timothy-weston · Oct 15, 2026 · Oct 15, 2026 · Oct 15, 2026 · Oct 15, 2026
diff --git a/.claude/skills/upload-file/SKILL.md b/.claude/skills/upload-file/SKILL.md
@@ -95,8 +95,16 @@ Publishing needs the AWS CLI, boto3, credentials and `WR_FILES_DISTRIBUTION_ID`
 7. **Rebuild and look.** `npm run build && npm run serve`, then the product page.
 
    A local build with the file still staged shows the **local** copy, badged
-   `staged`. To see what a customer gets, remove `static/_upload/` first and
-   rebuild — that is the build CI produces.
+   `staged`. To see what a customer gets — the build CI produces — move
+   `static/_upload/` outside the repository while you build, then move it back.
+   Do not delete it: it can hold staged documents that are not published yet.
+
+   ```bash
+   mv static/_upload ../_upload.hold
+   npm run build && npm run serve
+   # after stopping the server:
+   mv ../_upload.hold static/_upload
+   ```
 
 8. **Clean up.** Delete the staged copy once published; it is scaffolding, and
    the page now resolves from the store.

diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
@@ -123,7 +123,7 @@ curl -s https://download.westonrobot.net/index.json | grep wr65
 
 That is the same file the site build reads, so if a document is in there it is live.
 
-`npm run check:downloads` fails if a page still points into `static/_upload/`. That is not a lint rule — CI has no copy of your local file, so such a page would 404 for everyone but you. It also fails if a `<Downloads>` query matches nothing in the store.
+`npm run check:downloads` fails if a page still points into `static/_upload/`. That is not a lint rule — CI has no copy of your local file, so such a page would 404 for everyone but you. It also fails if a `<Downloads>` query matches nothing in the store's index. If the index can't be reached, it reports those queries as not checked and does not fail.
 
 The path convention is [ADR 0001 D4](docs/adr/0001-host-downloadable-documents-on-s3.md): `/<section>/<product>/<file>`, and a published path is **never renamed or deleted**. A customer's bookmark, a QR code printed on a chassis and a support email from 2024 all depend on that.
 

diff --git a/TODO.md b/TODO.md
@@ -440,21 +440,22 @@ A read-only audit of every Markdown file on 2026-10-07. The high-severity items
 - [ ] **Docking robots linked to hand-charging rules.** `solution/robot-platforms/index.md:91` links `guides/operational-safety.md:74-75` ("do not leave charging unattended overnight"). `robot-maintenance.md:52` says unattended where `operational-safety.md:75` says overnight.
 - [ ] **No emergency-stop subsection** on `h1-2.md`, `go2.md`, `b2.md`, `a2.md`. `g1.md` and `r1.md` have one.
 - [ ] **B2 sealing.** `robot/quadruped/b2.md:132` says "be familiar with the sealing procedure" and doesn't link one.
-- [ ] **WRL63 manual error.** The published `wrl63-user-manual-en-v1.1.pdf` contains RM65 text (an "RM65-B" spec heading, 610 mm radius, RM65 joint ranges). Raise it with the vendor.
-- [x] **WR65 J5 — 2026-10-08.** The WR65 manual gives J5 as ±128° in its installation section and ±180° in its spec table. The ±128° is likely a documentation error; `wr65.md` now uses ±180°, consistent with the spec table and the other joints.
-- [x] **WRL63 weight — 2026-10-08.** `wrl63.md` said 10.0 kg (`-B`); the published spec sheet (`wrl63-user-manual-en-v1.1.pdf`) says 10.2 kg. The page now follows the spec sheet.
-- [ ] **WRL63 force-sensing weight.** `wrl63.md` gives 10.1 kg for the force-sensing variants, from an unpublished 2023 document. The spec sheet gives no figure for them, and the page now shows them lighter than the `-B`. Source it or remove it.
+- [ ] **Manual errors to raise with Realman.** The manuals we publish disagree with Realman's current parameters ([RM65](https://develop.realman-robotics.com/en/robot/robotParameter/RM65OntologyParameters/), [RML63](https://develop.realman-robotics.com/en/robot/robotParameter/RML63OntologyParameters/)), which give the same figures for Gen 3 and Gen 4. Ask which figures apply to the arms we have shipped, and whether there are corrected manuals to republish.
+  - `wr65-user-manual-en-v2.8.pdf`: the specification table gives J1, J4 and J5 as ±180°, where Realman give ±178°, ±178° and ±128°. The manual's own installation section gives J5 as ±128°.
+  - `wrl63-user-manual-en-v1.1.pdf`: contains RM65 text (an "RM65-B" spec heading, 610 mm radius, RM65 joint ranges). Its specification table gives J1, J2, J4 and J5 as ±180°, J3 as +165° to −150° and a weight of 10.2 kg, where Realman give ±178°, −178° to +145° and 10 kg.
+- [x] **WR65 and WRL63 joint limits and weights — 2026-10-15.** Both pages follow Realman's current parameters again, which match the specifications restored from the archive in `6790cd8`. `8a37e9a` in PR #44 had switched them to the manuals' specification tables, which widened WR65 J5 to ±180° and WRL63 J3 to +165°. Each joint table now names its source and the manual's differing values. The 27 V maximum supply voltage from the manuals stays.
+- [x] **WRL63 force-sensing weight — sourced.** Realman's current parameters give 10.1 kg for the six-axis force version and 10 kg for the standard version.
 - [ ] **NanoPC 12 V claim.** `peripheral/computer/nanopc.md:35` "most robot bases we supply can feed 12 V directly" has no source. Check it against the bases before a customer over-volts a 5-20 V board.
 
