feat: raise the turn progress cadence 60% #920

Merged

coilysiren merged 1 commit from issue-901-raise-progress-cadence into main

2026-08-17 19:11:16 +00:00

coilyco-ops commented

2026-08-17 18:39:52 +00:00

Member

Both numbers up 60%, inside the 50 to 66% asked for, every value a whole number of seconds.

progress delay      5s ->  8s
beat               10s -> 16s
long-reply window  25s -> 40s

One edit

SIRENS_ECHO_PROGRESS_AFTER goes 5s to 8s. The beat is turnProgressAfter * 2 and the long-reply window is turnProgressAfter + turnProgressEvery*2, so both follow without a second decision. The derivation the issue asks to keep was already there, which is why the two requested numbers move by the same 60% rather than drifting apart.

agent/rendered/knobs.txt regenerated by just knobs.

The drift the issue is about, one layer up

The expected values were pinned as literals in three separate test files: cadencederivation_test.go, turnprogress_test.go, and thread_test.go. Raising the base failed all three, each with its own hardcoded number, which is exactly the drift the derivation exists to prevent.

They now read one var block in cadencederivation_test.go, so the next raise edits one file. TestTheCadenceIsDerivedAndNotThreeCoincidences still asserts the relationship independently, so a derivation that quietly stopped deriving still fails.

just gate PASS.

closes #901

Both numbers up 60%, inside the 50 to 66% asked for, every value a whole number of seconds. ``` progress delay 5s -> 8s beat 10s -> 16s long-reply window 25s -> 40s ``` ## One edit `SIRENS_ECHO_PROGRESS_AFTER` goes 5s to 8s. The beat is `turnProgressAfter * 2` and the long-reply window is `turnProgressAfter + turnProgressEvery*2`, so both follow without a second decision. **The derivation the issue asks to keep was already there**, which is why the two requested numbers move by the same 60% rather than drifting apart. `agent/rendered/knobs.txt` regenerated by `just knobs`. ## The drift the issue is about, one layer up The expected values were pinned as literals in three separate test files: `cadencederivation_test.go`, `turnprogress_test.go`, and `thread_test.go`. Raising the base failed all three, each with its own hardcoded number, which is exactly the drift the derivation exists to prevent. They now read one `var` block in `cadencederivation_test.go`, so the next raise edits one file. `TestTheCadenceIsDerivedAndNotThreeCoincidences` still asserts the relationship independently, so a derivation that quietly stopped deriving still fails. `just gate` PASS. closes #901

coilyco-ops added 1 commit

2026-08-17 18:39:53 +00:00

feat: raise the turn progress cadence 60%

ci / image-build (pull_request) Successful in 33s

Details

ci / test (pull_request) Successful in 58s

Details

ci / publish-echo-image (pull_request) Has been skipped

Details

ci / publish-observed (pull_request) Has been skipped

Details

0120a7f15d

The progress delay and the long-reply thread window both go up by 60%, inside
the 50 to 66% the issue asked for, and every value stays a whole number of
seconds.

One edit does it. SIRENS_ECHO_PROGRESS_AFTER goes 5s to 8s, and the beat and
the long-reply window are already expressions of it, so they follow to 16s and
40s without a second decision. That is the derivation the issue asked to keep,
and it was already there.

  progress delay      5s ->  8s
  beat               10s -> 16s
  long-reply window  25s -> 40s

The expected values were pinned as literals in three test files, which is the
drift the derivation exists to prevent, one layer up. They now come from one
var block in cadencederivation_test.go, so the next raise edits one file.

closes #901

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Kai Siren <coilysiren@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>

coilyco-ops requested review from coilysiren

2026-08-17 18:39:53 +00:00

coilyco-ops referenced this pull request

2026-08-17 18:40:07 +00:00

Increase progress indicator delay and long-reply thread threshold ~50-66% #901