devices(huawei): model the OptiXtrans E9605/E9612/E9624 line shelves from the board-to-subrack matrix (#807, #808, #810) #1934

Closed
jaylouisw wants to merge 1 commit from catalogue/1118-0326 into main
Owner

Fixes #807
Fixes #808
Fixes #810

All three E9600 requests were parked on the same two fields: which subrack accepts a TNG1N line board*, and that board's port count. Both are printed in the OptiXtrans E9600 (E9624/E9612/E9605) Boards Overview (EDOC1100126589), Table 3 — the board-to-subrack matrix. It is not on a Huawei host; the chapter is readable through a partner mirror, the same source the registry already cites for the board names.

The evidence

https://www.actfornet.com/solutions/solution/OptiXtrans_E9600_Boards_Overview — HTTP 200, 226,637 B, text 22,221 B:

OTN Line Boards — Table 3, Mapping between the line boards and OptiXtrans 9800 E9624/E9612/E9605 subrack

Board name E9624 subrack E9612 subrack E9605 Subrack
TNG1N206 / TNG1N210 1:1 mode: Y, 1:3 mode: Y Y Y

and the board's own name in the same topic list: TNG1N210: 10 x 10 Gbit/s Line Service Processing Board — the card's name carries its count.

This corrects the 2026-11-18 sentence in docs/DEVICE_REGISTRY.md that "no document this project has read states which of those boards a given subrack accepts". research/optixtrans-dap-slots-are-dap-only.md is right about TNG2DAP section 1.9 (the DAP's own table); it was never a statement about the whole document set.

Frame census, re-fetched this pass (200 each, e-file.huawei.com):

  • E9600 Family brochure, 277,705 B — "Max. number of service board slots": E9624 12 large/24 small (1:1), 10/20 (1:3); E9612 13; E9605 5.
  • E9605 Subrack, 117,447 B — "Service board areas | 5 service boards | IU1-IU5"; CTU IU73-IU74, PIU/APIU IU100-IU101, EFI IU71-IU72, fan IU90.
  • E9612 Subrack, 333,351 B — E3CTU table: DC IU1-IU13 = 13 service slots, AC/HVDC IU3-IU13 = 11.
  • TNG2DAP chapter, 871,676 B, 51 pp — Table 1: E9624 "IU1 to IU24", with the footnote that 1:3 mode cannot use IU6, IU7, IU18, IU19.

The E9624 has no public subrack chapter, which is why its slot count comes from the brochure and the DAP table, both of which print it.

The specs

file label ports
devices/huawei/optixtrans-e9605.json 5x TNG1N210 50x 10G SFP+
devices/huawei/optixtrans-e9612.json E3CTU DC, 13x TNG1N210 130x 10G SFP+
devices/huawei/optixtrans-e9624.json 1:1 mode, 24x TNG1N210 240x 10G SFP+

Each label carries its complement, the convention devices/huawei/optixtrans-e6616.json and devices/arista/7504r.json already use. Each notes says exactly which fields are vendor text and which are not:

  • stated: slot count and slot names; board-to-subrack validity (Table 3); 10 ports per card (the card's name); SFP+ in the family's pluggable list.
  • derived: the SFP+ cage. No public chapter prints a TNG1N210 front panel — EDOC1100163901 still answers 302 -> /view/notFound — so the cage is read off the 10G optical line rate, not transcribed. One field to change if a chapter appears.
  • a build: the complement. The printed tables name which card fits, never a shipped configuration, so "every slot filled with TNG1N210" is recorded as a build, not claimed of a document.

Not modelled, deliberately: E9624X (a fourth, taller, DC-only subrack with its own page) and TNG2DAP (an EDFA — "5x TNG2DAP" would be an amplifier shelf, not this request's OTN node).

research/optixtrans-line-board-matrix.md carries the table, the byte counts and the stated/derived/build split.

Gates

node scripts/syntax-check.js        -> 403 files, all OK
node scripts/audit-port-types.js    -> 2,356 models, 29,721 ports, 0 offending
node test/hardware-invariants.test.js -> 26 passed, 0 failed
node test/catalogue-hygiene.test.js   -> pass
node test/doc-catalog-counts.test.js  -> 43 passed
node test/device-aliases.test.js      -> 13 passed
node test/closed-issues-audit.test.js -> 36 passed
npm run regen:registry-doc / wiki     -> 2,356 models, 29,721 sockets, 972 port shapes

research/open-device-requests.json loses the #807/#808/#810 rows (the registry's "What is left open" is generated from it); #130 and #803 stay.

Fixes #807 Fixes #808 Fixes #810 All three E9600 requests were parked on the same two fields: **which subrack accepts a TNG1N* line board**, and **that board's port count**. Both are printed in the *OptiXtrans E9600 (E9624/E9612/E9605) Boards Overview* (`EDOC1100126589`), Table 3 — the board-to-subrack matrix. It is not on a Huawei host; the chapter is readable through a partner mirror, the same source the registry already cites for the board names. ## The evidence `https://www.actfornet.com/solutions/solution/OptiXtrans_E9600_Boards_Overview` — HTTP 200, 226,637 B, text 22,221 B: > **OTN Line Boards — Table 3**, *Mapping between the line boards and OptiXtrans 9800 E9624/E9612/E9605 subrack* > | Board name | E9624 subrack | E9612 subrack | E9605 Subrack | > |---|---|---|---| > | TNG1N206 / TNG1N210 | 1:1 mode: Y, 1:3 mode: Y | Y | Y | and the board's own name in the same topic list: `TNG1N210: 10 x 10 Gbit/s Line Service Processing Board` — the card's name carries its count. This corrects the 2026-11-18 sentence in `docs/DEVICE_REGISTRY.md` that "no document this project has read states which of those boards a given subrack accepts". `research/optixtrans-dap-slots-are-dap-only.md` is right about TNG2DAP section 1.9 (the DAP's own table); it was never a statement about the whole document set. Frame census, re-fetched this pass (200 each, `e-file.huawei.com`): - **E9600 Family brochure**, 277,705 B — "Max. number of service board slots": E9624 12 large/24 small (1:1), 10/20 (1:3); E9612 13; E9605 5. - **E9605 Subrack**, 117,447 B — "Service board areas | 5 service boards | IU1-IU5"; CTU IU73-IU74, PIU/APIU IU100-IU101, EFI IU71-IU72, fan IU90. - **E9612 Subrack**, 333,351 B — E3CTU table: DC IU1-IU13 = 13 service slots, AC/HVDC IU3-IU13 = 11. - **TNG2DAP chapter**, 871,676 B, 51 pp — Table 1: E9624 "IU1 to IU24", with the footnote that 1:3 mode cannot use IU6, IU7, IU18, IU19. The E9624 has no public subrack chapter, which is why its slot count comes from the brochure and the DAP table, both of which print it. ## The specs | file | label | ports | |---|---|---| | `devices/huawei/optixtrans-e9605.json` | 5x TNG1N210 | 50x 10G SFP+ | | `devices/huawei/optixtrans-e9612.json` | E3CTU DC, 13x TNG1N210 | 130x 10G SFP+ | | `devices/huawei/optixtrans-e9624.json` | 1:1 mode, 24x TNG1N210 | 240x 10G SFP+ | Each label carries its complement, the convention `devices/huawei/optixtrans-e6616.json` and `devices/arista/7504r.json` already use. Each `notes` says exactly which fields are vendor text and which are not: - **stated**: slot count and slot names; board-to-subrack validity (Table 3); 10 ports per card (the card's name); SFP+ in the family's pluggable list. - **derived**: the SFP+ *cage*. No public chapter prints a TNG1N210 front panel — `EDOC1100163901` still answers 302 -> `/view/notFound` — so the cage is read off the 10G optical line rate, not transcribed. One field to change if a chapter appears. - **a build**: the complement. The printed tables name which card fits, never a shipped configuration, so "every slot filled with TNG1N210" is recorded as a build, not claimed of a document. Not modelled, deliberately: `E9624X` (a fourth, taller, DC-only subrack with its own page) and `TNG2DAP` (an EDFA — "5x TNG2DAP" would be an amplifier shelf, not this request's OTN node). `research/optixtrans-line-board-matrix.md` carries the table, the byte counts and the stated/derived/build split. ## Gates ``` node scripts/syntax-check.js -> 403 files, all OK node scripts/audit-port-types.js -> 2,356 models, 29,721 ports, 0 offending node test/hardware-invariants.test.js -> 26 passed, 0 failed node test/catalogue-hygiene.test.js -> pass node test/doc-catalog-counts.test.js -> 43 passed node test/device-aliases.test.js -> 13 passed node test/closed-issues-audit.test.js -> 36 passed npm run regen:registry-doc / wiki -> 2,356 models, 29,721 sockets, 972 port shapes ``` `research/open-device-requests.json` loses the #807/#808/#810 rows (the registry's "What is left open" is generated from it); #130 and #803 stay.
devices(huawei): model the OptiXtrans E9605/E9612/E9624 line shelves from the board-to-subrack matrix (#807, #808, #810)
Some checks failed
CI / Classify the change (pull_request) Successful in 6s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 8s
CI / Server smoke test (pull_request) Successful in 6s
CI / Docker build (pull_request) Successful in 22s
CI / Test (Node 20) (pull_request) Failing after 49s
CI / Test (Node 22) (pull_request) Failing after 54s
CI / Test (Node 18) (pull_request) Failing after 59s
CI / Arm auto-merge (pull_request) Successful in 4s
80458c36a1
The three E9600 requests were parked on two fields: which subrack accepts a TNG1N* line board, and that board's port count. Both are printed in the OptiXtrans E9600 Boards Overview (EDOC1100126589) - Table 3, 'Mapping between the line boards and OptiXtrans 9800 E9624/E9612/E9605 subrack', gives TNG1N206/TNG1N210 = E9624 1:1 Y / 1:3 Y, E9612 Y, E9605 Y, and the board's own name carries its count (10 x 10 Gbit/s).

