Skip to content

Release v0.47.2: stop shipping internal dirs; record the cypher pilots - #254

Merged
NewGraphEnvironment merged 1 commit into
mainfrom
246-pilot-record-v0472
Aug 31, 2026
Merged

Release v0.47.2: stop shipping internal dirs; record the cypher pilots#254
NewGraphEnvironment merged 1 commit into
mainfrom
246-pilot-record-v0472

Conversation

@NewGraphEnvironment

Copy link
Copy Markdown
Owner

Bundles the two already-merged changes into one release, plus the evidence behind them.

Why one release, not two

#235 stops comms/ and research/ shipping — verified against the built tarball, 15/22 → 0/0. That is the substantive user-visible change: 37 files leave the installed package.

It also settles the question #253 raised. That PR's only shipped file was a regenerated research/study_areas.md; with #235 in, research/ no longer ships at all, so #253 genuinely shipped nothing and did not warrant its own tag.

#253 packs distributed work by finish time rather than segment count — the dispatcher is 2.23× faster than a cypher, so balancing segments left it idle 96 minutes. Provincial modelling 191 → 150 min; end-to-end 5.0 h → 4.3 h.

The evidence, made findable

First instance of soul#129:

  • research/run_record_2026_08_31_cypher_pilots.md — question, method, measurement, conclusion, dead ends
  • data-raw/logs/README.md — what each filename pattern is, and that the logs are retained as evidence rather than accumulated by accident
  • the 11 pilot-3 logs, including stamps.tsv (parity across both hosts) and the compare CSV

The record keeps the wrong turns on purpose. The root cause was diagnosed correctly, retracted on a bad inference, then confirmed by measurement — presence taken as provenance, one level above the bug itself. And two of the four defects were in gates written to catch that class, both passing tests that structurally could not fail.

Verification

  • tarball: comms/ 0, research/ 0, run record not shipped
  • FAIL 0 | PASS 1621
  • every number in the record traces to a committed log

Relates to #246, soul#129

🤖 Generated with Claude Code

https://claude.ai/code/session_014kSEJoad4Bc553qU8eFiE6

Bundles #235 (comms/ and research/ out of the built package) and #253
(finish-time packing), plus the evidence behind the latter.

The run record and logs README are the first instance of soul#129 — a
convention proposing that measurement runs leave something findable, not just
a timestamped log. Today produced four paid spins, four defects and the
0.0391 vs 0.0872 min/1k-segment host ratio that rewrote the packer, and none
of that chain was findable without reading a chat transcript.

The record deliberately keeps the dead ends, because they are the evidence:
the root cause was diagnosed correctly, retracted on a bad inference
(presence taken as provenance, one level above the bug itself), then
confirmed by measurement; and two of the four defects were in gates written
to catch that class, both passing tests that structurally could not fail.

Note the two land together on purpose. #235 means research/ no longer ships,
which is what makes #253's regenerated study_areas.md a non-shipping change
and settles the question of whether it warranted its own release.

Relates to #246, soul#129

Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_014kSEJoad4Bc553qU8eFiE6
@NewGraphEnvironment
NewGraphEnvironment merged commit 19ab9c1 into main Aug 31, 2026
1 check passed
@NewGraphEnvironment
NewGraphEnvironment deleted the 246-pilot-record-v0472 branch August 31, 2026 21:04
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant