Extend capability coverage past effects: 49% of the gap is not a gap, 26% needs no new vocabulary

[jeffreylouden]

Follow-up to #653, which introduced capabilities and populated the Effects
category group. Coverage sits at 768 of 6,184 hardware entries (12.4%), and
schema/capabilities.yaml says outright that synthesis, recording and playback
operations are missing.

Before adding any vocabulary I bucketed the 5,416 unassessed entries by
primary_category, because "12.4%" turns out to be the wrong number to be
working from.

The gap is three different things

bucket	entries	share of gap
Not applicable — instruments, transducers, controllers	2,631	48.6%
Already expressible in the current vocabulary	1,390	25.7%
Needs new vocabulary — synthesis and sequencing	962	17.8%
Unclassified in this pass	433	8.0%

1. Half the gap should stay empty

An electric guitar (472), a speaker (393), a microphone (235), a pair of
headphones (218), a snare (175), a MIDI controller (161) perform no audio
processing operation. They produce, transduce or control sound. capabilities
answers "what operation does this perform on audio", so the honest answer for
these is no list at all, which is what they already have.

The metric is what needs fixing here, not the data. pnpm capability-coverage
reports against all hardware, so it conflates "not assessed" with "not
applicable" and will never reach a satisfying number no matter how much work is
done. Measured against the ~3,553 entries the axis can actually describe,
current coverage is 21.6%, not 12.4%.

Suggestion: declare non-applicable category groups once, and have the coverage
report show assessed / applicable with the excluded set named. That is also the
number Studio's redundancy signal should be reasoning about, since it never asks
whether two guitars overlap functionally.

2. A quarter of the gap needs no new values at all

Mixers (346), audio interfaces (195), preamps (144), pedals (126), guitar
amplifiers (131), outboard (58) and rack units (61) perform operations the
vocabulary already has: equalization, compression, gating, limiting,
overdrive, reverb, tremolo, amp-modeling, cabinet-simulation,
crossover.

They are unassessed only because the #653 pass was scoped to the Effects
category group. pedal is the clearest case: 126 entries whose
primary_category is the form factor rather than the function, so the derivation
never looked at them.

This is the cheapest work in the whole programme and should go first. Note it is
not purely a derive-capabilities run: a mixer does not have compression
by definition, so anything beyond the definitional mappings has to be read out of
each entry's own prose, the way the multi-effect corpus was hand-authored.
Where a category does imply an operation definitionally, add it to
CATEGORY_CAPABILITIES in scripts/derive-capabilities.ts so the floor comes
for free.

3. Synthesis is the real vocabulary gap

962 entries: synthesizers (529), modular (116), drum machines (91), samplers
(89), workstations (52), sequencers (36).

sampling, sequencing, granular and filter are already in. What is missing
is the synthesis side, and it wants the same discipline the effects half got:
describe the operation, never the marketing name, and one dimension only.
Candidate shape, to be argued over rather than merged as-is:

• generation: subtractive, FM, wavetable, additive, physical-modeling,
  phase-distortion, sample-playback
• shaping already partly covered by filter, plus envelope generation and LFO
  modulation as operations rather than as components
• polyphony, voice count and analog-versus-digital are not capabilities:
  those are what the product is, and belong in categories

The risk to watch is that "subtractive synthesis" is closer to a technology than
an operation, which is the exact line capabilities exists to hold. Worth
settling before writing 529 entries against it.

Suggested order

1. Fix the coverage metric so the target is the applicable set (small, and it
   changes what every subsequent number means).
2. Sweep bucket 2 with the existing vocabulary, biggest categories first.
3. Design the synthesis vocabulary, then sweep bucket 3.
4. Revisit the 433 unclassified (headphone amps, consoles, romplers, patchbays,
   summing mixers) once the first three are done.

Why it matters downstream

Studio scores G.A.S. purchase redundancy from this axis
(existential-engineering/racks#3344): a numeric coverage score replaces the crude
same-category count, and null falls back to it. Every entry assessed moves one
product from the fallback to the real signal, and buckets 2 and 3 are where
users actually feel it, since "do I already own something that does this" is
asked of synths and mixers at least as often as of pedals.