loopmath 0.4.0rc1__tar.gz → 0.4.2__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/LICENSE +1 -1
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/PKG-INFO +44 -21
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/README.md +42 -19
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/pyproject.toml +4 -3
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/__init__.py +1 -1
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/compose.py +5 -3
- loopmath-0.4.2/src/loopmath/belief/cost_recipe.py +185 -0
- loopmath-0.4.2/src/loopmath/belief/cost_sources.py +49 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/design.py +178 -40
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/fit.py +182 -44
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/forest.py +9 -3
- loopmath-0.4.2/src/loopmath/belief/level.py +132 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/lookahead.py +25 -7
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/outcome.py +108 -10
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/priors.py +52 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/rules.py +10 -7
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/state.py +36 -6
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/cli.py +43 -3
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/cli_registry.py +183 -13
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/compare/__init__.py +259 -33
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/compare/commands.py +5 -3
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/compare/units.py +9 -5
- loopmath-0.4.2/src/loopmath/flows/__init__.py +10 -0
- loopmath-0.4.2/src/loopmath/flows/check.py +717 -0
- loopmath-0.4.2/src/loopmath/flows/commands.py +1306 -0
- loopmath-0.4.2/src/loopmath/flows/edits.py +725 -0
- loopmath-0.4.2/src/loopmath/flows/folder.py +133 -0
- loopmath-0.4.2/src/loopmath/flows/judges.py +843 -0
- loopmath-0.4.2/src/loopmath/flows/next.py +120 -0
- loopmath-0.4.2/src/loopmath/flows/plan.py +822 -0
- loopmath-0.4.2/src/loopmath/flows/prefill.py +335 -0
- loopmath-0.4.2/src/loopmath/flows/record.py +474 -0
- loopmath-0.4.2/src/loopmath/flows/records.py +384 -0
- loopmath-0.4.2/src/loopmath/flows/simulate.py +816 -0
- loopmath-0.4.2/src/loopmath/flows/status.py +374 -0
- loopmath-0.4.2/src/loopmath/flows/tomlview.py +666 -0
- loopmath-0.4.2/src/loopmath/flows/walk.py +869 -0
- loopmath-0.4.2/src/loopmath/frontier/__init__.py +13 -0
- loopmath-0.4.2/src/loopmath/frontier/commands.py +437 -0
- loopmath-0.4.2/src/loopmath/frontier/cost.py +299 -0
- loopmath-0.4.2/src/loopmath/frontier/kg.py +432 -0
- loopmath-0.4.2/src/loopmath/frontier/model.py +806 -0
- loopmath-0.4.2/src/loopmath/frontier/pareto.py +185 -0
- loopmath-0.4.2/src/loopmath/frontier/record.py +45 -0
- loopmath-0.4.2/src/loopmath/frontier/state.py +133 -0
- loopmath-0.4.2/src/loopmath/frontier/strategy.py +99 -0
- loopmath-0.4.2/src/loopmath/frontier/tb_prior.csv +42 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/herdr/loopmath-harness/herdr-plugin.toml +1 -1
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/mod/loopmath-harness/.claude-plugin/marketplace.json +1 -1
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/mod/loopmath-harness/.claude-plugin/plugin.json +2 -2
- loopmath-0.4.2/src/loopmath/harness_runs/__init__.py +19 -0
- loopmath-0.4.2/src/loopmath/harness_runs/ocp.py +741 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness_runs/rows.py +224 -12
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/__init__.py +6 -3
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/canonical.py +69 -8
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/commands.py +3 -2
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/conformance.py +632 -34
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/emit.py +100 -15
- loopmath-0.4.2/src/loopmath/ocp/finite.py +93 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/migrate.py +17 -9
- loopmath-0.4.2/src/loopmath/ocp/schema/ocp-v0.4.schema.json +2778 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/onboard/commands.py +1 -1
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/prices.toml +29 -26
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/reduce.py +6 -2
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/show.py +9 -4
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/validate.py +4 -1
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend/commands.py +4 -1
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend/search.py +5 -4
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/skill/commands.py +17 -19
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/skill/doctor.py +35 -8
- loopmath-0.4.2/src/loopmath/skill/guide.py +53 -0
- loopmath-0.4.0rc1/src/loopmath/skill/skills/loopmath-update-fit/SKILL.md → loopmath-0.4.2/src/loopmath/skill/guides/fit.md +4 -10
- loopmath-0.4.2/src/loopmath/skill/guides/flow.md +409 -0
- loopmath-0.4.0rc1/src/loopmath/skill/skills/loopmath-import-runs/SKILL.md → loopmath-0.4.2/src/loopmath/skill/guides/import.md +1 -7
- loopmath-0.4.0rc1/src/loopmath/skill/skills/loopmath-onboard/SKILL.md → loopmath-0.4.2/src/loopmath/skill/guides/onboard.md +2 -8
- loopmath-0.4.0rc1/src/loopmath/skill/skills/loopmath-plan-task/SKILL.md → loopmath-0.4.2/src/loopmath/skill/guides/plan.md +3 -9
- loopmath-0.4.0rc1/src/loopmath/skill/skills/loopmath-record-run/SKILL.md → loopmath-0.4.2/src/loopmath/skill/guides/record.md +2 -8
- {loopmath-0.4.0rc1/src/loopmath/skill/skills → loopmath-0.4.2/src/loopmath/skill/guides}/reference.md +3 -3
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/skill/install.py +146 -71
- loopmath-0.4.2/src/loopmath/skill/skills/loopmath/SKILL.md +58 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/commands.py +150 -27
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/config.py +24 -5
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/finish.py +87 -13
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/home.py +78 -8
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/ids.py +29 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/report.py +6 -3
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/runs.py +70 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/types.py +79 -0
- loopmath-0.4.2/src/loopmath/update.py +299 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/posterior.py +12 -6
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/candidates.py +3 -2
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/format.py +75 -6
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/ocp.py +17 -3
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath.egg-info/PKG-INFO +44 -21
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath.egg-info/SOURCES.txt +40 -6
- loopmath-0.4.2/tests/test_cli_help_lists.py +72 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_ledger.py +6 -3
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_packaging_smoke.py +1 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_price.py +6 -6
- loopmath-0.4.2/tests/test_update_check.py +428 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_v027_small_fixes.py +1 -1
- loopmath-0.4.0rc1/src/loopmath/harness_runs/__init__.py +0 -16
- loopmath-0.4.0rc1/src/loopmath/harness_runs/ocp.py +0 -425
- loopmath-0.4.0rc1/src/loopmath/skill/skills/loopmath/SKILL.md +0 -35
- loopmath-0.4.0rc1/tests/test_cli_help_lists.py +0 -40
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/setup.cfg +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/spec/ocp-v0.2.schema.json +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/spec/ocp-v0.schema.json +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/spec/ocp_conformance.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/__main__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapter_host.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/atrium.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/base.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/bb.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/bb_support.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/evolve.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/harness2.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/omp.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/omp_facts.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/omp_links.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/omp_parse.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/omp_types.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/opencode.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/opencode_store.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/openevolve.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/orca.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/otel_genai.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/otel_genai_graph.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/otel_genai_parse.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/otel_genai_types.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/otel_genai_usage.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/paseo.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/pi.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/pi_support.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/plugin_eval.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/registry.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/adapters/shinkaevolve.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/block.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/check_pymc.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/commands.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/features.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/gaussian.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/integrated.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/pricing.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/belief/sampled.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/builder/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/builder/cli.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/builder/context.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/builder/predict.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/builder/server.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/cli_errors.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/cli_graph.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/cli_support.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/compare/_thresholds.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/compare/_thresholds_sw.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/compare/calibrate.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/compare/calibration.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/compare/checks.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/compare/model.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/compare/value.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/e0/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/e0/arms.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/e0/cli_e0.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/e0/estimate.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/e0/estimate_rework.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/e0/figures.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/e0/io.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/e0/verb.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/explain.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/fit.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/fit_assembly.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/fit_bayes.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/fit_masks.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/fit_predict.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/fit_pricing.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/fit_transfer.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/gaps.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/gate/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/gate/commands.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/gate/core.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/gate/inputs.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/gate/winner.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/gate/winner_cli.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/gate/winner_model.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/gitwalk.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/grade.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/grade_support.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/artifact_git.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/artifact_join.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/artifact_kinds.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/artifacts.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/bashparse.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/bashread_ops.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/bashwrite_ops.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/bashwrites.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/cache.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/codexio.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/codexio_parse.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/codexio_scan.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/dataset.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/dataset_build.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/dataset_core.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/dataset_edges.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/dataset_nodes.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/extract.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/html_common.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/html_costcurve.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/html_css.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/html_data.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/html_force.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/html_render.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/html_swimlanes.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_cli.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_common.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_parse.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_pricing.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_prompt.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_prompt_build.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_prompt_dispatch.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_prompt_evidence.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_prompt_node.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_prompt_records.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_prompt_result.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_prompt_session.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_prompt_tools.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_prompt_validation.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_prompt_workflow.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_report.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labeler_score.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/labels.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/launch.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/launch_origin.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/ocp.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/ocp_emit.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/ocp_support.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/render.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/runfile.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/runfile_export.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/runfile_finish.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/scan.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/schema.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/graph/token_completeness.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/layouts.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/mod/loopmath-harness/.gitignore +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/mod/loopmath-harness/README.md +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/mod/loopmath-harness/hooks/events.ts +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/mod/loopmath-harness/hooks/hooks.json +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/mod/loopmath-harness/hooks/register.test.ts +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/mod/loopmath-harness/hooks/register.tsx +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/mod/loopmath-harness/types/index.d.ts +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/commands.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/config.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/contracts.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/delivery.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/design.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/editors.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/edits.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/effects.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/engine.