 ### Publishing tools
 
 - [ ] **`publish-files.py` overwrites and over-uploads.** `scripts/publish-files.py:340` uploads `! differs` items, overwriting a published key and its `.sha256` under an immutable cache header. `:312-315` treats an unreadable store as empty, so `--publish` would upload everything. Refuse both without an explicit override, and say in CONTRIBUTING what to do on `!`.
 - [ ] **No manual deploy.** `.github/workflows/deploy.yml:3-7` runs only on push and pull_request, so a publish or retire doesn't reach the download tables until the next merge. Add `workflow_dispatch` and a "trigger a deploy" step to CONTRIBUTING and both skills.
-- [ ] **`check-downloads.py` passes too easily.** `scripts/check-downloads.py:143-147` passes when the index is unreachable; `unsatisfied()` (`:100-106`) ignores `retired`, so an all-retired product passes with an empty table. `docs/design/file-hosting.md:86`, `:117` say "the build fails", but it's a separate script. `CONTRIBUTING.md:126` says it "also fails if a `<Downloads>` query matches nothing", which doesn't hold in either case.
+- [ ] **`docs/design/file-hosting.md` says "the build fails".** `:86` and `:117` say an unresolved `<Downloads>` query fails the build, but it is `scripts/check-downloads.py`, a separate CI step, that fails. Two behaviours of that script stay as they are (decided 2026-10-15): an unreachable index passes with its queries reported as not checked (see the `fetch_index()` docstring), and a query that matches only retired documents passes, because retiring a document is deliberate. `CONTRIBUTING.md` now states the unreachable-index case.
 - [ ] **`LOCAL_REF` misses link forms.** `scripts/check-downloads.py:32` doesn't match reference-style links, `pathname:///_upload/`, unquoted `href` or `static/_upload/`.
 - [ ] **`wrfiles.py` accepts bad names.** `scripts/wrfiles.py:171-245` accepts a subject with spaces or capitals, an uppercase extension (the key keeps `.PDF`) and firmware tagged `en`. CONTRIBUTING says bad names are refused.
 - [ ] **`--list` without credentials** ends in a `NoCredentialsError` traceback; `CONTRIBUTING.md:90` says it reports what's missing.
-- [ ] **upload-file skill.** `.claude/skills/upload-file/SKILL.md` should say a wrong version is also retired (as `retire-file/SKILL.md:21` does); `:94-96` "remove `static/_upload/` first" deletes staged files not yet published; `:37` has a garbled sentence.
+- [ ] **upload-file skill.** `.claude/skills/upload-file/SKILL.md` should say a wrong version is also retired (as `retire-file/SKILL.md:21` does); `:37` has a garbled sentence. (Step 7 now moves `static/_upload/` aside instead of deleting it, 2026-10-15.)
 - [ ] **`check-video-budget.sh` extensions.** `scripts/check-video-budget.sh:41` patterns are case-sensitive and miss `.gif` and `.avi`.
 - [ ] **First checks need more than python3.** `CONTRIBUTING.md:18` says so, but the video check needs bash and git.
 - [ ] **What the store is for.** ADR 0001:21 and ADR 0002:42 disagree, and the store holds vendor manuals while the template says to link the vendor. Record the real rule.

diff --git a/robot/manipulator/wr65.md b/robot/manipulator/wr65.md
@@ -85,7 +85,8 @@ and their figures for the force-sensing variant differ from ours — 627 mm reac
 
 Reach and force-sensor accuracy are the two values to confirm from the manual supplied with
 your arm, or from [Realman's current page](https://www.realman-robotics.com/en/products/rm65.html),
-before you rely on them. The figures below are the ones that are consistent across both.
+before you rely on them. Where Realman's current figures and our manual differ, the
+figures below follow Realman.
 