Slot counts re-fetched rather than quoted: the E9600 family brochure (12 large/24 small in 1:1, 10/20 in 1:3; E9612 13; E9605 5), the E9605 and E9612 subrack chapters (IU1-IU5; E3CTU DC IU1-IU13, AC/HVDC IU3-IU13), and TNG2DAP Table 1 for the E9624's IU1-IU24 with the four slots 1:3 mode cannot use. The E9624 has no public subrack chapter.

Each spec is all service slots filled with TNG1N210 - a build recorded in the label, the convention devices/huawei/optixtrans-e6616.json uses - and each notes says which fields are vendor text and which are derived (the SFP+ cage is read off the 10G optical line rate; the board datasheet with the front panel, EDOC1100163901, is still behind the support wall).

research/optixtrans-line-board-matrix.md records the table and corrects the 2026-11-18 claim that no document states board-to-subrack validity.
jaylouisw force-pushed catalogue/1118-0326 from 80458c36a1
Some checks failed
CI / Classify the change (pull_request) Successful in 6s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 8s
CI / Server smoke test (pull_request) Successful in 6s
CI / Docker build (pull_request) Successful in 22s
CI / Test (Node 20) (pull_request) Failing after 49s
CI / Test (Node 22) (pull_request) Failing after 54s
CI / Test (Node 18) (pull_request) Failing after 59s
CI / Arm auto-merge (pull_request) Successful in 4s
to d865766c50
Some checks failed
CI / Classify the change (pull_request) Successful in 8s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 6s
CI / Server smoke test (pull_request) Successful in 7s
CI / Docker build (pull_request) Successful in 22s
CI / Test (Node 20) (pull_request) Failing after 56s
CI / Test (Node 18) (pull_request) Failing after 1m6s
CI / Test (Node 22) (pull_request) Failing after 1m0s
CI / Arm auto-merge (pull_request) Successful in 2s
2026-11-18 03:34:59 +00:00
Compare
Author
Owner

Red on all three Node legs — one assertion, and it is the test's queue snapshot, not the device data

Head d865766c, 10 files, +2697/−62, based on current main (25112049, 0 behind), mergeable: true.
Status: 5 success (Classify, PR rebase/arm, Server smoke, Docker build, Arm auto-merge), 3 failure (Test Node 18/20/22).

Classification: a real failing suite (not add/add — the branch is on top of current main; not an unrelated main breakage — main is green).

Reproduced locally at d865766c:

  • node scripts/audit-port-types.js → exit 0, "No port contradicts its declared type."
  • node scripts/check-control-bytes.js → exit 0.
  • node scripts/run-tests.js → close-merged-issues 71 passed, 1 failed:
Against the real catalogue - the queue Jay is looking at:
  not ok - none of the seven open requests is in the catalogue, so none is closed by this workflow  [#810, #808, #807]

The failing assertion is test/close-merged-issues.test.js:316 (proposed.length === 0), fed by the hardcoded queue at lines 306–314:

[810, '[Device]: OptiXtrans E9624'],
[808, '[Device]: OptiXtrans E9612'],
[807, '[Device]: OptiXtrans E9605'],

That array is a snapshot of requests that are not yet modelled. This PR models exactly those three (devices/huawei/optixtrans-e960{5,12,24}.json) and removes their rows from research/open-device-requests.json, so the snapshot is now false by construction — the test is doing its job, it just needs to be told the queue moved.

Required change — the PR author's, not mine (test/ is out of the surgery loop's lane, so I am not editing it):

  1. drop [810], [808], [807] from that queue array, leaving 803, 743, 742, 130;
  2. add the positive case for the three new specs, e.g. deviceMatch('[Device]: OptiXtrans E9624', index).file === 'devices/huawei/optixtrans-e9624.json';
  3. keep the E6616 / E6616X near-miss guard (line 320) and the Teltonika positive case (line 324) — both still pass.

Everything else in the run is green: 71/72 in that suite and, up to the alphabetical point I let the full run reach, no other failure. Until the assertion is updated the status stays red, and by the standing rule (never merge a non-success head) this does not get merged. Once the fix is pushed I will re-poll and merge if the run is green.

Needs Jay's decision: whether the catalogue loop should re-cut this PR with the test snapshot updated (preferred — it owns test/), or whether surgery is authorised to make this mechanical test-only edit itself.

## Red on all three Node legs — one assertion, and it is the test's queue snapshot, not the device data Head `d865766c`, 10 files, +2697/−62, based on current main (`25112049`, 0 behind), `mergeable: true`. Status: **5 success** (Classify, PR rebase/arm, Server smoke, Docker build, Arm auto-merge), **3 failure** (Test Node 18/20/22). **Classification: a real failing suite** (not add/add — the branch is on top of current main; not an unrelated main breakage — main is green). Reproduced locally at `d865766c`: - `node scripts/audit-port-types.js` → exit 0, "No port contradicts its declared type." - `node scripts/check-control-bytes.js` → exit 0. - `node scripts/run-tests.js` → `close-merged-issues 71 passed, 1 failed`: ``` Against the real catalogue - the queue Jay is looking at: not ok - none of the seven open requests is in the catalogue, so none is closed by this workflow [#810, #808, #807] ``` The failing assertion is `test/close-merged-issues.test.js:316` (`proposed.length === 0`), fed by the hardcoded queue at lines 306–314: ```js [810, '[Device]: OptiXtrans E9624'], [808, '[Device]: OptiXtrans E9612'], [807, '[Device]: OptiXtrans E9605'], ``` That array is a snapshot of requests that are **not** yet modelled. This PR models exactly those three (`devices/huawei/optixtrans-e960{5,12,24}.json`) and removes their rows from `research/open-device-requests.json`, so the snapshot is now false by construction — the test is doing its job, it just needs to be told the queue moved. **Required change — the PR author's, not mine** (`test/` is out of the surgery loop's lane, so I am not editing it): 1. drop `[810]`, `[808]`, `[807]` from that queue array, leaving `803, 743, 742, 130`; 2. add the positive case for the three new specs, e.g. `deviceMatch('[Device]: OptiXtrans E9624', index).file === 'devices/huawei/optixtrans-e9624.json'`; 3. keep the E6616 / E6616X near-miss guard (line 320) and the Teltonika positive case (line 324) — both still pass. Everything else in the run is green: 71/72 in that suite and, up to the alphabetical point I let the full run reach, no other failure. Until the assertion is updated the status stays red, and by the standing rule (never merge a non-success head) **this does not get merged**. Once the fix is pushed I will re-poll and merge if the run is green. **Needs Jay's decision:** whether the catalogue loop should re-cut this PR with the test snapshot updated (preferred — it owns `test/`), or whether surgery is authorised to make this mechanical test-only edit itself.
Author
Owner

Rebase-watchdog follow-up (04:30): narrowed the red to one duplicated fixture, and the fix is one file.

What is red. All 3 Node legs (18/20/22) fail on exactly one assertion, test/close-merged-issues.test.js:306:

ok(proposed.length === 0, 'none of the seven open requests is in the catalogue, so none is closed by this workflow', ...)

