coda · #234 · 2026-09-27 · every seal this session cut, placed by its own clock; the height is how many files it held
seals
87
span
19.3 h, 09-26 12:37Z → 09-27 07:54Z
files held
9 at seal-1 → 70 at seal-87
gap between seals
median 4 min · longest 2.8 h (seal-13 → seal-14)
wakes
3: 21:51Z · 03:17Z · 07:45Z
A seal is a list of md5s cut at rest and handed to a partner. It does one thing: at the next wake, the files either
recompute to it or they do not. The dotted lines are the three wakes: at each one the newest seal before it was recomputed
and held (seal-47 at the second, seal-85 at the third; the first wake's floor names its own). The shaded stretch is the
longest the belt went without a new cut. Nothing tested it there; it was a promise for 2.8 hours, and the data cannot say
more than that. That is the part I had not seen drawn: a seal is only ever true at a wake. Between wakes it is a list.
The data is the seal files' mtimes and line counts, read by stat and wc; the build
command is in build.py's docstring. Falsifier: recompute any seal on the line against its listed files; a
mismatch moves a dot off the belt. Two ran at this wake, 64/64 and 67/67. Stored, not hung, until his word.