 :::
 
@@ -108,16 +109,18 @@ before you rely on them. The figures below are the ones that are consistent acro
 
 #### Joint limits and speeds
 
-From our manual, and identical across the variants. Worth having in front of you when you
-are planning trajectories — a pose that looks reachable can sit outside J2 or J3.
+From [Realman's current parameters](https://develop.realman-robotics.com/en/robot/robotParameter/RM65OntologyParameters/),
+and identical across the variants. The specification table in the manual we publish gives
+±180° for J1, J4 and J5; plan to the ranges here. Worth having in front of you when you
+are planning trajectories — a pose that looks reachable can sit outside J2 or J5.
 
 | Joint | Motion range | Maximum speed |
 | --- | --- | --- |
-| J1 | ±180° | 180°/s |
+| J1 | ±178° | 180°/s |
 | J2 | ±130° | 180°/s |
 | J3 | ±135° | 225°/s |
-| J4 | ±180° | 225°/s |
-| J5 | ±180° | 225°/s |
+| J4 | ±178° | 225°/s |
+| J5 | ±128° | 225°/s |
 | J6 | ±360° | 225°/s |
 
 ### Control interface

diff --git a/robot/manipulator/wrl63.md b/robot/manipulator/wrl63.md
@@ -85,7 +85,8 @@ and their figures for the force-sensing variant differ from ours — 917 mm reac
 
 Reach and force-sensor accuracy are the two values to confirm from the manual supplied with
 your arm, or from [Realman's current page](https://www.realman-robotics.com/en/products/rml63.html),
-before you rely on them. The figures below are the ones that are consistent across both.
+before you rely on them. Where Realman's current figures and our manual differ, the
+figures below follow Realman.
 
 :::
 
@@ -96,7 +97,7 @@ before you rely on them. The figures below are the ones that are consistent acro
 | Degrees of freedom | 6 |
 | Payload | 3 kg |
 | Repeatability | ±0.05 mm |
-| Body weight | 10.2 kg (`-B`) · 10.1 kg (force-sensing variants) |
+| Body weight | 10.0 kg (`-B`) · 10.1 kg (force-sensing variants) |
 | Working radius | 900 mm (`-B`) — see the caution above for the others |
 | Power supply | DC 24 V rated, 27 V maximum |
 | Power consumption | ≤ 200 W maximum, ≤ 100 W comprehensive |
@@ -106,20 +107,23 @@ before you rely on them. The figures below are the ones that are consistent acro
 | Communication | WiFi · network interface · Bluetooth · USB serial · RS485 |
 | Control modes | Drag-and-drop teaching · teaching pendant · API · JSON |
 
-Note the trade against the [WR65](/robot/manipulator/wr65): half again the reach, but **3 kg payload rather than 5 kg** and 10.2 kg of arm rather than 7.2 kg. Both matter when you are sizing a mobile base.
+Note the trade against the [WR65](/robot/manipulator/wr65): half again the reach, but **3 kg payload rather than 5 kg** and 10 kg of arm rather than 7.2 kg. Both matter when you are sizing a mobile base.
 
 #### Joint limits and speeds
 
-From our manual, and identical across the variants. Worth having in front of you when you
-are planning trajectories — note that **J3 is asymmetric**, unlike every other joint.
+From [Realman's current parameters](https://develop.realman-robotics.com/en/robot/robotParameter/RML63OntologyParameters/),
+and identical across the variants. The specification table in the manual we publish gives
+±180° for J1, J2, J4 and J5 and +165° to −150° for J3; plan to the ranges here. Worth
+having in front of you when you are planning trajectories — note that **J3 is asymmetric**,
+unlike every other joint.
 
 | Joint | Motion range | Maximum speed |
 | --- | --- | --- |
-| J1 | ±180° | 180°/s |
-| J2 | ±180° | 180°/s |
-| J3 | +165° to −150° | 225°/s |
-| J4 | ±180° | 225°/s |
-| J5 | ±180° | 225°/s |
+| J1 | ±178° | 180°/s |
+| J2 | ±178° | 180°/s |
+| J3 | +145° to −178° | 225°/s |
+| J4 | ±178° | 225°/s |
+| J5 | ±178° | 225°/s |
 | J6 | ±360° | 225°/s |
 
 ### Control interface