decay71 · Oct 3, 2026 · Oct 3, 2026
@@ -1,21 +1,28 @@
 # mUlt1ACE 
 
-> [!IMPORTANT]
-> **multiACE is in the running for the Snapmaker U1 Innovation Fund.** The fund backs
-> open-source projects for the U1. Community voting is open until Sep 30. Pick your
-> favorite projects.
->
-> [![Vote here](https://img.shields.io/badge/Vote%20here-0d9488?style=for-the-badge)](https://www.snapmaker.com/innovation-fund#vote)
-
 Started as a SnapACE fork, it has grown to over 5 times the original size, with around 90% of the code now its own and many unique features:
 
 [![ko-fi](https://ko-fi.com/img/githubbutton_sm.svg)](https://ko-fi.com/K3K610R4F9)
 
 [![Guides & Downloads](visitbutton.png)](https://postapocalyptic-diy.com/multiace/)
 
-## What's new in multiACE 1.11b
+## What's new in multiACE 1.20b-pre (prerelease)
+
+This is a prerelease: new features to try out before the next release. Please report anything odd.
+
+**A big thank you to our new collaborators.** Three developers joined the project and brought their own features with them:
+
+- **Simon-CR** - the ACE2-Open community firmware for the ACE 2 Pro, which lets multiACE read and write spool tags. multiACE now also runs on other Klipper printers, not only the Snapmaker U1 (his Voron is the first one), and a watchdog notices when an ACE silently stops answering and reconnects it.
+- **Tareku99** - multiACE as a ready-made package for the paxx Extended Firmware, so the firmware installs and updates it for you. A big thank you to [paxx12](https://github.com/paxx12) for the plan to ship multiACE with the rolling releases of the Extended Firmware.
+- **Godless50** - the ACE Pro gets its community firmware: flash it straight from the web UI, and read the tags of your spools on an ACE Pro too. Auto-dry can also follow an external humidity sensor.
+
+**New look for the web UI.** The web UI opens in a new design - the classic one is still there, one click in the Config tab.
 
-Fixes a regression in 1.10b: The Snapmaker and Polymaker vendors in the filament picker show again on firmware 1.6.0 and newer. 
+**Read spool tags on the ACE Pro.** With the community firmware, press *Read* in the filament picker and the spool's tag is identified and matched to your spool list. (Needs the community firmware on the ACE Pro.)
+
+**More robust connection.** If an ACE drops off, multiACE keeps trying to reconnect it in the background; a disconnected unit is shown as offline instead of a confusing state, and you get no pause when nothing is printing.
+
+**Also:** clearer messages when a spool has no tag code, small web UI and log fixes.
 
 
 ## What's new in multiACE 1.10b
@@ -664,6 +671,19 @@ All content is reviewed by humans before inclusion.
 
 - **Popstar (forum.snapmaker.com)**
   - **the default PLA swap temperature is set to 220 °C**
+
+- **[Simon-CR](https://github.com/Simon-CR)**
+  - **Other Klipper printers** - multiACE runs on printers other than the Snapmaker U1.
+  - **Connection watchdog** - notices a silent ACE and reconnects it.
+  - **ACE2-Open community firmware** for the ACE 2 Pro - reading and writing spool tags.
+
+- **[Tareku99](https://github.com/Tareku99)**
+  - **paxx Extended Firmware package** - multiACE as a package the firmware installs and updates.
+
+- **[Godless50](https://github.com/Godless50)**
+  - **ACE Pro firmware flashing** from the web UI.
+  - **Tag reading on the ACE Pro** through his community firmware.
+  - **External humidity for auto-dry.**
 
 
 

@@ -1 +1 @@
-1.11b
+1.20b-pre
@@ -88,16 +88,6 @@ pa_sync: true
 #tag_write_format: openspool
 #tag_write_uid_sku: true
 
-# Gen-1 (ACE Pro) tag tunnel: when true, a slot the unit's own reader could
-# not identify gets ONE opportunistic page read through the community
-# firmware's RC522 tunnel (CV1.3.87x builds), so third-party spools surface
-# with their card UID (OpenSpool tags also with material/colour). false =
-# the unit keeps its normal status-only tag path; ACE_TAG_READ remains the
-# explicit probe either way. On a unit without the tunnel this option only
-# produces one klippy.log line.
-# Live: ACE_SET_TAG_TUNNEL ENABLE=0|1
-#gen1_tag_tunnel: false
-
 # Humidity-controlled drying. Only an ACE 2 measures; an ACE Pro can only
 # FOLLOW a master (picked per unit in the web drying panel). These are the
 # defaults; per-unit overrides via ACE_SET_AUTO_DRY (web drying panel).