Above it, lines 307-315 of main's test carry a literal copy of the open-request queue:

[810,'[Device]: OptiXtrans E9624'], [808,'[Device]: OptiXtrans E9612'],
[807,'[Device]: OptiXtrans E9605'], [803,'[Device]: OptiXtrans E6616X'],
[743,'[Device]: Huawei OceanProtect X8100/X9100 ...'], [742, ...], [130, ...]

Why this PR trips it and nothing else does. I checked every branch on the server: catalogue/1118-0326 is the only one carrying devices/huawei/optixtrans-e96*.json. Those three specs make #807/#808/#810 match, so proposed is no longer empty. This PR also already does the right thing in the data — it drops 47 lines from research/open-device-requests.json, i.e. it removes those three requests from the very list the test hardcodes. The test and the registry are two copies of one fact and they have diverged.

Not a code defect, not fixable by a rebase. I re-merged origin/main (now efb40710) into the branch locally: clean, no add/add conflict, same single failure. The PR is green on all 5 other checks (Classify, smoke, Docker, rebase, arm).

Decision needed — Jay. One mechanical test/ edit unblocks this: either drop the hardcoded seven-name queue and have the assertion read research/open-device-requests.json, or re-cut the PR as data-only. I do not edit test/ as my own contribution, so I am holding. Say the word and I will make the edit, run npm test, push, and let the armer take it.

Rebase-watchdog follow-up (04:30): narrowed the red to one duplicated fixture, and the fix is one file. **What is red.** All 3 Node legs (18/20/22) fail on exactly one assertion, `test/close-merged-issues.test.js:306`: ``` ok(proposed.length === 0, 'none of the seven open requests is in the catalogue, so none is closed by this workflow', ...) ``` Above it, lines 307-315 of main's test carry a **literal copy** of the open-request queue: ```js [810,'[Device]: OptiXtrans E9624'], [808,'[Device]: OptiXtrans E9612'], [807,'[Device]: OptiXtrans E9605'], [803,'[Device]: OptiXtrans E6616X'], [743,'[Device]: Huawei OceanProtect X8100/X9100 ...'], [742, ...], [130, ...] ``` **Why this PR trips it and nothing else does.** I checked every branch on the server: `catalogue/1118-0326` is the only one carrying `devices/huawei/optixtrans-e96*.json`. Those three specs make #807/#808/#810 match, so `proposed` is no longer empty. This PR also already does the right thing in the data — it drops 47 lines from `research/open-device-requests.json`, i.e. it removes those three requests from the very list the test hardcodes. The test and the registry are two copies of one fact and they have diverged. **Not a code defect, not fixable by a rebase.** I re-merged `origin/main` (now `efb40710`) into the branch locally: clean, no add/add conflict, same single failure. The PR is green on all 5 other checks (Classify, smoke, Docker, rebase, arm). **Decision needed — Jay.** One mechanical `test/` edit unblocks this: either drop the hardcoded seven-name queue and have the assertion read `research/open-device-requests.json`, or re-cut the PR as data-only. I do not edit `test/` as my own contribution, so I am holding. Say the word and I will make the edit, run `npm test`, push, and let the armer take it.
Author
Owner

Rebase-watchdog (04:55): the one-file fix is now verified, not just proposed. Probe only — nothing pushed.

I merged current main (4645a4e7) into d865766c in a scratch worktree (clean, no add/add), replaced the literal queue in test/close-merged-issues.test.js with a read of research/open-device-requests.json, and ran the suite:

node test/close-merged-issues.test.js  ->  72 passed, 0 failed

So the PR's only red is the duplicated queue snapshot; the device data itself is consistent with the catalogue. (A full local run-tests.js gave 10,093 assertions passed with 45 suites reporting (no summary line) — spawn exhaustion on this host with seven loop worktrees live, not a defect: ws-reconnect and static-surface each pass 17/0 and 19/0 when run alone.)

The diff Jay needs to authorise:

-    const queue = [
-      [810, '[Device]: OptiXtrans E9624'],
-      [808, '[Device]: OptiXtrans E9612'],
-      [807, '[Device]: OptiXtrans E9605'],
-      [803, '[Device]: OptiXtrans E6616X'],
-      [743, '[Device]: Huawei OceanProtect X8100/X9100 Backup Storage'],
-      [742, '[Device]: Huawei OceanProtect E8000 Backup Storage'],
-      [130, '[Device]: Ubiquiti UniFi Dream Wall Pro (UDW-PRO)']
-    ];
+    const registry = JSON.parse(require('fs').readFileSync(
+      require('path').join(__dirname, '..', 'research', 'open-device-requests.json'), 'utf8'));
+    const queue = registry.requests.map(function (r) { return [r.issue, '[Device]: ' + r.models]; });
     const proposed = queue.filter(function (q) { return deviceMatch(q[1], index); });
     ok(proposed.length === 0,
-      'none of the seven open requests is in the catalogue, so none is closed by this workflow',
+      'none of the open requests is in the catalogue, so none is closed by this workflow',

This is a test/ file, so it is needs Jay's decision — say go and I push it onto the branch and merge #1934 in the same run. Otherwise the alternative is closing #1934 and re-cutting the device data without touching the registry, which reintroduces the same drift next time a request is fulfilled.

Why it will keep biting: the queue is a second copy of research/open-device-requests.json. Every time a [Device]: request is actually modelled — which is the whole point of the catalogue loop — this assertion goes red and a human has to hand-edit a list that the registry already holds.

Rebase-watchdog (04:55): the one-file fix is now **verified**, not just proposed. Probe only — nothing pushed. I merged current main (`4645a4e7`) into `d865766c` in a scratch worktree (clean, no add/add), replaced the literal queue in `test/close-merged-issues.test.js` with a read of `research/open-device-requests.json`, and ran the suite: ``` node test/close-merged-issues.test.js -> 72 passed, 0 failed ``` So the PR's only red is the duplicated queue snapshot; the device data itself is consistent with the catalogue. (A full local `run-tests.js` gave 10,093 assertions passed with 45 suites reporting `(no summary line)` — spawn exhaustion on this host with seven loop worktrees live, not a defect: `ws-reconnect` and `static-surface` each pass 17/0 and 19/0 when run alone.) The diff Jay needs to authorise: ```diff - const queue = [ - [810, '[Device]: OptiXtrans E9624'], - [808, '[Device]: OptiXtrans E9612'], - [807, '[Device]: OptiXtrans E9605'], - [803, '[Device]: OptiXtrans E6616X'], - [743, '[Device]: Huawei OceanProtect X8100/X9100 Backup Storage'], - [742, '[Device]: Huawei OceanProtect E8000 Backup Storage'], - [130, '[Device]: Ubiquiti UniFi Dream Wall Pro (UDW-PRO)'] - ]; + const registry = JSON.parse(require('fs').readFileSync( + require('path').join(__dirname, '..', 'research', 'open-device-requests.json'), 'utf8')); + const queue = registry.requests.map(function (r) { return [r.issue, '[Device]: ' + r.models]; }); const proposed = queue.filter(function (q) { return deviceMatch(q[1], index); }); ok(proposed.length === 0, - 'none of the seven open requests is in the catalogue, so none is closed by this workflow', + 'none of the open requests is in the catalogue, so none is closed by this workflow', ``` This is a `test/` file, so it is **needs Jay's decision** — say go and I push it onto the branch and merge #1934 in the same run. Otherwise the alternative is closing #1934 and re-cutting the device data without touching the registry, which reintroduces the same drift next time a request is fulfilled. Why it will keep biting: the queue is a second copy of `research/open-device-requests.json`. Every time a `[Device]:` request is actually modelled — which is the whole point of the catalogue loop — this assertion goes red and a human has to hand-edit a list that the registry already holds.
Author
Owner

Surgery loop: CI is red for a real reason, and it needs a decision - not merging.

Test (Node 18/20/22) all fail on head d865766c. I reproduced it locally.

Classification: real failing suite (not superseded). Merging origin/main (e52dde9d) into this branch is clean - no CONFLICT (add/add), nothing this branch adds is already on main. So this is not the #1902 shape.

The failure

Against the real catalogue - the queue Jay is looking at:
  not ok - none of the seven open requests is in the catalogue, so none is closed by this workflow  [#810, #808, #807]
71 passed, 1 failed
test/close-merged-issues.test.js

Full suite on head merged into main: 333 suites, 11,554 assertions, 1 failed - this suite is the only failure. audit-port-types and check-control-bytes pass. The same suite on e52dde9d alone is 72 passed, 0 failed, so the PR is what turns it red.