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/events.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/executor.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/herdr.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/isolation.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/ledger.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/library.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/modinstall.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/plan.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/prompts.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/reflect.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/schema.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/score.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/sessions.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/skilldoc.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/state.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness/tune/stream.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness_runs/commands.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness_runs/costs.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/harness_runs/streams.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ingest/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ingest/base.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ingest/claude_code.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ingest/claude_code_support.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ingest/codex.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ingest/codex_support.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ingest/ocp.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ingest/ocp_analysis.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ingest/ocp_artifact.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ingest/ocp_common.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ingest/ocp_convert.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ingest/ocp_projection.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/integrations/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/integrations/dspy.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/logmatch/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/logmatch/artifacts.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/logmatch/clip.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/logmatch/codex_window.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/logmatch/costs.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/logmatch/match.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/logmatch/settle.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/logmatch/tariff.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/contractv3.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/graph_contract.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/membership.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/schema/ocp-v0.2.schema.json +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/schema/ocp-v0.3.schema.json +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/schema/ocp-v0.schema.json +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/ocp/version.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/onboard/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/onboard/history.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/onboard/label.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/onboard/usual.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/output.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/pool.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/price.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/price_validation.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/benchmarks.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/benchmarks.toml +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/build.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/bundle/README.md +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/bundle/e0.jsonl.gz +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/bundle/lanes.jsonl.gz +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/bundle/manifest.json +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/bundle/rq1.jsonl.gz +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/bundle/sweep.jsonl.gz +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/commands.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/e0.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/history.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/lanes.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/ocpdoc.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/registry.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/rq1.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/priors/sweep.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/receipts.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend/backlog.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend/curve.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend/engine.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend/gain.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend/message.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend/receipts.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend/stats.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend/storeread.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend/variants.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/recommend_role.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/report/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/report/eligibility.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/report/html.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/report/terminal.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/report/terminal_exclusions.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/report/terminal_format.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/report/terminal_read.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/research_defaults.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/research_paths.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/scoring.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/scoring_comparisons.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/scoring_predictive.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/screen/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/screen/commands.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/screen/design.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/screen/edits.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/screen/model.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/screen/record.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/share/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/share/commands.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/share/export.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/share/import_.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/skill/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/budget.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/fitjob.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/lock.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/record.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/rule_change.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/run_names.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/status.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/store/verify.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/surface.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/surface_bootstrap.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/surface_estimation.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/surface_frame.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/surface_presentation.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/surface_support.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/taskmodel.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/tasks/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/tasks/commands.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/tasks/data.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/tasks/fit.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/tasks/posterior.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/tasks/select.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/tasks/sim.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/README.md +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/base.css +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/builder.css +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/builder.html +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/builder.js +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/common.js +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/estimates.css +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/estimates.js +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/graph.css +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/graph.js +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/plans.css +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/plans.js +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/results.css +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/results.js +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/runs.js +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/viz.css +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/assets/viz.js +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/common.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/plans.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/views/runs.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/__init__.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/catalog/best_of_n.toml +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/catalog/implement_review.toml +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/catalog/plan_implement.toml +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/catalog/plan_implement_review.toml +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/catalog/solo.toml +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/catalog/swarm.toml +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/commands.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/diff.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/graphview.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/ids.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/infer.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/models.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath/workflows/shapes.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath.egg-info/dependency_links.txt +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath.egg-info/entry_points.txt +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath.egg-info/requires.txt +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/src/loopmath.egg-info/top_level.txt +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_adapter_atrium.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_adapter_bb.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_adapter_harness2.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_adapter_omp.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_adapter_opencode.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_adapter_opencode_output_tokens.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_adapter_orca.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_adapter_otel_genai.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_adapter_paseo.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_adapter_services.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_adapters.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_adapters_pi.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_analyze_cli.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_arms.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_cli_graph.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_cli_labeler.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_cli_main_module.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_cli_plan_hidden.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_cli_prices.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_cli_startup.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_color_proof.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_duplicate_guards.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_estimate.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_explain.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_fit.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_gaps.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_gitwalk.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_grade.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_artifact_recording.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_artifacts.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_bashwrites.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_cache.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_codexio.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_dataset.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_extract.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_glue.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_html.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_labeler.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_labeler_evidence_matrix.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_launch.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_ocp.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_render.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_runfile.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_graph_schema.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_cli.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_config.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_delivery.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_design.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_editors.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_edits.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_effects.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_engine.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_executor.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_herdr.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_isolation.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_library.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_modinstall.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_plan.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_score.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_sessions.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_harness_tune_skilldoc.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_home_guard.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_html.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_ingest.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_ingest_claude_code.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_ingest_codex.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_ingest_ocp.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_io.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_not_run_summary.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_ocp_conformance.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_ocp_golden.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_ocp_open_enums.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_ocp_versioning.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_output_html_path.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_packaging_metadata.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_pool.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_preview_script.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_receipts.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_release_paths.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_report_eligibility.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_research_cli.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_runfile_privacy_n1b.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_scoring.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_split_rebindings.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_starting_prior.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_surface.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_taskmodel_features.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_terminal.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_types_control.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_types_piece_prediction.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_v01_scaffold.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_v027_p3.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_v028_sol_6_1.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_v031_help_labels.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_v031_json_errors_and_costs.py +0 -0
- {loopmath-0.4.0rc1 → loopmath-0.4.2}/tests/test_v031_new_findings.py +0 -0
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: loopmath
|
|
3
|
-
Version: 0.4.
