Operability, vocabulary teach-back, language pin, reframe, and dry-run fixes

[daveh-beep]

What setup experience does this improve?

One bundle, five independently reviewable commits. Gaps addressed: (1) #12 landed OPERABILITY.md but nothing routed to it; (2) setup Q4 surfaced everyday synonyms with nowhere to put them; (3) nothing let another repository consume a chalkline system; (4) the README undersold the problem scope; (5) the first end-to-end dry run surfaced seven protocol ambiguities.

Commit 1 — Wire the operability loop into the docs agents and users actually read

• README — describe the current loop (consult, surface gaps or conflicts, draft or stop, review and decide, revise, recalibrate) and mention the reviewer-facing usage report.
• AGENTS.md — new Usage reports section pointing at OPERABILITY.md: optional, kept separate from requested copy, never proof of compliance.
• setup/PROTOCOL.md — Phase 5 growth paths include the operability handoff.

Commit 2 — Vocabulary and teach-back

Slimmed re-cut of the red-teamed teaching-loop experiment:

• New optional reference: references/vocabulary.md — everyday phrases → canonical terms; governs conversation, not output.
• AGENTS.md teach-back contract — fulfill first; one teach-back per reply max; conversational framing only; urgency/opt-out/teaching: "off" gates; never a mistake record.
• Calibration pairs gain a one-line "What this teaches" note; Meridian example gains a completed vocabulary.md.

Kept from red-team hardening: countable cap, deliverable fence, urgency gate. Dropped: the in-repo conformance harness — template stays runtime-free.

Commit 3 — Language pin: let other repos declare this writing system

A consuming repository adds a short block to its own AGENTS.md naming the language repo and a commit. Agents there consult the pinned revision (unreadable = reported gap, not a guess), cite it in usage reports, and surface multi-repo conflicts to a human. Bumping the commit is how approved language changes reach consuming repos. Wired in README, AGENTS.md, OPERABILITY.md, and setup Phase 5.

Commit 4 — Reframe the problem (two sentences)

Tagline scopes chalkline to the part of a team's language it can actually record, inside the larger truth that teams brief agents, reach customers, and define products in words. Problem statement names prompt-restatement drift as the failure mode.

Commit 5 — Protocol fixes from the first end-to-end dry run

Scripted setup run (fresh fictional brand so the shipped Meridian example couldn't be cribbed) plus a consumer-phase test through a language pin. Full field report on #2. Fixes:

• Q2 probes skip what the answer already settled
• Q6: channels.md requires channel-specific rules, not named surfaces
• Q7: record where canonical exact-wording copy lives (stable link/path, or owning source when paper-only)
• Phase 2: edited rules count as approved-as-edited; product-fact candidates routed to product docs, never offered as writing rules; absence-inferred patterns are weak candidates
• Phase 3: vocabulary disambiguation note when one everyday phrase maps to two canonical terms
• Phase 4: ambiguous supplied facts ("06/02") — ask or keep original form; reformatting is language, reinterpreting is a fact decision
• Phase 5: diff review pairs the full diff with a one-line-per-file summary
• AGENTS.md: frontmatter contract now states the file-default + inline (must) enforcement model the examples already practiced

Verified in the dry run: fact/language boundary held both directions (setup refused product facts; consumer flagged missing closing time instead of inventing it); exactly one teach-back fired, conversationally, never in copy; pin consumer verified the revision and cited it in its usage report.

Docs only across all five commits; no runtime, schema, or new tooling. Replaces #13.

Safety and verification

• I used only fictional, redacted, or explicitly publishable material.
• I did not include secrets, personal or customer data, confidential material, or private links.
• Examples preserve supplied facts and do not invent product state, causes, dates, deadlines, guarantees, names, amounts, eligibility, or actions.
• I ran or documented the relevant setup or harness check (end-to-end dry run: setup, generation, calibration pair, pin consumption, teach-back, usage reports — report on Dry-run: Meridian setup end-to-end #2).
• My commits include a DCO sign-off (git commit -s).

[daveh-beep]

🤖 Added commit 4: two-sentence README reframe — the tagline scopes chalkline to 'the part of your team's language you can agree on' (voice, terminology, vocabulary, channel rules), and the problem statement names prompt-restatement drift as the failure mode. No new claims.