; direct edit UI absent (only adjustment flows)	Vitest asserts no PUT /parts-inventory/{sku} with on-hand
F9-12	#722	react	A*	N	Completed unharvested job → Harvest action visible on job/task/attention surfaces	Vitest
F9-13	#722	react	A*	N	Already-harvested job displays as complete; Harvest button absent or disabled	Vitest
F9-14	#722	react	A*	N	Idempotent replay: user clicks twice → server returns same result, UI shows single increment	Vitest with mocked 200-with-prior-body
F9-15	#722	react	A*	N	Multi-SKU harvest shows every adjustment in one atomic confirmation	Vitest

iOS

ID	Source	Layer	Auto	Proactive	Case	Evidence
F9-16	#714	ios-ui	A*	N	Plate-done → harvested + stock incremented in ≤3 taps with labeled bins, zero typing	UI test tap count + keyboard-absent assertion
F9-17	#714	ios-ui	A*	N	Wrong-bin scan warns before commit	UI test
F9-18	#714	ios-ui	A*	N	Recent scans list unified across spool/bin/part/printer types	UI test
F9-19	#714	manual	M	N	Camera permission denied path: user sees actionable instruction, can navigate away without crash	Physical device

---

10. F10 — Offline tolerance and write queue (#715)

ID	Source	Layer	Auto	Proactive	Case	Evidence
F10-01	#715	int	A*	N	Idempotency-Key on parts-adjust, harvest, task complete, spool bind: replay in window returns same result, single ledger row	Int test
F10-02	#715	int	A*	N	Replay outside dedupe window either rejected or creates second adjustment — behavior explicit and documented	Int test asserts documented behavior
F10-03	#715	ios-unit	A*	N	Write queue persists across app restart; replays on reconnect in original order	XCTest with fake network
F10-04	#715	ios-unit	A*	N	Conflict from server (409 / mismatch) surfaced to user with retry or discard, never silently dropped	XCTest
F10-05	#715	ios-unit	A	Y	Live-only actions (pause/cancel/filament ops) refused offline with clear state; never enqueued	XCTest asserts queue never gains these types
F10-06	#715	ios-ui	A*	N	Staleness banner shows last-updated timestamp when offline	UI test
F10-07	#715	manual	M	N	Airplane mode → scan/log/harvest → reconnect → single server-side effect per action	Physical device + server log
F10-08	#715	manual	M	N	Degraded Wi-Fi (packet loss / high latency): replays succeed without duplicates	Physical device with network conditioner

---

11. Multi-server / multi-user isolation (cross-cutting, ties #705 gate)

ID	Source	Layer	Auto	Proactive	Case	Evidence
MS-01	Epic invariant	ios-unit	A	Y	APNs device token stored per server config; switching server rotates registered token	XCTest using ServerRegistry + PushNotificationManager
MS-02	Epic invariant	ios-unit	A	Y	Local cache keyed by server id; switching servers never reads other server's cached farm state	XCTest
MS-03	#715, epic	ios-unit	A*	N	Write queue partitioned per server; queued write against server A never sent to server B after switch	XCTest
MS-04	Epic invariant	ios-unit	A	Y	Deep link opened while different server active prompts server switch or is rejected; never executes against wrong server	XCTest with deep-link fixture
MS-05	#708	int	A*	N	Backend does not fan out APNs across tenants; token registered by user A never receives user B's payload	Int test with two auth contexts
MS-06	Epic invariant	ios-ui	A*	N	Logout unregisters APNs token and clears queued writes for that server	UI test with fake API
MS-07	Epic invariant	manual	M	N	Physical device with two servers registered: switch server, run swap flow, verify all traffic goes to the newly-selected server (proxy capture)	Physical device + mitmproxy

---

12. Accessibility (WCAG 2.2 AA — epic implicit via team a11y instructions)

ID	Source	Layer	Auto	Proactive	Case	Evidence
A11-01	#706	ios-ui	A*	N	All 5 tabs and Attention overflow are VoiceOver-labeled with unique identifiers	UI test iterates accessibilityLabel
A11-02	#707	ios-ui	A*	N	Attention card actions have accessible traits (button) and speak severity	UI test
A11-03	#710	ios-ui	A*	N	Swap-flow mismatch alert is announced (Dynamic Type Large + VoiceOver on)	UI test
A11-04	React issues	react	A*	N	Coverage/fallback/harvest UIs meet axe-core baseline: no critical violations	Vitest + @axe-core/react
A11-05	React issues	react	A	Y	Every new interactive element has role + accessible name	RTL query getByRole
A11-06	#706	ios-ui	A*	N	Tab bar respects Dynamic Type at XXL without truncating labels	UI snapshot at accessibility category
A11-07	React	manual	M	N	Contrast: new badges/chips (covers/runout/harvest) hit ≥4.5:1 across all 7 themes	Manual review with contrast tool

