Add driftline to Observability - #811
Open
catidegla wants to merge 1 commit into
Open
Conversation
Signed-off-by: catidegla <4922726+catidegla@users.noreply.github.com>
catidegla
force-pushed
the
add-driftline
branch
from
September 12, 2026 11:34
751551a to
ebd9a42
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Adds driftline under Observability, alphabetically between Deepchecks and Evidently, with a star badge.
It answers one question the rest of the section does not: did the provider change the model, or did I change something? You pin the model name, set temperature to 0, and the outputs still move one Tuesday, and there is currently no good way to tell those two apart.
Three kinds of answer, kept separate. Providers already report what actually served a request, so when
system_fingerprintor the resolved snapshot changes there is nothing to infer, and that can fire on a probe whose outputs were byte for byte identical. When nobody reports one, it samples committed probes and tests today's answer distribution against the baseline with a G-test, with an effect floor and a Benjamini-Hochberg correction across the run so thirty probes do not produce a false alarm a day. And a run it could not read exits 3 rather than reporting drift, because "the model changed" and "we could not find out" call for different responses.No prompt text or model output is ever written to the history file, only counts and hashes. Zero dependencies, MIT, 117 tests that never call a paid API. I am the author.