BurnyCoder's picture
Publish the verified paper-review protocol before model inference
387a4cf verified
|
Raw History Blame Contribute Delete
2.03 kB

Research plan

Implement HyperAgents E.5's editable newest-node policy in the October RQGM two-role paper domain. Source audit · Methods and decision criteria · Exact run order

  1. Verify the publications, both independent RQGM implementations, the released selector and the data. Publish the initial review, split hashes, registration and README.
  2. Start from the pinned OpenRQGM archive, evidence model, posterior statistics and persistence. Align the gate, measurement sampler, feedback timing, eligibility, and terminal reviewer selection with the October paper. Isolate executable workspace programs and give selectors only immutable archive aggregates.
  3. Build the trusted local inference broker, full-paper reader, domain scoring, checkpoint and resume, deterministic event and token meters. Tests exercise a branching tree where local and clade evidence point to different parents, selector-owner/parent inheritance, timeout fallback, selective erasure and sealed-data isolation.
  4. Push the implementation and tests. Run tests and an offline end-to-end fixture. Publish the prespecified paper and website and initialize the Hugging Face collection before inference.
  5. Preflight GPU residency and run the root and four pilot conditions. Inspect the literal model outputs, admission rates, selector behavior, evaluator turnover and full-paper coverage; publish negative outcomes, repairs and progress.
  6. Fix the confirmatory budget using timing alone, execute paired seeds and controls and fork the first editable midpoint. Freeze endpoints, run sealed evaluation and generate a paired crossed interval, matched-token comparison, examples, diagrams and paper from raw results.
  7. Update GitHub, GitHub Pages and the single Hugging Face collection after completed searches and evaluations. Monitor long work at 5, 15 or 30 minutes as appropriate to the estimated duration. Capture all experiments and lessons. Robotics is a separate follow-up.