Completes the anti-fraud/safety core coverage on a fresh in-memory DB:
- occupancy.test.ts (12): the ledger-fold count, the capacity/full gate, and the
reserved-subscriber-spots model — never double-count a parked subscriber, reserve
tightens only the TRANSIENT gate.
- pay-station.test.ts (12): quote math against the frozen tariff, the signed-payment
side effect (+ chain verify), no-session / no-tariff errors, the booth lookup view,
active-session listing.
- exit-flow.test.ts (9): the GATE — refuse unknown / unpaid / grace-expired (no exit
signed); a paid-within-grace session signs the exit; the booth transient path has NO
subscription bypass; a prepaid subscriber leaves via the assist (reopenBarrier) path.
- shift-service.test.ts (14): site-wide single-open invariant, the takings SPLIT by
source (subscription sales vs out-of-window vs transient tickets), drawer carry-
forward + cash_in/out vouchers, Z-report sign + listShifts read-back.
- entry-flow.test.ts (5): the exported validateTicketCode Luhn typo-guard. (The
capacity-gate/print-hold/sign-before-open paths need device fakes — covered in the
device + route phases.)
Adds test-helpers.ts (real EventLog, silent logger, tariff seeder). server 68/68 green.
Note: apps/vision has 2 PRE-EXISTING failures (test_app.py) — environment drift now
that fast_alpr + the ONNX model are installed (the "stub mode" assertions are stale).
Untouched here; to be fixed in the vision phase.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Foundation for testing every service. Adds @parking/db/testing — createTestDb()
spins a fresh in-memory SQLite and applies the real Drizzle migrations, so server
tests run against the production schema with zero live-DB risk.
Wires Vitest into apps/server (test script + config; test signing keys via env)
and adds the first Phase-1 suites against the anti-fraud core:
- signer.test.ts (10): sign/verify round-trip, tamper + forgery rejection,
malformed-signature guard, determinism, keyId rotation (buildVerifier).
- event-log.test.ts (12): monotonic index, prevHash linkage, payload-in-signature,
append serialization, and verifyChain() catching every tamper class — edited
payload, deleted row (index gap), broken prevHash, unknown keyId — plus
canonicalize byte-stability.
Also stops *.test.ts leaking into shipped dist/ (tsconfig exclude in server +
shared; shared had been emitting compiled tests all along).
server 22/22, shared 87/87 green.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
A barcode/QR scanner is an HID "keyboard wedge" — it types the id + Enter into
whatever holds focus. Previously that only worked while the ticket <input> was
focused; a scan with focus elsewhere (or nowhere) went nowhere.
New useScanner hook (apps/web/src/lib/use-scanner.ts): a document-level keydown
listener that detects the scanner's FAST keystroke burst ended by Enter and opens
the pay/exit modal via setActiveTicket — regardless of focus. A gap > 50ms resets
the buffer, so human-paced typing with nothing focused never registers as a scan
(min length 3 guards stray Enters). Keystrokes into an input/textarea/select/
contenteditable are ignored, so the manual ticket field still works by hand. The
hook is paused while a modal is already open — a scan must not abandon an
in-progress payment; the operator finishes/closes, then scans the next car.
Verified at runtime (Playwright): a fast burst with focus on BODY opens the modal;
a second scan while the modal is open is ignored; slow (120ms) human typing does
NOT open it; the manual input submit still opens it. build+lint 14/14.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Three changes to the shift hub, addressing the report:
1. Route rename /shift → /shifts (matches the plural "Turnet" label and the
section). /shift and /setup/shifts both redirect to /shifts; the header link
and the operator-landing fallback point at /shifts.
2. The activity-log rows are now CLICKABLE and open the same read-only
event-detail modal the booth live feed uses (full signed payload + entry/exit
snapshots + chain provenance) — previously they were static rows. Extracted
EVENT_STYLE, the feed row, the detail modal, and their helpers out of
BoothScreen into a shared apps/web/src/ui/event-detail.tsx imported by both the
booth and the shift log, so the two render and behave identically and can't
drift.
3. Reworked the /shifts layout to fill the viewport like /booth: a fixed
title + filters, then a two-pane area (shift list | activity log) where each
pane scrolls independently (min-h-0/flex-1 + overflow-y-auto) instead of the
whole page growing. ShiftActivityLog is now a flex column with a fixed header
and a scrollable list.
Verified at runtime (Playwright): /shift redirects to /shifts, an activity row
opens the detail modal, the layout fills height, and the booth still works (0
console errors after the extraction). build+lint 14/14.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The header shift link used nav.shift (singular: Turni/Shift) but points at the
/shift HISTORY hub, so it now uses nav.shifts (plural: Turnet/Shifts).
The Setup "Turnet" tab was a duplicate — /setup/shifts and the standalone /shift
both rendered ShiftsHistory. Removed the Setup tab + its child route; /setup/shifts
redirects to /shift for old bookmarks, and the operator-landing fallback (a
shift:read user opening /setup) now points at /shift. The orphaned nav.shift key is
left in both catalogs (harmless).
Verified at runtime (Playwright): header reads Kabina·Turnet·Abonimet·Konfigurimi,
Setup no longer lists Turnet, /setup/shifts redirects to /shift. build+lint 14/14.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Subscriptions, Plans, and Tariff Lab were tabs under /setup. Moved them into a
standalone /subscriptions section with its own header nav entry (between Turni and
Konfigurimi) and a tab bar: Abonimet (/subscriptions), Planet
(/subscriptions/plans), Lab Tarife (/subscriptions/tariff-lab).
- New SubscriptionsLayout (tab bar + <Outlet>); the three screens are now its
child routes at the top level, not under setupRoute.
- Removed Subscriptions/Plans/Tariff-Lab from SetupLayout and SETUP_TABS. Setup
now holds Devices/Tariff/Site/Users/Roles/Shifts/Logs.
- Header gains the "Abonimet" link, gated on subscription:read OR subscription:plan
OR tariff:read (shown if the user can reach any sub-tab).
- Tabs are permission-gated; the /subscriptions index redirects a user lacking
subscription:read to the first sub-tab they can see (or the booth).
- Legacy redirects: /setup/subscriptions → /subscriptions, /setup/plans →
/subscriptions/plans, /setup/tariff-lab → /subscriptions/tariff-lab. Dropped the
old /subscriptions → /setup redirect (it's a real route now).
- The Tariff COMPOSER stays in Setup; only the Tariff LAB simulator moved.
