The generated whitespace battery as a standing gate, with three perturbation populations, a diff-derived deletion gate and an address-count script

[mergeconflict]

What this is

Standing instruments only; no src/ change. Each item either lands a measurement that a later change's gate cites, or retires a hand-maintained instrument that the measurement replaces.

Deliverables

1. The generated whitespace battery as a standing gate under scripts/: the cross-product generator (92 templates, 198 whitespace slots, perturbed in LENGTH, TAB and NEWLINE) renders every case through Asciidoctor Ruby 2.0.26 unconditionally and through @asciidoctor/core 4.0.11 as the cheap check, under the exact conformance lens of tests/helpers.ts (the fold is proved equal to the TypeScript one, not re-implemented loosely). Its derived table is the source of the whitespace binding rows. Runs at batched integration points alongside Measure the instrument against the reference: render the corpus through Asciidoctor Ruby 2.0.26 and @asciidoctor/core 4.0.11 and pin the disagreements #235's reference diff. The prototype and its output are in the local design record (sp2-battery-v2.md and the battery2/ generator).
2. Two more perturbation populations over the same alphabet: the tracker's issue witnesses (every reflow-hazard issue's repro), and the registry deep grid's own shapes, the one population nobody wrote by hand.
3. Templates the roster lacks: the implicit menu "File > New" under :experimental: (rx.rb:570), the bare-URL left boundary (rx.rb:526), and the three hyphen-fused-to-reference shapes (-{h}, measured by hand so far).
4. A deletion gate derived from a diff, not from a brief: a script that lists the symbols a change deletes and requires, for each, the closed issue whose fix it was, that issue's witness verbatim, the binding row that makes the witness a fixed point, the test that reddens under a mutant of that row (the fact's arm flipped, scoped to one file), the direction of the change (wider or narrower than today) and any cross-change hold.
5. The parse-to-print address count as a script over crossings-registry.json and graph.ts: four addresses today; the target after the reader-records-facts changes is three plus one shared module.
6. Measure the instrument against the reference: render the corpus through Asciidoctor Ruby 2.0.26 and @asciidoctor/core 4.0.11 and pin the disagreements #235 lands here, with the seam-window census retirement it names (720 lines out; one instrument in, one out).
7. The instrument's anchor: reftext transcription bug (its regex reads reftext across a newline where the Ruby does not) is recorded as an instrument-fidelity row in Measure the instrument against the reference: render the corpus through Asciidoctor Ruby 2.0.26 and @asciidoctor/core 4.0.11 and pin the disagreements #235's ledger. Nothing is reported to any other repository.

Gate

bun run gates green; the battery at zero mismatches between the two programs except the rows #235's ledger pins; the census retirement's 720 lines deleted in the same push. Expected 6 to 8 files.

Sequencing: nothing blocks this. The link-macro conversion (#219) waits for the battery run over its domain; the whitespace-fact change cites the battery's table but does not wait for the gate to land.

From solution-proposal-v3.md section 4.7 item 1 (local design record).