devices(ubiquiti): the LTE Backup Pro's 15.4 W passthrough port was dead (refs #130) #2150

Merged
jaylouisw merged 2 commits from catalogue/1119-1222 into main 2026-11-20 00:08:38 +00:00
Owner

refs #130 (records the catalogue defect found underneath this request; does not close it)

#130 asks for a Dream Wall Pro and has been blocked on every Ubiquiti surface for four passes; this pass did not re-walk them. It spent the increment on the lane the request's own lab names - NAT, DHCP, VPN, cellular WAN - because the two rows a requester is handed for cellular WAN are the LTE modems, and the 11:0x pass fixed only the UMR.

devices/ubiquiti/u-lte-backup-pro.json declared poe-pse, and its description promised a passthrough port, but eth1 carried no poeOut and the row had no poe block at all. engine/device.js therefore fell through to the model-name parser, which found no wattage in "U-LTE-Backup Pro" and set the budget to 0, and engine/simulator.js opens its allocator with if (!dev.poeBudget) continue; - so the box was skipped, not reported unknown. The documented port could power nothing and show power inline said nothing about why, the same defect class as umr.json in #2128.

Measured before/after (U6-Lite cabled to eth1):

poeBudget PSE ports computePoe()
before 0 (source: name) 0 no entry - not even a denial
after 15.4 (source: research) 1 powers the AP at 12.95 W of 15.4 W

Sources (all Ubiquiti-hosted, read 2026-11-19)

  • https://techspecs.ui.com/unifi/integrations/u-lte-backup-pro - 200, 624,978 B: "Networking interface (2) GbE RJ45 ports"; "PoE interface (1) PoE in (PoE output not used) / (1) PoE passthrough output (Requires a PoE+ switch)"; "Max. power consumption Excluding PoE passthrough output: 8.5W"; "Max. PoE passthrough wattage per port 15.4W*" with "*Requires a PoE+ switch"; "Supported voltage range 44-57V DC".
  • https://dl.ui.com/ds/u-lte-backup-pro_ds.pdf - 200, application/pdf, 329,546 B, 2 pp, text layer: the same census, including "15.4W*" and the PoE+ footnote, plus "Throughput speeds Download 150 Mbps / Upload 50 Mbps".
  • https://store.ui.com/us/en/products/u-lte-backup-pro - 200, 402,625 B (landed on its collection URL): repeats the spec table, and its diagram labels the output "PoE+ GbE Port (Passthrough)"; the power-method row reads "PoE+ (Pairs 1, 2+; 3, 6- | Pairs 4, 5+; 7, 8-)" - four-pair Type 2 on the input.

What changed, and the judgement calls

  • eth1: poeOut: true, poeOutClass: "PoE+" - the vendor labels the port PoE+ - while poe.budgetW stays at the vendor's own 15.4 W per-port cap. The two are both vendor statements; the note records that the class label is the store's and the cap is the datasheet's.
  • capabilities: 802.3at added, because the 8.5 W own draw plus 15.4 W passed through cannot come from an 802.3af source - which is exactly what "*Requires a PoE+ switch" means.
  • wwan0 speed 100M -> 150M: the old note said the datasheet "names the category rather than a measured rate", but it prints Download 150 Mbps / Upload 50 Mbps. 150M is also what umr.json and the other Cat 4 rows use.
  • Not guessed: neither surface numbers the two RJ45 ports, so eth0 sinks and eth1 sources on the existing catalogue convention (same as umr.json, zyxel/gs1900-24ep.json). Recorded in the notes rather than presented as a vendor fact.

Gates: scripts/syntax-check.js 453 files OK; scripts/audit-port-types.js "No port contradicts its declared type"; test/hardware-invariants.test.js 26/0; test/poe.test.js 21/0; test/poe-inference.test.js 17/0 (including "a device that claims a PoE budget actually sources power"); test/catalogue-hygiene.test.js 25/0; test/doc-catalog-counts.test.js 43/0; scripts/regen-registry-doc.js "catalogue pages are current".

#130 stays open and blocked, unchanged: Ubiquiti publishes no UDW-PRO port census on any surface checked (techspecs, store, datasheets, FCC, Wayback, and the 701-record controller device database). Unblock is still any Ubiquiti-hosted page/datasheet/QSG/FCC exhibit/storefront/KB article/device record naming UDW-PRO, or the requester confirming the base UDW.

refs #130 (records the catalogue defect found underneath this request; does not close it) #130 asks for a Dream Wall Pro and has been blocked on every Ubiquiti surface for four passes; this pass did not re-walk them. It spent the increment on the lane the request's own lab names - *NAT, DHCP, VPN, **cellular WAN*** - because the two rows a requester is handed for cellular WAN are the LTE modems, and the 11:0x pass fixed only the UMR. `devices/ubiquiti/u-lte-backup-pro.json` declared `poe-pse`, and its description promised a passthrough port, but `eth1` carried no `poeOut` and the row had **no `poe` block at all**. `engine/device.js` therefore fell through to the model-name parser, which found no wattage in "U-LTE-Backup Pro" and set the budget to 0, and `engine/simulator.js` opens its allocator with `if (!dev.poeBudget) continue;` - so the box was skipped, not reported unknown. The documented port could power nothing and `show power inline` said nothing about why, the same defect class as `umr.json` in #2128. **Measured before/after** (U6-Lite cabled to `eth1`): | | poeBudget | PSE ports | computePoe() | |---|---|---|---| | before | 0 (`source: name`) | 0 | no entry - not even a denial | | after | 15.4 (`source: research`) | 1 | powers the AP at 12.95 W of 15.4 W | **Sources** (all Ubiquiti-hosted, read 2026-11-19) - `https://techspecs.ui.com/unifi/integrations/u-lte-backup-pro` - 200, 624,978 B: "Networking interface (2) GbE RJ45 ports"; "PoE interface (1) PoE in (PoE output not used) / (1) PoE passthrough output (Requires a PoE+ switch)"; "Max. power consumption Excluding PoE passthrough output: 8.5W"; "**Max. PoE passthrough wattage per port 15.4W***" with "*Requires a PoE+ switch"; "Supported voltage range 44-57V DC". - `https://dl.ui.com/ds/u-lte-backup-pro_ds.pdf` - 200, `application/pdf`, 329,546 B, 2 pp, text layer: the same census, including "15.4W*" and the PoE+ footnote, plus "Throughput speeds Download 150 Mbps / Upload 50 Mbps". - `https://store.ui.com/us/en/products/u-lte-backup-pro` - 200, 402,625 B (landed on its collection URL): repeats the spec table, and its diagram labels the output **"PoE+ GbE Port (Passthrough)"**; the power-method row reads "PoE+ (Pairs 1, 2+; 3, 6- | Pairs 4, 5+; 7, 8-)" - four-pair Type 2 on the input. **What changed, and the judgement calls** - `eth1`: `poeOut: true`, `poeOutClass: "PoE+"` - the vendor labels the port PoE+ - while `poe.budgetW` stays at the vendor's own **15.4 W** per-port cap. The two are both vendor statements; the note records that the class label is the store's and the cap is the datasheet's. - `capabilities`: `802.3at` added, because the 8.5 W own draw plus 15.4 W passed through cannot come from an 802.3af source - which is exactly what "*Requires a PoE+ switch" means. - `wwan0` speed 100M -> **150M**: the old note said the datasheet "names the category rather than a measured rate", but it prints **Download 150 Mbps / Upload 50 Mbps**. 150M is also what `umr.json` and the other Cat 4 rows use. - **Not guessed:** neither surface numbers the two RJ45 ports, so `eth0` sinks and `eth1` sources on the existing catalogue convention (same as `umr.json`, `zyxel/gs1900-24ep.json`). Recorded in the notes rather than presented as a vendor fact. **Gates:** `scripts/syntax-check.js` 453 files OK; `scripts/audit-port-types.js` "No port contradicts its declared type"; `test/hardware-invariants.test.js` 26/0; `test/poe.test.js` 21/0; `test/poe-inference.test.js` 17/0 (including "a device that claims a PoE budget actually sources power"); `test/catalogue-hygiene.test.js` 25/0; `test/doc-catalog-counts.test.js` 43/0; `scripts/regen-registry-doc.js` "catalogue pages are current". **#130 stays open and blocked, unchanged:** Ubiquiti publishes no UDW-PRO port census on any surface checked (techspecs, store, datasheets, FCC, Wayback, and the 701-record controller device database). Unblock is still any Ubiquiti-hosted page/datasheet/QSG/FCC exhibit/storefront/KB article/device record naming `UDW-PRO`, or the requester confirming the base UDW.
devices(ubiquiti): the LTE Backup Pro's 15.4 W passthrough port was dead (refs #130)
Some checks failed
CI / Classify the change (pull_request) Successful in 13s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 9s
CI / Server smoke test (pull_request) Has been skipped
CI / Docker build (pull_request) Has been skipped
CI / Test (Node 22) (pull_request) Failing after 54s
CI / Test (Node 20) (pull_request) Failing after 55s
CI / Test (Node 18) (pull_request) Failing after 58s
CI / Arm auto-merge (pull_request) Successful in 3s
e95b22d382
poe-pse was declared and the description promised a passthrough, but eth1 carried no poeOut and the row had no poe block, so device.js parsed a budget of 0 from the name and simulator.computePoe() skipped the box entirely: the documented port could power nothing and show power inline said nothing about it.

