The box program carried 160 lines of dead code and a public repo held hotspot passwords

A tidy-up of main.py with the same MQTT, NVS and screen contract, so that
the next person can read it. Verified on the PC with a stubbed device
environment driving every branch (65 checks: boot from empty storage, each
card type, positive / negative / foreign / undecodable replies, the 10 s
timeout, lamp and reset topics, MQTT / WiFi / reader recovery, heartbeat,
OTA accept and reject paths); nothing here can run on the device itself.

Removed without changing behaviour: every commented-out line, duplicate and
unused imports, the timestamp helper and reset-UID constant nobody used,
the seen_uids set with its uid_count/uid_<n> NVS bookkeeping (never read for
anything; old keys stay in flash, ignored), labels that were "updated" but
never created, the serial recomputed on every pass, callback locals leaking
into globals, and the non-blocking WiFi reconnect that the blocking one on
the next line always overrode. Widgets now live in `ui` under plain names
(label_set / set_msg tolerate a widget not yet drawn, which is what lets
setup() read NVS before the screen exists); lamp colours are a table.

Deliberate small changes, all in the box's favour:

- Reads only card blocks 10 (type) and 12 (operator id). Blocks 4-9 held
  fields that stopped being published long ago. RFIDUnit.read() authenticates
  the block's own sector each call, so the reads do not depend on each other.
- A Reset card fires once per tap (tap_handled). Before, one left on the
  reader re-sent RESET after every server reply.
- A failed heartbeat publish drops the MQTT client so the throttled reconnect
  runs. Before, it flagged the RFID reader for a rebuild and never touched
  MQTT.
- No NVS write on every Product tap - it rewrote identical values.
- The serial is drawn at boot instead of "SR: LOADING..." until WiFi
  finished; "RECIEVED" fixed on screen.
- The personal hotspots (EHTISHAM, A16, okay., StormFiber) are gone: this is
  a public repo. One alternative network and one alternative broker remain
  commented, as the convention says.

Kept on purpose: an Operator message still carries the PREVIOUS card's uid.
That is a server-facing contract and changes separately.

version.txt is NOT bumped - nothing reaches a box until it is.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

CLAUDE.md

@@ -21,18 +21,18 @@ Devices update themselves at boot from this repo's `main` branch on Gitea, over
 
 ## How the program works
 
-Single-file, cooperative main loop: `setup()` once, then `while True: loop()`. No threads or asyncio; all state is module-level globals (functions declare `global`). Anything added to the loop must stay non-blocking: timers use `utime.ticks_ms()` deadlines (`ack_deadline`, the 5 s throttles in `check_wifi()` / `reconnect_mqtt()`), not sleeps. The exception is `connect_wifi()`, which blocks and retries recursively up to `MAX_RETRIES`; `loop()` falls back to it whenever WiFi is down.
+Single-file, cooperative main loop: `setup()` once, then `while True: loop()`. No threads or asyncio; all state is module-level globals (functions declare `global`). Anything added to the loop must stay non-blocking: timers use `utime.ticks_ms()` deadlines (`ack_deadline`, the 5 s throttle in `reconnect_mqtt()`), not sleeps. The exception is `connect_wifi()`, which blocks and retries recursively up to `MAX_RETRIES`; `loop()` falls back to it whenever WiFi is down.
 
 ### Identity and persistence
 
 - `serial` = hex of `machine.unique_id()`. It is the MQTT client id, the `device_serial` in every published message, and what every incoming message is matched against. Shown on screen as `SR:`.
-- NVS namespace `rfid_data` (`load_from_nvs()` / `save_to_nvs()`): `count` (product counter), `color` (last lamp colour), `operator_id`, and `uid_count` + `uid_<n>` (the `seen_uids` set, no longer used for counting). These survive reboots; a `Reset` card or the reset topic clears them.
+- NVS namespace `rfid_data` (`load_from_nvs()` / `save_to_nvs()`): `count` (product counter), `color` (last lamp colour) and `operator_id`. Older builds also wrote `uid_count` + `uid_<n>` (a set of seen UIDs that never affected anything); those keys are left in place and ignored. These survive reboots; a `Reset` card or the reset topic clears them.
 
