Say what justifies a release, and what does not - #81
Merged
Merged
Conversation
The convention said "cut one when there is a measured change to report" and left the three cases anyone actually faces unanswered: time passing, the benchmark changing, the product changing. Three triggers, any one sufficient. The product moved — something shipped or a finding came out of a run. The instrument changed what it measures and a full run has been measured under it, so the numbers and the notes describing them agree. Or the snapshot has gone stale, which has a deadline attached: transcripts expire at ninety days, so a snapshot nobody released is evidence nobody can check later. And one explicit non-trigger, because it is the tempting one. A run whose numbers moved is not a reason. The weak model's skills delta read -1 on 1 September and +4 on 14 September with nothing changed between them; eleven cells flipped and two of them moved in both directions across the arms of the same pair. Releasing on movement publishes variance and spends the changelog on non-events. Also records that the monthly matrix is the better thing to package: a weekly covers four experiments, so its snapshot carries the frontier -no-skills arms forward and a release calling it "one run" would be wrong. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MQzUoMAwEBJWpEGVvVzSjK
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
From the question "are we due a release?" — which the convention could not answer.
It said "cut one when there is a measured change to report, not on a schedule", and left the three cases anyone actually faces open: time passing, the benchmark changing, the product changing.
Three triggers, any one sufficient
One explicit non-trigger
A run whose numbers moved. Movement is the default here. The weak model's delta read −1 on 1 September and +4 on 14 September with nothing changed between them; eleven cells flipped and two moved in both directions across the arms of the same pair (#2). Releasing on movement publishes variance and spends the changelog on non-events.
Package the monthly, not a weekly
A weekly covers four experiments, so its snapshot carries the frontier
-no-skillsarms forward from whenever they last ran. Publishable, but a release whose notes say "one run" would be wrong — which is the property v0.4.0 was cut for.🤖 Generated with Claude Code
https://claude.ai/code/session_01MQzUoMAwEBJWpEGVvVzSjK