AggregateError and the energy_ setting prefix
Verdict: refuted as the cause of Athom AggregateError.
energy_source is a real validator defect. homey-lib 4.4.1 (node_modules/homey-lib/lib/App/index.js, RESERVED_SETTING_PREFIXES) warns when a setting id starts with homey:, zw_, zb_, mtr_, thread_, zone_, energy_, satellite_mode_, or homekit_. The warning is console.warn, not a thrown error, which is why a local homey app validate --level publish can exit 0 and still print drivers.<id> invalid setting id: energy_source. That defect is renamed to tuya_energy_source and blocked by npm run check:p3285. It does not separate the refused Athom builds from the accepted ones.
Counts below are git grep of the setting id "id": "energy_source" in drivers/*/driver.compose.json and driver.settings.compose.json at the commit that was published. Driver ids energy_meter_3ch_bidirectional, energy_meter_3phase, and energy_meter_din match the prefix energy_ and are not settings. No other reserved prefix occurred as a setting id on the compared commits.
Evidence matrix
| Version | Athom build | Result | Commit | energy_source files | Payload drivers | What the ledger says changed |
|---|---|---|---|---|---|---|
| 9.0.1365 | 3483 | AggregateError | bcbc9a7c4e | 0 | not recorded | first build with deprecated: true on wall_remote_4_gang; 9.0.1364 / #3482 reached Test |
| 9.0.1599 | 3682 | Test | 11ca32b731 | 0 | 347 | payload accepted, draft then Test |
| 9.0.1602 | 3684 | AggregateError | 49be689153 | 93 | 361 | 14 never-published driver ids added over the 347-driver Test set |
| 9.0.1603 | 3685 | AggregateError | cc97563484 | 103 | 353 | 6 never-published ids still released; settings, flow cards, images, and assets matched the accepted tree |
| 9.0.1604 | 3686 | Test | 83fa78e9b4 | 103 | 347 | same driver set as 9.0.1599, hold list restored |
| 9.0.1606 | 3688 | AggregateError | e2120e3367 | 103 | 347 + released holds | P3095 bisect, held drivers released |
| 9.0.1615 | 3697 | Test | 86628bcc39 | 103 | 347 set | release_held=none; ledger says #3687–#3696 came from released held drivers |
| 9.0.1616 | not stored | AggregateError | 8725d07254 | 103 | 347 + 1 released id | isolated water_leak_aoyan_ay222z |
| 9.0.1618 | not stored | AggregateError | 2718d08ef3 | 103 | 347 + 2 released ids | water_leak_blitzwolf_bw_is9 and sauna_thermostat_ae5503 |
Source of the result column: docs/knowledge/athom-publish-outcomes.jsonl. The setting was introduced by 85ddc0b480 (2026-10-26, P3067, 92 files). Its parent 6ecb7d1d82 has 0. Every AggregateError in this table from 9.0.1365 through 9.0.1599 is therefore before the setting existed.
347 / 348 boundary
The figure 347 is the last accepted payload driver count (#3682, #3686, #3697). It is not a count of drivers that carry energy_source. That count moved 0 → 92 → 93 → 103 and then stayed at 103 on both sides of the boundary. The refused payloads at 353 and 361 drivers have the same setting the accepted 347-driver payload has. Releasing one never-published driver on top of that accepted set was enough to get AggregateError again (9.0.1616), with no change to the setting count.
Build #3699 is not a row in the outcomes ledger. #3697 is 9.0.1615 and is the last numbered Test build there. The next recorded AggregateError versions, 9.0.1616 and 9.0.1618, still have 103 energy_source files, the same number as #3697.
Other AggregateError candidates
These are the candidates the same ledger actually distinguishes. They are not a claim that any one of them is the only server-side check.
| Candidate | Evidence | Relation to AggregateError |
|---|---|---|
| Never-published driver id in the payload | 347-driver set accepted three times; 353, 361, and single released hold ids refused | Strongest correlation in this file |
deprecated: true | #3483 refused, #3482 and the revert reached Test, no energy_source yet | Explains that one pair, not the later streak |
| Manifest byte size | Accepted 9.0.1604 app.json is 3,328,185 bytes; refused 9.0.1616 manifest is recorded at 3,110,579 bytes | Does not rise with the failures |
| Flow card ids, images, assets | 9.0.1603 note: those diffs were clean against accepted 9.0.1599 | Not the delta of that refusal |
| Identical bytes, opposite result | 9.0.1424 AggregateError and 9.0.1425 Test are recorded as the same content | At least one historical AggregateError was flaky |
| Socket hang up | Many processing_failed rows, including 9.0.1596–9.0.1598 at 347 drivers | Different stateMeta; not AggregateError |
| Diagnostic staging path | Run 43487854250 wrote /tmp/homey-diag-catalogue (372 drivers) and the size gate read /tmp/homey-publish-temp | Died before upload. athom-final had no build for 9.0.1619, so this run never became an AggregateError |
Reserved prefixes stay blocked because the SDK rejects them. The Athom accept/refuse split in this ledger follows the published driver set, not the presence or the count of energy_source.
Follow-up (2026-11-15)
The five drivers the P3095 bisect isolated are compared field by field in aggregateerror-isolated-drivers.md. No content trait separates them from the accepted ones; every AggregateError since 9.0.1602 has a payload of 348+ drivers and every accepted build has 347.