---

13. Migration & provider validation (release gate #705 explicit)

ID	Source	Layer	Auto	Proactive	Case	Evidence
MG-01	Epic	migration	A*	N	dotnet ef migrations has-pending-model-changes clean for AppDbContext on PostgreSQL and SQL Server after F3/F5/F6/F9 land	CI gate command
MG-02	Epic	migration	A*	N	Same clean check for SlicerDbContext if slicer schema touched by F5 per-tool requirements	CI gate command
MG-03	Epic	int	A*	N	Idempotent up + down: apply migration, downgrade, reapply produces identical schema	dotnet ef scripted round-trip
MG-04	Epic	int	A*	N	Legacy RequiredMaterialType still readable after F5 migration (data preserved)	Int test with pre-migration seed
MG-05	Epic	int	A*	N	Provider parity: same integration test suite passes against SQLite (dev), PostgreSQL, SQL Server for any endpoint touched	Matrix CI run

---

14. Performance & resource limits

ID	Source	Layer	Auto	Proactive	Case	Evidence
PF-01	#707	perf	A*	N	Batch GET /api/attention p95 ≤ 300 ms for 50-printer farm fixture	BenchmarkDotNet or timed int test
PF-02	#709	perf	A*	N	Batch GET /api/printers/filament-coverage p95 ≤ 500 ms for 50-printer / 200-queued-job fixture	Timed int test
PF-03	#707	perf	A*	N	Attention SignalR broadcast on state change fires ≤ 1 s from source event	Timed int test
PF-04	#708	perf	A*	N	APNs delivery p95 ≤ 10 s from attention creation (measured server→APNs relay)	Timed test with mock relay
PF-05	Epic	perf	M	N	iOS cold-launch to Attention feed rendered ≤ 2 s on baseline iPhone 15	Instruments trace
PF-06	#715	perf	A*	N	Write queue replay of 100 offline scans completes ≤ 30 s with zero duplicates	XCTest performance metric
PF-07	#709	perf	A*	N	Coverage service memory: no per-request allocation growth over 1000 iterations	BenchmarkDotNet memory diagnoser

---

15. Rollback / disable paths (epic gate: "tested disable/rollback path")

ID	Source	Layer	Auto	Proactive	Case	Evidence
RB-01	#708	int	A*	N	APNs sender: master disable via config stops delivery within one refresh, preserves tokens for re-enable	Int test toggling config
RB-02	#707	int	A*	N	Attention endpoint feature flag off → 404 or empty; iOS falls back to prior source screens without crash	Int + iOS unit test
RB-03	#715	ios-unit	A*	N	Write queue disable flag → mutations attempt direct online only; queued items drained cleanly	XCTest
RB-04	#714	int	A*	N	Parts inventory feature flag off: harvest returns 404 gracefully; existing job completion still works	Int test
RB-05	Epic	manual	M	N	Full rollback rehearsal: disable F3+F9+F10 flags on staging, run wave-0 smoke, verify nothing crashes	Manual staging exercise

Gap RB-G1: No feature-flag mechanism is named in any feature issue. Ask Dallas whether we're using appsettings toggles, LaunchDarkly-style flags, or Feature.Enabled config classes — this materially affects RB-01..04.

---

16. End-to-end operator journeys (release gate)

Each journey exercises multiple features and produces a single pass/fail. All are A* (need implementation) except where noted; several journeys are M because they require a physical device and printer.

