Three signals on, all at once.

The reference set, all twelve.

The small reference set I hand-picked for calibration: six community-tagged Type IIs spanning three decades, two unremarkable 2010s versions, three 2024 baselines, and one monster from Phish's 2003 return tour after the 2000-02 hiatus. Twelve performances is not a survey of Bathtub Gin. There are hundreds. It is a working sample chosen to span the kinds of behavior I want the detectors to discriminate. What I use it for: confirm that the three jamcharts-tagged Type IIs (UIC '98, Virginia Beach '98, Bethel '24) all fire A + B + C, and confirm that the unremarkable versions and baselines do not. Both checks pass. Worth flagging that no single subtype is what makes UIC '98 the calibration anchor: Worcester '95 actually has a slightly higher drone ratio (4.05 vs UIC's 3.98), and Bethel '24 edges UIC on key departure (0.43 vs 0.34). UIC is canonical because it fires the strongest conjunction of all three together, and the conjunction is the definition of textbook Type II by the consensus. The detectors agree on UIC because the thresholds (0.10 for A, 0.18 for B, 1.5× for C) were calibrated against eight canonical Phish Type II performances I had already accepted as the reference. That is construct-validity, not a discovery. The same thresholds also don't fire on the unremarkable and baseline performances I tested, which is a sanity check that calibration didn't over-fit catastrophically, but it is not predictive validity. One more disclosure: the twelve performances span 1995 to 2024 across audience tapes, soundboards, and the 2024 Sphere production, all unified under the relisten provider but otherwise heterogeneous in recording style. The chroma-based metrics (A and B) are mostly robust to that, since the chromagram is a relative pitch-class profile. The drone ratio (C) is more exposed. The interesting work happens later, when I point the same chain at tape nobody has tagged.

^† A note on the Bethel call. Bethel '24's whole-jam chord-vocabulary divergence averages to 0.11, below the 0.18 threshold the table is checking. I score it as Subtype B anyway because the score for individual thirty-second windows crosses 0.18 during the F-major passage in the back half. The whole-jam average smears that excursion out; the window-by-window view catches it. That is a measurement-method choice, not a free pass. It is the same choice that promoted Bethel from “A + C partial” on an earlier multi-metric scorecard to “A + B + C” here. A different reasonable threshold would leave it as A + C. I flag it because that kind of borderline call is exactly what the dispatch should surface, not bury.

Take the calibrated detector, aim it at unseen tape.

A reference example is only useful if the detector tuned against it generalises. So once I had anchored A + B + C against UIC '98 (and cross-checked against Virginia Beach '98 and Bethel '24), I ran the same chain over recordings the community has not flagged. Two early candidates came back as triple-fires from outside the Bathtub Gin set. These are working candidates, not pronouncements. Both still need close listening review before I would call either a confirmed Type II. The reason to show them here is methodological: this is what the validation step looks like, with all the rough edges visible. Denominators matter, so: the Great Woods '24 Tweezer surfaced from a small set of seven baseline Tweezers where three of seven triple-fired, which is high enough that it could be the chain miscalibrating against 2024-era recordings rather than three real discoveries. The Goose IWD4U surfaced from a forty-eight-performance Goose 2025-2026 application where one performance fired all three, which is rare enough to be its own signal. Different evidence weight on the two candidates.

This is the first dispatch.

A calibration note, not a textbook. Each subtype has its own anatomy. Each candidate match has its own minute-by-minute story. The reason to write in public, slowly, against a small bench of recordings, is to keep the reasoning visible while it is still being figured out. A short list of what is next on the bench:

• Anatomy of a key departure: what to listen for when the band leaves home, with the metric attached.
• The 2024-07-21 Great Woods Tweezer, in detail. A candidate still pending close listening review.
• Tweezer through the eras (Tweezer is another long-form Phish jam vehicle, in the same family as Bathtub Gin). Thirty years of one song, on the bench.

I post when something is worth filing, not on a schedule. If you want to hear about new dispatches, send a note: zabriskieapp@gmail.com.