@@ -82,7 +82,7 @@ sequence, but nothing on the wire:
 
 | entry id | image | size | CRC-16 | md5 |
 |---|---|---|---|---|
-| `1.3.863-opencubic` | OpenCubic ACE 1 Pro CFW v1.1.1 release asset `ACE_V1.3.863_20260716.bin` | 113720 | `0xC110` | `9f7b9a678a96caf98d6a08842d3ff971` |
+| `1.3.871-tunnel` | Godless50/ACE-PRO-v1.-NFC-UID asset `ACE_V1.3.863_tunnel_ops.bin` (OpenCubic CFW base + UID stub + RC522 tunnel, reports CV1.3.871) | 114632 | `0x1AC9` | `219df3df77f7c7e1e15a79d580a2379e` |
 | `1.3.863-stock` | clean stock `ACE_V1.3.863_20250518.bin` (rollback target) | 105652 | `0xDEFB` | `dcd04589dcadd5b4feab66d33e772531` |
 
 To add an image: flash a dry run with it, read size/CRC/MD5 off the

@@ -1,4 +1,4 @@
-# Gen-1 (ACE Pro) tag tunnel - `ACE_TAG_READ`, `ACE_SET_TAG_TUNNEL`
+# Gen-1 (ACE Pro) tag tunnel - `ACE_TAG_READ`
 