ID	Journey	Layer	Auto	Evidence
EJ-01	Cold-launch → Attention → tap failure card → Resume (2 taps) → job resumes on real printer	manual	M	Screen recording + printer state
EJ-02	APNs lock-screen alert (runout risk) → Snooze 15 → banner clears → returns after 15 min	manual	M	Physical device
EJ-03	APNs lock-screen alert (failure) → PAUSE action → app never foregrounds → backend log shows pause fired	manual	M	Physical device + server log
EJ-04	Runout attention → Start swap → scan wrong material → mismatch block → scan correct material → load → coverage badge updates within 5 s	manual	M	Physical device
EJ-05	Multi-tool U1 job: T0 runs out mid-print, F5 mid-print variant pauses, swap, resume; F4 shows correct predicted end	manual	M	Physical device
EJ-06	AMS printer with fallback chain: primary bay empties, auto-switch fires, F2 shows info-severity, no timed task; when last bay exhausts, timed task appears	manual	M	Physical device
EJ-07	Plate completes → Attention harvest card → scan bin → confirm quantity → adjust stock → task auto-completes, attention resolves (3 taps, zero typing)	manual	M	Physical device
EJ-08	Multi-SKU plate: single harvest confirmation shows all SKU/bin rows, single atomic API call	ios-ui + manual	A*/M	UI test + physical confirm
EJ-09	Airplane mode: 5 scans + 1 harvest + 1 task complete queued → reconnect → all replayed exactly once, no duplicate stock	manual	M	Physical device + server log
EJ-10	Two servers registered: switch A→B mid-session, queued writes from A never post to B; APNs from A silenced or routed correctly	manual	M	Physical device + proxy
EJ-11	Shift start: Tasks tab shows Now/timed/anytime; complete swap via F5 → runout task auto-completes; complete harvest → stock and attention update	ios-ui + manual	A*/M	UI test + physical
EJ-12	Advanced disclosure: jog and preheat reachable only under PrinterDetail → Advanced; visible controls match hardware capabilities	ios-ui	A*	UI test
EJ-13	Notification preferences: opt out of runout category → no APNs for runout, other categories still deliver	manual	M	Physical device
EJ-14	Onboarding a new operator on both iPhone and iPad simulators: complete first swap + first harvest without external instruction	manual	M	Usability walk

---

17. Automatable vs manual summary

Category	Automatable now	Automatable when impl lands	Manual/physical only
Contract & serialization	6	1	0
F1 shell	1	7	0
F2 attention	4	10	0
F3 APNs	3	7	6
F4 coverage	6	9	0
F5 swap	1	9	2
F6 fallback / per-tool	1	11	0
F7 detail	0	6	0
F8 tasks	3	9	0
F9 inventory	3	15	1
F10 offline	1	5	2
Multi-server	3	3	1
Accessibility	1	5	1
Migrations	0	5	0
Performance	0	6	1
Rollback	0	4	1
E2E journeys	0	2	12
Total	33	114	27

---

18. Proactive work I can start now (no waiting on implementation)

1. Contract/serialization fixtures. Add golden-JSON tests for every new DTO described in F2 — Attention feed (API + client) #707/F4 — Spool coverage and runout prediction (API + client) #709/F6 — Multi-toolhead / AMS slot UI and fallback chains #711/F8 — Shift plan compiler (Tasks tab) #713/F9 — Scan station and printed-parts inventory #714 using the shapes already frozen in the issue bodies. If backend changes the shape at implementation, the fixture becomes the review conversation.
2. iOS decoder tests. Write AttentionItem, Coverage, ShiftTask, PartInventory, FallbackGroup decoder tests against the same fixtures — same fail-loud effect.
3. Enum-as-string guards for attention.kind, attention.severity, task.anchor, adjustment.reason, toolheadType, APNs platform.
4. SignalR lowercase-only assertion — a grep-based test that fails if any PascalCase listener is added for new hubs.
5. AppTab unit test locking the 5-value enum order.
6. Coverage math unit tests — the algorithm is specified in F4 — Spool coverage and runout prediction (API + client) #709 (spoolId, remainingGrams, currentJobRequiredGrams, currentJobRemainingGrams prorated by progress, queuedRequiredGrams, covers, predictedRunoutAt, runoutLayer). I can write these against a stub service today; when the real CoverageService lands, tests either pass or expose a contract drift.
7. Shift compiler math unit tests for anchor selection, dedupe, and burn-rate reorder trigger.
8. Idempotency contract test scaffolding — an int test that will run once endpoints exist; the harness (same-key double-post helper, dedupe-window fixture) can be built now.
9. Multi-server key isolation tests for PushNotificationManager + ServerRegistry + cache — the existing services already exist; I don't need F3 to write MS-01/02/04.
10. React coverage-badge and harvest button component tests driven by fixture props — the components don't exist yet, but the fixture shapes and the covers/runout/unknown state matrix (F4-10, F9-13) are frozen; I can commit test files under src/Web/ReactApp/src/features/…/__tests__/ that will light up when Ripley adds the component.
11. Accessibility grep tests — A11-05 (every new interactive element has role + accessible name via RTL) can run as a linter-style test against any new tsx file.
12. APNs category id lock test — enumerate PAUSE/RESUME/CANCEL/SNOOZE_15/OPEN_SWAP in a Swift XCTest so registration drift fails loudly.

