[GreenGrassBlueOcean]

Part of #735

What is the problem

1. The gateway profile oracle check could not fail for most profiles. scripts/update_gateway_profiles.py and custom_components/myhome/where_grammar.py each carried a shim that ran at import and rewrote OWNd's auth_measured, audio_measured and supported_who for F453, F453AV, F454, F461, MyHomeServer1, MH200 and MH200N with hard-coded values taken from the firmware oracle. verify_profiles_against_oracle() then compared those rewritten values with the same oracle, so a regression in OWNd for any of those seven models would pass test_gateway_profiles_match_oracle_verdicts and sync_documentation.py --check. The script also held a third, older copy of the shim.
2. The live F461 2026-10-18 captures were not checked against the oracle. Their frames were only checked to parse, which OWNMessage.parse() does for any well-formed frame, including an unknown WHO.

How is it solved

1. Shims removed. All three copies are gone. Every value they wrote is already OWNd's own value in 2.0.0b11 (the minimum in pyproject.toml and the manifest.json pin) and in OWNd master, checked profile by profile. The verification now reads OWNd's real profile data.
2. Guard against a new shim. verify_profiles_against_oracle() reports any OWNd profile whose class __init__ was replaced, or whose registered instance no longer equals a fresh one. Two tests in tests/test_gateway_profiles.py patch an instance and a class and expect the violation.
3. F461 traces tied to the oracle (test_f461_live_trace_agrees_with_oracle, one case per trace):
   • every event the capture recorded with an ownd_class must still decode to that class (73 + 58 events);
   • every command the gateway answered that the F461 020011 oracle also replayed must agree with it: live ACK when the oracle verdict is out, NACK when it is silent (6 frames per trace).

How is it tested

• Full suite against OWNd 2.0.0b11: 4392 passed, 100.0 % statement coverage.
• The changed test files also pass against OWNd master.
• ruff, mypy, scripts/verify_ha_standards.py and scripts/sync_documentation.py --check pass.

Where are the traces

• tests/fixtures/traces/f461_live_2026-10-18/myhome_trace_F461_live_2026-10-18.json (physical F461, firmware 2.0.11)
• tests/fixtures/traces/f461_who14_live_2026-10-18/myhome_trace_F461_who14_live_2026-10-18.json (physical F461, firmware 2.0.11)

Both were already on v2-phase1-architecture; this PR adds no fixtures.

What does the oracle say

Frame	Live F461 2.0.11	Oracle F461 020011 reply / verdict
*#13**15##, *#13**16##, *#13**23##, *#13**24##	ACK	ack / out
*#1*0##, *#18*51*113##	ACK	ack / out
*1*1*0##, *#2*0##, *#18*51*#1200#1*0##	ACK	nack / out
*#18*51*1200##	NACK	nack / silent

The live ACK/NACK follows the oracle's verdict in every case. Its reply field disagrees with the hardware on three frames, so the test does not compare it. That field is also what test_golden_corpus_against_firmware_oracle and KNOWN_GATEWAY_DISCREPANCIES rely on, which is worth a separate look.

Not in this PR

The earlier description also listed an oracle resync, README/docs calibration and the repair-issue catalog. None of that is in this diff (the base branch already passes the oracle integrity test and sync_documentation.py --check), so this PR no longer closes #735.

🤖 Generated with Claude Code

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!