{ "columns": [ "Note id / path", "Title", "Type", "Used in sheets" ], "widths": [ 70, 60, 30, 40 ], "rows": [ [ "in-this-work-we-present-leanstral-a-series-of-generalist-code-agent-models-for", "Leanstral technical report (LeanstralSafeVerify repo PDF)", "vault primary", "Recipe_Stack, WBS" ], [ "interim-report-leanstral-base-midtraining-and-compute-provenance", "Leanstral architecture, mid-training, compute provenance", "vault interim (our pipeline)", "README, Recipe_Stack, WBS" ], [ "interim-report-data-vs-rl-lever-and-8b-4xa10g-reproducibility", "Data vs RL lever; 8B on 4x A10G box-hours", "vault interim (our pipeline)", "README, Recipe_Stack, Experiment_Plan" ], [ "interim-report-rl-adds-vs-sharpens-on-lean-passk", "Does RL add or sharpen Lean pass@k", "vault interim (our pipeline)", "Recipe_Stack, WBS" ], [ "interim-report-objective-choice-cispo-vs-grpo-effect-size-controlled", "CISPO vs GRPO family, objectives written out", "vault interim (our pipeline)", "Recipe_Stack, WBS" ], [ "interim-report-multiturn-compiler-feedback-vs-singleturn-and-sampling", "Multi-turn compiler feedback vs single-turn and sampling", "vault interim (our pipeline)", "Recipe_Stack" ], [ "interim-report-headline-benchmark-protocol-comparability", "Benchmark protocol comparability", "vault interim (our pipeline)", "Evals_Gates" ], [ "research/runs/leanstral-recipe-lineage-c14fca/comparisons.md", "Cross-locus tensions", "run artefact", "README" ], [ "/apps/development/rlhf/align/docs/SOTA_PLAN.md", "8B SOTA plan: protocol classes, levers, gates, calendar", "repo doc", "WBS, Experiment_Plan, README" ], [ "/apps/development/rlhf/align/docs/SOTA_TARGETS.md", "Verified ~8B prover targets", "repo doc", "Evals_Gates, WBS" ], [ "/apps/development/rlhf/align/docs/SELF_DISTILLATION.md", "Self-distillation plan", "repo doc", "WBS, Experiment_Plan" ], [ "/apps/development/rlhf/align/docs/P2.md", "P2 baseline build and protocol", "repo doc", "WBS" ] ] }