|
|
3
|
+
Version: 0.4.2
|
|
4
4
|
Summary: Plan agent workflows, record how they went, and learn from your own Claude Code and Codex logs
|
|
5
|
-
Author:
|
|
5
|
+
Author: loopmath
|
|
6
6
|
License-Expression: MIT
|
|
7
7
|
Project-URL: Homepage, https://github.com/loopmath/loopmath
|
|
8
8
|
Project-URL: Issues, https://github.com/loopmath/loopmath/issues
|
|
@@ -42,7 +42,7 @@ loopmath helps an orchestrator agent choose how to run a coding task, and learns
|
|
|
42
42
|
|
|
43
43
|
Before a task, it predicts how the user's usual agent workflow will do: the chance of an accepted result, and the cost in tokens and dollars. It shows the other workflows on a success-cost curve and offers two exploration picks. Each pick comes with its gain on every future similar run, its one-time price, and the number of runs it takes to pay back. After the task, the orchestrator records what ran and what came of it. loopmath fills in the token costs from the Claude Code and Codex logs, writes a receipt of predicted against actual, and refits.
|
|
44
44
|
|
|
45
|
-
loopmath never starts or coordinates agents. Claude Code, Codex, herdr or your own tool does that; loopmath is a CLI they call. Everything runs locally. loopmath connects to no other machine: `loopmath builder` serves its page on 127.0.0.1 only, and no other command opens a network connection. The logs are opened read only, and loopmath writes only to its store (`~/.loopmath`, which holds its caches too), to the skill folders `skill install` names, and to paths you name.
|
|
45
|
+
loopmath never starts or coordinates agents, except `loopmath harness` (experimental), which runs Claude Code sessions to tune a repo's harness once you agree to its cost. Otherwise Claude Code, Codex, herdr or your own tool does that; loopmath is a CLI they call. Everything runs locally. loopmath connects to no other machine but pypi.org: `loopmath builder` serves its page on 127.0.0.1 only, and no other command opens a network connection, except the daily update check. Once a day loopmath asks pypi.org whether a newer loopmath exists; it sends only its version and uploads nothing. `LOOPMATH_NO_UPDATE_CHECK=1` turns the check off, and it is off when `CI` is set. A harness tune's sessions are Claude Code sessions, and they call their model as Claude Code does. The logs are opened read only, and loopmath writes only to its store (`~/.loopmath`, which holds its caches too), to the skill folders `skill install` names, and to paths you name.
|
|
46
46
|
|
|
47
47
|
The package installs two commands: `loopmath` and its short alias `loop`. If another `loop` is already on your PATH, use `loopmath`.
|
|
48
48
|
|
|
@@ -64,6 +64,8 @@ python3 -m venv .venv
|
|
|
64
64
|
|
|
65
65
|
Installing needs the network: pip, pipx and uv download the build backend (setuptools 77 or newer) and the dependencies (jsonschema, numpy, pandas, scipy), from a clone too, so `pip install --no-index -e .` fails offline.
|
|
66
66
|
|
|
67
|
+
To upgrade, `loopmath update` says whether a newer loopmath is out and prints the command for how you installed it (`uv tool upgrade loopmath`, `pipx upgrade loopmath`, or pip); run it, then `loopmath skill install`. loopmath never upgrades itself.
|
|
68
|
+
|
|
67
69
|
## First run
|
|
68
70
|
|
|
69
71
|
```sh
|
|
@@ -73,20 +75,22 @@ loopmath skill install # teach your coding agents how to use loopmat
|
|
|
73
75
|
|
|
74
76
|
Then ask your agent to onboard, for example "onboard loopmath". The agent runs a dry run, asks you one question (which model labels your sessions, with the cost of each option), onboards, fits, and opens the results page.
|
|
75
77
|
|
|
76
|
-
`skill install` writes
|
|
78
|
+
`skill install` writes one skill, `loopmath`. It tells the agent to check the store with `loopmath status --json`, pick the job, and read that job's guide from the CLI, so the instructions always match the loopmath you have installed. Mention loopmath in what you ask, and the skill picks the guide:
|
|
77
79
|
|
|
78
|
-
|
|
|
80
|
+
| Guide | Ask your agent |
|
|
79
81
|
|---|---|
|
|
80
|
-
| `loopmath` |
|
|
81
|
-
| `loopmath
|
|
82
|
-
| `loopmath
|
|
83
|
-
| `loopmath
|
|
84
|
-
| `loopmath
|
|
85
|
-
| `loopmath
|
|
82
|
+
| `loopmath guide onboard` | "onboard", "set up", "start" or "try" loopmath; or loopmath has no runs yet |
|
|
83
|
+
| `loopmath guide import` | OCP files (`*.ocp.json`), a folder of finished runs, or the output of an orchestrator or experiment harness that writes OCP |
|
|
84
|
+
| `loopmath guide plan` | "plan a task with loopmath"; which workflow, model or effort to use |
|
|
85
|
+
| `loopmath guide record` | a task planned with loopmath is done; "record", "log" or "save" a run; both runs of a pair need a blind judge |
|
|
86
|
+
| `loopmath guide fit` | "update", "refresh" or "rerun" the fit; what loopmath learned; runs imported or recorded without a refit |
|
|
87
|
+
| `loopmath guide flow` | an experiment on a fixed multi-step workflow (a flow): set it up, run its arms, record them, read the verdicts |
|
|
88
|
+
|
|
89
|
+
`loopmath guide` alone lists them, and `loopmath guide reference` prints the command reference they share, with every command, flag and JSON field they use. A guide is read only: it needs no store and writes nothing.
|
|
86
90
|
|
|
87
|
-
By default the
|
|
91
|
+
By default the skill goes to every agent whose home exists: Claude Code (`~/.claude/skills/loopmath/`, or under `$CLAUDE_CONFIG_DIR`) and Codex (`~/.codex/skills/loopmath/`, or under `$CODEX_HOME`). `--target claude-code`, `codex` or `both` picks them yourself. If your Codex has no skills folder, the skill goes to `~/.codex/loopmath/loopmath/` instead, with a short marked block in `~/.codex/AGENTS.md` that points to it. Add `--scope project` to install into the current repository (or `--dir PATH` for another one).
|
|
88
92
|
|
|
89
|
-
It is safe to run again. It records the files it wrote, with their hashes, in `.loopmath-skills.json` beside the skill
|
|
93
|
+
It is safe to run again. It records the files it wrote, with their hashes, in `.loopmath-skills.json` beside the skill folder, and it replaces or removes only those. A skill file you changed is kept, with a note. Any other file where the skill goes stops the install before it writes anything, and names the path to move aside. The single skill that 0.1 installed is replaced when it is as 0.1 wrote it. Upgrading from 0.2 or 0.3 removes the five job skills those versions installed and the `reference.md` copy in each skill folder, where loopmath wrote them and they are unchanged; a file you changed is kept, with a note. `--dry-run` lists every file install would write or remove. `loopmath skill show` prints the skill, and `loopmath skill uninstall` removes the files it wrote. `loopmath doctor` shows what loopmath can see: logs, agents, store, fit, prices and the skill, and names any job skill an older loopmath left behind. To read the logs it builds the log parse cache in the store and keeps it, so the next read is fast, and it says how big the cache is; it changes nothing else in the store.
|
|
90
94
|
|
|
91
95
|
To onboard by hand instead:
|
|
92
96
|
|
|
@@ -106,13 +110,15 @@ The dry run makes no labelling call and writes no runs or fit; like `doctor`, it
|
|
|
106
110
|
|
|
107
111
|
Reading the history uses up to 8 worker processes (one per CPU, at most 8). Set `LOOPMATH_WORKERS=1` to keep everything in one process, or another number to change the count.
|
|
108
112
|
|
|
109
|
-
loopmath ships with a prior built from our own sweeps and experiments, so `recommend` gives an answer before you have any history. Your own runs then move it.
|
|
113
|
+
loopmath ships with a prior built from our own sweeps and experiments, so `recommend` gives an answer before you have any history. Your own runs then move it. Your store's overall pass rate is its own, with a wide prior that only your runs move: the shipped runs give their costs and gate rows but no success rows, and the public benchmarks say which models are stronger (`belief.level`, `store` by default; `loopmath config set belief.level shared`, then `loopmath fit`, gives 0.3.5's fit).
