fix: write the report prompt in plain English - #4
Merged
Conversation
The prompt told the model to write clearly while being written badly itself, and the reports picked that up. "Windows have a bulk", "the job is selection rather than coverage", "name the shape of the work". Those are aphorisms, not instructions, and the Exchangeability report came back ending a paragraph with "feeding the common endings". Most of this is rewording. Three changes are substantive. The link cap is inverted. It said "at most three links in the whole report", which capped exactly the wrong thing: the two reports so far cite 43 and 15 pull request numbers against 11 and 2 documentation links. A link tells a reader what a result is, which is what they came for; a pull request number only says where it was written. So link every named result, and cite pull requests sparingly. The proportion sentence is gone. It asked for a count of how much of the window was routine, and a count tells a reader nothing. The Exchangeability report had already found the better version by itself, in thirteen words: "much of the remaining window is product-kernel and cylinder API feeding the common endings". Say what the rest of the work was about, not how much of it there was. New: write names the way you would say them. `deFinetti_RyllNardzewski_equivalence` is a Lean identifier, and belongs in the link target rather than the sentence. Also drops a paragraph explaining that an earlier version of this prompt produced 932 words and a reader complained. That is this file's history, not guidance, and it does not help anyone write. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
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.
The prompt told the model to write clearly while being written badly itself,
and the reports picked that up. "Windows have a bulk", "the job is selection
rather than coverage", "name the shape of the work". Those are aphorisms, not
instructions, and the Exchangeability report came back ending a paragraph with
"feeding the common endings".
Most of this is rewording. Three changes are substantive.
The link cap is inverted. It said "at most three links in the whole report",
which capped exactly the wrong thing: the two reports so far cite 43 and 15
pull request numbers against 11 and 2 documentation links. A link tells a
reader what a result is, which is what they came for; a pull request number
only says where it was written. So link every named result, and cite pull
requests sparingly.
The proportion sentence is gone. It asked for a count of how much of the window
was routine, and a count tells a reader nothing. The Exchangeability report had
already found the better version by itself, in thirteen words: "much of the
remaining window is product-kernel and cylinder API feeding the common
endings". Say what the rest of the work was about, not how much of it there was.
New: write names the way you would say them.
deFinetti_RyllNardzewski_equivalenceis a Lean identifier, and belongs in the link target rather than the sentence.
Also drops a paragraph explaining that an earlier version of this prompt
produced 932 words and a reader complained. That is this file's history, not
guidance, and it does not help anyone write.
Co-Authored-By: Claude Opus 5 (1M context) noreply@anthropic.com
🤖 Prepared with Claude Code
After review, the same motivation is now applied to both STATUS-generation paths: write theorem names as spoken English, link every selected result or definition when the supplied material has an exact documentation URL, and use at most two pull-request citations in the whole status snapshot. The branch has also been merged forward to include #3.