Kenwood TM-D710A support: channels, two-transport settings, ladder 1-5 (#113)

.claude/skills/new-radio/SKILL.md

@@ -37,6 +37,38 @@ not the opening move, and this has paid off on every radio where it was tried.
 | GitHub RE repos for the model or its siblings | Config-file structure, command tables, enum tables |
 | **The radio's own manual** | Menu order, option lists, defaults — and it is a *published source*, not a fallback |
 
+### ★ First enumerate what the RADIO has, not what a source describes
+
+Do this before reading any source in depth, and write it down. A published table
+covers what its author needed; the gap between that and the radio is invisible
+unless you have the radio's own list to hold it against.
+
+1. **What is this radio FOR?** Read the model's feature list — the manual's
+   first pages, the manufacturer's product page. A built-in TNC, a GPS, D-STAR,
+   a second receiver, cross-band repeat: each is a whole family of settings.
+   Manufacturer naming often carries it (Kenwood's `D` in TM-**D**710 and
+   TH-**D**75 means the data/APRS half; the TM-V71 is the same radio without it).
+2. **Enumerate the complete menu map** from the manual — every group, and how
+   many menus are in each. This is the denominator for everything after it.
+3. **For each group, name the transport that reaches it.** A group with no
+   transport is a **finding**, not an omission, and it belongs in `PLAN.md`
+   before a line of code.
+
+⚠ **One command's coverage is not the radio's settings.** The TM-D710 (#113)
+shipped a 35-field settings schema built on its `MU` command, every field
+measured on the radio and correct — and **no APRS at all**, on a radio whose
+headline feature is APRS. `MU` carries menus 000-5xx; the APRS and TNC settings
+are the 600-series and there is no `MU` parameter for one of them. The radio's
+own `aprs_capable` flag was set to `true` in the same session. Nobody counted
+the menus, so nobody noticed the settings stopped at 500.
+
+The earlier note that "`MU` is not exhaustive — menus 504, 505 and 506 have no
+parameter" was already in `FINDINGS.md`. It was read as a three-menu gap instead
+of the question it actually was: *what else is missing, and how would we know?*
+
+**Ask that question out loud in `PLAN.md`, with a number.** "The manual lists N
+menus in G groups; this transport reaches M of them; the other N-M are <where>."
+
 Then classify, because it decides how much of this process applies:
 
 - **Clone of a family already supported** — AT-D868UV/D578 against the D890UV,
@@ -49,9 +81,14 @@ Then classify, because it decides how much of this process applies:
   record-by-record programming (AnyTone).
 
 **Gate:** a `PLAN.md` in `scratchpad/<driver_key>/` naming the sources found,
-the programming medium, the family, and what the user owns. Template in
-`templates/PLAN.md`. Nothing is written before this exists — it is also the
-thing that makes a resumed session cheap.
+the programming medium, the family, and what the user owns — **plus the menu
+census above: how many menus the radio has, how many the chosen transport
+reaches, and where the rest live.** Nothing is written before this exists; it is
+also the thing that makes a resumed session cheap.
+
+⚠ If the census cannot be completed because a group's transport is unknown, that
+is the finding to report, not a detail to settle later. A radio shipped with a
+whole feature's settings missing looks finished from the inside.
 
 ## 2. Anchor on a file the radio wrote
 
@@ -178,7 +215,19 @@ Then wire *both* ends, and check each off explicitly:
 - [ ] **`apply_settings` called by the export path**
 - [ ] the table↔schema agreement test
 
-**Gate:** a test proving an export carries memories **and** settings together.
+- [ ] **the coverage check against step 1's menu census** — the schema's field
+      count and groups reconciled against the menus the radio actually has, with
+      every absence named
+
+**Gate:** a test proving an export carries memories **and** settings together,
+and a **stated count**: N of the radio's M menus are exposed, and the M-N are
+listed with a reason. "35 fields" is not a result; "35 of the 42 this transport
+reaches, and the transport reaches 42 of the radio's ~90" is.
+
+⚠ A cheap mechanical version of that reconciliation: the seed row already
+asserts what the radio can do. A model with `aprs_capable: true` and no APRS
+field in its settings schema is a contradiction the test suite can catch on its
+own, and the TM-D710 shipped exactly that pairing for a whole session.
 
 ⚠ The fourth box is the one that nearly shipped broken. The read path worked and
 the form filled correctly, so nothing looked wrong — the values simply never
@@ -235,6 +284,14 @@ if the folder is empty, the process above still stands on its own.
 
