catalogue(usw-ultra): the PoE budget is a function of the power input, and port 8 takes power rather than giving it #2005

Merged
jaylouisw merged 1 commit from catalogue/1118-1626 into main 2026-11-18 18:00:04 +00:00
Owner

Related to #130, which found the same defect one layer up: a PoE budget that exists in a description but not in poe.budgetW, so the engine falls through to the model-name parser. This PR does not close #130 - UDW-PRO still has no Ubiquiti source.

What the three specs said

devices/ubiquiti/usw-ultra.json, -60w and -210w each carried the description "42 W PoE budget" - including the 210 W one - and none carried poe.budgetW:

spec before show power inline why
USW-Ultra budget not modelled: the ports are reported, the power they could supply is not UDW-style parser gets nothing from the name
USW-Ultra-60W 60 W parsed from the name - that is the PSU rating
USW-Ultra-210W 210 W parsed from the name - that is the max draw including PoE output

All three also gave the switch 8 PSE ports, the eighth being eth7 at poeOutClass: "PoE++" (60 W).

What Ubiquiti publishes

One row, "Total PoE Availability", with five values keyed on how the unit is powered:

power input PoE availability
PoE++ input 42 W
PoE+ input 16 W
60 W PoE adapter input 42 W
60 W AC adapter input 52 W
210 W AC adapter input 202 W

All three print the identical five-row table: the SKU suffix names the adapter, not a different PoE engine. The store page says the same in prose - "PoE input is supported for switch operation, with PoE+/PoE++ input or AC power required for PoE output" - and offers the one product with "No AC Adapter" / "60W AC Adapter" / "210W AC Adapter" as power options (https://store.ui.com/us/en/products/usw-ultra, 200, 673,337 B; .../usw-ultra-210w, 200, 660,250 B).

So each spec now holds the row its own name selects: 42 W / 52 W / 202 W, with all five figures in poe.note so a lab powered from a different adapter takes that row instead of inheriting this one - the same shape as the ECS-24-PoE PSU table.

The census error the same table corrected

The port layout is "8 (7 PoE+; 1 PoE++ input)" and "Max. PoE Output: Up to PoE+", with "Max. PoE Wattage per Port by PSE: PoE+ 30 W" and no PoE++ PSE row. Port 8 is where the unit takes power from an upstream switch - which is why its power method reads "(1) AC/DC adapter (Excluded) / (1) PoE++" and why the store requires PoE+/PoE++ input or AC power for any PoE output at all.

eth7 is therefore "poeOut": 0, "poeIn": 9 (9 W is the vendor's "max. power consumption excluding PoE output with PoE input"), so the model reports 7 PSE ports rather than 8. This is the phantom-PoE class again - the extra UDM draw - this time on the source side of one port.

The eight ports also regained the 10M cap that the vendor's own "(1G/100M/10M)" line gives them and the specs had dropped.

Verified, not assumed

== USW-Ultra        budget=42 W   PSE ports=7   PD-input ports=eth7(9W)
Total: 12.95 W of 42 W in use, 29.05 W left, nothing denied
== USW-Ultra-60W    budget=52 W   PSE ports=7   PD-input ports=eth7(9W)
Total: 12.95 W of 52 W in use, 39.05 W left, nothing denied
== USW-Ultra-210W   budget=202 W  PSE ports=7   PD-input ports=eth7(9W)
Total: 12.95 W of 202 W in use, 189.05 W left, nothing denied

Gates

  • node scripts/syntax-check.js - 418 files, all OK
  • node scripts/audit-port-types.js - no port contradicts its declared type
  • node scripts/audit-poe-budget.js - silent 130 -> 129, name-inferred 22 -> 20, orphans 0
  • node test/hardware-invariants.test.js - 26 passed, 0 failed
  • node test/poe.test.js - 18 passed, 0 failed; duplicate-models 21, port-roles 19, combo 19, apply-research 10 - all 0 failed
  • research/poe-budget-field-census.md gains section 4 recording the family and the corrected cohorts

Trap for the next pass: https://dl.ui.com/ds/usw-ultra_ds.pdf 302s to https://techspecs.ui.com/ and serves the same 609,911-byte home page as every missing dl.ui.com document - there is no Ultra datasheet, and that page is not one.