Vendor: techspecs 'PoE interface (1) PoE in (PoE output not used) / (1) PoE passthrough output (requires a PoE+ switch)' and 'Max. PoE passthrough wattage per port 15.4W*'; the datasheet prints the same row; the store names the port 'PoE+ GbE Port (Passthrough)'. eth1 is now poeOut/PoE+ with poe.budgetW 15.4; 802.3at added because the input must be Type 2 (8.5 W own draw + 15.4 W passthrough). wwan0 corrected 100M -> 150M: the datasheet prints download 150 Mbps, not just the Cat 4 class.

Verified: before, poeBudget 0 and computePoe() returned no entry; after, a U6-Lite on eth1 is powered at 12.95 W of 15.4 W.
Author
Owner

Red classified: real failing suite, this PR's own file — not staleness, not a superseded merge. mergeable=true, and a local origin/main merge has no conflict, so this is not the add/add case.

Local repro (node test/spec-vs-description.test.js, head e95b22d3 merged with main 053b3995), 1 of 449 failing:

a device described as PoE-powered does not source power: 1 device(s) claim power they cannot supply:
      ubiquiti/U-LTE-Backup Pro: says PoE-in, 1 port(s) source

Cause — a regex gap, not the device data. test/spec-vs-description.test.js:124 exempts a box from that rule only when its description matches both POE_IN and POE_OUT; POE_OUT needs a literal passthrough, poe out, N poe+, or poe … out. The old description read "one passing PoE through" — also a non-match — but the old JSON had no poeOut on eth1, so the rule never fired. This PR adds poeOut: true / poeOutClass: "PoE+" to eth1 (correctly — the vendor's "PoE+ GbE Port (Passthrough)"), and the new description says "one passing up to 15.4 W of PoE+ through", which still matches neither passthrough nor poe … out. So the exemption misses and the device is judged as claiming power it cannot supply.

Also note the PR does two things the description doesn't claim: it raises wwan0 from 100M to 150M and adds a poe budget block — both consistent with the cited datasheet rows.

Needs a decision — the author's (-firmware), or Jay's if it is a policy call:

  1. keep the description's wording and say the word the vendor uses — e.g. "one passing up to 15.4 W of PoE+ through (passthrough)" — so the existing exemption recognises the two-direction box; or
  2. widen POE_OUT to accept "passing … PoE+ through", which changes a gate that every device in the catalogue is judged by.

(1) is one clause and leaves the invariant untouched; (2) is the honest fix if the regex is considered too literal, but it loosens the check for all 700-odd devices, so I am not making that call in surgery.

Verified: full suite on the merged tree — 449 tests, 1 fail (this one); every other leg green locally. Branch left untouched, nothing rebased, nothing merged.

**Red classified: real failing suite, this PR's own file — not staleness, not a superseded merge.** `mergeable=true`, and a local `origin/main` merge has no conflict, so this is not the add/add case. Local repro (`node test/spec-vs-description.test.js`, head `e95b22d3` merged with main `053b3995`), 1 of 449 failing: ``` a device described as PoE-powered does not source power: 1 device(s) claim power they cannot supply: ubiquiti/U-LTE-Backup Pro: says PoE-in, 1 port(s) source ``` **Cause** — a regex gap, not the device data. `test/spec-vs-description.test.js:124` exempts a box from that rule only when its *description* matches both `POE_IN` and `POE_OUT`; `POE_OUT` needs a literal `passthrough`, `poe out`, `N poe+`, or `poe … out`. The old description read "one passing PoE through" — also a non-match — but the old JSON had no `poeOut` on `eth1`, so the rule never fired. This PR adds `poeOut: true` / `poeOutClass: "PoE+"` to `eth1` (correctly — the vendor's "PoE+ GbE Port (Passthrough)"), and the new description says "one passing up to 15.4 W of PoE+ through", which still matches neither `passthrough` nor `poe … out`. So the exemption misses and the device is judged as claiming power it cannot supply. Also note the PR does two things the description doesn't claim: it raises `wwan0` from `100M` to `150M` and adds a `poe` budget block — both consistent with the cited datasheet rows. **Needs a decision — the author's (`-firmware`), or Jay's if it is a policy call:** 1. keep the description's wording and say the word the vendor uses — e.g. "one passing up to 15.4 W of PoE+ through (passthrough)" — so the existing exemption recognises the two-direction box; or 2. widen `POE_OUT` to accept "passing … PoE+ through", which changes a gate that every device in the catalogue is judged by. (1) is one clause and leaves the invariant untouched; (2) is the honest fix if the regex is considered too literal, but it loosens the check for all 700-odd devices, so I am not making that call in surgery. **Verified:** full suite on the merged tree — 449 tests, 1 fail (this one); every other leg green locally. Branch left untouched, nothing rebased, nothing merged.
Author
Owner