|
|
114
|
+
|
|
115
|
+
In a workflow with several pieces, the piece that writes the change counts most in the predicted chance of success: its model, effort and harness count 1, the planner's 0.22 and the reviewers' 0.11 between them. These three weights are a placeholder from one experiment on 8 coding tasks; they are fixed, not learned from your runs, and will change. `loopmath config set belief.seat_weights average`, then `loopmath fit`, counts every piece the same, as 0.4.1 did.
|
|
110
116
|
|
|
111
117
|
The prior ships inside the package: our runs, in the shared form (no titles, paths, commands or free text), and the published benchmark results. They are the same files we fit with, and each release updates them. `onboard` and the results page say in one line which prior they start from, and `loopmath prior show` lists it source by source.
|
|
112
118
|
|
|
113
119
|
## The loop
|
|
114
120
|
|
|
115
|
-
The plan and record
|
|
121
|
+
The plan and record guides take an agent through the loop in three commands:
|
|
116
122
|
|
|
117
123
|
```sh
|
|
118
124
|
loopmath recommend --type bug_fix --repo acme/api --title "Fix the parser crash" --json --brief --html
|
|
@@ -155,7 +161,7 @@ A pair runs the goal workflow and an exploration pick from the same base commit,
|
|
|
155
161
|
|
|
156
162
|
A run counts as a success when it is accepted under your rule. By default that means the task's tests pass. A run that `onboard` brought in from your logs, or that `run import` stored, with no tests verdict but a review or accepted verdict, is judged by the review rule instead: a success when its latest review or accepted verdict is accept, approve or pass. `loopmath runs` lists other finished runs such a verdict could settle, with the command that moves them to the review rule, `loopmath outcome --rule review RUN...`. `loopmath outcome --undo-rule RUN...` undoes that, restoring each run's rule as it was; naming that rule (`tests` when it had none) does too, unless the run's own rule was also named review. A review verdict on a reviewer's last attempt then counts twice: as the review gate's pass, and as the run's success. A rule can also be a score target: `loopmath recommend ... --target 'heldout_perf>=2400'` or `--target 'runtime_s<=200'` gives the expected score and the chance of reaching the target. Set standing rules and other options with `loopmath config set KEY VALUE` (`loopmath config get` lists every key, set or not, `config set` refuses a key loopmath does not read, and `loopmath config unset KEY` returns a key to its default), and the spending cap with `loopmath budget --usd X --period month`. Spend counts your runs and onboard's labelling calls; runs onboard brought in from your logs are shown apart and not counted. A run you record with `--source habit` counts.
|
|
157
163
|
|
|
158
|
-
Runs recorded by another tool that writes OCP v0.3, such as an experiment runner, come in with `loopmath run import DIR --finish`: several files or a directory, with one refit at the end (`--no-fit` skips it). It says how many runs were new, unchanged or replaced; importing finished runs into a store with no fit starts a fit. Importing the same documents again changes nothing and keeps the outcomes you added since, and with `--finish` it finishes the runs left open (without it, the runs are stored open and it says to import again with `--finish`). A document whose run id is already in the store with different content is refused by name, with the fields that differ; `--replace` takes it and keeps the outcomes added after the import, and refuses a copy that would overwrite one. Runs in your store always count as yours, whatever source their documents name. `loopmath fit --without SOURCE` leaves out one shipped prior source, and `loopmath status` names the options of the current fit.
|
|
164
|
+
Runs recorded by another tool that writes OCP v0.3 (or v0.4, for a run that names a flow or states checks), such as an experiment runner, come in with `loopmath run import DIR --finish`: several files or a directory, with one refit at the end (`--no-fit` skips it). It says how many runs were new, unchanged or replaced; importing finished runs into a store with no fit starts a fit. Importing the same documents again changes nothing and keeps the outcomes you added since, and with `--finish` it finishes the runs left open (without it, the runs are stored open and it says to import again with `--finish`). A document whose run id is already in the store with different content is refused by name, with the fields that differ; `--replace` takes it and keeps the outcomes added after the import, and refuses a copy that would overwrite one. Runs in your store always count as yours, whatever source their documents name. `loopmath fit --without SOURCE` leaves out one shipped prior source, and `loopmath status` names the options of the current fit.
|
|
159
165
|
|
|
160
166
|
## The three views
|
|
161
167
|
|
|
@@ -175,16 +181,19 @@ The workflow builder is the one page that is served: `loopmath builder --type bu
|
|
|
175
181
|
|---|---|
|
|
176
182
|
| Plan | `task-types`, `workflows` (list, show, validate, diff), `recommend`, `builder` |
|
|
177
183
|
| Record | `run` (start, record, import, attempt, artifact, finish, sessions), `outcome`, `budget`, `config` (get, set, unset) |
|
|
184
|
+
| Flows | `flow` (init, show, export, import, plan, next, record, score, status, check) |
|
|
185
|
+
| Frontier | `frontier` (init, next, record, report) (experimental) |
|
|
178
186
|
| Learn | `fit`, `onboard`, `share`, `prior` |
|
|
179
|
-
| View | `runs`, `posterior`, `status`, `report`, `doctor`, `verify-receipts` |
|
|
187
|
+
| View | `runs`, `posterior`, `status`, `report`, `doctor`, `verify-receipts`, `update` |
|
|
180
188
|
| Harness testing | `check-change`, `gate` (experimental), `screen` (experimental) |
|
|
189
|
+
| Harness tuning | `harness` (experimental) |
|
|
181
190
|
| Task subsets | `tasks pick` (experimental) |
|
|
182
191
|
| Evolution loops | `adapt openevolve` (experimental), `adapt shinkaevolve` (experimental), `recommend --role` (experimental) |
|
|
183
|
-
| Skill | `skill` (install, uninstall, show) |
|
|
192
|
+
| Skill | `skill` (install, uninstall, show), `guide` |
|
|
184
193
|
| OCP | `ocp` (validate, migrate) |
|
|
185
194
|
| Log analysis | `analyze`, `graph`, `adapt`, `prices`, `validate-prices` |
|
|
186
195
|
|
|
187
|
-
These are experimental: `gate`, `screen`, `tasks pick`, the harness-part estimates (`recommend --harness-variants` and `--harness`, `posterior --level harness` and its `factors`), `recommend --role`, `adapt harness2`, `adapt plugin-eval`, `adapt openevolve`, `adapt shinkaevolve` and the ADE adapters (`adapt NAME`). Each works and has tests, but has had little use on real work so far, and its output may change. Their help says so too. The gate at DSPy's GEPA and MIPROv2 accept step (`loopmath.integrations.dspy`, Python) is experimental too: it is tested against fakes written from their source, not yet with DSPy installed, and its guide says so. So is the opt-in cost model for arms that mix models, `loopmath.compare.model.COST_MODEL = "sw"`, a Python setting with no help of its own; the [changelog](CHANGELOG.md) describes it.
|
|
196
|
+
These are experimental: `harness`, `gate`, `screen`, `tasks pick`, `frontier`, the harness-part estimates (`recommend --harness-variants` and `--harness`, `posterior --level harness` and its `factors`), `recommend --role`, `adapt harness2`, `adapt plugin-eval`, `adapt openevolve`, `adapt shinkaevolve` and the ADE adapters (`adapt NAME`). Each works and has tests, but has had little use on real work so far, and its output may change. Their help says so too. The gate at DSPy's GEPA and MIPROv2 accept step (`loopmath.integrations.dspy`, Python) is experimental too: it is tested against fakes written from their source, not yet with DSPy installed, and its guide says so. So is the opt-in cost model for arms that mix models, `loopmath.compare.model.COST_MODEL = "sw"`, a Python setting with no help of its own; the [changelog](CHANGELOG.md) describes it.
