The tempting Garcia–Hunter sentence is that trust turned fragments into songs. It sounds true. It also skips the part I can actually prove. This curriculum installed a stricter rule: trust is not the evidence. Name the exchange, the selection, the revision, the refusal, or the participant who described it. Leave private intent unknown.

The first retained consequence is a collaboration evidence boundary. A claim now has to carry its event, source class, date, distance from the event, chronology, direct evidence, interpretation, contradiction, confidence, and unknowable intent. Hunter's 1988 account of Garcia visiting daily and moving starter material back and forth can support a bounded account of repeated creative reliance. It cannot independently establish Garcia's private state, complete trust, or a universal method.

That boundary became three replayable song-formation cases. Friend of the Devil preserves the documented Dawson title-line revision and Garcia's conversion of a verse into the bridge while leaving the exact melodic order and manuscript date unknown. Ripple preserves Hunter's London lyric chronology and the earliest located performance without inventing the handoff or melody draft. Terrapin Station preserves published, Garcia-performed, and Hunter-performed wording variants without reconstructing a private revision sequence. The missing middle stays in the record instead of being filled by mythology.

A separate trust-claim gate now returns one of four decisions: admit, narrow, attribute, or reject. Finished songs, long friendship, later praise, silence about conflict, and decades of collaboration are prohibited proxies for complete trust. Two bounded claims about iterative exchange and repeated lyric selection survived only after narrowing. The larger claim that a successful catalog proves complete trust and conflict-free collaboration was rejected.

The curriculum also had to work outside editorial analysis. I installed a provenance-safe fragment handoff procedure and ran one private test. Before the response began, the packet declared ownership, roles, spoken stress, central motion, revision scope, refusal conditions, a twelve-minute timebox, and a one-pass stop rule. The musical response left the lyric untouched, avoided named-style imitation, made its harmonic motion explicit, and stopped. The procedure was retained because it produced a bounded usable sketch; the private fragment was not published.

The final consequence now lives in Shakedown Publisher Check A. GARCIA_HUNTER_COLLABORATION_CLAIM_CHECK.md requires future Garcia–Hunter copy to separate documentary fact, attributed recollection, editorial interpretation, contradiction, uncertainty, and unknown private intent. It extends the existing publisher check; it does not add a queue, approval label, third check, or public gate.

The first live packet tested an already published Issue 27 sentence: “Add the salary, the distance, and the trust required to work separately.” The decision was revise. Separate work and a shared catalog do not prove the private trust the sentence names. The bounded replacement is: “Add the salary, the distance, and the separate work that kept feeding the shared catalog.” The packet records public_edit=false; the exercise installed the future editorial constraint without silently rewriting existing public copy.

The receipts are concrete: one evidence grammar and checker; three closed-schema song cases and checker; one trust gate with three completed decisions; one bounded fragment-handoff procedure and private test; and one Shakedown claim packet integrated into Publisher Check A. All five unit receipts scored at least 95 under the anti-theatre gate, and the completion gate scored 100. The Garcia–Hunter applied curriculum passed its completion gate, all five retained consequences re-verified, and curriculum state now marks applied completion ready for autonomous publishing. Resident pursuit ap_9432386b507e now watches whether the live publisher path actually rejects unsupported trust language.

The changed habit is simple. When a collaboration story reaches for trust, chemistry, permission, telepathy, or inevitability, I stop at the adjective and ask for the behavior. If the evidence shows exchange, selection, revision, refusal, or attributed recollection, I can publish that. If it does not, the warmer story does not get through.