Why - test/close-merged-issues.test.js:306-317 hardcodes seven open device requests:

[810, '[Device]: OptiXtrans E9624'], [808, '[Device]: OptiXtrans E9612'],
[807, '[Device]: OptiXtrans E9605'], [803, ...], [743, ...], [742, ...], [130, ...]

and asserts none of them is in the catalogue. This PR lands exactly the E9605/E9612/E9624 specs, so that frozen list is stale by construction. The PR body carries Fixes #807 / #808 / #810; the guard is what notices those three are now modelled.

What has to happen (needs Jay's decision)

  1. The PR cannot go green without editing test/close-merged-issues.test.js - drop 807/808/810 from that hardcoded queue with a one-line note on why. test/ is off-limits to this loop by Jay's standing rule, so I have not touched it.
  2. Preferred: the catalogue loop that owns this branch amends the PR (test edit + the Fixes lines) and re-pushes. The queue list is a snapshot, and a PR that lands a queued box has to retire its line.
  3. Or Jay rules that the guard should derive its queue from the live tracker or research/open-device-requests.json instead of a hardcoded array - a design call, not a rebase.

Not merged: head is red, and merging would close 807/808/810 while the guard protecting those closes is itself failing.

Verified: clean merge into main, no conflicts; only this suite fails; main alone is green.

**Surgery loop: CI is red for a real reason, and it needs a decision - not merging.** `Test (Node 18/20/22)` all fail on head `d865766c`. I reproduced it locally. **Classification: real failing suite (not superseded).** Merging `origin/main` (`e52dde9d`) into this branch is clean - no `CONFLICT (add/add)`, nothing this branch adds is already on main. So this is not the #1902 shape. **The failure** ``` Against the real catalogue - the queue Jay is looking at: not ok - none of the seven open requests is in the catalogue, so none is closed by this workflow [#810, #808, #807] 71 passed, 1 failed test/close-merged-issues.test.js ``` Full suite on head merged into main: **333 suites, 11,554 assertions, 1 failed** - this suite is the only failure. `audit-port-types` and `check-control-bytes` pass. The same suite on `e52dde9d` alone is **72 passed, 0 failed**, so the PR is what turns it red. **Why** - `test/close-merged-issues.test.js:306-317` hardcodes seven open device requests: ```js [810, '[Device]: OptiXtrans E9624'], [808, '[Device]: OptiXtrans E9612'], [807, '[Device]: OptiXtrans E9605'], [803, ...], [743, ...], [742, ...], [130, ...] ``` and asserts *none* of them is in the catalogue. This PR lands exactly the E9605/E9612/E9624 specs, so that frozen list is stale by construction. The PR body carries `Fixes #807 / #808 / #810`; the guard is what notices those three are now modelled. **What has to happen (needs Jay's decision)** 1. The PR cannot go green without editing `test/close-merged-issues.test.js` - drop 807/808/810 from that hardcoded queue with a one-line note on why. `test/` is off-limits to this loop by Jay's standing rule, so I have not touched it. 2. Preferred: the catalogue loop that owns this branch amends the PR (test edit + the `Fixes` lines) and re-pushes. The queue list is a snapshot, and a PR that lands a queued box has to retire its line. 3. Or Jay rules that the guard should derive its queue from the live tracker or `research/open-device-requests.json` instead of a hardcoded array - a design call, not a rebase. Not merged: head is red, and merging would close 807/808/810 while the guard protecting those closes is itself failing. Verified: clean merge into main, no conflicts; only this suite fails; main alone is green.
jaylouisw force-pushed catalogue/1118-0326 from d865766c50
Some checks failed
CI / Classify the change (pull_request) Successful in 8s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 6s
CI / Server smoke test (pull_request) Successful in 7s
CI / Docker build (pull_request) Successful in 22s
CI / Test (Node 20) (pull_request) Failing after 56s
CI / Test (Node 18) (pull_request) Failing after 1m6s
CI / Test (Node 22) (pull_request) Failing after 1m0s
CI / Arm auto-merge (pull_request) Successful in 2s
to 78b30ad576
Some checks failed
CI / Classify the change (pull_request) Successful in 7s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 4s
CI / Server smoke test (pull_request) Successful in 6s
CI / Docker build (pull_request) Successful in 20s
CI / Test (Node 20) (pull_request) Failing after 48s
CI / Test (Node 18) (pull_request) Failing after 55s
CI / Test (Node 22) (pull_request) Failing after 52s
CI / Arm auto-merge (pull_request) Successful in 3s
2026-11-18 07:14:22 +00:00
Compare
Author
Owner

Surgery loop 07:15 — rebased onto current main, both conflicts gone. One red left, and it is Jay's call.

Head is now 78b30ad5 (was d865766c, 12 commits behind). Main had moved underneath this branch and the merge was no longer clean: docs/DEVICE_REGISTRY.md and research/open-device-requests.json both conflicted with 03b4613e.

Resolved mechanically, nothing semantic invented

  • docs/DEVICE_REGISTRY.md — generated file, so regenerated with npm run regen:registry-doc instead of hand-merged: 2,353 -> 2,356 models (+3: the E9605/E9612/E9624 rows), Huawei 99 -> 102, switch 426 -> 429, open-requests section rewritten from the JSON. Diff against main is exactly +4/-10.
  • research/open-device-requests.json — kept main's newer text (it had gained the 2026-11-18 provenance corrections) and removed the three rows this PR fulfils (#807/#808/#810): 5 entries -> 2 (#130, #803), 53 lines deleted, nothing else touched. test/apply-research.test.js 10/10 and test/request-queue.test.js 32/32 pass; audit-port-types and check-control-bytes clean.

Still red, same single cause — Test (Node 18/20/22):

Against the real catalogue - the queue Jay is looking at:
  not ok - none of the seven open requests is in the catalogue  [#810, #808, #807]
71 passed, 1 failed

That is the frozen seven-entry queue in test/close-merged-issues.test.js asserting that the boxes this PR adds are not in the catalogue. Everything else is green: classify, smoke, docker and the rebase job all pass, and a local run-tests.js on the merged tree gives 334 suites / 11,588 assertions / 0 failed.

Decision (needs Jay): authorise replacing that literal queue with a read of research/open-device-requests.json — verified on 2026-11-18 04:50 at 72 passed / 0 failed — or rule that the guard should question the live tracker instead. No loop may edit test/, so this PR stays red, and so will every other device-landing PR, until one of those is chosen.

**Surgery loop 07:15 — rebased onto current main, both conflicts gone. One red left, and it is Jay's call.** Head is now `78b30ad5` (was `d865766c`, 12 commits behind). Main had moved underneath this branch and the merge was no longer clean: `docs/DEVICE_REGISTRY.md` and `research/open-device-requests.json` both conflicted with `03b4613e`. **Resolved mechanically, nothing semantic invented** - `docs/DEVICE_REGISTRY.md` — generated file, so regenerated with `npm run regen:registry-doc` instead of hand-merged: `2,353 -> 2,356 models` (+3: the E9605/E9612/E9624 rows), Huawei 99 -> 102, switch 426 -> 429, open-requests section rewritten from the JSON. Diff against main is exactly +4/-10. - `research/open-device-requests.json` — kept main's newer text (it had gained the 2026-11-18 provenance corrections) and removed the three rows this PR fulfils (#807/#808/#810): 5 entries -> 2 (#130, #803), 53 lines deleted, nothing else touched. `test/apply-research.test.js` 10/10 and `test/request-queue.test.js` 32/32 pass; `audit-port-types` and `check-control-bytes` clean. **Still red, same single cause** — `Test (Node 18/20/22)`: ``` Against the real catalogue - the queue Jay is looking at: not ok - none of the seven open requests is in the catalogue [#810, #808, #807] 71 passed, 1 failed ``` That is the frozen seven-entry queue in `test/close-merged-issues.test.js` asserting that the boxes this PR adds are *not* in the catalogue. Everything else is green: classify, smoke, docker and the rebase job all pass, and a local `run-tests.js` on the merged tree gives 334 suites / 11,588 assertions / 0 failed. **Decision (needs Jay):** authorise replacing that literal queue with a read of `research/open-device-requests.json` — verified on 2026-11-18 04:50 at 72 passed / 0 failed — or rule that the guard should question the live tracker instead. No loop may edit `test/`, so this PR stays red, and so will every other device-landing PR, until one of those is chosen.
jaylouisw force-pushed catalogue/1118-0326 from 78b30ad576
Some checks failed
CI / Classify the change (pull_request) Successful in 7s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 4s
CI / Server smoke test (pull_request) Successful in 6s
CI / Docker build (pull_request) Successful in 20s
CI / Test (Node 20) (pull_request) Failing after 48s
CI / Test (Node 18) (pull_request) Failing after 55s
CI / Test (Node 22) (pull_request) Failing after 52s
CI / Arm auto-merge (pull_request) Successful in 3s
to 27cd12cc95
Some checks failed
CI / Classify the change (pull_request) Successful in 6s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 3s
CI / Server smoke test (pull_request) Successful in 5s
CI / Docker build (pull_request) Successful in 20s
CI / Test (Node 20) (pull_request) Failing after 46s
CI / Test (Node 18) (pull_request) Failing after 53s
CI / Test (Node 22) (pull_request) Failing after 51s
CI / Arm auto-merge (pull_request) Successful in 2s
2026-11-18 07:42:27 +00:00
Compare
Author
Owner

Surgery loop 07:40 — rebased again onto a7ff99ab; conflicts were the same two files, still one red.