|
|
188
197
|
|
|
189
198
|
Each command prints a short summary by default. With `--json`, it prints exactly one JSON object instead. One exception: `loopmath adapt harness2`, `adapt harness-runs`, `adapt plugin-eval`, `adapt openevolve` and `adapt shinkaevolve` without `--out` print the run documents on stdout, one per line, so their `--json` summary goes to stderr. Exit codes: 0 ok, 1 user error, 2 not found, 3 not implemented, 4 store locked, 5 no fit yet. For flags, `loopmath <command> --help` is the source of truth. Commands, pages and the `runs` table show times in your local time with the zone; the store and `--json` keep UTC. A cost under a cent shows one significant digit (`$0.004`), and an exact zero is `$0`.
|
|
190
199
|
|
|
@@ -314,15 +323,29 @@ loopmath gate --base arm:without --cand arm:with --runs rows.jsonl --rule accept
|
|
|
314
323
|
|
|
315
324
|
Runs of one case are not new tasks: to learn whether the plugin helps on tasks like your cases, add cases rather than runs. With grades (`--outcome score`), `check-change from-runs` compares case by case. The import's summary prints the last two filled in for your suite. Which cases are compared, what the suite's one cost can and cannot say, how far the answer reaches and the provisional input layout are in [`docs/plugin-eval.md`](docs/plugin-eval.md).
|
|
316
325
|
|
|
326
|
+
## Finding a cost and pass-rate frontier (experimental)
|
|
327
|
+
|
|
328
|
+
`loopmath frontier` looks for the settings (model and effort) that give the best pass rate for their cost on your own tasks, by ask and tell: `next` says which setting to run on which task, `record` takes the result back, `report` shows each setting's pass rate, cost per run and P, the model's probability that it is on the frontier.
|
|
329
|
+
|
|
330
|
+
```sh
|
|
331
|
+
export LOOPMATH_HOME=$(mktemp -d)
|
|
332
|
+
loopmath frontier init --tasks tasks.jsonl --settings settings.toml --budget 50
|
|
333
|
+
loopmath frontier next --json
|
|
334
|
+
loopmath frontier record TRL --passed PASSED --usd USD
|
|
335
|
+
loopmath frontier report
|
|
336
|
+
```
|
|
337
|
+
|
|
338
|
+
`tasks.jsonl` holds one task per line (`{"id": "t1"}`); `settings.toml` holds `[[settings]]` blocks with `name`, `model` and `effort`. TRL is the trial id `next` gave, PASSED is 1 for a pass and 0 for a fail, and USD is what the run cost in dollars. It is experimental: on simulated fresh DeepSWE runs it found the right frontier in about a quarter of runs at a $8,000 budget, level with the study design it is ported from. P is not yet calibrated, so do not stop on it.
|
|
339
|
+
|
|
317
340
|
## The store and privacy
|
|
318
341
|
|
|
319
342
|
The store is `~/.loopmath`. Set `LOOPMATH_HOME` or pass `--home` to put it elsewhere; the log parse cache moves with it, to `cache/` inside the store, unless you set `LOOPMATH_CACHE_DIR`. It holds:
|
|
320
343
|
|
|
321
|
-
- one OCP v0.3
|
|
344
|
+
- one OCP document per recorded run: v0.3, or v0.4 for a run that names a flow or states checks;
|
|
322
345
|
- signals, receipts and fits;
|
|
323
346
|
- your config and your workflows;
|
|
324
347
|
- the views you write;
|
|
325
|
-
- the caches: the log parse cache, and the folder where `fit --full` compiles its model (when that compile fails, it also leaves the C code it tried in a `pytensor_compilation_error_*` file in the system temp folder and prints where).
|
|
348
|
+
- the caches: the log parse cache, the daily update check's last answer (`update.json`), and the folder where `fit --full` compiles its model (when that compile fails, it also leaves the C code it tried in a `pytensor_compilation_error_*` file in the system temp folder and prints where).
|
|
326
349
|
|
|
327
350
|
Do not edit the store by hand; use the commands. `loopmath verify-receipts` checks the receipts in your store; give a file to verify a research receipt ledger instead.
|
|
328
351
|
|
|
@@ -4,7 +4,7 @@ loopmath helps an orchestrator agent choose how to run a coding task, and learns
|
|
|
4
4
|
|
|
5
5
|
Before a task, it predicts how the user's usual agent workflow will do: the chance of an accepted result, and the cost in tokens and dollars. It shows the other workflows on a success-cost curve and offers two exploration picks. Each pick comes with its gain on every future similar run, its one-time price, and the number of runs it takes to pay back. After the task, the orchestrator records what ran and what came of it. loopmath fills in the token costs from the Claude Code and Codex logs, writes a receipt of predicted against actual, and refits.
|
|
6
6
|
|
|
7
|
-
loopmath never starts or coordinates agents. Claude Code, Codex, herdr or your own tool does that; loopmath is a CLI they call. Everything runs locally. loopmath connects to no other machine: `loopmath builder` serves its page on 127.0.0.1 only, and no other command opens a network connection. The logs are opened read only, and loopmath writes only to its store (`~/.loopmath`, which holds its caches too), to the skill folders `skill install` names, and to paths you name.
|
|
7
|
+
loopmath never starts or coordinates agents, except `loopmath harness` (experimental), which runs Claude Code sessions to tune a repo's harness once you agree to its cost. Otherwise Claude Code, Codex, herdr or your own tool does that; loopmath is a CLI they call. Everything runs locally. loopmath connects to no other machine but pypi.org: `loopmath builder` serves its page on 127.0.0.1 only, and no other command opens a network connection, except the daily update check. Once a day loopmath asks pypi.org whether a newer loopmath exists; it sends only its version and uploads nothing. `LOOPMATH_NO_UPDATE_CHECK=1` turns the check off, and it is off when `CI` is set. A harness tune's sessions are Claude Code sessions, and they call their model as Claude Code does. The logs are opened read only, and loopmath writes only to its store (`~/.loopmath`, which holds its caches too), to the skill folders `skill install` names, and to paths you name.
|
|
8
8
|
|
|
9
9
|
The package installs two commands: `loopmath` and its short alias `loop`. If another `loop` is already on your PATH, use `loopmath`.
|
|
10
10
|
|
|
@@ -26,6 +26,8 @@ python3 -m venv .venv
|
|
|
26
26
|
|
|
27
27
|
Installing needs the network: pip, pipx and uv download the build backend (setuptools 77 or newer) and the dependencies (jsonschema, numpy, pandas, scipy), from a clone too, so `pip install --no-index -e .` fails offline.
|
|
28
28
|
|
|
29
|
+
To upgrade, `loopmath update` says whether a newer loopmath is out and prints the command for how you installed it (`uv tool upgrade loopmath`, `pipx upgrade loopmath`, or pip); run it, then `loopmath skill install`. loopmath never upgrades itself.
|
|
30
|
+
|
|
29
31
|
## First run
|
|
30
32
|
|
|
31
33
|
```sh
|
|
@@ -35,20 +37,22 @@ loopmath skill install # teach your coding agents how to use loopmat
|
|
|
35
37
|
|
|
36
38
|
Then ask your agent to onboard, for example "onboard loopmath". The agent runs a dry run, asks you one question (which model labels your sessions, with the cost of each option), onboards, fits, and opens the results page.
