Wire consensus panel into engine, add verified execution and evaluation - #1
Open
Kendiukhov wants to merge 1 commit into
Open
Wire consensus panel into engine, add verified execution and evaluation#1Kendiukhov wants to merge 1 commit into
Kendiukhov wants to merge 1 commit into
Conversation
Make the consensus reviewer a live part of the loop and characterise the orchestrator quantitatively. Orchestration: - dsl: compile a consensus_merger node into a single consensus step that carries the full reviewer panel instead of dropping parallel fan-out edges. - engine: execute the panel concurrently, account panel tokens/cost, format merged feedback, and support an optional advancement gate that blocks convergence while a Critical critique is unresolved. Implement on_flag edges. - consensus: configurable panel and agreement threshold, deterministic tie-break, optional per-role weighting, and a consensus report. Verified execution: - code_executor: run fenced Python from an executor artifact in an isolated subprocess (timeout, output cap); opt-in via code_execution_enabled. Engine folds the execution report into feedback so reviewers see executed results. Mock adapter: - role-authoritative detection from the prompt's role variable (fixes a misdetection of the executor prompt), differentiated per-lens critique sets with controlled overlap, fixed-iteration determinism, and a runnable snippet in the executor output. Evaluation: - analysis/experiments.py plus scripts: panel-size ablation, reviewer-lens overlap, convergence profile, and concurrent-run load benchmark; figure and results generators; a live-Gemini consensus corroboration probe. Paper: - paper/plos: PLOS ONE manuscript, figures, data, and cover letter. Adds 17 tests; full suite green (358 passing). Co-Authored-By: Claude Opus 4.8 <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
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.
Summary
Makes the consensus reviewer a live part of the execution loop (previously it existed only as an unused module), adds sandboxed verified execution, and adds a quantitative evaluation of the orchestrator. The PLOS ONE manuscript is updated to match.
Orchestration
compile_loopnow compiles aconsensus_mergernode into one consensus step carrying the full reviewer panel, instead of following only the firstalwaysedge and silently dropping the other reviewers. Linear loops compile unchanged.asyncio.gather), accounts panel tokens/cost, formats merged feedback, and supports an optional advancement gate (blocks convergence while a Critical critique is unresolved).on_flag:edges are now honored from run config.compute_consensus_report.Verified execution
python -Isubprocess (fresh tempdir, wall-clock timeout, output cap). Opt-in viacode_execution_enabled; the engine folds the execution report into feedback so reviewers critique executed results.Mock adapter
rolevariable (fixes a latent misdetection of the executor prompt asidea_generator), differentiated per-lens critique sets with controlled overlap, fixed-iteration determinism, and a runnable snippet in the default executor output.Evaluation
scripts/: panel-size ablation, reviewer-lens overlap, convergence profile, concurrent-run load benchmark; figure/results generators; and a live-Gemini consensus corroboration probe.Paper
paper/plos/: PLOS ONE manuscript, figures, results data, and cover letter.Tests
test_consensus_wiring.py,test_code_executor.py,test_experiments.py(17 tests). Full suite green: 358 passing (1 pre-existing unrelatedasyncio.get_event_looptest deselected).🤖 Generated with Claude Code