Head is now 27cd12cc (was 78b30ad5). Two merges landed while the last run was reporting (#1951 mac-table, #1952 EUCC/#810), both inside this PR's conflict surface, so Forgejo had flipped it to mergeable=false — a real content conflict, not the uncomputed case.

Resolved mechanically, nothing semantic invented

  • docs/DEVICE_REGISTRY.md — generated, so regenerated (npm run regen:registry-doc): 2,356 models, and --check is clean afterwards; the open-requests section rebuilt from the JSON.
  • research/open-device-requests.json — kept this PR's side (the three rows it fulfils, #807/#808/#810, stay deleted). #1952's new EUCC sentence was an edit inside the #810 row that this PR closes, and the text it added lives on main in research/optixtrans-euuc-security-target-brand-join.md, which this PR does not touch.

Verified after rebase: mergeable=true on 27cd12cc; 10 files, +2697/−69. regen-registry-doc --check, audit-port-types and check-control-bytes all exit 0. Suites covering the touched files: catalogue-hygiene 25/0, doc-catalog-counts 43/0, device-aliases 13/0, spec-survives-load 32/0, spec-vs-description 3/0, missing-devices 66/0, faceplate-names-every-connector 33/0.

The red is unchanged and is still Jay's call — test/close-merged-issues.test.js 71 passed, 1 failed:

not ok - none of the seven open requests is in the catalogue, so none is closed by this workflow  [#810, #808, #807]

That is the seven-entry literal queue at lines 307-315 of main's test, tripped by three specs that are correctly in the catalogue. I swept every remote branch: none carries a change to that test file, so no other stream is about to land the fix. Decision (comment 10887, unchanged): authorise replacing the literal queue with a read of research/open-device-requests.json, or rule that the guard should read the live tracker.

**Surgery loop 07:40 — rebased again onto `a7ff99ab`; conflicts were the same two files, still one red.** Head is now **`27cd12cc`** (was `78b30ad5`). Two merges landed while the last run was reporting (#1951 mac-table, #1952 EUCC/`#810`), both inside this PR's conflict surface, so Forgejo had flipped it to `mergeable=false` — a real content conflict, not the uncomputed case. **Resolved mechanically, nothing semantic invented** - `docs/DEVICE_REGISTRY.md` — generated, so regenerated (`npm run regen:registry-doc`): 2,356 models, and `--check` is clean afterwards; the open-requests section rebuilt from the JSON. - `research/open-device-requests.json` — kept this PR's side (the three rows it fulfils, #807/#808/#810, stay deleted). #1952's new EUCC sentence was an edit *inside* the `#810` row that this PR closes, and the text it added lives on main in `research/optixtrans-euuc-security-target-brand-join.md`, which this PR does not touch. **Verified after rebase:** `mergeable=true` on `27cd12cc`; 10 files, +2697/−69. `regen-registry-doc --check`, `audit-port-types` and `check-control-bytes` all exit 0. Suites covering the touched files: catalogue-hygiene 25/0, doc-catalog-counts 43/0, device-aliases 13/0, spec-survives-load 32/0, spec-vs-description 3/0, missing-devices 66/0, faceplate-names-every-connector 33/0. **The red is unchanged and is still Jay's call** — `test/close-merged-issues.test.js` 71 passed, 1 failed: ``` not ok - none of the seven open requests is in the catalogue, so none is closed by this workflow [#810, #808, #807] ``` That is the seven-entry literal queue at lines 307-315 of main's test, tripped by three specs that are correctly in the catalogue. I swept every remote branch: **none** carries a change to that test file, so no other stream is about to land the fix. Decision (comment 10887, unchanged): authorise replacing the literal queue with a read of `research/open-device-requests.json`, or rule that the guard should read the live tracker.
Author
Owner

Surgery loop 07:55 — corrected diagnosis: main is green, and the snapshot has drifted in both directions. Still Jay's call.

Head unchanged at 27cd12cc (rebased at 07:40, mergeable=true). No merge: head status is failure on all three Node legs.

Correction to my 07:40 note. I said main's queue was the thing tripping this. It is not — main at a7ff99ab is green, and node test/close-merged-issues.test.js on a clean origin/main worktree gives 72 passed, 0 failed. The red is caused by this PR's own correct content: it adds devices/huawei/optixtrans-{e9605,e9612,e9624}.json, so the guard's premise — "none of the seven open requests is in the catalogue" — becomes false by design.

not ok - none of the seven open requests is in the catalogue, so none is closed by this workflow  [#810, #808, #807]
71 passed, 1 failed

Classification: real failing suite, rule 3 — not add/add (branch sits on current main), not an inherited main breakage.

New: the literal queue has drifted in both directions. At 27cd12cc:

  • research/open-device-requests.json → [130, 803] (this PR deletes 807/808/810 because it models them).
  • test/close-merged-issues.test.js:307-315 → seven entries, including #743 and #742 (OceanProtect X8100/X9100, E8000) which are not open requests in the tracker at all on either main or this head.

So trimming only the three modelled entries would leave the guard asserting about two issues nobody is tracking.

Two verified one-file fixes (probe only, nothing pushed, worktree reverted):

  • A — trim the queue to [803, 743, 742, 130]: 1 insertion, 4 deletions → 72 passed, 0 failed. Verified locally just now. Keeps the stale 743/742 rows.
  • B — read the live tracker instead of the literal copy → 72 passed, 0 failed (verified at 04:50, comment 10811). Queue becomes [130, 803], which is the truth.

Recommendation: B, and it is a one-file change to test/close-merged-issues.test.js. It is test content, so I will not land it as my own contribution.

Blast radius if it stays. The guard reds any PR that models a queued request. #803 (E6616X) and #130 (UDW-PRO) are both actively being chased by the issues/docs streams; whichever models one first hits this exact wall, and auto-merge cannot fire. This is the pipeline's next stop, not just this PR's.

Decision needed from Jay: replace the literal seven-entry queue with a read of research/open-device-requests.json (option B). Say yes and it is a one-line change on this branch plus one rebase.

**Surgery loop 07:55 — corrected diagnosis: main is green, and the snapshot has drifted in *both* directions. Still Jay's call.** Head unchanged at `27cd12cc` (rebased at 07:40, `mergeable=true`). No merge: head status is `failure` on all three Node legs. **Correction to my 07:40 note.** I said main's queue was the thing tripping this. It is not — main at `a7ff99ab` is green, and `node test/close-merged-issues.test.js` on a clean `origin/main` worktree gives **72 passed, 0 failed**. The red is caused by this PR's own correct content: it adds `devices/huawei/optixtrans-{e9605,e9612,e9624}.json`, so the guard's premise — *"none of the seven open requests is in the catalogue"* — becomes false by design. ``` not ok - none of the seven open requests is in the catalogue, so none is closed by this workflow [#810, #808, #807] 71 passed, 1 failed ``` Classification: **real failing suite, rule 3** — not add/add (branch sits on current main), not an inherited main breakage. **New: the literal queue has drifted in both directions.** At `27cd12cc`: - `research/open-device-requests.json` → `[130, 803]` (this PR deletes 807/808/810 because it models them). - `test/close-merged-issues.test.js:307-315` → seven entries, including **#743 and #742 (OceanProtect X8100/X9100, E8000) which are not open requests in the tracker at all** on either main or this head. So trimming only the three modelled entries would leave the guard asserting about two issues nobody is tracking. **Two verified one-file fixes** (probe only, nothing pushed, worktree reverted): - **A — trim the queue** to `[803, 743, 742, 130]`: `1 insertion, 4 deletions` → **72 passed, 0 failed**. Verified locally just now. Keeps the stale 743/742 rows. - **B — read the live tracker** instead of the literal copy → **72 passed, 0 failed** (verified at 04:50, comment 10811). Queue becomes `[130, 803]`, which is the truth. **Recommendation: B**, and it is a one-file change to `test/close-merged-issues.test.js`. It is test content, so I will not land it as my own contribution. **Blast radius if it stays.** The guard reds *any* PR that models a queued request. #803 (E6616X) and #130 (UDW-PRO) are both actively being chased by the issues/docs streams; whichever models one first hits this exact wall, and auto-merge cannot fire. This is the pipeline's next stop, not just this PR's. **Decision needed from Jay:** replace the literal seven-entry queue with a read of `research/open-device-requests.json` (option B). Say yes and it is a one-line change on this branch plus one rebase.
Author
Owner

Surgery loop 08:10 — reproduced the failure exactly; it is a landmine under main too, not just this PR. Still Jay's call.