Related to #130, which found the same defect one layer up: a PoE budget that exists in a description but not in `poe.budgetW`, so the engine falls through to the model-name parser. **This PR does not close #130** - UDW-PRO still has no Ubiquiti source. ## What the three specs said `devices/ubiquiti/usw-ultra.json`, `-60w` and `-210w` each carried the description "42 W PoE budget" - including the 210 W one - and none carried `poe.budgetW`: | spec | before `show power inline` | why | |--|--|--| | `USW-Ultra` | `budget not modelled: the ports are reported, the power they could supply is not` | `UDW`-style parser gets nothing from the name | | `USW-Ultra-60W` | **60 W** | parsed from the name - that is the PSU rating | | `USW-Ultra-210W` | **210 W** | parsed from the name - that is the max draw *including* PoE output | All three also gave the switch **8 PSE ports**, the eighth being `eth7` at `poeOutClass: "PoE++"` (60 W). ## What Ubiquiti publishes One row, "Total PoE Availability", with five values keyed on **how the unit is powered**: | power input | PoE availability | |--|--| | PoE++ input | 42 W | | PoE+ input | 16 W | | 60 W PoE adapter input | 42 W | | 60 W AC adapter input | 52 W | | 210 W AC adapter input | 202 W | - <https://techspecs.ui.com/unifi/switching/usw-ultra> - 200, 628,685 B - <https://techspecs.ui.com/unifi/switching/usw-ultra-60w> - 200, 628,751 B - <https://techspecs.ui.com/unifi/switching/usw-ultra-210w> - 200, 628,783 B All three print the identical five-row table: the SKU suffix names the adapter, not a different PoE engine. The store page says the same in prose - *"PoE input is supported for switch operation, with PoE+/PoE++ input or AC power required for PoE output"* - and offers the one product with "No AC Adapter" / "60W AC Adapter" / "210W AC Adapter" as power options (<https://store.ui.com/us/en/products/usw-ultra>, 200, 673,337 B; `.../usw-ultra-210w`, 200, 660,250 B). So each spec now holds **the row its own name selects**: 42 W / 52 W / 202 W, with all five figures in `poe.note` so a lab powered from a different adapter takes that row instead of inheriting this one - the same shape as the ECS-24-PoE PSU table. ## The census error the same table corrected The port layout is **"8 (7 PoE+; 1 PoE++ input)"** and "Max. PoE Output: **Up to PoE+**", with "Max. PoE Wattage per Port by PSE: PoE+ 30 W" and no PoE++ PSE row. Port 8 is where the unit *takes* power from an upstream switch - which is why its power method reads "(1) AC/DC adapter (Excluded) / (1) PoE++" and why the store requires PoE+/PoE++ input *or* AC power for any PoE output at all. `eth7` is therefore `"poeOut": 0, "poeIn": 9` (9 W is the vendor's "max. power consumption excluding PoE output with PoE input"), so the model reports 7 PSE ports rather than 8. This is the phantom-PoE class again - the extra UDM draw - this time on the source side of one port. The eight ports also regained the `10M` cap that the vendor's own "(1G/100M/10M)" line gives them and the specs had dropped. ## Verified, not assumed ``` == USW-Ultra budget=42 W PSE ports=7 PD-input ports=eth7(9W) Total: 12.95 W of 42 W in use, 29.05 W left, nothing denied == USW-Ultra-60W budget=52 W PSE ports=7 PD-input ports=eth7(9W) Total: 12.95 W of 52 W in use, 39.05 W left, nothing denied == USW-Ultra-210W budget=202 W PSE ports=7 PD-input ports=eth7(9W) Total: 12.95 W of 202 W in use, 189.05 W left, nothing denied ``` ## Gates - `node scripts/syntax-check.js` - 418 files, all OK - `node scripts/audit-port-types.js` - no port contradicts its declared type - `node scripts/audit-poe-budget.js` - silent 130 -> **129**, name-inferred 22 -> **20**, orphans 0 - `node test/hardware-invariants.test.js` - 26 passed, 0 failed - `node test/poe.test.js` - 18 passed, 0 failed; `duplicate-models` 21, `port-roles` 19, `combo` 19, `apply-research` 10 - all 0 failed - `research/poe-budget-field-census.md` gains section 4 recording the family and the corrected cohorts **Trap for the next pass:** `https://dl.ui.com/ds/usw-ultra_ds.pdf` 302s to `https://techspecs.ui.com/` and serves the same 609,911-byte home page as every missing `dl.ui.com` document - there is no Ultra datasheet, and that page is not one.
catalogue(usw-ultra): the PoE budget is a function of the power input, and port 8 takes power rather than giving it (#130)
All checks were successful
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 7s
CI / Classify the change (pull_request) Successful in 7s
CI / Server smoke test (pull_request) Successful in 9s
CI / Docker build (pull_request) Successful in 19s
CI / Test (Node 20) (pull_request) Successful in 50s
CI / Test (Node 22) (pull_request) Successful in 54s
CI / Test (Node 18) (pull_request) Successful in 59s
CI / Arm auto-merge (pull_request) Successful in 20s
1effe910d2
Three specs carried the sentence "42 W PoE budget" and no poe.budgetW, so the
engine fell through to the model-name parser: USW-Ultra reported nothing at all
over 8 PSE ports, and the 60W/210W siblings reported their PSU ratings (60 W,
210 W) as if they were budgets.

