Adversarial review: I cited a source I had not read - #281
Merged
Merged
Conversation
The comment on the fragility threshold said Reactome's user guide makes the point about hits resting on few entities. It may well. I did not check before writing it, and I could not check afterwards -- the user guide here is a Chroma database, not text. So the citation is withdrawn rather than left standing, and what is actually true is stated instead: five is chosen, taken from this feature's own spec using 2 of 3 as its example of what is not evidence, and setting it properly is a curator's judgement. An invented citation is worse than none. It survives review precisely because it looks like the thing that ends an argument. The second attack found the design sound, which is worth recording too. A flag that fires on everything is as useless as one that never fires, and this one flagged 12 of 12 pathways on the four-identifier analysis I had been testing with. Against a hundred-gene analysis on beta it flags 0 of 12, with found counts of 13 to 67. So it discriminates: it fires where a hit really does rest on nothing and stays quiet on ordinary inputs. Both ends are now pinned with those real shapes rather than invented ones. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
adamjohnwright
force-pushed
the
011-phase5-review
branch
from
September 20, 2026 17:55
1648f55 to
0841501
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Follow-up review of #280. Builds on it, so merge that first.
An invented citation
The comment on the fragility threshold said "Reactome's own user guide makes this point" about hits resting on few entities. It may well. I did not check before writing it, and could not check afterwards — the user guide here is a Chroma database, not text.
Withdrawn rather than left standing. What is actually true is now stated: five is chosen, taken from this feature's own spec using 2 of 3 as its example of what is not evidence, and setting it properly is a curator's judgement, not a programmer's.
An invented citation is worse than none, because it survives review precisely by looking like the thing that ends an argument.
The second attack found the design sound
A flag that fires on everything is as useless as one that never fires — and this one flagged 12 of 12 pathways on the four-identifier analysis I had been testing with, which is exactly what a useless flag looks like.
Against a hundred-gene analysis on beta it flags 0 of 12, with found counts of 13 to 67. So it discriminates: it fires where a hit really does rest on nothing, and stays quiet on ordinary inputs. Both ends are pinned by a test using those real shapes rather than invented ones.
All checks gated with
set -ethis time.🤖 Generated with Claude Code