Head unchanged at 27cd12cc, mergeable=true, 2 commits behind 3d33b7ca. origin/main is green at 3d33b7ca (push CI, all three Node legs success), so nothing here is inherited.

The one failing assertion, verbatim (CI run 6038, job 2; reproduced locally in a scratch worktree at head):

test/close-merged-issues.test.js:305-318
  not ok - none of the seven open requests is in the catalogue,
           so none is closed by this workflow  [#810, #808, #807]
71 passed, 1 failed

The block hardcodes a seven-row queue ([810, 808, 807, 803, 743, 742, 130]) and asserts deviceMatch(title, loadDeviceIndex()) is empty for every row. This PR's three correct specs make three of those rows match, by design. Per the red-PR rule: not add/add, not inherited — a real failing suite (rule 3).

Why it is bigger than this PR. That literal list cannot update itself. The moment a spec for any of #810/#808/#807/#803/#743/#742/#130 lands on main by any route, main's own push CI goes red for all seven streams — and ubnt/130 is live in another worktree right now. This PR is just the first arrival.

Two fixes, both one file, both verified by me locally, neither landed (test content is not mine to change):

  • A (recommended) read research/open-device-requests.json instead of the literal array — 71/0 at head, queue becomes [130, 803].
  • B keep the literal but drop rows as their specs land — 72/0, needs an edit every time.

Needs Jay's decision: A, B, or neither. If the guard fix goes in as its own PR (owned by the issues stream, since it guards scripts/close-merged-issues.js), it should land first; #1934 then goes green on the next push. I will not touch test/.

**Surgery loop 08:10 — reproduced the failure exactly; it is a landmine under *main* too, not just this PR. Still Jay's call.** Head unchanged at `27cd12cc`, `mergeable=true`, 2 commits behind `3d33b7ca`. `origin/main` is **green** at `3d33b7ca` (push CI, all three Node legs success), so nothing here is inherited. **The one failing assertion, verbatim** (CI run 6038, job 2; reproduced locally in a scratch worktree at head): ``` test/close-merged-issues.test.js:305-318 not ok - none of the seven open requests is in the catalogue, so none is closed by this workflow [#810, #808, #807] 71 passed, 1 failed ``` The block hardcodes a seven-row queue (`[810, 808, 807, 803, 743, 742, 130]`) and asserts `deviceMatch(title, loadDeviceIndex())` is empty for every row. This PR's three correct specs make three of those rows match, by design. Per the red-PR rule: **not** add/add, **not** inherited — a real failing suite (rule 3). **Why it is bigger than this PR.** That literal list cannot update itself. The moment a spec for **any** of #810/#808/#807/#803/#743/#742/**#130** lands on main by any route, main's own push CI goes red for all seven streams — and `ubnt/130` is live in another worktree right now. This PR is just the first arrival. **Two fixes, both one file, both verified by me locally, neither landed (test content is not mine to change):** - **A (recommended)** read `research/open-device-requests.json` instead of the literal array — 71/0 at head, queue becomes `[130, 803]`. - **B** keep the literal but drop rows as their specs land — 72/0, needs an edit every time. **Needs Jay's decision:** A, B, or neither. If the guard fix goes in as its own PR (owned by the issues stream, since it guards `scripts/close-merged-issues.js`), it should land **first**; #1934 then goes green on the next push. I will not touch `test/`.
Author
Owner

Surgery loop 08:25 — correcting my 08:17 note: the landmine did not go off, and #1955 has now changed this PR's shape. mergeable=false is real.

Two things happened while I was writing the last comment.

1. My landmine claim was wrong — retracted. I said main would go red the moment any of those seven requests got a spec. It did not: #1955 (commit 1304b0bd, merged 08:16) landed devices/huawei/optixtrans-e9605.json and dropped [807, ...] from the guard's literal queue in the same commit — six rows now, plus a new positive assertion that #807 resolves. node test/close-merged-issues.test.js on current main: 73 passed, 0 failed. Push CI green on Node 20, 18/22 running. So the literal queue is maintained per-spec by whoever lands the spec; it is not an unattended landmine, and I was wrong to call it one.

2. #1955 collides with this PR. Main now carries its own E9605 — a stated 5x TNG1N210 build — where this branch carries a derived one. Merging origin/main locally into 27cd12cc gives seven conflicts, and the decisive one is:

CONFLICT (add/add): devices/huawei/optixtrans-e9605.json

plus content conflicts in README.md, docs/CONNECTORS.md, docs/DEVICE_REGISTRY.md, docs/wiki/Home.md, research/open-device-requests.json, scripts/build-wiki.js. test/close-merged-issues.test.js merged cleanly, so the guard now carries rows for 808 and 810 — this branch's E9612/E9624 specs will still red it until those two rows go, which is test content and not mine to write.

Needs Jay's decision, and it is now a content question, not a rebase:

  • Which E9605 wins? Adopt main's 1304b0bd and re-cut this branch as E9612 + E9624 only (my recommendation — main's is the stated build, and it is already landed, cited and green), or keep this branch's derived framing and revert 1304b0bd. I will not choose between two vendor-sourced specs.
  • Once the branch is re-cut with E9612/E9624, the same commit has to drop rows 808/810 from the guard queue, exactly as #1955 did for 807 — that is the pattern the stream has already established, and it needs the catalogue stream (or Jay) to write it.

Nothing merged, nothing rebased, nothing dropped: an add/add on a file another stream landed 17 minutes ago is not a mechanical conflict. Standing by for the call.

**Surgery loop 08:25 — correcting my 08:17 note: the landmine did not go off, and #1955 has now changed this PR's shape. `mergeable=false` is real.** Two things happened while I was writing the last comment. **1. My landmine claim was wrong — retracted.** I said main would go red the moment any of those seven requests got a spec. It did not: `#1955` (commit `1304b0bd`, merged 08:16) landed `devices/huawei/optixtrans-e9605.json` **and** dropped `[807, ...]` from the guard's literal queue in the same commit — six rows now, plus a new positive assertion that #807 resolves. `node test/close-merged-issues.test.js` on current main: **73 passed, 0 failed**. Push CI green on Node 20, 18/22 running. So the literal queue is maintained per-spec by whoever lands the spec; it is not an unattended landmine, and I was wrong to call it one. **2. #1955 collides with this PR.** Main now carries its own E9605 — a *stated* 5x TNG1N210 build — where this branch carries a derived one. Merging `origin/main` locally into `27cd12cc` gives seven conflicts, and the decisive one is: ``` CONFLICT (add/add): devices/huawei/optixtrans-e9605.json ``` plus content conflicts in `README.md`, `docs/CONNECTORS.md`, `docs/DEVICE_REGISTRY.md`, `docs/wiki/Home.md`, `research/open-device-requests.json`, `scripts/build-wiki.js`. `test/close-merged-issues.test.js` merged cleanly, so the guard now carries rows for **808** and **810** — this branch's E9612/E9624 specs will still red it until those two rows go, which is test content and not mine to write. **Needs Jay's decision, and it is now a content question, not a rebase:** - **Which E9605 wins?** Adopt main's `1304b0bd` and re-cut this branch as E9612 + E9624 only (my recommendation — main's is the stated build, and it is already landed, cited and green), or keep this branch's derived framing and revert `1304b0bd`. I will not choose between two vendor-sourced specs. - Once the branch is re-cut with E9612/E9624, the same commit has to drop rows 808/810 from the guard queue, exactly as `#1955` did for 807 — that is the pattern the stream has already established, and it needs the catalogue stream (or Jay) to write it. Nothing merged, nothing rebased, nothing dropped: an add/add on a file another stream landed 17 minutes ago is not a mechanical conflict. Standing by for the call.
Author
Owner

Surgery loop 08:55 — new duplicate on the way in, and the guard trap is systemic.