 The ACE Pro's own reader understands **Anycubic tags only**: a third-party
 spool (OpenSpool NDEF, a blank NTAG, a Bambu/Snapmaker MIFARE chip) arrives
@@ -34,17 +34,25 @@ Companion documents:
   NTAG capability container additionally triggers a bounded read of the
   OpenSpool user area (pages 4..39), decoded by the shared `ace_rc522`
   OpenSpool decoder.
-* **Falls back automatically** (when enabled): a slot that is occupied and
-  whose tag the firmware did not identify - or whose SKU matches no table
-  entry - gets **one opportunistic read per insert**. The result lands in
+* **Reads only on demand.** Every tunnel TRANSCEIVE blocks the firmware
+  for about 7 s (its lamps blink, its gates flicker), and a full read
+  holds about ten of them. So nothing reads automatically, neither at
+  connect nor after an insert (a session during an insert broke the
+  insert and took the unit off the bus, HW 2026-10-03): a tag is read
+  only by `ACE_TAG_READ`, i.e. the web Read button. The result lands in
   multiACE's own per-unit store, never in `_info_per_ace`. Binding is
   **gated on attribution**: the two slots of an antenna pair share one RF
   path, so the card UID is offered to the **existing** tag-bind path
   (`_spool_bind_by_tag(..., unbind=False)`) **only when the partner slot
-  (`slot ^ 1`) reads empty in the same status**. An occupied OR
-  unknown/absent partner still stores and surfaces the read but does not
-  bind (a wrong first binding has no repair path on a Gen 1); the operator
-  probe `ACE_TAG_READ` is the explicit override and always binds.
+  (`slot ^ 1`) reads empty in the same status, or when the partner's
+  card UID is already known** (its last tunnel read, or a UID code on the
+  spool bound there). A known partner card is put to sleep with HLTA
+  during the read, so the next SELECT answers with the other card; a
+  result can then never be the partner's. An occupied partner with an
+  unknown card still stores and surfaces the read but does not bind (a
+  wrong first binding has no repair path on a Gen 1); the operator probe
+  `ACE_TAG_READ` is the explicit override, always binds and also skips a
+  known partner card.
   `unbind=False` on purpose: a tunnel read must never release a binding
   the vendor path owns.
 * **Surfaces it.** `get_status`: the slot's `uid` / `tag_format` are filled
@@ -76,38 +84,89 @@ ACE_TAG_READ ACE=<n> SLOT=<0..3> [PAGE=<n>]
   [multiACE] ACE 0 slot 2 page 0: 04 22 52 FC 51 C8 2A 81 32 48 00 00 E1 10 6D 00
   [multiACE] ACE 0 slot 2: UID 04225251C82A81 (ntag) - third-party tag
   ```
-* It **moves nothing** (no lane rotation) and is safe while printing; the
-  tag must face the coil, so a read can fail on a stationary spool.
-* It works with the enable flag **off** - the command itself is the
-  explicit consent. The result feeds the same store/bind/status path as an
-  automatic read.
+* When no card answers and the lane may move (no print running, the slot
+  feeds no head), it rotates the spool forward in 20 mm steps until a card
+  answers (at most 650 mm, below the shortest path to a combiner), reads
+  it and pulls the lane back by the same distance. Otherwise it reads in
+  place.
+* It is the only tunnel read; nothing reads automatically.
 * On an ACE 2 the command keeps its previous meaning (rotate + read +
   bind); the Gen-1 branch is only taken for a non-V2 protocol.
 
-## The enable flag
+## Always on
 
-Automatic fallback traffic needs an explicit opt-in:
-
-```
-ACE_SET_TAG_TUNNEL ENABLE=0|1 [PERSIST=0|1]
-```
-
-* Default **off** (`#gen1_tag_tunnel: false` in `[ace]`); the setter is
-  live + write-through, `PERSIST=0` = RAM only until restart.
-* With it off, the **only** tunnel traffic is a command the user typed.
-* On a unit without the tunnel the flag is inert (one `klippy.log` line,
-  no console noise, no repeated probes).
+`ACE_TAG_READ` uses the tunnel on every ACE Pro whose firmware matches
+the tunnel builds (`CV1.3.87x`) **and** answers a probe op through the
+stub; a stock unit never receives tunnel traffic, so there is no switch. To go
+without it, flash the stock firmware. The former `gen1_tag_tunnel` option
+is still read and ignored, so an existing config line does not halt
+Klipper; `ACE_SET_TAG_TUNNEL` only reports that the switch is gone.
 
 ## Limits
 
 * **Community firmware required.** The tunnel exists only on the
   community build `CV1.3.87x` (verified reference `CV1.3.871`). Stock and
   the UID-only community image are never touched.
-* **The tag must face the coil.** A Gen-1 has no host-side motor control
-  (`feed_filament`/`unwind_filament` are refused by the stock firmware), so
-  an automatic read cannot rotate a spool to find the tag. The one attempt
-  per insert is timed to the firmware's own insert procedure, which already
-  rotates the spool; if the read misses, use `ACE_TAG_READ` to retry.
+* **Searching for the tag.** The tag only answers while it faces the coil.
+  When `ACE_TAG_READ` finds no card and the lane may
+  move, it rotates the spool forward (`feed_filament`, 20 mm
+  steps, at most 650 mm, after the slot reports `ready`), stop at the first
+  card and pull the lane back by the distance moved (`unwind_filament`;
+  the ACE Pro has no decoder, so it is the commanded distance). Never
+  during a print, never on a slot that feeds a head. A read starts only
+  once no slot of the unit reports `preload` (the firmware's own pull-in,
+  17-25 s measured). The
+  pull-back waits up to 60 s while the unit is busy with another slot
+  (it answers FORBIDDEN then).
+* **Unreadable card at the field edge.** A card that answers SELECT but
+  whose page read fails is re-read further in (+20, +40 mm) and just
+  before the hit (-20 mm), like the ACE 2 centring.
+* **Whose card?** When the partner slot holds an unknown card, the
+  partner lane is rotated (20 mm steps, at most 200 mm, then back): the
+  card leaves the field -> it is the partner's (stored for the partner,
