Integrate C1 and evidence-driven benchmark candidate selection by huangruiteng · Pull Request #8255 · loopx-project/loopx

huangruiteng · 2026-11-07T06:42:31Z

Long-running benchmark workers should spend fewer interactions recovering control context and use observed candidate outcomes when choosing their next experiment. This research integration combines C1 with candidate-selection guidance and optional process-group observations, including the real native benchmark entry paths.

This PR targets the preceding research integration branch. It does not promote this treatment to the default product or claim a measured task-score gain.

Changes

Integrate the reviewed C1 head from C1: unify work sources and validate recovery #7430 (8e0a4799ce05935fd555e80b30df78f0973f9d1c) and current main. Preserve Turn, capture, recovery and settlement identity while consuming the shared work contract.
Reuse the committed quota capture in native benchmark preflight #8251 reuses the committed original capture for native benchmark admission instead of copying and retransmitting the complete decision. The context-off route remains unchanged.
feat(explore): ground work choices in paired goal outcomes #8250 compares distinct, finite candidate measurements only within matching work and measurement scope. A matching proxy/native ordering reversal and bounded comparison alternatives are advisory; the authored next-step reason remains intact.
Observe aggregate memory in local work probes #8249 adds candidate-bound Linux process-group observations through the existing local execution provider. PSS is a non-atomic sampled estimate, not certified concurrent memory or container headroom. Sidecar failures preserve the command result.
Opt in to local work probes from SForge heartbeat #8252 explicitly exposes that producer to SForge heartbeat runs and verifies the exact staged entry point as the worker user before model entry. Omission preserves existing profiles.

Evidence and limits

The complete native checkpoint/context regression selection passed 96 cases, with 11 platform skips. The candidate-selection checks exercised actual SQLite work and periodic projections as well as typed measurement scopes. The resource producer was independently checked on Linux against phase-shifted allocation, output-file failure and feature-off cases; the real wrapper installer was checked against missing and shadowed producers. Component PRs retain their exact-head reviews and required local gates.

A paired deterministic control journey reduced normal commands from four to three and model-visible bytes by about 26.7%. Including refusal and replay reduced bytes by about 17.6%; replay itself grew. Single-sample tool latency regressed, so these measurements do not establish lower total latency, billed token cost or better task outcomes. Host-internal capture transport savings are reported separately from model-visible savings.

Natural model adoption, sustained score growth and full cost remain unqualified. New single-treatment runs can test operational adoption and provide diagnostic comparisons with historical runs; they do not form a new matched four-arm experiment. Native account routing remains unchanged.

The combined branch is a research artifact. The upstream C1 default-retirement and broader natural-provider qualification gates remain open; component merge-readiness does not settle those gates.

…h main Retain the complete checked decision across production callers, reuse main Progressive/Delta presentation and keep late transport options in derived views. Preserve original capture, numeric identity, failure and explicit-off semantics. Co-authored-by: aihop <4408344+hhyykk@users.noreply.github.com> Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>

Cover checked source fidelity, real File/SQLite caller settlement, native transport recovery, numeric observations and source-entrypoint parity. Retain independent semantic oracles and historical failure boundaries. Co-authored-by: aihop <4408344+hhyykk@users.noreply.github.com> Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>

…on limits Keep paired RFC checkpoints, production caller contracts and historical negative cost evidence. Integration and retry results do not establish original-workload net benefit or installed delivery. Co-authored-by: aihop <4408344+hhyykk@users.noreply.github.com> Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>