Re-verified against today's main (df8e41d9), unchanged: still exactly one failing assertion.

Clean origin/main merge (no conflicts), full suite locally: 382 suites, 12,889 assertions, 1 failed —

✗ test/spec-vs-description.test.js
    a device described as PoE-powered does not source power:
      ubiquiti/U-LTE-Backup Pro: says PoE-in, 1 port(s) source

Everything else (port-type audit, control bytes, validation, hardware invariants, poe) is green on the merged tree, and the PR's own device JSON is not at fault — the gap is the POE_OUT exemption in test/spec-vs-description.test.js not accepting "passing … PoE+ through".

So the two options from my 12:40Z comment still stand, and the fix is one clause, not structural:

  1. author (-firmware): say the vendor's word — "…through (passthrough)" — the existing exemption then matches;
  2. Jay, if the regex is judged too literal: widen POE_OUT to accept "passing … through", which loosens a gate applied to every device.

test/ is not surgery's to edit, so this stays with the author or Jay. Nothing rebased, branch untouched (e95b22d3, now 32 behind main).

**Re-verified against today's main (`df8e41d9`), unchanged: still exactly one failing assertion.** Clean `origin/main` merge (no conflicts), full suite locally: **382 suites, 12,889 assertions, 1 failed** — ``` ✗ test/spec-vs-description.test.js a device described as PoE-powered does not source power: ubiquiti/U-LTE-Backup Pro: says PoE-in, 1 port(s) source ``` Everything else (port-type audit, control bytes, validation, hardware invariants, poe) is green on the merged tree, and the PR's own device JSON is not at fault — the gap is the `POE_OUT` exemption in `test/spec-vs-description.test.js` not accepting "passing … PoE+ through". So the two options from my 12:40Z comment still stand, and the fix is one clause, not structural: 1. **author (`-firmware`)**: say the vendor's word — "…through (passthrough)" — the existing exemption then matches; 2. **Jay, if the regex is judged too literal**: widen `POE_OUT` to accept "passing … through", which loosens a gate applied to every device. `test/` is not surgery's to edit, so this stays with the author or Jay. Nothing rebased, branch untouched (`e95b22d3`, now 32 behind main).
Author
Owner

Surgery re-verification, 18:5xZ — main is now e0fa4221 (five commits on from the last check).

Rebased cleanly onto e0fa4221 in a scratch worktree (no conflicts, mechanically clean), then ran the suite that owns this failure:

$ node test/spec-vs-description.test.js
  ✗ a device described as PoE-powered does not source power: 1 device(s) claim power they cannot supply:
      ubiquiti/U-LTE-Backup Pro: says PoE-in, 1 port(s) source
2 passed, 1 failed

Identical single failure — nothing new, no regression from the eleven UniFi devices that landed in #2192 (that PR touched UCG/UDM/UDR/UX models, not the LTE Backup Pro). I did not push the rebase: a clean rebase cannot change this outcome, and the repo would only gain another red run.

Still needs Jay's decision: either the port data is wrong (a PoE-in-only box should not have poeOut: true on eth1) or the rule needs a passthrough exception (a device may pass through power it does not itself supply). Surgery can implement whichever you choose; it cannot choose for you.

