fix(plc-modbus): RegisterData drops every named VFD register — match REGISTER_NAMES (#161)

[Mikecranesync]

Fixes #161 — the isolated bug-fix PR required by PR 2 of MIRA PRD #3048 (machine-evidence handoff, canonical-source correctness).

The bug — verified live against current main before fixing

read_io() (backend/services/plc_connection.py) builds its registers dict from REGISTER_NAMES:

ItemCount, ConveyorHz, MotorCurrentX10, MotorTempX10, VFDStatus, ErrorCode

but RegisterData (backend/models/plc_models.py) declared placeholder register_101..105 fields. Pydantic v2 silently drops unknown fields, so every named VFD value was stripped from /api/plc/io and replaced with the 0 default — diagnosis read zeros while the PLC reported real data.

The fix

RegisterData now declares exactly the names read_io() produces. Repo-wide grep confirms the placeholder names had zero consumers outside the model itself, so this is a pure correction — no caller that worked before changes behavior.

Regression coverage (tests/unit/test_backend_models.py, new)

• Drift guard: model fields pinned equal to REGISTER_NAMES parsed from plc_connection.py source (parsed, not imported, so the pure-model test carries no pymodbus dependency) — the two can never silently diverge again
• Round-trip test proving named values survive the model
• Guard against reintroducing register_* placeholders

154 unit tests pass (151 pre-existing + 3 new), run locally on the PLC laptop.

Slice coordination

Claimed on MIRA #3048: PLC laptop (LAPTOP-0KA3C70H), session f985f0fe. Next in this slice: rebase/test the canonical-tag mapping from #188, then generate factorylm.machine-snapshot.v1 fixtures. BRAVO holds MIRA-side PR 1.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Dgej4vU7mMnYTmU6QXsH3n

[Mikecranesync]

Repair pass complete for #161 docs drift (commit 59abf9a):

• Fixed [P2] update plc-modbus CLAUDE.md register table after #161 #204: services/plc-modbus/CLAUDE.md now lists registers 101-105 as named VFD telemetry (ConveyorHz / MotorCurrentX10 / MotorTempX10 / VFDStatus / ErrorCode) instead of unused placeholders; the session-history line claiming only register 100 is used is corrected, and a 2026-08-02 entry records the fix(plc-modbus): RegisterData model strips named VFD register fields → diagnosis reads zeros #161 repair.
• Added the optional fix(plc-modbus): RegisterData model strips named VFD register fields → diagnosis reads zeros #161 acceptance test: test_io_response_serializes_named_vfd_fields proves the serialized IOResponse (the /api/plc/io boundary) keeps every named VFD register value nonzero, constructed from the exact registers dict shape read_io()/mock mode produce — no plc_connection/pymodbus import.

Tests (local, since only brain-ingest runs on PRs until #205):

PYTHONPATH=src pytest tests/unit/test_backend_models.py -q
4 passed in 0.36s

PYTHONPATH=src pytest tests/unit -q
155 passed in 0.46s   (baseline on this branch: 154)

No behavior beyond the #161 model correction.