|
|
37
39
|
|
|
38
|
-
`skill install` writes
|
|
40
|
+
`skill install` writes one skill, `loopmath`. It tells the agent to check the store with `loopmath status --json`, pick the job, and read that job's guide from the CLI, so the instructions always match the loopmath you have installed. Mention loopmath in what you ask, and the skill picks the guide:
|
|
39
41
|
|
|
40
|
-
|
|
|
42
|
+
| Guide | Ask your agent |
|
|
41
43
|
|---|---|
|
|
42
|
-
| `loopmath` |
|
|
43
|
-
| `loopmath
|
|
44
|
-
| `loopmath
|
|
45
|
-
| `loopmath
|
|
46
|
-
| `loopmath
|
|
47
|
-
| `loopmath
|
|
44
|
+
| `loopmath guide onboard` | "onboard", "set up", "start" or "try" loopmath; or loopmath has no runs yet |
|
|
45
|
+
| `loopmath guide import` | OCP files (`*.ocp.json`), a folder of finished runs, or the output of an orchestrator or experiment harness that writes OCP |
|
|
46
|
+
| `loopmath guide plan` | "plan a task with loopmath"; which workflow, model or effort to use |
|
|
47
|
+
| `loopmath guide record` | a task planned with loopmath is done; "record", "log" or "save" a run; both runs of a pair need a blind judge |
|
|
48
|
+
| `loopmath guide fit` | "update", "refresh" or "rerun" the fit; what loopmath learned; runs imported or recorded without a refit |
|
|
49
|
+
| `loopmath guide flow` | an experiment on a fixed multi-step workflow (a flow): set it up, run its arms, record them, read the verdicts |
|
|
50
|
+
|
|
51
|
+
`loopmath guide` alone lists them, and `loopmath guide reference` prints the command reference they share, with every command, flag and JSON field they use. A guide is read only: it needs no store and writes nothing.
|
|
48
52
|
|
|
49
|
-
By default the
|
|
53
|
+
By default the skill goes to every agent whose home exists: Claude Code (`~/.claude/skills/loopmath/`, or under `$CLAUDE_CONFIG_DIR`) and Codex (`~/.codex/skills/loopmath/`, or under `$CODEX_HOME`). `--target claude-code`, `codex` or `both` picks them yourself. If your Codex has no skills folder, the skill goes to `~/.codex/loopmath/loopmath/` instead, with a short marked block in `~/.codex/AGENTS.md` that points to it. Add `--scope project` to install into the current repository (or `--dir PATH` for another one).
|
|
50
54
|
|
|
51
|
-
It is safe to run again. It records the files it wrote, with their hashes, in `.loopmath-skills.json` beside the skill
|
|
55
|
+
It is safe to run again. It records the files it wrote, with their hashes, in `.loopmath-skills.json` beside the skill folder, and it replaces or removes only those. A skill file you changed is kept, with a note. Any other file where the skill goes stops the install before it writes anything, and names the path to move aside. The single skill that 0.1 installed is replaced when it is as 0.1 wrote it. Upgrading from 0.2 or 0.3 removes the five job skills those versions installed and the `reference.md` copy in each skill folder, where loopmath wrote them and they are unchanged; a file you changed is kept, with a note. `--dry-run` lists every file install would write or remove. `loopmath skill show` prints the skill, and `loopmath skill uninstall` removes the files it wrote. `loopmath doctor` shows what loopmath can see: logs, agents, store, fit, prices and the skill, and names any job skill an older loopmath left behind. To read the logs it builds the log parse cache in the store and keeps it, so the next read is fast, and it says how big the cache is; it changes nothing else in the store.
|
|
52
56
|
|
|
53
57
|
To onboard by hand instead:
|
|
54
58
|
|
|
@@ -68,13 +72,15 @@ The dry run makes no labelling call and writes no runs or fit; like `doctor`, it
|
|
|
68
72
|
|
|
69
73
|
Reading the history uses up to 8 worker processes (one per CPU, at most 8). Set `LOOPMATH_WORKERS=1` to keep everything in one process, or another number to change the count.
|
|
70
74
|
|
|
71
|
-
loopmath ships with a prior built from our own sweeps and experiments, so `recommend` gives an answer before you have any history. Your own runs then move it.
|
|
75
|
+
loopmath ships with a prior built from our own sweeps and experiments, so `recommend` gives an answer before you have any history. Your own runs then move it. Your store's overall pass rate is its own, with a wide prior that only your runs move: the shipped runs give their costs and gate rows but no success rows, and the public benchmarks say which models are stronger (`belief.level`, `store` by default; `loopmath config set belief.level shared`, then `loopmath fit`, gives 0.3.5's fit).
|
|
76
|
+
|
|
77
|
+
In a workflow with several pieces, the piece that writes the change counts most in the predicted chance of success: its model, effort and harness count 1, the planner's 0.22 and the reviewers' 0.11 between them. These three weights are a placeholder from one experiment on 8 coding tasks; they are fixed, not learned from your runs, and will change. `loopmath config set belief.seat_weights average`, then `loopmath fit`, counts every piece the same, as 0.4.1 did.
|
|
72
78
|
|
|
73
79
|
The prior ships inside the package: our runs, in the shared form (no titles, paths, commands or free text), and the published benchmark results. They are the same files we fit with, and each release updates them. `onboard` and the results page say in one line which prior they start from, and `loopmath prior show` lists it source by source.
|
|
74
80
|
|
|
75
81
|
## The loop
|
|
76
82
|
|
|
77
|
-
The plan and record
|
|
83
|
+
The plan and record guides take an agent through the loop in three commands:
|
|
78
84
|
|
|
79
85
|
```sh
|
|
80
86
|
loopmath recommend --type bug_fix --repo acme/api --title "Fix the parser crash" --json --brief --html
|
|
@@ -117,7 +123,7 @@ A pair runs the goal workflow and an exploration pick from the same base commit,
|
|
|
117
123
|
|
|
118
124
|
A run counts as a success when it is accepted under your rule. By default that means the task's tests pass. A run that `onboard` brought in from your logs, or that `run import` stored, with no tests verdict but a review or accepted verdict, is judged by the review rule instead: a success when its latest review or accepted verdict is accept, approve or pass. `loopmath runs` lists other finished runs such a verdict could settle, with the command that moves them to the review rule, `loopmath outcome --rule review RUN...`. `loopmath outcome --undo-rule RUN...` undoes that, restoring each run's rule as it was; naming that rule (`tests` when it had none) does too, unless the run's own rule was also named review. A review verdict on a reviewer's last attempt then counts twice: as the review gate's pass, and as the run's success. A rule can also be a score target: `loopmath recommend ... --target 'heldout_perf>=2400'` or `--target 'runtime_s<=200'` gives the expected score and the chance of reaching the target. Set standing rules and other options with `loopmath config set KEY VALUE` (`loopmath config get` lists every key, set or not, `config set` refuses a key loopmath does not read, and `loopmath config unset KEY` returns a key to its default), and the spending cap with `loopmath budget --usd X --period month`. Spend counts your runs and onboard's labelling calls; runs onboard brought in from your logs are shown apart and not counted. A run you record with `--source habit` counts.
|
|
119
125
|
|
|
120
|
-
Runs recorded by another tool that writes OCP v0.3, such as an experiment runner, come in with `loopmath run import DIR --finish`: several files or a directory, with one refit at the end (`--no-fit` skips it). It says how many runs were new, unchanged or replaced; importing finished runs into a store with no fit starts a fit. Importing the same documents again changes nothing and keeps the outcomes you added since, and with `--finish` it finishes the runs left open (without it, the runs are stored open and it says to import again with `--finish`). A document whose run id is already in the store with different content is refused by name, with the fields that differ; `--replace` takes it and keeps the outcomes added after the import, and refuses a copy that would overwrite one. Runs in your store always count as yours, whatever source their documents name. `loopmath fit --without SOURCE` leaves out one shipped prior source, and `loopmath status` names the options of the current fit.
|
|
126
|
+
Runs recorded by another tool that writes OCP v0.3 (or v0.4, for a run that names a flow or states checks), such as an experiment runner, come in with `loopmath run import DIR --finish`: several files or a directory, with one refit at the end (`--no-fit` skips it). It says how many runs were new, unchanged or replaced; importing finished runs into a store with no fit starts a fit. Importing the same documents again changes nothing and keeps the outcomes you added since, and with `--finish` it finishes the runs left open (without it, the runs are stored open and it says to import again with `--finish`). A document whose run id is already in the store with different content is refused by name, with the fields that differ; `--replace` takes it and keeps the outcomes added after the import, and refuses a copy that would overwrite one. Runs in your store always count as yours, whatever source their documents name. `loopmath fit --without SOURCE` leaves out one shipped prior source, and `loopmath status` names the options of the current fit.