Verified at runtime (Playwright): header order Kabina·Turni·Abonimet·Konfigurimi,
the three sub-tabs render, Setup no longer lists them, /setup/subscriptions
redirects cleanly. build+lint 14/14.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Three changes:
1. Confirm-before-close. The header shift button closed the shift directly — a
stray click would sign the irreversible Z-report. It now opens a confirm modal
showing the live X-report (takings split by source + expected drawer) with
Cancel / End-shift. Opening a shift stays immediate (no such risk).
2. Split takings by SOURCE. The report separates Tickets (transient) from
Subscriptions (monthly sales + a subscriber's out-of-window charge), so the
operator sees subscriber money apart from ticket money. Buckets are derived
from the signed payment payload flags (subscriptionSale /
subscriptionWindowCharge) and always reconcile to cash + card (a payment with
neither flag is a ticket). Computed in #summariseWindow, carried on the signed
shift_z_report payload, and shown in the X-report, the close modal, the shift
history detail, and the printed Z-report. Reports predating the fields default
subscription to 0 (ticket absorbs the whole take), so old shifts still
reconcile.
3. Fix dark-theme native <select> popups rendering WHITE on WebKitGTK (the Tauri
Linux WebView): set color-scheme dark/light on <html> per theme + explicit
<option> colours, so the OS-drawn dropdown list follows the theme.
Verified the split on a read-only DB copy: tickets 0, subscriptions 10,200
(10,000 sale + 200 out-of-window), reconciles to cash+card. build+lint 14/14,
i18n parity (sq+en).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
A subscription froze its planVersionId at sale (reproducible pricing). There was
no way to move a sold sub onto a different VERSION of the SAME plan — needed when
an admin publishes v2 with different timeframes (e.g. mujor-naten-cdo-dite v1
"every day" → v2 "weekdays only") and wants an existing subscriber on it, or back
on v1.
Backend (PUT /api/subscriptions/:id):
- accept planVersionId; honored only with the subscription:plan permission
(stronger than subscription:update — a plan-management action). Non-privileged
caller sending a change → 403, not silently dropped.
- validated to belong to the sub's EXISTING planId (a different plan = a
different price basis = a re-sale → 400).
- price/currency/period/planId stay frozen; only planVersionId moves. The swap is
server-logged for audit (the row is mutable master data, not on the ledger).
Past signed entry/exit events keep their own windowTariffVersionId, so history
reprices identically — only future access uses the new version's windows.
Frontend (SubscriptionManager):
- pass the session user through the route (like RolesManager).
- admin-only "Versioni" picker in the edit modal: lists every version of the
sub's plan by effective date + a timeframe summary (days + window, or 24/7),
current pre-selected. The plan itself stays read-only. Sends planVersionId only
when it changed.
- i18n: subs.version/versionHint/versionCurrent/versionOnlyOne/everyDay/allDay
in both sq + en.
Verified on a writable DB copy: version changed, price + planId frozen,
cross-plan version rejected. Live DB untouched. build+lint 14/14.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The previous width fix also forced ALIGN_LEFT inside code128(), which moved the
slip's barcode to the left. But the no-print bug was the barcode WIDTH (too wide
to fit the head at module width 3), not the centering — at width 2 it fits and
centers fine. So code128() no longer touches alignment; the caller controls it.
The out-of-window slip block is ALIGN_CENTER, so the Code128 + QR center as they
did before, just narrow enough (width 2, ~510 dots) to actually print. The
voucher receipt barcode likewise centers as it originally did.
Verified: alignment-in-effect at the barcode = CENTER, module width = 2, QR
present; build+lint 14/14.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The slip printed the QR but NOT the Code128 barcode on the Rongta. Root cause:
the ~20-char occurrence id (SUBSESS-…) at module width 3 is ~765 dots wide —
over the 80mm head's ~576 printable dots — so the firmware silently aborts the
barcode (prints nothing). It was also emitted while ALIGN_CENTER (set for the
title) was active, which shifts the start point right and makes it overflow
even sooner. The QR, being compact, rendered fine — hence QR-only output.
code128() now takes a moduleWidth (default 3, so the shorter entry-ticket id is
unchanged) and forces ALIGN_LEFT (a wide barcode must hug the margin). The slip
passes width 2 (~510 dots — fits with quiet zones) and re-centers the QR/text
after. renderReceipt's voucher barcode re-asserts ALIGN_CENTER for the lines
that follow it.
Verified: width n=2 in the byte stream, est 510 dots; Code128 + QR both present;
build+lint 14/14.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
An out-of-window subscriber entry stamped a FIXED windowOwedMinor = the whole
gap to window-open (e.g. 800 ALL for a 13:21 arrival to a 20:00 window) and
deferred it to exit. That over-charged anyone who left before the window
opened — a 1-hour visit was billed as 6.5 hours.
The amount isn't knowable at entry: a subscriber may enter early, leave after
an hour, come and go several times before the window opens, and linger past
window-close. They should pay only for the minutes actually parked outside the
window (capped at the window edges) — exactly what minutesOutsideWindow already
computes.
So the entry now stamps a MARKER only (outOfWindow: true + windowTariffVersionId
for reproducible pricing), no fixed amount. The exit gate and booth quote price
it live via windowOwedBetween(entry → settle-time), which already caps at the
window edges (early entry stops accruing at window-open; the in-window portion
of a crossing stay is free; the late-exit tail keeps accruing until payment).
Both already called that one function, so they agree.
- subscription-flow: entry stamps outOfWindow marker; the advisory slip is now a
scannable out-of-window TICKET (Code128 + QR of the occurrence id).
- shared LedgerPayload: add outOfWindow; mark windowOwedMinor/windowGap*/
windowCurrency deprecated read-only (historic signed events still type-check).
- BoothScreen: window-charge badge keys on outOfWindow (or the old stamp).
- ActiveSessions: drop the always-on "Open barrier" for subscribers — the
assist-open / window-charge payment live in the pay modal, so the list can't
one-click past an unpaid out-of-window charge.
Verified the live model on a DB copy: 13:21→14:30 = 200 ALL; 19:55(in grace)→
23:00 = 0; 19:00→21:30 (crosses into window) = 100 ALL. Existing signed
occurrences left untouched (immutable). build+lint 14/14, shared 87/87.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The advisory out-of-window slip for a subscriber had two problems:
1. Faulty character codes. It rendered via the generic text printReport,
which has no CP852 mapping for the em dash, ellipsis, or warning sign in
the composed strings — so they printed as "?" ("PARKIM ? JASHTE ORARIT").