 ### Card handling (`loop()`)
 
-- `read_all_fields()` reads the card's text blocks positionally with lengths `[4, 5, 6, 8, 9, 10, 12]` -> `(sku, color, size, article, remarks, cardtype, operator_id)`. Only `cardtype` and `operator_id` still matter; the first five are read but no longer published.
+- `read_card_fields()` reads two MIFARE Classic blocks from the card: block 10 (`CARD_TYPE_BLOCK`, the type as text) and block 12 (`OPERATOR_ID_BLOCK`). Blocks 4-9 once held sku / colour / size / article / remarks and are no longer read; each `RFIDUnit.read(block)` authenticates its own sector, so the reads are independent. The values are written into the card by the desk writer that the UiMetrix "Register Card Type" page drives - the box never consults the database.
 - `cardtype` selects the behaviour: `Operator` sets and persists `Operator_ID`, `Product` publishes a scan, `Reset` zeroes the counter and operator and publishes a `RESET` message. Anything else is ignored.
-- The `card` flag edge-detects a tap: set when a card is handled, cleared only once no card is present, so one tap publishes once.
+- `tap_handled` edge-detects a tap: set when a card is acted on (all three types, Reset included since 2026-09-15), cleared only once no card is present, so one tap publishes once however long the card rests on the reader.
 - After every publish the device enters `waiting_for_ack` for 10 s: further cards are ignored ("WAITING FOR RESPONSE") until a `response_data` (or `lamp_topic`) message addressed to this serial arrives, or the deadline passes ("NO RESPONSE RECIEVED"). `tag_counter` increments only on a positive ACK from the server, never on the scan itself, so the on-screen count is the server's count.
 - Quirk: the `Operator` branch publishes before `last_uid_str` is updated, so an operator message carries the previous card's UID (`-` after boot); a `Product` message carries the current UID.
 
@@ -56,16 +56,16 @@ Subscribes (each handler first checks the serial in the payload; messages for ot
 ### Connectivity and the light bar
 
 - Blue on the RGB unit means "not connected" (boot, WiFi down, MQTT down); once MQTT connects the persisted `lamp_color` (default green) is restored. After that, colours come only from `lamp_topic`.
-- A publish failure calls `reconnect_mqtt()`, which tears the client down and re-runs `init_mqtt()`, throttled to once per 5 s. A `check_msg()` error sets `mqtt_client = None`, which takes the same path on the next loop.
+- A publish failure calls `reconnect_mqtt()`, which tears the client down and re-runs `init_mqtt()`, throttled to once per 5 s. A `check_msg()` or heartbeat-publish error sets `mqtt_client = None`, which takes the same path on the next loop.
 - Status icons (WiFi ok/error, MQTT cloud/error, charging) are "updated" by drawing a new `Widgets.Image` over the old one at fixed coordinates.
 
 ### Screen and hardware
 
-- `init_ui()` draws everything with `M5.Widgets` at absolute 320x240 coordinates and keeps the handles in the `ui_elements` dict; change text through `safe_label_update()`. `label3` and `label5` are referenced but never created, so updates to them are no-ops. The battery bar is an `m5ui.M5Bar`.
+- `init_ui()` draws everything with `M5.Widgets` at absolute 320x240 coordinates and keeps the handles in the `ui` dict under plain names (`operator`, `count`, `msg`, `uid`, `serial`, `ip`, `battery`, `version`, the icons); change text through `label_set(name, text)` / `set_msg(text)`, which tolerate a widget that has not been created yet (that is what lets `setup()` load NVS before drawing). The battery bar is an `m5ui.M5Bar`.
 - Images are loaded from `/flash/res/img/` on the device (`Logo`, `emp`, `mach`, `rfid-tag-log`, `bar_c`, `wifi`, `wifi_error`, `MQTT_Cloud`, `mqtt_error`, `charging`, `not_charging`, all `.jpg`); they are not in this directory.
-- RFID Unit on I2C bus 0 (SCL pin 1, SDA pin 2, 100 kHz); if any RFID call throws, `rfid_re_init` makes the next loop pass rebuild the bus and reader. RGB Unit on pins 8/9 with 10 LEDs.
+- RFID Unit on I2C bus 0 (SCL pin 1, SDA pin 2, 100 kHz); if any RFID call throws, `rfid_needs_init` makes the next loop pass rebuild the bus and reader (`init_rfid()`). RGB Unit on pins 8/9 with 10 LEDs.
 
 ## Conventions
 
-- WiFi SSID/password and the MQTT broker/user/password are hardcoded near the top of the file, with previously used networks left commented out; switching networks means commenting/uncommenting those lines. The file already holds live credentials, so don't echo them into other files or chat.
+- WiFi SSID/password and the MQTT broker/user/password are hardcoded near the top of the file, with one alternative each left commented out (`UI-Matrix`, `scada.utopia.pk`); switching means commenting/uncommenting those lines. Personal hotspots that used to be listed were removed on 2026-09-15 - this is a public repo, do not add them back. The file already holds live credentials, so don't echo them into other files or chat.
 - The serial console never prints the card UID (only MQTT traffic, NVS saves/loads, connection state and errors), which is why the enrolment tool reads UIDs from MQTT instead of the USB port.