|
|
121
127
|
|
|
122
128
|
## The three views
|
|
123
129
|
|
|
@@ -137,16 +143,19 @@ The workflow builder is the one page that is served: `loopmath builder --type bu
|
|
|
137
143
|
|---|---|
|
|
138
144
|
| Plan | `task-types`, `workflows` (list, show, validate, diff), `recommend`, `builder` |
|
|
139
145
|
| Record | `run` (start, record, import, attempt, artifact, finish, sessions), `outcome`, `budget`, `config` (get, set, unset) |
|
|
146
|
+
| Flows | `flow` (init, show, export, import, plan, next, record, score, status, check) |
|
|
147
|
+
| Frontier | `frontier` (init, next, record, report) (experimental) |
|
|
140
148
|
| Learn | `fit`, `onboard`, `share`, `prior` |
|
|
141
|
-
| View | `runs`, `posterior`, `status`, `report`, `doctor`, `verify-receipts` |
|
|
149
|
+
| View | `runs`, `posterior`, `status`, `report`, `doctor`, `verify-receipts`, `update` |
|
|
142
150
|
| Harness testing | `check-change`, `gate` (experimental), `screen` (experimental) |
|
|
151
|
+
| Harness tuning | `harness` (experimental) |
|
|
143
152
|
| Task subsets | `tasks pick` (experimental) |
|
|
144
153
|
| Evolution loops | `adapt openevolve` (experimental), `adapt shinkaevolve` (experimental), `recommend --role` (experimental) |
|
|
145
|
-
| Skill | `skill` (install, uninstall, show) |
|
|
154
|
+
| Skill | `skill` (install, uninstall, show), `guide` |
|
|
146
155
|
| OCP | `ocp` (validate, migrate) |
|
|
147
156
|
| Log analysis | `analyze`, `graph`, `adapt`, `prices`, `validate-prices` |
|
|
148
157
|
|
|
149
|
-
These are experimental: `gate`, `screen`, `tasks pick`, the harness-part estimates (`recommend --harness-variants` and `--harness`, `posterior --level harness` and its `factors`), `recommend --role`, `adapt harness2`, `adapt plugin-eval`, `adapt openevolve`, `adapt shinkaevolve` and the ADE adapters (`adapt NAME`). Each works and has tests, but has had little use on real work so far, and its output may change. Their help says so too. The gate at DSPy's GEPA and MIPROv2 accept step (`loopmath.integrations.dspy`, Python) is experimental too: it is tested against fakes written from their source, not yet with DSPy installed, and its guide says so. So is the opt-in cost model for arms that mix models, `loopmath.compare.model.COST_MODEL = "sw"`, a Python setting with no help of its own; the [changelog](CHANGELOG.md) describes it.
|
|
158
|
+
These are experimental: `harness`, `gate`, `screen`, `tasks pick`, `frontier`, the harness-part estimates (`recommend --harness-variants` and `--harness`, `posterior --level harness` and its `factors`), `recommend --role`, `adapt harness2`, `adapt plugin-eval`, `adapt openevolve`, `adapt shinkaevolve` and the ADE adapters (`adapt NAME`). Each works and has tests, but has had little use on real work so far, and its output may change. Their help says so too. The gate at DSPy's GEPA and MIPROv2 accept step (`loopmath.integrations.dspy`, Python) is experimental too: it is tested against fakes written from their source, not yet with DSPy installed, and its guide says so. So is the opt-in cost model for arms that mix models, `loopmath.compare.model.COST_MODEL = "sw"`, a Python setting with no help of its own; the [changelog](CHANGELOG.md) describes it.
|
|
150
159
|
|
|
151
160
|
Each command prints a short summary by default. With `--json`, it prints exactly one JSON object instead. One exception: `loopmath adapt harness2`, `adapt harness-runs`, `adapt plugin-eval`, `adapt openevolve` and `adapt shinkaevolve` without `--out` print the run documents on stdout, one per line, so their `--json` summary goes to stderr. Exit codes: 0 ok, 1 user error, 2 not found, 3 not implemented, 4 store locked, 5 no fit yet. For flags, `loopmath <command> --help` is the source of truth. Commands, pages and the `runs` table show times in your local time with the zone; the store and `--json` keep UTC. A cost under a cent shows one significant digit (`$0.004`), and an exact zero is `$0`.
|
|
152
161
|
|
|
@@ -276,15 +285,29 @@ loopmath gate --base arm:without --cand arm:with --runs rows.jsonl --rule accept
|
|
|
276
285
|
|
|
277
286
|
Runs of one case are not new tasks: to learn whether the plugin helps on tasks like your cases, add cases rather than runs. With grades (`--outcome score`), `check-change from-runs` compares case by case. The import's summary prints the last two filled in for your suite. Which cases are compared, what the suite's one cost can and cannot say, how far the answer reaches and the provisional input layout are in [`docs/plugin-eval.md`](docs/plugin-eval.md).
|
|
278
287
|
|
|
288
|
+
## Finding a cost and pass-rate frontier (experimental)
|
|
289
|
+
|
|
290
|
+
`loopmath frontier` looks for the settings (model and effort) that give the best pass rate for their cost on your own tasks, by ask and tell: `next` says which setting to run on which task, `record` takes the result back, `report` shows each setting's pass rate, cost per run and P, the model's probability that it is on the frontier.
|
|
291
|
+
|
|
292
|
+
```sh
|
|
293
|
+
export LOOPMATH_HOME=$(mktemp -d)
|
|
294
|
+
loopmath frontier init --tasks tasks.jsonl --settings settings.toml --budget 50
|
|
295
|
+
loopmath frontier next --json
|
|
296
|
+
loopmath frontier record TRL --passed PASSED --usd USD
|
|
297
|
+
loopmath frontier report
|
|
298
|
+
```
|
|
299
|
+
|
|
300
|
+
`tasks.jsonl` holds one task per line (`{"id": "t1"}`); `settings.toml` holds `[[settings]]` blocks with `name`, `model` and `effort`. TRL is the trial id `next` gave, PASSED is 1 for a pass and 0 for a fail, and USD is what the run cost in dollars. It is experimental: on simulated fresh DeepSWE runs it found the right frontier in about a quarter of runs at a $8,000 budget, level with the study design it is ported from. P is not yet calibrated, so do not stop on it.
|
|
301
|
+
|
|
279
302
|
## The store and privacy
|
|
280
303
|
|
|
281
304
|
The store is `~/.loopmath`. Set `LOOPMATH_HOME` or pass `--home` to put it elsewhere; the log parse cache moves with it, to `cache/` inside the store, unless you set `LOOPMATH_CACHE_DIR`. It holds:
|
|
282
305
|
|
|
283
|
-
- one OCP v0.3
|
|
306
|
+
- one OCP document per recorded run: v0.3, or v0.4 for a run that names a flow or states checks;
|
|
284
307
|
- signals, receipts and fits;
|
|
285
308
|
- your config and your workflows;
|
|
286
309
|
- the views you write;
|
|
287
|
-
- the caches: the log parse cache, and the folder where `fit --full` compiles its model (when that compile fails, it also leaves the C code it tried in a `pytensor_compilation_error_*` file in the system temp folder and prints where).
|
|
310
|
+
- the caches: the log parse cache, the daily update check's last answer (`update.json`), and the folder where `fit --full` compiles its model (when that compile fails, it also leaves the C code it tried in a `pytensor_compilation_error_*` file in the system temp folder and prints where).
|
|
288
311
|
|
|
289
312
|
Do not edit the store by hand; use the commands. `loopmath verify-receipts` checks the receipts in your store; give a file to verify a research receipt ledger instead.