Added ASCII transliterations for that typographic punctuation in the
ESC/POS encoder (— → -, ⚠ → !, … → ..., curly quotes/bullet), so they
degrade to a readable glyph instead of "?".
2. Not scannable. The slip printed only "Nr: SUBSESS-…" as plain text, so
the operator had to hand-key it. Gave the notice its own render function
(renderWindowChargeNotice) + a printWindowChargeNotice device method that
prints the occurrence id as a Code128 AND a QR — the same scan path as a
transient ticket, so the operator scans it straight into the booth pay
modal, which then quotes the combined window charge. Implemented on both
the rongta and cashino drivers.
Also fixed the booth pay modal: "Open barrier" no longer shows by default
for a subscriber. A prepaid subscriber with nothing owed sees only a small
"assist open" reveal (the audited manual open for a faulty reader / lost
card stays available, just not the default). A subscriber owing an
out-of-window charge is now two steps — take payment first, then "Open
barrier" appears — instead of an always-on open button.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Point the Tauri updater at the real self-hosted Gitea "latest release" path:
https://git.infra.msai.al/mca/parking_solution/releases/latest/download/latest.json
— redirects to the newest tag's latest.json published by release.yml. Verified
against tauri-plugin-updater: it GETs the endpoint (200 + manifest / 204 = up to
date) and reads platforms.linux-x86_64.{signature,url}.
Document the desktop WS origin: the Tauri window loads from tauri://localhost
(Linux may also send http://tauri.localhost), which is NOT same-origin with the
backend, so WS_ALLOWED_ORIGINS must include both or the live feed won't connect.
Added both to apps/server/.env.example.
Updated the as-built in wiki/decisions/desktop-shell-tauri.md. Also carries an
unrelated plans.namePlaceholder copy tweak already in the tree. turbo build lint
14/14 green.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Mirror the house Gitea Actions pattern (cf. trm/processor): corepack pnpm +
frozen install on ubuntu-latest.
ci.yml — push/PR to dev → pnpm turbo run build lint + test (covers tsc, vite
build, i18n catalog type-parity, and the shared vitest suite).
release.yml — on a v* tag → install Tauri Linux deps (webkit2gtk-4.1, libsoup-3,
gtk-3, appindicator, rsvg, patchelf) + rustup, cache cargo/target, then
`pnpm --filter @parking/desktop bundle` signed with the updater key from Gitea
secrets (TAURI_SIGNING_PRIVATE_KEY + _PASSWORD). Collects .deb/.rpm/.AppImage +
their .sig, assembles latest.json (platform key linux-x86_64 — verified against
the tauri-plugin-updater target format), and publishes a Gitea Release via the
API with the built-in token (no marketplace release action needed).
Both workflows validated (PyYAML parse). No secret values committed — only
${{ secrets.* }} references.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Add apps/desktop, a thin Tauri v2 shell wrapping the SAME @parking/web SPA so
the desktop and browser UIs never drift: dev loads the Vite dev server (HMR),
prod bundles the web app's dist/. No business logic in the shell (device/auth/
ledger stay in @parking/server); deny-by-default capabilities.
apps/web (single UI source of truth):
- lib/origin.ts: centralize the backend origin (API_BASE/apiUrl/wsUrl from
VITE_API_BASE); no-op in the browser, lets the desktop build target Fastify.
- lib/kiosk.ts: block the right-click context menu in PROD only (dev keeps it +
devtools).
- lib/desktop-updater.ts: prompt-on-update auto-update (no-op in browser/offline)
→ downloadAndInstall + relaunch; i18n update.* keys (sq+en).
- .env.production: VITE_API_BASE wired to the Fastify origin for the bundle.
Desktop:
- window starts maximized (not fullscreen — operator keeps OS access).
- auto-update via tauri-plugin-updater + -process; self-hosted endpoint is a
PLACEHOLDER to fill in. Updater keypair: pubkey embedded in tauri.conf.json;
private key + password kept OUTSIDE the repo (~/.parking-updater-keys) and as
TAURI_SIGNING_* build secrets.
- Turbo build is a no-op; the real signed bundle is `pnpm --filter
@parking/desktop bundle` (verified → .deb/.rpm/.AppImage + .sig signatures).
Verified: cargo check clean; turbo run build lint 14/14 green; i18n parity holds;
no key/sig/bundle artifacts in the repo.
Wiki (security + desktop analysis recorded alongside):
- new concepts/tpm.md (TPM 2.0: how it works, sealed-LUKS auto-unlock + non-
extractable signing key, limits — live-root, bus-sniff — TPM-vs-ATECC608 by
platform).
- new decisions/desktop-shell-tauri.md (Tauri v2 over Electron; best-case Ubuntu
26.04 LTS, worst-case Windows+WSL → kiosk browser; full as-built).
- pull-the-disk attack trace on append-only-event-chain; ATECC608 not-in-a-PC
caveat; cross-links from disk-os-hardening / threat-model.
- open-questions #11 (appliance WebKitGTK), #12 (TPM hardening impl), #13
(startup verifyChain self-check); index/overview/log/standing-decisions.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Shift screen:
- The standalone ShiftControl block is gone from /shift. The open/CURRENT shift now
appears at the TOP of the shift list (CURRENT badge, live figures synthesized from
the X-report), unified with history. Selecting it shows its live activity log.
- Shift ACTIONS moved into the current shift's detail pane, each opening a MODAL:
End shift (confirm → signed Z-report result), drawer voucher (Mandat in/out),
takings-so-far (X-report). When no shift is open, a Start-shift button shows.
- The current shift's log auto-refreshes (5s); a closed shift is bounded by its
window. /setup/shifts stays read-only history (no manage props). Deleted the now-
orphaned ShiftControl.tsx.
Layout:
- Every screen is now full-width like /booth — stripped the per-screen
`mx-auto max-w-*` caps (Logs, Subscriptions, Plans, Tariff, Users, Roles, Setup
layout, Shifts). The shell <main> already provides padding.
Build+lint 12/12 (i18n parity). Verified a live open shift surfaces as the CURRENT
list entry.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Rework the shift screen into a master/detail view on /shift: the shift CONTROL
(open/close, drawer vouchers, X-report) on top, then a two-pane history below —
shift list on the LEFT, the selected shift's signed activity log on the RIGHT.
