How much of the HC3 does flua cover?

flua simulates a slice of the HC3 — the parts QAs, tools, and the viewer actually touch — and forwards the rest to the real controller in online mode. Until now "a slice" was a feeling. flua audit turns it into numbers:

.venv/bin/flua audit
# flua audit
# REST routes (docs/specs/hc3-api.json)
#   78 of 356 simulated (21%)
#   278 not simulated (forwarded to the HC3 online)
#   not simulated by area: energy 24, panels 30, scenes 11, settings 30, ...
# Device catalog (src/flua/devices/devices.json)
#   176 types: 26 light-modeled, 29 specific-device captures with light actions, 52 other, 69 inert
# Type hierarchy (devices/hierarchy.json): 175 types (firmware 5.210.12), 174 in the catalog, 175 resolvable
# QA templates (quickapp-types): 30 templates, 30 with resolvable types

Three numbers matter. Routes: 356 endpoints in the distilled swagger, 78 answered offline — the floor is pinned by a test, so coverage can only go up. Templates: all 30 QA scaffolds must actually boot, or CI fails. And types: 176 cataloged device types, 26 of them light-modeled. The specific captures (an FGR dimmer, a Davis weather station) are data for the house — for a QA you'd use the parent type instead, and the parents are the modeled ones.

Device types flua has never heard of

New HC3 firmware introduces new types. The catalog can lag — but flua now ships the HC3's own type hierarchy (distilled from the undocumented /api/devices/hierarchy), and an unknown type falls back to its nearest cataloged ancestor. Declare a type the catalog predates and it still runs, as its parent's skeleton, keeping its own name:

--%%type:com.fibaro.FGD223
print("TYPE", api.get("/devices/".._FLUA.qaId).type)
exit(0)

.venv/bin/flua qa.lua
# [qa5000]: TYPE com.fibaro.FGD223

Types the hierarchy never heard of still fail loudly with "unknown device type" — a typo stays a typo. The hierarchy file is regenerable from any HC3 (scripts/extract_hierarchy.py), so updating flua's knowledge is one command, and flua audit --json hands the same report to your agent or CI.

Next time: the structured trace stream — what your AI agent should read while it debugs your QA.

flua audit: USAGE.md · hierarchy: devices/hierarchy.json · pip: fibaro-flua 0.1.83 · flua: https://github.com/jangabrielsson/flua