+  our lane is searched with it skipped); it still answers -> ours, bind.
+  For `ACE_TAG_READ` only; the partner must be idle
+  and feed no head.
+* **Unit-wide wait.** The ACE Pro refuses motor commands while any slot
+  of the unit still moves, so a search starts only when every slot is
+  ready or empty.
+* **Firmware identity is vetted.** The community firmware reads tags on
+  its own, but cannot say which bay of the shared antenna answered (or
+  replays an old record). While the partner slot is occupied, our own
+  read decides: our OpenSpool read replaces the firmware identity, and a
+  card of ours whose user pages were ALL read without an OpenSpool record
+  hides a firmware OpenSpool identity. Until such a read exists (pending,
+  running, or only the UID came through) the firmware identity stays
+  shown.
+* **Never during an insert.** A tunnel session holds the reader the
+  firmware needs to pull in and identify a new spool. No session starts
+  while any slot of the unit reports `preload` or `shifting`, and a
+  running session stops at its next op when an insert starts or a gate
+  of the unit changes; the reader is released at once and the read is
+  retried when the unit is quiet. A session overlapping an insert made
+  the preload drop the new spool and crashed the ACE off the USB.
+  Lanes the session moved itself are not counted: the ACE Pro reports
+  our own search step as `shifting` too, and a real re-insert on such a
+  lane still flips its gate. While a stopped read waits for its retry,
+  the slot keeps the identity it showed before.
+* **Only real moves count.** A search step counts only when the ACE
+  answers `success`; `empty` ends the search. A spool pulled or
+  re-inserted during the search ends it, and the new spool is never
+  pulled back.
+* **Survives a restart.** The reads are persisted (`ace__gen1_tag_reads`,
+  save_variables) and restored at startup; an empty slot evicts them, a
+  fresh read replaces them. A spool swapped while the printer was off
+  keeps the old UID until its card is read again (same as the ACE 2).
+* **Never during a print.** A tunnel session holds the unit's reader;
+  `ACE_TAG_READ` stays available during a print (read in place, no
+  rotation).
+* **A TRANSCEIVE blocks the firmware ~6.9 s** on the tunnel build (every
+  page read, data or not; FA log 2026-10-03). The reply wait for op 3 is
+  10 s, and any op without a reply ends the session (release only), so
+  no further op is queued into a unit that is still blocked.
+* **One read per unit at a time.** A manual `ACE_TAG_READ` issued while
+  another read runs on the same unit is queued behind it (at most 180 s)
+  instead of being refused.
 * **Two antennas, two bays each; reader channel = bit-swap.** Antenna 1
   covers slots 0 and 1 (reader channels 0 and 2), antenna 2 covers slots 2
   and 3 (channels 1 and 3). The reader **channel** for a slot is
@@ -116,7 +175,7 @@ ACE_SET_TAG_TUNNEL ENABLE=0|1 [PERSIST=0|1]
   `slot ^ 1`, whose channel is `channel ^ 2`. An earlier live run saw
   readers 0/2 and 1/3 answer with the same UID - that is the shared RF
   path, and the reason the automatic bind is gated on the partner slot
-  reading empty (see above). Two slots reading the same UID cannot both
+  reading empty or carrying a known card (see above). Two slots reading the same UID cannot both
   bind: the existing duplicate guard refuses the second one, and an
   unattributed read is stored but never bound.
 * **Read cost.** One page read is ~25 tunnel commands (~0.5-1 s on the
@@ -148,29 +207,27 @@ packing/signed conversion, the slot -> reader-channel bit-swap map
 nothing; a matching-but-silent firmware gets one probe; a dead link times
 out bounded), both genuine live captures (`04 22 52 FC ...` and
 `53 42 70 E9 ...`) yielding their bytes and UIDs, an OpenSpool NDEF decode
-through the reused `ace_rc522` decoder, and the ace.py wiring (flag
-gating, one attempt per occupancy, the shared-antenna bind gate - partner
-occupied/unknown/absent -> stored but not bound - own store, no
+through the reused `ace_rc522` decoder, and the ace.py wiring (one
+attempt per occupancy, the shared-antenna bind gate - partner occupied
+with an unknown card -> stored but not bound, known card -> halted - own store, no
 `_info_per_ace` writes, get_status surfacing, unchanged V2 path).
 
 ## Open points for the maintainer
 
-1. **Flag name / default.** Implemented as `gen1_tag_tunnel`, default
-   `false`, setter `ACE_SET_TAG_TUNNEL`. Happy to rename or default it on.
-2. **Surface third-party spools automatically?** The automatic read binds
+1. **Surface third-party spools automatically?** The automatic read binds
    by card UID only when a table entry already carries that UID **and the
    partner slot on the shared antenna reads empty**; it never creates
    entries and never releases a vendor binding. If the maintainer prefers
    report-only (no bind call at all), the `bind=False` path is already the
    non-binding one.
-3. **Web UI.** The new `tag_tunnel` status block and the slot `uid` /
+2. **Web UI.** The new `tag_tunnel` status block and the slot `uid` /
    `tag_format` fill are already in `get_status`; the web backend passes
    the slot `uid`/`tag_format` through today. Whether to add a Config tab
    toggle and a "third-party tag" badge is a UI decision.
-4. **Retry policy.** One attempt per insert is deliberate and cheap; a
+3. **Retry policy.** One attempt per insert is deliberate and cheap; a
    bounded retry ladder (or a "read on next rotation" hook) would raise the
    hit rate on spools whose tag parks away from the coil.
-5. **Firmware acceptance.** The version pre-gate accepts `CV1.3.87x`.
+4. **Firmware acceptance.** The version pre-gate accepts `CV1.3.87x`.
    A later tunnel build (e.g. `CV1.3.872`) would pass the gate; the probe
    then decides. If a future build changes the op contract, the constant
    needs revisiting (the reference is `CV1.3.871`).