- Timeframe presets replace the bare from/to inputs: Yesterday / Last week /
Last month / All / Custom (custom reveals the date pickers). Filters the shift
list by start time.
- Activity log = every ledger event in the selected shift's [start, end] window
(entries, exits, payments, vouchers, anomalies, the Z-report), rendered like the
booth live feed (same EVENT_STYLE), with the shift's drawer reconciliation in the
pane header.
- Scope unchanged + enforced SERVER-SIDE: an operator sees only their own shifts
(no operator filter); an admin (shift:cash) sees all + the operator filter. The
list auto-selects the newest shift.
API: /api/events gains an optional `until` (ISO) upper bound so a shift's window
can be fetched ([start,end]); fetchEvents passes it. Verified on live data: a
closed shift window returns just its 20 events out of 260.
Build+lint 12/12 (i18n parity). The same component also backs /setup/shifts.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
A subscriber entering outside their plan's window owes a deferred charge, but
nothing printed — they had no paper proof a fee was pending. Print a best-effort
ADVISORY slip at entry ("PARKIM — JASHTË ORARIT"): holder, entry time, "entered
out-of-window (window opens HH:MM)", and the key line "⚠ fee computed at exit"
+ the occurrence number. It is NOT a payable ticket and carries NO amount — the
total is computed at the booth on settlement, combining early-entry AND any
late-exit time into one number (windowOwedBetween over the whole stay).
Best-effort like the Z-report / subscription card: printed AFTER the barrier
opens and fully swallowed, so a missing/failed printer never blocks entry. New
printWindowChargeNotice (booth-print.ts) via the generic printReport; wired into
the subscription entry flow when an out-of-window entry charge applies.
(The "both charges at the booth" requirement was already satisfied by the
windowOwedBetween fix — verified: early-entry + late-exit minutes combine in one
calc at lookup/exit. This commit only adds the entry paper trail.) Build+lint 12/12.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The exit-gate refusal for an unpaid out-of-window subscriber charge rendered as
the raw code `reason.sub.refused.unpaidWindow` — the code + English fallback
existed in @parking/shared but the reason.* catalogs had no entry. Add it to
both catalogs with the {{amount}}/{{currency}} params the gate passes.
EN: "Exit refused — out-of-window charge unpaid ({{amount}} {{currency}}); pay at the booth"
SQ: "Dalja u refuzua — detyrim jashtë orarit i papaguar ({{amount}} {{currency}}); paguaje në kabinë"
Build+lint 12/12 (i18n parity).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The tariff-bridge owed amount summed TWO charges — the early-entry gap +
a "late-exit" gap — and the exit gap (outOfWindowGap edge:"exit") always
measured back to the PREVIOUS window close, even for a subscriber still BEFORE
their window. So a car that entered ~30 min early showed ~12h owed (4,100 ALL)
the moment it was looked up, instead of ~100 ALL.
Replace the two-gap sum with a single correct primitive,
minutesOutsideWindow(timeframes, tz, from, to): the minutes within the actual
stay [entry, now] that fall outside the allowed window (covering early entry AND
late exit, bounded by the stay, weekend/off-days free). windowOwedBetween prices
those minutes once as a transient stay (so increments + daily cap apply) against
the tariff in force at entry. Both the exit gate (subscription-flow) and the
booth quote (pay-station) now use this one source of truth — they can't disagree.
Verified on the live occurrence: was 4,100 ALL, now 100 ALL (9 min outside →
one increment). 87 shared tests (6 new regression cases incl. the phantom span).
Build+lint 12/12.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
A subscriber entering outside their plan's allowed window gets a deferred
transient charge (windowOwedMinor, collected/gated at exit) — but it was
SILENT at the booth: the entry showed as a plain subscriber pass with no hint
money is owed, so the operator only discovers it at exit.
Surface it: add a "out-of-window — owes fee" badge on any entry/exit event
carrying windowOwedMinor > 0, so the operator sees immediately that this
subscriber owes a fee. Also type the window-charge fields on LedgerPayload
(were riding the open-ended index signature).
Behaviour is otherwise unchanged and correct — verified the live "Mon Kukaleshi"
entry: entered 20:29 local (before the 21:00 Mon–Sat window, grace 5m), owes
100 ALL for 18:29–18:55Z, stamped on the signed entry, still owed, gated at
exit. Subscribers get no ticket by design. Build+lint 12/12.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Addresses three issues with the plan catalog screen:
1. Retired plans had NO actions (the action cell was gated on "current
version", which a retired plan lacks) — so there was no way to make one
in-force again. Add POST /:planId/reactivate (inverse of retire) + a
Reactivate button on retired plans.
2. No delete. Add DELETE /:planId, allowed ONLY when zero subscriptions
reference the planId (any version) — a referenced plan version must survive
for reproducible repricing/audit, so an in-use delete returns 409 and the UI
says "retire it instead". The Delete button only shows when the plan has 0
subscribers.
3. The 6-column table overflowed max-w-3xl: action buttons overlapped and the
status badges wrapped to a second line. Replace it with a CARD list (one card
per planId, grouped across versions): name + status on top, price · hours ·
effective on a wrap row, "used by N" expandable to holder names, and actions
on their own bordered row — nothing overlaps, badges stay inline.
Build+lint 12/12 (i18n parity). Verified on a DB copy: unused plans report
deletable; retire→reactivate flips active back.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Deleting versioned plans is unsafe (a plan version referenced by a subscription's
planVersionId must survive for reproducible repricing/audit) — so instead of
delete, give the admin the VISIBILITY they actually needed:
- Hours column: a compact timeframes summary ("Hën–Pre 21:00–08:00" / "24/7"),
so two same-priced plans are distinguishable at a glance.
- Period + currency are already in the price cell; the hours column removes the
remaining ambiguity between night/day plans.
- "Used by" column: a count of subscriptions on each (current) plan (active /
total), expandable to the holder names — so you can see what depends on a plan
before retiring or replacing it. Computed client-side from the existing
subscriptions list (both screens are admin-grade; no new endpoint).
Build+lint 12/12 (i18n parity).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Selling/quoting a subscription resolved the plan version against `validFrom`,
but validFrom is a DATE (midnight UTC for "starts today"). A plan published
later the same day (effectiveFrom 15:22) then failed `effectiveFrom ≤ validFrom`
(00:00), so resolvePlanVersion returned null → "no active plan for that planId",
and the form's selectedPlan went null (hiding the new count field too).