 ## Traps, each of which has already cost time
 
+- ★ **A source's coverage is not the radio's.** Every field measured off one
+  command can be right and the set still be badly incomplete — the TM-D710
+  shipped a correct 35-field settings schema with no APRS on an APRS radio,
+  because `MU` stops at menu 500 and nobody counted the menus. Enumerate what
+  the radio HAS first, then hold every source against it.
+- ★ **A noted gap is a question, not a footnote.** "`MU` is not exhaustive —
+  three menus have no parameter" sat in the findings for two sessions. It was
+  the same fact as "an entire feature is unreachable", written small.
 - A working **read** path hides a dead **write** path. Verify the write.
 - A printed option list is **display** order, not the stored index. One radio
   prints High/Medium/Low and stores Low as 0.

.claude/skills/new-radio/templates/PLAN.md

@@ -25,6 +25,33 @@ re-measured.
 **What the user has:** radio / cable / microSD card / programming software
 (RT Systems, OEM CPS, none) / availability for hardware steps.
 
+## ★ What this radio IS — the census
+
+Fill this in **before** reading any source in depth. A published table covers
+what its author needed; the gap is invisible without the radio's own list.
+
+**Headline features** (from the manual's first pages, not from a driver): TNC /
+APRS · GPS · D-STAR · DMR · second receiver · cross-band repeat · weather alert
+· … Each one is a whole family of settings, and the model name often says so
+(Kenwood's `D` in TM-**D**710 is the data/APRS half; the TM-V71 is the same
+radio without it).
+
+**Menu census** — the denominator for everything downstream:
+
+| menu group | what it covers | how many | transport that reaches it |
+|---|---|---|---|
+| 0xx | | | |
+| 1xx | | | |
+| … | | | |
+| **total** | | **N** | **M reached, N-M elsewhere** |
+
+⚠ A group with **no** transport is a finding, not an omission — write it here
+and say so out loud, with the number. "This command reaches M of N menus; the
+other N-M are in \<where\>." The TM-D710 shipped a correct 35-field settings
+schema with **no APRS at all** on an APRS radio, because `MU` stops at menu 500
+and nobody ever counted. Every field in it was measured and right; the set was
+the problem.
+
 ## Shape of the work
 
 `driver_key = "<manufacturer>_<model>"`, `export_format = "<key>"`,

CLAUDE.md

@@ -21,6 +21,21 @@ substantial work under `src-tauri/src/radios/`.
 
 ## Standing workflow
 
+- **⚠⚠ Build. Do not commit, do not push, do not run `/code-review`. Stop and
+  say the work is ready.** Tim, 2026-09-06: *"Just build the thing and when I'm
+  ready to push I'll ask for code review."* Committing and pushing are **his
+  calls, on his timing** — `/code-review` is a command he invokes when he is ready
+  to push, not a step for this agent to run on its own.
+
+  ⚠ The failure this replaces is real: in s134 this agent finished each piece of
+  work, ran `npm run ci`, committed and pushed — five times, unasked — and Tim's
+  objection ("you just haul off and start doing CI") was about being moved past,
+  not about which checks ran. A green `npm run ci` is not permission to commit and
+  is not a review; it runs the tests that already exist, so it cannot find the bug
+  nobody thought to test for.
+
+  So: finish the work, leave it in the working tree, and report what changed and
+  what is unverified. Commit only when asked, and push only when asked.
 - **Verify in dev, then commit, then push.** `npm run tauri:dev` runs against a
   separate `.dev` app identifier, so dev never shares the production database.
 - **CI is free and unmetered.** The repo went public on 2026-08-22, so standard
@@ -31,6 +46,11 @@ substantial work under `src-tauri/src/radios/`.
   three OSes before it lands, which is the point: a branch that has never been
   verified anywhere but the author's Mac should not reach `main`. This reverses
   the old rule, which existed only because a PR cost metered minutes.
+- **⚠ A radio model is finished work only when the whole model is done.** Keep
+  pushing the branch — that is what runs CI — but do not open a PR per phase or
+  per hardware step, and do not treat an open one as something to keep
+  merge-ready commit by commit. One PR, opened when the radio is essentially
+  complete: channels and settings both working, the hardware ladder climbed.
 - **`main` is still verified on its own.** CI runs on push to `main` as well, so
   a merge of two green branches gets checked as the combination — this project
   has shipped bugs that existed nowhere else. Landing by local merge is still