Estimated proactive coverage: 33 test cases (the "Automatable now" column) plus scaffolding for another ~15 that flip to green the moment the corresponding implementation PR lands.

---

19. Consolidated gap list for Dallas / coordinator

Gap	Feature	What is missing	Why it matters
CT-G1	#707	SignalR event name(s) for attention changes not specified	CT-03/F2-06 cannot be deterministic; risk of duplicate PascalCase listener
F2-G1	#707	"Updates within 5 s" lacks measurement anchor	Automation vs manual gate ambiguity
F2-G2	#707	No pagination/cap for attention feed	Perf/UX risk on large farms
F3-G1	#708	Deep-link URL scheme not stated	DeepLinkHandler tests non-deterministic
F3-G2	#708	No explicit rollback/disable criterion, but epic gate requires it	RB-01 becomes a release blocker without an owner
F4-G1	#709	Coverage SignalR event name not stated	F4-09 ambiguous
F4-G2	#709	Config surface for runout lead-time thresholds not stated	Can't test "thresholds configurable" acceptance
F4-G3	#709	Acceptance criteria don't restate the epic invariant on "explicitly assigned queued jobs only"	Risk of implementer counting all queued demand
RB-G1	Epic-wide	No feature-flag mechanism named	RB-01..04 depend on this; cross-feature
MG-G1	Epic-wide	No explicit acceptance criterion that provider parity (SQLite/PostgreSQL/SQL Server) is verified per feature	CI gate exists but per-feature signoff is undefined
F9-G1	#714	Job → SKU mapping precedence not stated (project vs gcode vs manual)	F9-08 needs a defined precedence to test
F10-G1	#715	Behavior for replay outside the dedupe window is undefined	F10-02 cannot pass/fail objectively
A11-G1	Epic-wide	No stated a11y acceptance floor (AA?)	I will hold the line at WCAG 2.2 AA per team a11y instructions; needs Dallas ratification

---

20. Next action

• I will not write product code and will not create a branch. This matrix is the contract.
• I can begin the 33 proactive test cases immediately (Section 18). Requesting go:yes on that scope only — the full qualification pass still waits on implementation completion per the epic gate.
• Awaiting Dallas / coordinator response on the 13 gaps above. Any gap left open at implementation-complete becomes a P1 defect I will file rather than waive.

[jpapiez]

Dallas disposition — Section 19 requirement gaps

All 13 gaps now have objective owners/contracts:

• CT-G1, F2-G1, F2-G2: F2 — Attention feed (API + client) #707 addendum fixes attentionchanged, timing anchors, and cursor/limit rules.
• F3-G1, F3-G2: F3 — Native push with actionable categories (API + client) #708 fixes printfarmer:// routes and nativePushEnabled semantics; Release: prepare operator-first iOS beta and APNs deployment #724 owns staging rehearsal.
• F4-G1, F4-G2, F4-G3: F4 — Spool coverage and runout prediction (API + client) #709 fixes filamentcoveragechanged, FilamentCoverageSettings, and assigned-queue-only demand.
• RB-G1: new prerequisite Backend: add operator feature gates and capability exposure #725 defines OperatorFeatureSettings / IOperatorFeatureGate and exact flag names.
• MG-G1: Epic: Mobile app operator-first redesign #705 now requires SQLite/PostgreSQL/SQL Server evidence per persistence feature; active schema issues repeat the requirement.
• F9-G1: F9 — Scan station and printed-parts inventory #714 fixes explicit-request → job snapshot → project mapping → G-code mapping → 409 precedence.
• F10-G1: F10 — Offline tolerance and write queue #715 fixes a 7-day replay/dedupe window and expiredNeedsReview behavior outside it.
• A11-G1: Epic: Mobile app operator-first redesign #705 ratifies WCAG 2.2 AA-equivalent acceptance; F1 — 5-tab operator shell and navigation restructure #706/F2 — Attention feed (API + client) #707/F5 — Guided filament swap flow #710/F9 — Scan station and printed-parts inventory #714 include feature-specific criteria.

Kane may update matrix rows to reference these addenda. The only product blocker remains Jeff's APNs topology decision in #708. #725 is a new implementation/release prerequisite, not a product-decision blocker.