The plan/price in force is determined by WHEN THE SALE HAPPENS, not by the
coverage start — like a tariff, the customer buys today's published rate. Resolve
at new Date() in all three sites (validate, priceSale, /quote); validFrom is kept
only for span pricing. Verified the two live plans now resolve.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
When subscriptions moved to the plan model the span became start + end dates,
which lost the simple "renew for N months/weeks/days" input — the operator had
to hand-compute the end date. (quantity is CARS, a separate axis, not periods.)
Add a count field to the sell form: the operator types e.g. 3, and validTo is
auto-derived as validFrom + count × the plan's period (day/week/month), with the
same month-overflow clamp the server uses (Jan 31 +1mo → Feb 28) so the preview
matches what's stored + charged. The end-date field stays directly editable for
an irregular span (the hotel checkout case), and editing it isn't overwritten by
the count effect. The count row shows the plan's unit ("× month").
Build+lint 12/12 (i18n parity).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The timeframes model was a coarse weekday/weekend split, which couldn't express
"open Saturdays" or different rules on a specific day — and it didn't match the
V2 tariff, which already has a proper per-day-of-week picker (Hën–Die).
Replace PlanTimeframes { weekday, weekend } with { days[], fromMin, toMin }: the
allowed window applies only on the selected days (0=Sun..6=Sat; empty = every
day); on unselected days the subscriber parks free. A "night plan, free
weekends" is just days [Mon..Fri] with a 20:00→08:00 window — the exact case
from before, now expressible alongside any other day combination.
outOfWindowGap reworked to the days model (per-day membership test instead of
the weekend helper); the plans editor reuses the tariff composer's Mon-first
checkbox row and the shared tariff.dow0..6 labels. No production plans carry
timeframes yet (feature shipped today), so the shape changed directly with no
migration. Unit tests updated + extended (Saturday-only, every-day, weekday
night); 81 shared tests pass. Build+lint 12/12.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
`db:migrate` (and drizzle-kit's config default) resolved DATABASE_URL to
`./parking.sqlite` relative to packages/db — a stray, half-empty leftover DB,
not the real store at apps/server/parking.sqlite. Running `pnpm --filter
@parking/db db:migrate` with no env therefore migrated the wrong file and
failed on its broken state, while the real DB went untouched.
Default DATABASE_URL to ../../apps/server/parking.sqlite in both the db:migrate
script and drizzle.config.ts (an explicit DATABASE_URL still overrides). The
stray packages/db/parking.sqlite was untracked + already gitignored (*.sqlite);
deleted it from disk. Now `pnpm --filter @parking/db db:migrate` targets the
appliance DB out of the box.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Three subscriber enhancements driven by real scenarios (migration 0011, all
additive columns — backward-compatible).
1. QUANTITY. One subscription covers N cars (a family pays once for two). Sale
amount = span price × quantity; maxConcurrent defaults to the quantity so all
N cars can be inside. Quantity rides in the payment payload.
2. PLAN TIMEFRAMES → TARIFF BRIDGE. A plan may restrict WHEN a subscriber may
park (e.g. weekday 20:00→08:00, weekend all-day). A scan outside the window is
NOT refused — the out-of-window minutes are charged at the normal TRANSIENT
tariff (the subscriber is a transient for that time):
- early entry: arrival → window-open, DEFERRED (signed as windowOwedMinor on
the vehicle_entry payload), collected at exit;
- late exit: window-close → departure, and exit is GATED
(sub.refused.unpaidWindow) until paid at the booth.
Pure, tz-aware outOfWindowGap in @parking/shared (12 unit tests); pricing
reuses computeFee + the active tariff version
(apps/server/src/subscription-window.ts). The exit refusal is a host-ONLINE
business gate — the fail-open rule still governs the offline path.
3. RESERVED SPOTS. Site toggle reserve_subscriber_spots: occupancy holds
max(0, quantity − itsCarsInside) per active subscription, so transients see
"full" sooner; effectiveFree = capacity − count − reserved. Subscribers are
never gated by full.
UI: quantity field + ×N quote (SubscriptionManager); timeframes editor
(SubscriptionPlansManager); reserve checkbox (SiteSettings); booth pay modal
shows an "OUT-OF-WINDOW" charge and takes payment to clear the exit gate.
Verified on a copy of the live DB: qty 2 = 2× price; a night-plan 19:30 entry →
30min/15,000 ALL owed, stamped + paid → gate clears, chain verifies; the reserve
toggle holds a qty-2 sub's 2 spots. Build+lint 12/12; 80 shared tests pass.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Re-model subscription pricing from per-row, operator-typed prices into an
admin-composed, versioned PLAN CATALOG (the tariff pattern). The operator now
SELLS by picking a plan over a date span; the price is LOOKED UP, never typed —
removing the fat-finger risk on a money field — and day/week/month periods make
the hotel "guest stays 1–N days" case a daily plan over a check-in→check-out span.
- Schema/migration 0010: new `subscription_plans` (immutable, effective-dated,
keyed by a stable planId; period day/week/month + per-period price + active
flag). `subscriptions` gains planId/planVersionId; period enum widened. Seeds a
"Monthly" plan from the existing site default price (no data loss).
- Pricing (pure, unit-tested in @parking/shared): periods = ceil(span / period),
amount = periods × per-period price. Ceil = any started period is full (hotel
practice). `resolvePlanVersion` picks the latest active version ≤ sale instant.
- Backend: new admin-only plan CRUD (`subscription:plan` permission); reworked
sell path derives the amount from the plan; `POST /api/subscriptions/quote`
returns a server-computed quote so the operator can't override it. The
signed-payment sale fix is unchanged — only the amount SOURCE moved; payload
now carries planId/planVersionId/periods. Updates never re-sell (price frozen).
- Frontend: SubscriptionManager sell form swaps the price field for a plan
picker + start/end dates + a live quote line. New SubscriptionPlansManager
(Setup tab) for the admin catalog. i18n (sq+en) for both.
Verified on a copy of the live DB: 0010 applies (existing subs intact), a
3-night hotel sale prices to 2,400 ALL, appends one signed payment with
planVersionId, chain verifies. Build+lint 12/12; 68 shared tests pass.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The "X-REPORT / Z-REPORT" labels are till-accounting jargon operators don't
recognize. Relabel the user-facing strings to plain wording in both catalogs —
SQ: "ARKËTIMET DERI TANI" / "MBYLLJA E TURNIT"; EN: "TAKINGS SO FAR" /
"SHIFT CLOSE". The X/Z naming stays in code (xReport/zReport keys, the
shift_z_report event, currentReport) and the wiki.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Let the operator see, on demand during an open shift, the opening float
inherited, cash/card collected so far, pay-ins/pay-outs, and the current
expected drawer balance — without closing.
