Catalogue the whole-run cases a harness eval suite must cover - #182
Merged
Conversation
Thirty eight cases in ten areas, derived from the decision log rather than invented, and ordered by the cost of the regression each would catch times how likely that regression is. Every case names what it proves, the scripted provider behaviour it needs, and the code path or decision entry it defends. Cases already proved by the in-process loop tests are listed separately so nobody writes them twice. Claude-Session: https://claude.ai/code/session_01PG2vuux5tMsjnZBLWJzMyp
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 work list for a deterministic, whole-run evaluation suite. Thirty eight cases in ten areas: provider transport and retries, streaming and re-asks, the context window and compaction, tool calls, approvals, the workspace boundary, the store and replay, the trace file, the process surface, and build gates.
Every case is derived from the decision log or the code, not invented. Each names what it proves, the scripted provider behaviour it needs, the assertion that decides it, and the decision entry or code path it defends. The order is by the cost of the regression it would catch times how likely that regression is, judged from the failures this repo actually had.
Two sections keep the later work honest: what is already proved by the in-process loop tests and must not be written again, and what this suite deliberately leaves to the model facing half.
Documentation only. No code changes.
https://claude.ai/code/session_01PG2vuux5tMsjnZBLWJzMyp