PLN named this himself while testing the remap: "i see the hud shows d9 on A8 so
didnt we migrate parvagues HUD to new convention? why 2 sources of truth tbh"

It was five. The CC -> physical-control -> owning-orbit mapping was hardcoded
independently in:

    tools/surface-columns.py     GRID
    tools/migrate-columns.py     KNOB_A/B/C, BUTTONS, ARDOUR, label(), and the
                                 destination ARITHMETIC (28 + orbit, 48 + orbit)
    tools/lcxl-leds.py           12 + (orbit - 8) for the d9-d12 level knobs
    tools/pvlint/rules.py        PV008's BT_CCS / BL_CCS / FAMILY_CCS
    <hud>/lib/render.js          ORBIT_CONVENTION

After the 2026-07-29 remap the Python copies moved and the HUD's did not, so the
topbar drew d9 on A8 while the LED board and the .tidal files both said A1. The
display contradicted the hardware under his hands, mid-test, and NOTHING FAILED —
which is the property that guaranteed it would happen again on the next remap.

NOW
tools/lcxl_grid.py is the only place the grid is written: the row/column table,
the role of every cell (level / fx / fx2 / gate / gate2 / family_filter /
family_mute), the Ardour-owned set, and gPanic. Everything else derives from it,
and the two non-Python consumers read a GENERATED artifact:

    tools/lcxl_grid.json                        for anything outside Python
    <hud>/lib/lcxl-grid.generated.js            imported by render.js

Same pattern as the fleet colour language (models.py -> gen_tokens -> tokens.css):
author the ontology once in Python, generate for every other language.

Two details worth keeping:
  * ROW ALIASES. The consumers had each invented their own names — D/fader,
    E/btn1/BT, F/btn2/BL. Forcing one vocabulary would have churned five files
    and PLN's own muscle memory for zero benefit, so every row carries all its
    names and each tool keeps printing what it always printed.
  * ROWS E AND F ARE ONE ROW OF EIGHT, not two of four. They are non-contiguous
    on the hardware (41-44 then 57-60) and modelling that as two rows is exactly
    what put d6's second button in column 5 in the old map. Asserted directly.

THE TEST IS THE DELIVERABLE
14 new tests. Half assert the authored table is coherent (48 controls, every row
covers columns 1-8, every orbit has a level/fx/gate, no Tidal slot lands on an
Ardour-learned control, d1-d3 own exactly one knob and one button because C1-3
and F1-3 are the family controls). The other half assert every CONSUMER still
agrees, and that regenerating the artifacts is a no-op — so a remap that forgets
one copy fails the suite instead of shipping a lying topbar. One test simply
checks render.js has not re-grown a literal ORBIT_CONVENTION.

VALIDATION — behaviour must be bit-identical, this is a refactor
  surface-columns    83/83 aligned, 0 renumbers   (unchanged)
  migrate-columns    --plan: 0 moves, 0 overflow  (nothing left to do)
  pvlint             13 tracks, 0 errors, 9 pre-existing warnings
  pytest             490 passed (was 476 + 14 new)
  HUD specs          smoke / lcxl-leds / scene-directive all pass
  lcxl_grid --check  48 controls, every row 1-8, every orbit housed

Also fixed while here: the LED watcher was still running the process started at
16:39, i.e. code from before the A1-lights-for-d9 feature existed. That is why
PLN saw no A1 LED while d9 was declared — not a mapping bug, a stale daemon.
Restarted. Worth remembering as its own class: for gear that runs as a service,
"I fixed the code" is not "the rig picked it up".