README.md

@@ -26,6 +26,7 @@ card — or exported as CSV for tools that expect it.
 | **Icom ID-52** | D-STAR + Analog | VHF / UHF TX, 108–174 / 225–479 MHz RX | microSD — patches the radio's own `.icf` file | 1000 memories in 100 groups; memories and menu settings restore in one operation |
 | **Kenwood TH-D72** | APRS + Analog | 2 m / 70 cm TX, 118–174 / 320–524 MHz RX | Direct USB — read, write, settings | 1000 memories; 113 menu settings over the radio's own `MU` command |
 | **Kenwood TH-D75** | D-STAR + APRS + Analog | VHF / 1.25 m / UHF TX, 0.1–524 MHz RX | microSD — patches the radio's own `.d75` file | 1000 memories in 30 groups; memories and menu settings, including the APRS setup |
+| **Kenwood TM-D710** | APRS + Analog | 2 m / 70 cm TX, 118–524 MHz RX | Serial cable — read, write, settings | 1000 memories; 95 settings across **two transports** — 35 over the radio's `MU` command and 60 more, including the APRS position, status texts and station icon, out of the settings image `MU` cannot reach. Programmed live, one memory at a time over the COM port on the rear of the **operation panel** (not the main unit), so there is no image file and a backup is a transcript of the radio's own lines |
 | **Binteradio BT-9000** | Analog FM/NFM | 18–64 / 136–174 / 200–260 / 400–520 MHz TX, 18–520 MHz RX | Direct USB — read, write, settings | 960 channels in 15 fixed zones; 42 menu settings. Also sold as the Radtel RT-950 Pro, Bajeton BJ-9000 and Tenway TP-900 Pro — the radio reports itself as `RT-950` |
 
 Direct USB programming reads the radio's current image, applies your changes, backs up the
@@ -59,7 +60,6 @@ settings together, then verify on the actual radio before shipping.
 | **AnyTone AT-D578UV** | DMR + Analog mobile | [#47](https://github.com/ww8l/codeplug-magic/issues/47) |
 | **AnyTone AT-D868UV** | DMR + Analog handheld | [#51](https://github.com/ww8l/codeplug-magic/issues/51) |
 | **Icom ID-51** | D-STAR + Analog handheld | [#50](https://github.com/ww8l/codeplug-magic/issues/50) |
-| **Kenwood TM-D710** | APRS + Analog mobile | [#113](https://github.com/ww8l/codeplug-magic/issues/113) |
 | **Icom ID-5100** | D-STAR + Analog mobile | [#49](https://github.com/ww8l/codeplug-magic/issues/49) |
 | **Icom IC-9100** | HF / VHF / UHF base | [#45](https://github.com/ww8l/codeplug-magic/issues/45) |
 | **Icom IC-7610** | HF / 6 m SDR base | [#46](https://github.com/ww8l/codeplug-magic/issues/46) |

src-tauri/src/db.rs

@@ -60,18 +60,18 @@ mod tests {
         // Models are reintroduced one at a time (migration 0005 trimmed the
         // original set): currently the Baofeng UV-5R, TIDRADIO TD-H3, AnyTone
         // AT-D890UV, Yaesu FT5D, Icom ID-52, Kenwood TH-D75, Kenwood TH-D72
-        // and the Binteradio BT-9000. (0015 removed
-        // the Vero VR-N76 placeholder.) None of the last three has a migration
-        // of its own — seeding INSERTs new (manufacturer, model) rows, so a new
-        // model reaches existing databases on the next startup without one.
+        // the Binteradio BT-9000 and Kenwood TM-D710. (0015 removed the Vero
+        // VR-N76 placeholder.) None of the last four has a migration of its
+        // own — seeding INSERTs new (manufacturer, model) rows, so a new model
+        // reaches existing databases on the next startup without one.
         let count: (i64,) = sqlx::query_as("SELECT COUNT(*) FROM radio_models")
             .fetch_one(&pool)
             .await
             .unwrap();
         assert_eq!(
-            count.0, 8,
-            "expected the UV-5R, TD-H3, AT-D890UV, FT5D, ID-52, TH-D75, TH-D72 and BT-9000 \
-             seeded models"
+            count.0, 9,
+            "expected the UV-5R, TD-H3, AT-D890UV, FT5D, ID-52, TH-D75, TH-D72, BT-9000 \
+             and TM-D710 seeded models"
         );
 
         let models: Vec<(String,)> =
@@ -82,7 +82,10 @@ mod tests {
         let names: Vec<&str> = models.iter().map(|m| m.0.as_str()).collect();
         assert_eq!(
             names,
-            vec!["AT-D890UV", "BT-9000", "FT5D", "ID-52", "TD-H3", "TH-D72", "TH-D75", "UV-5R"]
+            vec![
+                "AT-D890UV", "BT-9000", "FT5D", "ID-52", "TD-H3", "TH-D72", "TH-D75", "TM-D710",
+                "UV-5R"
+            ]
         );
 
         // Seeding twice must remain idempotent.
@@ -91,7 +94,7 @@ mod tests {
             .fetch_one(&pool)
             .await
             .unwrap();
-        assert_eq!(count2.0, 8, "seeding should be idempotent");
+        assert_eq!(count2.0, 9, "seeding should be idempotent");
 
         // A new database starts with NO talkgroups. The BrandMeister list used
         // to be compiled in and seeded here; it is downloaded on request now,