Ubiquiti publishes one row, "Total PoE Availability", with five values keyed on
how the unit is powered - PoE++ input 42 W, PoE+ input 16 W, 60 W PoE adapter
42 W, 60 W AC adapter 52 W, 210 W AC adapter 202 W - and the store offers the
same chassis with No AC Adapter / 60W / 210W as power options. Each spec now
carries the row its own name selects, and all five live in poe.note so a lab
powered differently takes that row instead of inheriting this one.

The same table corrects a census error: the port layout is "8 (7 PoE+; 1 PoE++
INPUT)" with "Max. PoE Output: Up to PoE+" and no PoE++ PSE wattage row, so
eth7 was a 60 W source that does not exist. It is now poeOut 0 / poeIn 9 W
(9 W is the vendor's draw excluding PoE output on PoE input) and the model
reports 7 PSE ports. All eight ports regained the 10M cap the vendor's
"1G/100M/10M" line gives them.

Verified: show power inline now prints 42/52/202 W with 7 PSE ports.
Gates: syntax-check 418 OK, audit-port-types clean, audit-poe-budget silent
130->129 and inferred 22->20, hardware invariants 26 passed, poe 18 passed.
forgejo-actions force-pushed catalogue/1118-1626 from 1effe910d2
All checks were successful
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 7s
CI / Classify the change (pull_request) Successful in 7s
CI / Server smoke test (pull_request) Successful in 9s
CI / Docker build (pull_request) Successful in 19s
CI / Test (Node 20) (pull_request) Successful in 50s
CI / Test (Node 22) (pull_request) Successful in 54s
CI / Test (Node 18) (pull_request) Successful in 59s
CI / Arm auto-merge (pull_request) Successful in 20s
to 567f07aae2 2026-11-18 16:50:15 +00:00
Compare
forgejo-actions force-pushed catalogue/1118-1626 from 567f07aae2 to cb44083f88 2026-11-18 16:54:55 +00:00 Compare
forgejo-actions force-pushed catalogue/1118-1626 from cb44083f88 to b2c0eeb11e 2026-11-18 17:09:48 +00:00 Compare
Author
Owner

CI has never dispatched for this branch — four SHAs, zero checks, while a sibling PR from the same loop got 8/8.

Statuses for 567f07aa, cb44083f and the current head b2c0eeb1 all report total_count: 0. #2007 (catalogue/1118-1656, created the same minute) got its full 8. So this is branch-specific, not repo-wide.

Likely mechanism: the "PR rebase / arm auto-merge" workflow force-pushes the branch when it is behind, and a bot-token push appears to suppress the resulting event — so each rebase lands a new SHA that starts no run, checks stay empty, the PR stays unmergeable, and the loop rebases again. b2c0eeb1 already contains current main (0df1bef6), so a further rebase is a no-op; what is needed is a human/loop-origin push of the same tree so the pull_request event fires.

