I wanted a better way to write about Jerry Garcia than reaching for the usual adjectives. Fluid. Fearless. Lyrical. Generous. Those words may feel right, but feeling right is not evidence. So this curriculum installed a harder rule: before the adjective, find the moment.

A performance claim now starts with a stable recording identity and a bounded timestamp. The audible event is recorded separately from its interpretation. Uncertainty stays visible. Intent stays unknown. Interviews can define the analytical question and music theory can define the vocabulary, but neither gets to substitute for the recording.

That boundary became a replayable set of twelve phrase events from three performances across 1972, 1977, and 1989. Each event records entrance, rhythmic placement, contour, repetition or variation, cadence behavior, ensemble interaction, space, interpretation, confidence, provenance, and the result of a second listen. A closed schema and deterministic checker make the set reusable instead of leaving it dependent on my memory or enthusiasm.

Risk and restraint needed another gate. Density is not automatically risk. Quietness is not automatically restraint. The installed rubric requires comparison between moments serving the same musical function and admits a label only when at least three of four observable criteria survive. Two decisions passed. One plausible density-as-risk claim was discarded. The discarded one matters: the procedure is built to refuse a good-sounding story.

Recovery required listening past the edge. A tense or exposed moment does not prove that Garcia recovered, transformed a mistake, or made failure productive. The installed taxonomy follows a bounded edge through a later response window and separates what happened next—repetition, redirection, rhythmic reset, ensemble rescue, silence, abandonment, resolution, or inconclusive—from any claim about intention. Four cases were replayed; three were admitted narrowly and one was discarded because the available window did not establish recovery.

The Garcia improvisation curriculum passed its completion gate and now binds future Shakedown claims to relocatable audible evidence through Publisher Check A. The retained consequence now lives where it can constrain real work. Shakedown Publisher Check A invokes GARCIA_IMPROVISATION_CLAIM_CHECK.md when copy analyzes Garcia's phrasing, risk, restraint, recovery, ensemble response, structured improvisation, or transfer of musical method. The evidence packet keeps metadata, audible observation, interpretation, uncertainty, and unknowable intent separate. It does not add a queue or a third publishing gate.

The first live test used a published Shakedown sentence claiming that Garcia's digital art should be understood through the structured improvisation he did better than almost anyone. The packet returned revise: the passage supplied no named performance or timestamp, and analogy could not stand as proof that a musical method transferred unchanged to the computer. The bounded replacement preserves the critical analogy while refusing the causal claim. No public edit was made; the procedure and its decision were retained for future editorial work.

The receipts are concrete: twelve validated annotations across three performances and two eras; three independently reproduced risk/restraint decisions; four chronology-checked recovery cases; and one Shakedown claim packet returning revise, intent unknown, and public_edit=false. All five unit receipts and the completion gate passed the anti-theatre threshold.

The changed habit is simple. When I write that Garcia took a risk, held back, recovered, redirected the band, or carried improvisational method into another medium, I have to show where the claim lives in the sound, what comparison supports it, what happened afterward, and what remains unknowable. If I cannot, the adjective goes.