8.28 drum kit, what is actually verified

An audit of the claims on the live drumkit page, measured from Jacks-iMac on 2026-09-06 and recounted against Google Drive on 2026-09-07. Every number in the measured column was produced by reading the wav files, not copied from a prior report.

Claimed against measured, take 13

The live page claims a peak for each of 66 lanes. This is the one take whose full six-piece kit sits on a disk this Mac can read, at /Volumes/T7 Shield/New Folder With Items 3/New Folder With Items/song13/drumkit/.

Zero clipped samples across all six. Every file full length against its header. Every peak above the −40 dBFS presence floor. Gate A on the take's drum stem reads 3.03 onsets per second at −26.5 dBFS, which is a drummer playing.

Every take, and where it actually lives

Google Drive, actual contents

Read with rclone ls gdrive:BandPractice/8-28-26-drumkit on 2026-09-07. The folder holds 132 objects, 13.776 GiB across ten take folders: song11, 13, 14, 15, 17, 18, 19, 20, 21 and 22.

Two gaps survive the completed upload

song10 is absent. Ten take folders landed and song10 is not among them, so the Drive holds 60 kit lanes across 10 takes rather than the 66 across 11 that the drumkit page and the handoff filename both claim. The other six lanes are on the MacBook Pro and have never been verified from here.

The scratch exclude was only half applied. Every .asd and .DS_Store is gone, which the earlier pass flagged and this one confirms at zero. The _blocks/*.npy exclude did not take: 39 objects, 4.35 GiB, spread across song11 (9), song17 (6), song19 (10), song20 (4), song21 (5) and song22 (5). Real wav payload is 8.69 GiB of the 13.776 GiB, so roughly a third of the folder is still intermediate arrays.

Per the never-delete rule those move to gdrive:BandPractice/delete/ rather than being purged. Queued as q-2026-09-07-1c7faf.

Defects found, and what was done

The first four are visible on the page itself and needed no other machine to confirm. All four are corrected in this pass. The fifth cannot be closed from here.

1. Prose still described two takes fixed

   The body read “Two takes from the 8.28 tape”, “The two songs”, “On both takes” and “Both tapes record to a single mono channel”, while the stat strip and the take list presented 11 takes across 4 tapes. Nine takes were injected into the list and the surrounding prose was never rewritten.

2. The file paths named no machine fixed

   Where the files are pointed at ~/Projects/_outputs/band-practice-stems/8-28/ and the footer at ~/Music/Band-Practice/songs-8-28-LOSSLESS/. Both resolve on the MacBook Pro and neither resolves on Jacks-iMac. A bare path in a cross-Mac project reads as the machine you are sitting at, which is how a session concludes the files are gone when they are one machine away. Both are now labelled, with the iMac-reachable location named.

3. The tilt paragraph generalised from two takes fixed

   It gave the low-band share for song 13 and song 11 and called them “both source files”. The per-take cards carry that figure for all 11, so the paragraph was stale and its numbers were sound. It now gives the range across all 11 and points at the cards.

4. The handoff peak table dropped a row fixed

   Its take list names 11 and its table has 10 rows. song10 is listed as delivered and is absent from the table. The live page does carry song10, so the numbers exist. The handoff now carries a correction block recording this.

5. The Drive folder was reported as an upload in progress closed 09-07

   At 16:24 on 2026-09-06 no rclone process was running, no upload_drive.sh existed under ~/Projects on this Mac, and the folder had held one take since 15:53. The upload then ran from the MacBook Pro that evening, 15:53 to 19:36, and landed ten take folders. Queue item q-2026-09-06-e4e98c is closed with that evidence.

6. song10 never reached Drive open

   Ten folders landed and song10 is not one of them, so Drive holds 60 kit lanes across 10 takes against the 66 across 11 that the drumkit page and the handoff filename assert. Its six lanes have never been readable from this Mac. Queued as q-2026-09-07-e0e494, with the count reconciliation as q-2026-09-07-71f680.

7. The scratch exclude was half applied open

   Every .asd and .DS_Store is gone, confirmed at zero. The _blocks/** exclude did not take, leaving 39 .npy objects and 4.35 GiB across six takes. Queued as q-2026-09-07-1c7faf.

8. song14 was called an empty drum chair, and its lanes disagree resolved

   Gate A on the whole take reads 1.62 onsets/s at −40.8 dBFS, under both the 2.0 onset arm and the −40.0 chair arm, which is what produced the EMPTY_DRUM_CHAIR verdict. Pulling the six lanes back down from Drive and measuring them individually tells a different story: kick peaks at −7.04 dBFS, louder than song13's −7.75, and carries about 108 onsets across 2:43. Every lane clears the −40 floor. Somebody is working the kick pedal with the rest of the kit quiet. The lanes stay. Gate A measures whole-mix onset density, which reads a sparse take as an absent one. q-2026-09-07-d06d52 closed on that measurement.

What is left

Everything else on this page is closed. The five page defects are corrected, the take 13 kit is re-measured against its claimed table, and the upload that this page originally caught as stalled has since completed and been recounted.