a320-bench serve + loader value/domain validation (#19) - #82
Merged
Conversation
…tes values and set_controls domain serve productizes the subscription smoke: privileged setup + injection + validity gate in-process, then the benchmark tool surface on stdio for a client that owns its own agent loop (claude -p), with the harness verdict written to --result on shutdown. Demo/dev path, stated as such: the trajectory is the client's transcript and its system prompt is a confound for baselines, which stay on a320-bench run. Loader hardening (reviewer follow-ups from #75/#79): control values are validated by the core's own set() on a throwaway catalog Sim (the valid_values strings are for humans; BadValueError is the machine truth), and set_controls now rejects domain=world controls symmetrically to world_controls rejecting cockpit ones. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014JCrwRbtN7UA5ijn13vPmm
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 on the #19 epic after the real-model smokes.
What
a320-bench serve --scenario X --result out.json(bench/a320_bench/serve.py): productizes the path used for the subscription smoke — privileged setup + injection + validity gate in-process, then the benchmark tool surface over stdio for an external MCP client that owns its own agent loop (above allclaude -pon a Claude subscription). The harness's success evaluation is written to--resultat shutdown. Documented honestly as the demo/dev path: the trajectory is the client's transcript, and the client's system prompt is a confound for model-vs-model baselines, which stay ona320-bench run.set()on a throwaway catalog Sim — thevalid_valuescatalog strings are human-facing,BadValueErroris the machine truth (D-017 spirit);apu_gen: 7now fails at load time. (2)set_controlsrejectsdomain=worldcontrols, symmetric toworld_controlsrejecting cockpit ones.Tests
56 passed (
pytest bench/tests mcp/tests -q): new end-to-end serve test over the real stdio protocol (spawnsa320-bench serveas a subprocess, sees the benchmark surface + the failed aircraft, resolves it, and asserts the shutdown verdict JSON), parser coverage, and the two hardening cases.🤖 Generated with Claude Code
https://claude.ai/code/session_014JCrwRbtN7UA5ijn13vPmm