What I verified locally on b2c0eeb1 (so nobody has to re-derive it):

  • merge with main: already up to date, no conflict
  • test/poe.test.js: 18 passed, 0 failed
  • scripts/check-canonical-urls.js: OK — 3154 files, every project URL canonical
  • scripts/check-control-bytes.js: clean

The tree looks good. I am not merging it with an empty status (standing rule: green head SHA only). If it stays stale, an empty commit pushed from the loop's own credentials should be enough to get the event to fire.

**CI has never dispatched for this branch — four SHAs, zero checks, while a sibling PR from the same loop got 8/8.** Statuses for `567f07aa`, `cb44083f` and the current head `b2c0eeb1` all report `total_count: 0`. #2007 (`catalogue/1118-1656`, created the same minute) got its full 8. So this is branch-specific, not repo-wide. Likely mechanism: the "PR rebase / arm auto-merge" workflow force-pushes the branch when it is behind, and a bot-token push appears to suppress the resulting event — so each rebase lands a new SHA that starts no run, checks stay empty, the PR stays unmergeable, and the loop rebases again. `b2c0eeb1` already contains current `main` (`0df1bef6`), so a further rebase is a no-op; what is needed is a *human/loop-origin* push of the same tree so the `pull_request` event fires. **What I verified locally on `b2c0eeb1` (so nobody has to re-derive it):** - merge with `main`: already up to date, no conflict - `test/poe.test.js`: 18 passed, 0 failed - `scripts/check-canonical-urls.js`: OK — 3154 files, every project URL canonical - `scripts/check-control-bytes.js`: clean The tree looks good. I am not merging it with an empty status (standing rule: green head SHA only). If it stays stale, an empty commit pushed from the loop's own credentials should be enough to get the event to fire.
forgejo-actions force-pushed catalogue/1118-1626 from b2c0eeb11e to 4b6f2ae22b
Some checks failed
CI / Classify the change (pull_request) Successful in 5s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 6s
CI / Server smoke test (pull_request) Successful in 8s
CI / Docker build (pull_request) Successful in 21s
CI / Test (Node 20) (pull_request) Failing after 48s
CI / Test (Node 22) (pull_request) Successful in 52s
CI / Test (Node 18) (pull_request) Successful in 55s
CI / Arm auto-merge (pull_request) Successful in 3s
2026-11-18 17:24:48 +00:00
Compare
forgejo-actions force-pushed catalogue/1118-1626 from 4b6f2ae22b
Some checks failed
CI / Classify the change (pull_request) Successful in 5s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 6s
CI / Server smoke test (pull_request) Successful in 8s
CI / Docker build (pull_request) Successful in 21s
CI / Test (Node 20) (pull_request) Failing after 48s
CI / Test (Node 22) (pull_request) Successful in 52s
CI / Test (Node 18) (pull_request) Successful in 55s
CI / Arm auto-merge (pull_request) Successful in 3s
to 4ec1a15cca
All checks were successful
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 4s
CI / Classify the change (pull_request) Successful in 5s
CI / Server smoke test (pull_request) Successful in 8s
CI / Docker build (pull_request) Successful in 22s
CI / Test (Node 20) (pull_request) Successful in 48s
CI / Test (Node 22) (pull_request) Successful in 50s
CI / Test (Node 18) (pull_request) Successful in 53s
CI / Arm auto-merge (pull_request) Successful in 31s
2026-11-18 17:45:24 +00:00
Compare
jaylouisw deleted branch catalogue/1118-1626 2026-11-18 18:00:05 +00:00
Author
Owner

Checks recovered the same way, then merged as c142e739. Same cause as #2008: 1effe910 was checked (run 6463), the rebase bot force-pushed 4b6f2ae2, no run for it. Close/reopen produced run 6497; Node 20 failed once there (Test (Node 20) only, 18 and 22 green) and a second close/reopen on 4ec1a15c passed 8/8 — so that failure was transient, not a defect. Nothing in the diff needed changing.

**Checks recovered the same way, then merged as `c142e739`.** Same cause as #2008: `1effe910` was checked (run 6463), the rebase bot force-pushed `4b6f2ae2`, no run for it. Close/reopen produced run 6497; Node 20 failed once there (`Test (Node 20)` only, 18 and 22 green) and a second close/reopen on `4ec1a15c` passed 8/8 — so that failure was transient, not a defect. Nothing in the diff needed changing.
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
2 participants
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!2005
No description provided.