GET /api/shift/report (shift:read; 204 when no shift is open) returns the same
drawer projection the Z-report computes. Factored that math into a shared
ShiftService.#summariseWindow(open, asOf) used by BOTH the X-report (asOf=now,
read-only) and close()'s Z-report (asOf=endedAt, signed), so the two can't
drift. The X-report appends NOTHING — it's a snapshot, not an accountability
mark; the Z-report at close remains the signed record.
UI: a "Takings so far" button on the shift control reveals a cyan X-report
panel; the header still shows the live drawer total for the at-a-glance figure.
Verified against a copy of the live DB: X figures match drawerBalance(), the
drawer identity holds, zero events appended, chain still verifies.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Replace the single signed-± cash_movement with two distinct financial
documents — the direction is the event TYPE, not the sign of an amount:
cash_in = Mandat Arkëtimi (receipt / pay-IN, +) voucher AR-NNNN
cash_out = Mandat Pagese (disbursement / pay-OUT, −) voucher PA-NNNN
Each carries a positive magnitude, voucher number, reason, the operator who
raised it and the admin who authorized it, and prints an Albanian slip.
Authorization changes from admin-only to operator-RAISED / admin-AUTHORIZED:
any shift:create holder raises the voucher, but POST /api/cash-voucher only
commits when authorizedBy is a real admin (shift:cash) re-entering their
password (verified server-side). Keeps the float control while letting the
operator do the booth paperwork.
Legacy cash_movement events are kept — they still verify and still fold into
the drawer (signed-±); the append-only chain is never rewritten. The drawer
fold and the Z-report window now sum all three types.
Verified against a copy of the live DB with the real signing modules:
cash_in 3000 + cash_out 5000 → drawer −2000, hash-chain verifies OK.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Creating a priced subscription wrote only the mutable `subscriptions`
master row and appended NOTHING to the signed ledger — so the cash an
operator collected showed in the live feed, drawer, and shift Z-report
nowhere, leaving no signed trace. A booth operator could sell
subscriptions and pocket the money untraceably — the exact
operator-as-adversary path the append-only signed ledger exists to close.
Found live: 3 priced subscriptions (27,000 ALL) had zero payment events.
Selling a priced subscription now appends a signed `payment` event at
create time: amount = priceMinor x months (full multi-month prepay),
operator-chosen tender (cash->drawer / card->bank), payload
{ subscriptionSale: true, permitId, operator, months }. Folds into the
shift Z-report/drawer with no new summing logic; the feed badges it
"subscription sale" and resolves the holder name. The create response
returns the recorded { sale }; subscriptionRoutes now takes the EventLog
and ShiftService.
Not hard-gated on an open shift (a sale can happen outside the booth money
path) — it warns instead. The 3 historical off-book sales are not
back-fillable (append-only forbids forging dated events) — reconcile via
cash_movement or a Z-report note.
Verified against a copy of the live DB with the real signing modules:
signed payment appended, hash-chain still verifies, lands in shift cash
totals. Build + lint 12/12.
Wiki: subscription "Collecting the fee" deferred -> BUILT (+ the off-book
hole and why); shift sale-folds-in; threat-model worked example
("store the price != account for the sale").
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Surface the advisory ANPR plate (device_events kind="read", keyed by
session identity — unsigned, prunable, never an access decision) next to
entry/exit events in the live feed and on active-session rows.
Resolved at serialize time (new plate-lookup.ts; prefers an entry read;
one device_events scan per page) like subscriber-name enrichment — the
signed ledger is untouched. Adds plate? to the shared LedgerEvent and to
ActiveSession/SessionLookup; a small amber badge in the UI.
Caveat: a vehicle_entry is signed + pushed over WS before the async ANPR
read lands, so a fresh feed row may show no plate until reload; always
present on active sessions.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
priceMinorPerIncrement is per BILLING INCREMENT, not per hour. Documented
the effective-hourly formula (price x 60/incrementMin) as a callout after
recurring "Lab is wrong" confusion (weekend 3h=600 not 300 was correct),
and filed a per-hour-preview composer UX candidate under Open.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
A stepped ("up-to") default card prices the whole stay as one total, so the V2
engine short-circuits to steppedFee and NEVER consults windowed cards — any
time/seasonal tiers would silently never fire. Found live: an active tariff had a
stepped base plus weekday-night + weekend tiers, and every 3h stay priced 600 ALL
regardless of hour/day because the tiers were dead.
- validateTariffV2 now rejects a stepped defaultCard combined with windowedCards,
with an actionable message (switch the base to ladder/flat, or remove the tiers).
- Composer shows an inline red warning the moment base mode is stepped and tiers
exist; publishing is blocked server-side regardless.
- ApiError now carries the server's problems[], so the publish error surfaces the
SPECIFIC reason instead of a generic "invalid tariff structure".
- 2 new validation tests (55 pass).
Wiki: tariff, log.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Owners often state rates as a total-by-duration matrix (0-1h=200, 0-3h=500,
0-6h=800, 0-9h=900, 0-12h=1000) that the marginal hourly ladder can't express
(the ladder sums per-increment rates; this is cumulative totals at thresholds).
Add STEPPED as a third pricing mode alongside the ladder and flat.
- @parking/shared: TariffStep {uptoMin, totalMinor} + a `steps[]` field on V1
structures and V2 cards (mutually exclusive with blocks/flatMinor). steppedFee():
smallest tier with uptoMin >= duration wins (INCLUSIVE boundary), the top tier
repeats as a per-day cap; wired into computeFeeV1 + computeFeeV2 (V2 default card
only — a whole-stay total can't be sliced per-increment by a windowed card).
Validation: ascending uptoMin, non-negative totals, no daily-cap-with-steps,
steps-only-on-default. priceSession/quote/booth/Lab price it via the shared core.
- Composer UI: a "By duration (up-to)" mode with an up-to/total table (base card
only). i18n modeStepped/steppedHint/stepUpTo/stepTotal/addStep (sq+en).