#1962 (ec3b431f, 0 behind main, mergeable) models the same E9612 as this PR: 13x TNG1N210, 130x 10G line, same board-to-subrack table. It differs only in metadata — its model is the lowercase slug optixtrans-e9612 (this PR's is Title Case, and this PR alone carries color/icon), plus different label/description/notes/sourceUrl. Whichever lands first leaves the other's E9612 a duplicate, and two specs of one box cannot both stay in the catalogue.

It is red for the same reason this PR is: test/close-merged-issues.test.js hardcodes the open-request queue and asserts none of the six has a spec. I reproduced it at ec3b431f:

not ok - none of the six open requests is in the catalogue ... [#808]
72 passed, 1 failed

So the trap is not this branch's fault — any device PR resolving a request trips it.

What is still uniquely this PR's, and what is not:

  • E9605 — landed on main as #1955 (1304b0bd), add/add conflict with this PR's version.
  • E9612 — now also in #1962.
  • E9624 — only here, nothing else on main or in the queue models it.

Jay's decision, stated small: (a) let #1962 carry E9612 and shrink this PR to E9624, or (b) keep this PR as the single E9600-family commit and close #1962, taking whichever E9605 model you prefer. Either way the branch needs a rebase afterwards, and the guard literal needs the 807/808/810 rows updated. I have not resolved the E9605 or E9612 files: both sides are defensible data, not adjacent-insertion noise.

**Surgery loop 08:55 — new duplicate on the way in, and the guard trap is systemic.** `#1962` (`ec3b431f`, 0 behind main, mergeable) models the **same E9612** as this PR: 13x TNG1N210, 130x 10G line, same board-to-subrack table. It differs only in metadata — its `model` is the lowercase slug `optixtrans-e9612` (this PR's is Title Case, and this PR alone carries `color`/`icon`), plus different `label`/`description`/`notes`/`sourceUrl`. Whichever lands first leaves the other's E9612 a duplicate, and two specs of one box cannot both stay in the catalogue. It is red for the same reason this PR is: `test/close-merged-issues.test.js` hardcodes the open-request queue and asserts none of the six has a spec. I reproduced it at `ec3b431f`: ``` not ok - none of the six open requests is in the catalogue ... [#808] 72 passed, 1 failed ``` So the trap is not this branch's fault — any device PR resolving a request trips it. **What is still uniquely this PR's, and what is not:** - E9605 — landed on main as `#1955` (`1304b0bd`), add/add conflict with this PR's version. - E9612 — now also in `#1962`. - E9624 — **only here**, nothing else on main or in the queue models it. **Jay's decision, stated small:** (a) let `#1962` carry E9612 and shrink this PR to E9624, or (b) keep this PR as the single E9600-family commit and close `#1962`, taking whichever E9605 model you prefer. Either way the branch needs a rebase afterwards, and the guard literal needs the 807/808/810 rows updated. I have not resolved the E9605 or E9612 files: both sides are defensible data, not adjacent-insertion noise.
Author
Owner

Rebase forensics, run 2026-11-18 09:1x against origin/main = 9d852ed8. I attempted the rebase in a scratch worktree and aborted it: it is not mechanically resolvable, and one conflict is a question I must not answer.

$ git rebase origin/main            # on 27cd12cc
CONFLICT (content): README.md
CONFLICT (add/add): devices/huawei/optixtrans-e9605.json
CONFLICT (content): docs/CONNECTORS.md
CONFLICT (content): docs/DEVICE_REGISTRY.md
CONFLICT (content): docs/wiki/Home.md
CONFLICT (content): research/open-device-requests.json
CONFLICT (content): scripts/build-wiki.js

The one that stops it — devices/huawei/optixtrans-e9605.json is add/add against a file main already ships. 1304b0bd (PR #1955) landed an E9605 on 2026-11-18. The two files are not a formatting difference; they disagree about the product:

main (1304b0bd) this PR (27cd12cc)
label OptiXtrans E9605 OTN line shelf, 5x TNG1N210 (50x 10G line) OptiXtrans E9605 OTN platform, DC, 5x TNG1N210 (50x 10G line)
keys includes icon, color neither key
interfaces 50 50

So "line shelf" vs "OTN platform, DC", plus the presentation keys. Choosing between them is a statement about the device, so I have left both sides alone. Needs Jay's decision: which E9605 model ships. Once that is answered the file is a one-line keep-theirs/keep-ours and I can finish the rebase.

The other six are mine and are mechanical — README.md counts, scripts/build-wiki.js (2,354 vs 2,356 models), and the generated docs/DEVICE_REGISTRY.md / docs/CONNECTORS.md / docs/wiki/Home.md / research/open-device-requests.json, which I would regenerate rather than hand-merge. I can do that in one pass the moment the e9605 question is settled.

Two more things in the way, both already reported:

  1. This PR duplicates #1962 on E9612. Main has neither optixtrans-e9612.json nor optixtrans-e9624.json; #1962 carries the E9612 alone, is 0 commits behind and mergeable: true. Only E9624 is unique to this PR. Both cannot land.
  2. The guard. node test/close-merged-issues.test.js here gives 71 passed, 1 failed: none of the seven open requests is in the catalogue ... [#810, #808, #807]. This PR edits that literal (the wording says "seven") but keeps #807, #808 and #810 in the queue while its own catalogue answers all three. Tracked as #1963.

What I would do, if it were mine to decide: close this in favour of #1962 for E9612, and re-open E9624 as its own PR off current main — E9624 is the only content here that nothing else carries, and it would not collide with #1962 or with main's E9605. Say the word and I will do exactly that, or finish the rebase if you would rather keep the E9605/E9612 models from this PR.

Rebase forensics, run 2026-11-18 09:1x against `origin/main` = `9d852ed8`. I attempted the rebase in a scratch worktree and aborted it: **it is not mechanically resolvable**, and one conflict is a question I must not answer. ``` $ git rebase origin/main # on 27cd12cc CONFLICT (content): README.md CONFLICT (add/add): devices/huawei/optixtrans-e9605.json CONFLICT (content): docs/CONNECTORS.md CONFLICT (content): docs/DEVICE_REGISTRY.md CONFLICT (content): docs/wiki/Home.md CONFLICT (content): research/open-device-requests.json CONFLICT (content): scripts/build-wiki.js ``` **The one that stops it — `devices/huawei/optixtrans-e9605.json` is add/add against a file main already ships.** `1304b0bd` (PR #1955) landed an E9605 on 2026-11-18. The two files are not a formatting difference; they disagree about the product: | | main (`1304b0bd`) | this PR (`27cd12cc`) | |---|---|---| | label | `OptiXtrans E9605 OTN line shelf, 5x TNG1N210 (50x 10G line)` | `OptiXtrans E9605 OTN platform, DC, 5x TNG1N210 (50x 10G line)` | | keys | includes `icon`, `color` | neither key | | interfaces | 50 | 50 | So "line shelf" vs "OTN platform, DC", plus the presentation keys. Choosing between them is a statement about the device, so I have left both sides alone. **Needs Jay's decision: which E9605 model ships.** Once that is answered the file is a one-line keep-theirs/keep-ours and I can finish the rebase. **The other six are mine and are mechanical** — README.md counts, `scripts/build-wiki.js` (2,354 vs 2,356 models), and the generated `docs/DEVICE_REGISTRY.md` / `docs/CONNECTORS.md` / `docs/wiki/Home.md` / `research/open-device-requests.json`, which I would regenerate rather than hand-merge. I can do that in one pass the moment the e9605 question is settled. **Two more things in the way, both already reported:** 1. **This PR duplicates #1962 on E9612.** Main has neither `optixtrans-e9612.json` nor `optixtrans-e9624.json`; #1962 carries the E9612 alone, is 0 commits behind and `mergeable: true`. Only **E9624** is unique to this PR. Both cannot land. 2. **The guard.** `node test/close-merged-issues.test.js` here gives 71 passed, 1 failed: `none of the seven open requests is in the catalogue ... [#810, #808, #807]`. This PR edits that literal (the wording says "seven") but keeps #807, #808 and #810 in the queue while its own catalogue answers all three. Tracked as #1963. **What I would do, if it were mine to decide:** close this in favour of #1962 for E9612, and re-open E9624 as its own PR off current main — E9624 is the only content here that nothing else carries, and it would not collide with #1962 or with main's E9605. Say the word and I will do exactly that, or finish the rebase if you would rather keep the E9605/E9612 models from this PR.
Author
Owner

Two of the three devices in this PR are already on main. That is why it is red on every Node leg, and no rebase fixes it.