**Surgery re-verification, 18:5xZ — main is now `e0fa4221` (five commits on from the last check).** Rebased cleanly onto `e0fa4221` in a scratch worktree (no conflicts, mechanically clean), then ran the suite that owns this failure: ``` $ node test/spec-vs-description.test.js ✗ a device described as PoE-powered does not source power: 1 device(s) claim power they cannot supply: ubiquiti/U-LTE-Backup Pro: says PoE-in, 1 port(s) source 2 passed, 1 failed ``` Identical single failure — nothing new, no regression from the eleven UniFi devices that landed in #2192 (that PR touched UCG/UDM/UDR/UX models, not the LTE Backup Pro). I did **not** push the rebase: a clean rebase cannot change this outcome, and the repo would only gain another red run. Still **needs Jay's decision**: either the port data is wrong (a PoE-in-only box should not have `poeOut: true` on `eth1`) or the rule needs a passthrough exception (a device may *pass through* power it does not itself supply). Surgery can implement whichever you choose; it cannot choose for you.
devices(ubiquiti): the LTE Backup Pro's passthrough port was sourced but not described (refs #130)
All checks were successful
CI / Classify the change (pull_request) Successful in 15s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 15s
CI / Server smoke test (pull_request) Successful in 17s
CI / Test (Node 20) (pull_request) Successful in 1m2s
CI / Test (Node 22) (pull_request) Successful in 1m2s
CI / Test (Node 18) (pull_request) Successful in 1m8s
CI / Docker build (pull_request) Successful in 25s
CI / Arm auto-merge (pull_request) Successful in 5s
da2ecc2634
PR #2150 modelled eth1 as the PoE+ passthrough port but left the
description saying only "passing up to 15.4 W of PoE+ through", which
the spec-vs-description comparator reads as a PoE-in-only box that
sources power (test/spec-vs-description.test.js, red on all three Node
legs). The vendor's own row calls it "(1) PoE passthrough output"; say
"passthrough" in the prose, and state 2x 1G RJ45 in the countable form
the port-claim check parses.
Author
Owner

2026-11-19 23:5x UTC - the red legs were this file's prose, not its ports. Fixed in da2ecc26; full suite now 374/374, 12,696 assertions, 0 failed.

All three Node legs failed test/spec-vs-description.test.js with ubiquiti/U-LTE-Backup Pro: says PoE-in, 1 port(s) source. The check exempts a dual-role box only when its description names the output side; the previous text said "one passing up to 15.4 W of PoE+ through", which matches none of the comparator's output phrasings. The vendor's own row is (1) PoE passthrough output (requires a PoE+ switch) — tech-spec page and datasheet p.1 text layer both - so the fix is the vendor's word, not a re-wording to quiet the suite:

... 2x 1G RJ45, one PoE-in port (whose output is not used) and one PoE+ passthrough output at up to 15.4 W, plus an LTE Cat 4 radio (150/50 Mbps) ...

The 2x 1G RJ45 form also puts the port count under the comparator's "promises more ports than it has" check (2 claimed, 2 present). No port, wattage, capability or source changed.

Gates on da2ecc26: scripts/syntax-check.js 453 files OK; scripts/audit-port-types.js no contradiction; test/spec-vs-description.test.js 3/3; test/hardware-invariants.test.js 26/26; scripts/audit-poe-budget.js 0 budgets without a PSE; scripts/run-tests.js 374 suites / 12,696 assertions / 0 failed.

**2026-11-19 23:5x UTC - the red legs were this file's prose, not its ports. Fixed in `da2ecc26`; full suite now 374/374, 12,696 assertions, 0 failed.** All three Node legs failed `test/spec-vs-description.test.js` with `ubiquiti/U-LTE-Backup Pro: says PoE-in, 1 port(s) source`. The check exempts a dual-role box only when its description names the output side; the previous text said *"one passing up to 15.4 W of PoE+ through"*, which matches none of the comparator's output phrasings. The vendor's own row is `(1) PoE passthrough output (requires a PoE+ switch)` — tech-spec page and datasheet p.1 text layer both - so the fix is the vendor's word, not a re-wording to quiet the suite: > ... 2x 1G RJ45, one PoE-in port (whose output is not used) and one PoE+ passthrough output at up to 15.4 W, plus an LTE Cat 4 radio (150/50 Mbps) ... The `2x 1G RJ45` form also puts the port count under the comparator's *"promises more ports than it has"* check (2 claimed, 2 present). No port, wattage, capability or source changed. Gates on `da2ecc26`: `scripts/syntax-check.js` 453 files OK; `scripts/audit-port-types.js` no contradiction; `test/spec-vs-description.test.js` 3/3; `test/hardware-invariants.test.js` 26/26; `scripts/audit-poe-budget.js` 0 budgets without a PSE; `scripts/run-tests.js` 374 suites / 12,696 assertions / 0 failed.
forgejo-actions scheduled this pull request to auto merge when all checks succeed 2026-11-20 00:03:59 +00:00
jaylouisw deleted branch catalogue/1119-1222 2026-11-20 00:08:40 +00:00
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!2150
No description provided.