- 8 new unit tests incl. the exact owner matrix, multi-day repeat, overstay, and
validation (53 pass). Verified end-to-end via the UI: authored + published the
matrix, Tariff Lab prices it exactly (3h->500, 6h->800, 12h->1000, 2d->2000).
Wiki: tariff (three pricing modes + stepped semantics), log.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Test rates "in time" (overnight windows, daily caps, overstay) in seconds against
any tariff version, instead of waiting hours/days. No real ledger writes.
- Extract priceSession() into @parking/shared: the grace/overstay wrapper over
computeFee (unpaid -> entry..now; within-grace -> settled 0; grace-expired ->
overstay, a fresh period from grace-expiry). PayStation.quote() now calls it so
the booth and the lab can never diverge.
- API (tariffs.ts, tariff:read, read-only): POST /api/tariff/simulate prices a
hypothetical session (active/any version/inline structure) and returns the
priceSession outcome + a 30m..3d duration curve (see where the daily cap flattens);
GET /api/tariff/simulate/session/:identity prefills from a real ledger session.
- UI TariffLab.tsx at Setup -> "Tariff Lab": version picker, entry/asOf times,
optional payment+grace, category, and load-a-real-ticket. Admin-gated, available
on-site (useful to quote a dispute).
- 4 new priceSession unit tests incl. the ticket-1245791632490 overstay-not-zero
regression (40 pass). i18n lab.* + nav.tariffLab (sq+en). Verified live via the UI.
Wiki: tariff (priceSession + Tariff Lab as-built), log.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Rework paid-but-grace-expired sessions and add booth filters.
Overstay (was "stuck"):
- Stop silently aging out a paid transient whose walk-back grace lapsed with no
signed exit. Keep it listed with an OVERSTAY badge — a new parking period began
(re-parked) or the car is faulty/abandoned; it is not a system fault.
- No free exit: reopenBarrier refuses server-side once a transient's payment grace
has expired (allow only subscription OR paid-and-within-grace); the UI hides the
Open-barrier button on overstay rows and routes to the pay/exit modal. Closes a
hole where a stale payment authorized a free multi-day exit (operator-as-adversary).
- Price the overstay as a NEW period from grace-expiry -> now with its own daily-cap
ladder, NOT "full stay minus paid" (which a daily cap collapsed to 0 — ticket
1245791632490 owed ALL 0; now owes its real overstay). quote() gains periodStart +
overstay; SessionLookup/ActiveSession gain `overstay`. handlePayAndExit charges
whenever the session is payable (was: only if !alreadyPaid, skipping the overstay).
Filters (new ui/FilterBar): Active Sessions — search + status
(unpaid/paid/exiting/overstay) + transient-vs-subscriber. Live feed — search +
event (entry/exit/pay/void/anomaly) + direction + source (booth=manual vs reader).
All client-side over already-fetched data; matched/total count shown.
i18n parity (sq+en). Wiki: booth-exit-flow updated (overstay model, naming history,
no-free-exit security fix, new-period pricing; open question on grace-renewal noted).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The session list filled to its content height instead of clipping, so a long list
(29 sessions) overflowed the column instead of scrolling — unlike the live feed.
The ActiveSessions Panel sat in a plain block wrapper, so it sized to content and its
inner `h-full overflow-y-auto` had no bounded height to scroll within. Make the wrapper
a flex column and give the Panel `flex-1` so it fills the column; the inner scroll area
is then bounded and scrolls — matching the live-feed treatment.
Verified live (Playwright): the scroll container is now 437px tall over 1118px of
content → scrollable, while the live feed is unchanged.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The ANPR plate was saved (device_events kind:"read") but had no UI. Extend
GET /api/snapshots/by-identity/:identity to also return plates[] (plate, confidence,
region, direction, snapshotId, at) for that session, and render each as a cyan
"Plate: AA558EE 100%" chip in the SnapshotStrip — so it shows in both the booth
event-detail modal and the pay modal, beside the evidence photo, no separate screen.
Deduped by plate+direction; session:read gated; i18n sq+en.
Verified: by-identity returns plates[] for a seeded read (200, AA558EE 0.999 Albania
entry). Build + lint green.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Rework the ANPR trigger to the real design: when a transient presses the button or a
subscriber passes QR/RFID, the entry/exit fires and takes its evidence snapshot — that
is the moment to recognize. snapshotAsync now takes the VisionClient and, after storing
each snapshot from an opt-in (config.anpr) camera, runs ANPR on the SAME image and
records the plate against the SAME session identity (device_events kind:"read" with
plate/confidence/region/snapshotId/source:"entry-exit-snapshot"). One image serves both
evidence and plate extraction; recognition fires only on a real entry/exit — no polling.
The entry/exit/subscription flows take an optional VisionClient and pass it through;
server.ts wires it. Removed the polling VisionReader and VISION_POLL_MS/VISION_DEDUPE_MS.
Advisory + fire-and-forget: a low-confidence/no-plate result records nothing, a vision
failure never delays or changes the open, and the plate does not feed the access
decision. Verified e2e: a simulated entry snapshot on an anpr camera (live fast_alpr)
stored the snapshot for the session and recorded {identity, plate:AA558EE, 0.999,
region:Albania, snapshotId}. Build + lint green.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Make the vision service genuinely configurable (was env-only).
- SetupWizard: an "ANPR" checkbox on the camera form (writes config.anpr; persisted
only when on; sq+en) — opt-in is no longer raw JSON.
- DeviceMonitor optionally takes the VisionClient and probes /health each tick, emitting
a "vision" pseudo-device → a Vision chip (ready/degraded/offline + recognizer) in the
booth footer when VISION_ENABLED, no chip when off. Widened the DeviceStatus category
union (server + web) + footer maps + devices.catVision. Verified: ready/fast_alpr when
up, 0 chips when disabled.
- apps/vision/.env.example (Python service) + a VISION_* block in apps/server/.env.example
(Node side) + a Configuration section in opencv-anpr-service.md covering all four
layers and the caveats: the two processes share the VISION_ prefix but need SEPARATE
.env files; bind /analyze to 127.0.0.1; cache model weights at deploy; an unbound anpr
camera recognizes but every read is refused.
Build + lint green.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Answers "is a recognized plate saved?" — now yes, for both transient and subscriber, as
an ANPR audit trail independent of whether it matched anything.