Head 27cd12cc: 5/8 checks green, failure on Node 18, 20 and 22. Classified the #1902 way — merged origin/main (23098286) into this branch in a scratch worktree:

CONFLICT (add/add): devices/huawei/optixtrans-e9605.json
CONFLICT (add/add): devices/huawei/optixtrans-e9612.json
CONFLICT (content): README.md, docs/CONNECTORS.md, docs/DEVICE_REGISTRY.md,
                   docs/wiki/Home.md, research/open-device-requests.json, scripts/build-wiki.js

Both add/add files are already on main: optixtrans-e9605.json landed in 1304b0bd (#1955), optixtrans-e9612.json in 3f58050a (#1962, merged 09:41 today). So the pull_request run is built from a merge that reintroduces files main carries — red on all three legs, not a code defect. docs/DEVICE_REGISTRY.md and docs/wiki/Home.md are generated, so those resolve by regeneration; the two device JSONs do not, because they are a different model of the same device than what landed, and choosing between them is a content decision, not a mechanical merge.

Superseded: #807 (E9605) and #808 (E9612) are answered on main by the two commits above.

What is actually left is #810, E9624. Worth re-reading before you re-cut it: #810's stated missing field is the cross-connect mode, and the board-to-subrack matrix you cite (EDOC1100126589 Table 3) gives which subrack takes a TNG1N* board and its port count — it does not state a cross-connect mode. If the E9624 spec asserts one, that is the "presence is not correctness" trap that produced the PBE-5AC-Gen2 correction: an issue closed against an unchecked spec ships the error and still marks the request answered.

Recommendation: re-cut E9624-only from current main as a new branch and PR; don't rebase this one. Left open rather than closed, since it is the only place the E9624 work exists.

Needs Jay's decision: does E9624 ship now as a partial model (no cross-connect mode, stated as absent), or wait for a source that prints the mode?

**Two of the three devices in this PR are already on main. That is why it is red on every Node leg, and no rebase fixes it.** Head `27cd12cc`: 5/8 checks green, failure on Node 18, 20 and 22. Classified the #1902 way — merged `origin/main` (`23098286`) into this branch in a scratch worktree: ``` CONFLICT (add/add): devices/huawei/optixtrans-e9605.json CONFLICT (add/add): devices/huawei/optixtrans-e9612.json CONFLICT (content): README.md, docs/CONNECTORS.md, docs/DEVICE_REGISTRY.md, docs/wiki/Home.md, research/open-device-requests.json, scripts/build-wiki.js ``` Both add/add files are already on main: `optixtrans-e9605.json` landed in `1304b0bd` (#1955), `optixtrans-e9612.json` in `3f58050a` (#1962, merged 09:41 today). So the `pull_request` run is built from a merge that reintroduces files main carries — red on all three legs, not a code defect. `docs/DEVICE_REGISTRY.md` and `docs/wiki/Home.md` are generated, so those resolve by regeneration; the two device JSONs do not, because they are a *different model of the same device* than what landed, and choosing between them is a content decision, not a mechanical merge. **Superseded:** #807 (E9605) and #808 (E9612) are answered on main by the two commits above. **What is actually left is #810, E9624.** Worth re-reading before you re-cut it: #810's stated missing field is the *cross-connect mode*, and the board-to-subrack matrix you cite (`EDOC1100126589` Table 3) gives which subrack takes a TNG1N* board and its port count — it does not state a cross-connect mode. If the E9624 spec asserts one, that is the "presence is not correctness" trap that produced the PBE-5AC-Gen2 correction: an issue closed against an unchecked spec ships the error and still marks the request answered. **Recommendation:** re-cut E9624-only from current `main` as a new branch and PR; don't rebase this one. Left open rather than closed, since it is the only place the E9624 work exists. **Needs Jay's decision:** does E9624 ship now as a partial model (no cross-connect mode, stated as absent), or wait for a source that prints the mode?
Author
Owner

This PR is now two-thirds superseded, and its remaining third has a single-device replacement: #1969 (catalogue/1118-0956, off main 23098286).

  • devices/huawei/optixtrans-e9605.json — landed on main as #1955 (1304b0bd).
  • devices/huawei/optixtrans-e9612.json — landed on main as #1962 (3f58050a).
  • devices/huawei/optixtrans-e9624.json — #1969 carries the same reading this PR does (1:1 mode, 24x TNG1N210, 240x 10G line ports, IU1-IU24), from the same Boards Overview Table 3 / Table 1 rows, with the TNG2DAP section 1.9 slot map and the EUCC Security Target's OSN 9800 M24 = E9624 brand join written into its notes.

So the add/add conflicts on the first two files are the PR's own content arriving twice, and there is nothing left here that #1969 does not carry. Nothing was pushed or merged by this pass — flagging it so the two do not both land.

This PR is now two-thirds superseded, and its remaining third has a single-device replacement: **#1969** (`catalogue/1118-0956`, off main `23098286`). - `devices/huawei/optixtrans-e9605.json` — landed on main as **#1955** (`1304b0bd`). - `devices/huawei/optixtrans-e9612.json` — landed on main as **#1962** (`3f58050a`). - `devices/huawei/optixtrans-e9624.json` — **#1969** carries the same reading this PR does (1:1 mode, 24x TNG1N210, 240x 10G line ports, IU1-IU24), from the same Boards Overview Table 3 / Table 1 rows, with the TNG2DAP section 1.9 slot map and the EUCC Security Target's `OSN 9800 M24` = `E9624` brand join written into its notes. So the add/add conflicts on the first two files are the PR's own content arriving twice, and there is nothing left here that #1969 does not carry. Nothing was pushed or merged by this pass — flagging it so the two do not both land.
jaylouisw closed this pull request 2026-11-18 10:22:16 +00:00
Author
Owner

Superseded in full — all three devices are on main now. Closing.

I said at 10:02 that this PR was two-thirds superseded and its remaining third had a single-device replacement in #1969. #1969 has since landed, so the third third is on main too:

  • devices/huawei/optixtrans-e9605.json — #1955, 1304b0bd
  • devices/huawei/optixtrans-e9612.json — #1962, 3f58050a
  • devices/huawei/optixtrans-e9624.json — #1969, 60990df8 (the stated 1:1 24x TNG1N210 reading, which is the mode question this PR could not answer)

Nothing unique is lost. The Table 3 board-to-subrack matrix this PR carried in research/optixtrans-line-board-matrix.md is already quoted on main in research/optixtrans-e9605-stated-build.md, research/optixtrans-e9612-stated-build.md and research/optixtrans-e9624-stated-build.md (all three cite EDOC1100126589 Table 3 and the 1:1 mode: Y / 1:3 mode: Y row). Its research/open-device-requests.json deletions were done by the PRs that landed.

Closed as superseded, not merged — head 27cd12cc stays red because it reintroduces specs main already carries. No decision needed from Jay on this one any more.

**Superseded in full — all three devices are on main now. Closing.** I said at 10:02 that this PR was two-thirds superseded and its remaining third had a single-device replacement in #1969. #1969 has since landed, so the third third is on main too: - `devices/huawei/optixtrans-e9605.json` — #1955, `1304b0bd` - `devices/huawei/optixtrans-e9612.json` — #1962, `3f58050a` - `devices/huawei/optixtrans-e9624.json` — #1969, `60990df8` (the stated 1:1 24x TNG1N210 reading, which is the mode question this PR could not answer) Nothing unique is lost. The Table 3 board-to-subrack matrix this PR carried in `research/optixtrans-line-board-matrix.md` is already quoted on main in `research/optixtrans-e9605-stated-build.md`, `research/optixtrans-e9612-stated-build.md` and `research/optixtrans-e9624-stated-build.md` (all three cite EDOC1100126589 Table 3 and the `1:1 mode: Y` / `1:3 mode: Y` row). Its `research/open-device-requests.json` deletions were done by the PRs that landed. Closed as superseded, not merged — head `27cd12cc` stays red because it reintroduces specs main already carries. No decision needed from Jay on this one any more.
Some checks failed
CI / Classify the change (pull_request) Successful in 6s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 3s
CI / Server smoke test (pull_request) Successful in 5s
CI / Docker build (pull_request) Successful in 20s
CI / Test (Node 20) (pull_request) Failing after 46s
CI / Test (Node 18) (pull_request) Failing after 53s
CI / Test (Node 22) (pull_request) Failing after 51s
CI / Arm auto-merge (pull_request) Successful in 2s

Pull request closed

Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
jaylouisw/packetforge!1934
No description provided.