VisionReader now stores the snapshot bytes in `snapshots` keyed by identity=PLATE — the
same identity the flow signs its anomaly/event with — so GET /api/snapshots/by-identity/:plate
(the booth event-detail modal's snapshot strip) shows the car's photo against that
anomaly with no UI changes. It also records an unsigned device_events{kind:"read"}
breadcrumb (plate, confidence, region, model, snapshotId, and the dispatch outcome) as a
queryable recognition log. Switched from emitRead to calling ReadDispatcher.dispatch
directly (like qr-reader) to capture that outcome.
Non-blocking: a refused read (no session / unpaid / unknown plate) just returns
rejected — no barrier hold — and is logged with its snapshot for investigation. Plate
stays advisory (exit demands payment; subscription matches only a bound plate).
Verified e2e: a recognized AL plate with no open session signed exit.refused.noSession
(identity=plate), stored a 555KB snapshot under that plate, recorded the read breadcrumb
(accepted:false, reason "no open session"), and by-identity returned the image — the
refused read is fully investigable with its picture. Build + lint green.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
A VisionReader polls each opt-in camera (config.anpr===true, off by default) every
VISION_POLL_MS, captures a snapshot, recognizes via VisionClient, and on a confident
plate emits deviceEvents.emitRead({kind:"plate", value}) — the same event a physical
plate reader sends, so the existing ReadDispatcher routes it to the subscription/exit
flow unchanged (no flow rewrite).
The plate stays advisory by construction: the exit flow still demands a covering
payment, the subscription flow only matches a bound plate. Guards: low-confidence reads
dropped; debounce (VISION_DEDUPE_MS) so a parked car doesn't re-fire; per-camera
in-flight guard; idle when vision is off or no camera opts in. #recognizeOn is public
for a future on-demand (loop-edge/API) trigger.
Verified end-to-end: an in-memory anpr camera (AL plate image) + live fast_alpr service
→ VisionReader emitted exactly one {kind:"plate",value:"AA558EE"} onto the bus; debounce
held it to 1 emit over 7 polls. Build + lint green. Updates opencv-anpr-service
(trigger-wiring + per-camera opt-in marked done).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Node-side adapter to the apps/vision ANPR microservice (localhost HTTP: POST /analyze
with snapshot bytes, GET /health), returning a normalised VisionResult or null. Enforces
"advisory, never sole authority" at the boundary: opt-in (VISION_ENABLED, default off),
fail-soft (any error/timeout/unreachable → null, never throws into the lane → ticket
fallback), and re-applies the confidence floor (VISION_MIN_CONFIDENCE) on top of the
service's own low_confidence flag. Per-request AbortController timeout so a slow call
can't hang the barrier. Constructed in server.ts.
Verified: fail-soft (disabled/unreachable → null, no throw) and live end-to-end (Node
client → running fast_alpr service → AA558EE 0.999, region=Albania). NOT yet wired into
the read bus — the opt-in snapshot→DeviceReadEvent{kind:"plate"} trigger is the next
step. Build + lint green. Updates opencv-anpr-service (adapter gap marked done).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Record the verdict: the ANPR service is worthy to consume NOW as an advisory plate
IDENTITY source (Job 1) — the flows already treat a kind:"plate" read as first-class
(exit signs source:"lpr"; subscription matches read plate vs bound plates), so it feeds
an existing input with no flow rewrite. It is NOT worthy as the sole authority to open a
transient barrier (a plate is not a payment; spoofing needs Job 2 vehicle verification,
unbuilt) — gated by the confidence floor with ticket/manual fallback. Lists the four
gaps before consumption (VisionClient adapter, opt-in trigger, field accuracy,
weight-provenance). Next step is the adapter, not more model work.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Benchmarked fast-alpr's four fast-plate-ocr models via the full pipeline on real AL
plates (AA558EE, AA687KE), CPU. All four read both correctly; the default
cct-xs-v2-global-model wins on confidence (0.999/1.000) AND speed (33-39ms) and returns
region=Albania. The "European 40+country" model is WORSE here (~0.77 confidence, one
synthetic misread) — overturning the "EU model better for AL" assumption from the prior
research. Decision: no config change. Resolves the AL-accuracy-benchmark open item
(results table + finding added to opencv-anpr-service); weight-provenance remains the
one open recognizer item. Re-benchmark on real on-site captures once cameras installed.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Add a dev CLI (uv run python -m vision_service.cli <image>) that runs a recognizer on
an image file and prints the parsed plate(s) + confidence + region — fast feedback with
no HTTP. Also a package.json `recognize` script and a vision-recognize entry point.
Verified fast-alpr for real: installed the `alpr` extra, downloaded the YOLOv9 + CCT
ONNX weights (~11MB, cached offline under ~/.cache), and ran recognition on the
project's test image → "5AU5341" at 1.000 confidence, region "Czech Republic", ~40ms
on CPU, via both the CLI and POST /analyze.
Fixes result parsing against the actual fast-alpr API: ocr.confidence is a LIST of
per-character confidences (not a scalar) — reduced to one plate confidence via the MIN
(a plate is only as trustworthy as its weakest character); also surface ocr.region.
Extracted the per-result mapping into a pure plate_from_alpr_result + _reduce_confidence
and unit-tested them (no model weights needed). 7 tests pass; ruff + mypy strict clean;
full turbo build/lint/test green.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Skeleton of the host-side vision service per the packaging decision: a Python/FastAPI
app at apps/vision/, uv-managed, wired into the Turbo graph via a thin package.json
shim (dev/lint/test/build → uv/uvicorn/ruff/pytest). A per-package turbo.json sets
build outputs [] so the no-op build is warning-free.
Endpoints: GET /health (readiness + model version) and POST /analyze (raw
octet-stream body, so Node POSTs Snapshot.bytes directly; empty→400, oversize→413,
recognizer-not-ready→503). The recognizer is a Protocol with a StubRecognizer (no
models, boots/tests offline — the dev/CI default) and a FastAlprRecognizer (the real
MIT YOLOv9+CCT/ONNX stack, lazily imported; missing models ⇒ ready=False, not a crash)
— the device-adapter pattern applied to the model. fast-alpr + onnxruntime are an
optional `alpr` extra, so `uv sync` needs no model download.
Verified: turbo run lint|test|build includes @parking/vision and stays green; uv run
mypy strict-clean; uvicorn boots and serves /health + /analyze live; pnpm workspace
6→7. Not built yet: the Node VisionClient adapter, a Dockerfile + model fetch, and
Job 2 (vehicle verification). Updates the packaging decision (As-scaffolded) + log.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
julian