|
|
290
313
|
|
|
@@ -4,14 +4,14 @@ build-backend = "setuptools.build_meta"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "loopmath"
|
|
7
|
-
version = "0.4.
|
|
7
|
+
version = "0.4.2"
|
|
8
8
|
description = "Plan agent workflows, record how they went, and learn from your own Claude Code and Codex logs"
|
|
9
9
|
readme = "README.md"
|
|
10
10
|
license = "MIT"
|
|
11
11
|
license-files = ["LICENSE"]
|
|
12
12
|
requires-python = ">=3.11"
|
|
13
13
|
authors = [
|
|
14
|
-
{ name = "
|
|
14
|
+
{ name = "loopmath" },
|
|
15
15
|
]
|
|
16
16
|
keywords = [
|
|
17
17
|
"agents",
|
|
@@ -87,8 +87,9 @@ where = ["src"]
|
|
|
87
87
|
[tool.setuptools.package-data]
|
|
88
88
|
loopmath = [
|
|
89
89
|
"prices.toml",
|
|
90
|
+
"frontier/tb_prior.csv",
|
|
90
91
|
"workflows/catalog/*.toml",
|
|
91
|
-
"skill/
|
|
92
|
+
"skill/guides/*.md",
|
|
92
93
|
"skill/skills/*/SKILL.md",
|
|
93
94
|
"priors/*.toml",
|
|
94
95
|
"priors/bundle/*",
|
|
@@ -90,7 +90,7 @@ def inv_rate_mean(m, v, floor: float | None) -> np.ndarray:
|
|
|
90
90
|
return np.where(certain, np.exp(-np.maximum(m, lf)), spread)
|
|
91
91
|
|
|
92
92
|
|
|
93
|
-
def compose(st: Structure, cost_eta: dict, rate_eta: dict, gate_eta: dict, *, sigma_usd: float,
|
|
93
|
+
def compose(st: Structure, cost_eta: dict, rate_eta: dict, gate_eta: dict, *, sigma_usd: float | Mapping,
|
|
94
94
|
sigma_rate: float, floors: Mapping[str, float] | None = None, rate_var: Mapping | None = None,
|
|
95
95
|
rng: np.random.Generator | None) -> RunDraws:
|
|
96
96
|
"""Compose one configuration. `cost_eta[(piece, k)]`, `rate_eta[(piece, k)]` and `gate_eta[(gate index, k)]`
|
|
@@ -99,7 +99,7 @@ def compose(st: Structure, cost_eta: dict, rate_eta: dict, gate_eta: dict, *, si
|
|
|
99
99
|
path passes the rate rows' known parts and adds their unseen variance there). With `rng` None nothing is
|
|
100
100
|
simulated: the `sim_*` arrays stay zero."""
|
|
101
101
|
n = len(next(iter(cost_eta.values())))
|
|
102
|
-
|
|
102
|
+
var_r = sigma_rate ** 2
|
|
103
103
|
floors = floors or {}
|
|
104
104
|
exp_u, exp_t = np.zeros(n), np.zeros(n)
|
|
105
105
|
sim_u, sim_t = np.zeros(n), np.zeros(n)
|
|
@@ -120,6 +120,8 @@ def compose(st: Structure, cost_eta: dict, rate_eta: dict, gate_eta: dict, *, si
|
|
|
120
120
|
def run_piece(piece: str, k: int, weight: np.ndarray, alive: np.ndarray):
|
|
121
121
|
w = st.widths[piece]
|
|
122
122
|
eu, er = cost_eta[(piece, k)], rate_eta[(piece, k)]
|
|
123
|
+
sigma_u = sigma_usd[(piece, k)] if isinstance(sigma_usd, Mapping) else sigma_usd
|
|
124
|
+
half_u = .5 * sigma_u ** 2
|
|
123
125
|
f = float(floors.get(piece) or 0.0)
|
|
124
126
|
cu = w * np.exp(eu + half_u)
|
|
125
127
|
ct = cu * inv_rate_mean(er, var_r if rate_var is None else rate_var[(piece, k)], f)
|
|
@@ -129,7 +131,7 @@ def compose(st: Structure, cost_eta: dict, rate_eta: dict, gate_eta: dict, *, si
|
|
|
129
131
|
su = np.zeros(n)
|
|
130
132
|
stt = np.zeros(n)
|
|
131
133
|
for _ in range(w if rng is not None else 0):
|
|
132
|
-
u = np.exp(eu +
|
|
134
|
+
u = np.exp(eu + sigma_u * rng.standard_normal(n))
|
|
133
135
|
r = np.exp(er + sigma_rate * rng.standard_normal(n))
|
|
134
136
|
su += u
|
|
135
137
|
stt += u / (np.maximum(r, f) if f > 0 else r)
|
|
@@ -0,0 +1,185 @@
|
|
|
1
|
+
"""The frontier's source-only setting cost law, expressed as Gaussian row features.
|
|
2
|
+
|
|
3
|
+
FrontierCost owns the sufficient statistics, run-variance shrinkage and joint
|
|
4
|
+
width refit. The features below factor exactly its four covariance components,
|
|
5
|
+
so unseen settings, prediction and lookahead use the same conditional law.
|
|
6
|
+
"""
|
|
7
|
+
|
|
8
|
+
from __future__ import annotations
|
|
9
|
+
|
|
10
|
+
import hashlib
|
|
11
|
+
import json
|
|
12
|
+
import math
|
|
13
|
+
import re
|
|
14
|
+
from functools import lru_cache
|
|
15
|
+
|
|
16
|
+
import numpy as np
|
|
17
|
+
from scipy import sparse
|
|
18
|
+
|
|
19
|
+
from ..frontier.cost import FrontierCost, anchors
|
|
20
|
+
from ..price import load_prices
|
|
21
|
+
from .cost_sources import TB_COSTS
|
|
22
|
+
from .forest import canonical_model_id
|
|
23
|
+
from .gaussian import HeadFit
|
|
24
|
+
|
|
25
|
+
METHOD = "frontier-setting-v1"
|
|
26
|
+
OPTION = "belief.cost_recipe"
|
|
27
|
+
CONSTANT = "fixed:cost_anchor"
|
|
28
|
+
PREFIXES = ("clevel:", "ctransfer:", "csetting:", "cstep:")
|
|
29
|
+
GROUPS = ("level", "transfer", "setting", "steps")
|
|
30
|
+
EFFORT = {"low": 0, "medium": 1, "high": 2, "xhigh": 3, "max": 4, "default": 2}
|
|
31
|
+
STRUCTURAL = {"model", "effort", "harness", "role", "step", "topology", "position", "round", "control", "hfactor"}
|
|
32
|
+
|
|
33
|
+
|
|
34
|
+
@lru_cache(maxsize=65536)
|
|
35
|
+
def _describe(terms: tuple) -> tuple[str, str, int] | None:
|
|
36
|
+
model = next((n.split(":", 1)[1] for n, _, v in terms if n.startswith("model:") and v), None)
|
|
37
|
+
if model is None:
|
|
38
|
+
return None # the task part contributes no setting-cost effect
|
|
39
|
+
effort = next((n.split(":", 1)[1] for n, _, _ in terms if n.startswith("effort:")), "default")
|
|
40
|
+
setting = sorted((n, float(v)) for n, _, v in terms if n.split(":", 1)[0] in STRUCTURAL)
|
|
41
|
+
key = hashlib.sha256(json.dumps(setting, separators=(",", ":")).encode()).hexdigest()
|
|
42
|
+
return key, canonical_model_id(model), EFFORT.get(effort, 4)
|
|
43
|
+
|
|
44
|
+
|
|
45
|
+
def describe(terms) -> tuple[str, str, int] | None:
|
|
46
|
+
return _describe(tuple(tuple(t) for t in terms))
|
|
47
|
+
|
|
48
|
+
|
|
49
|
+
def source_prior() -> dict:
|
|
50
|
+
"""Freeze source anchors and one shared price level, independent of query order.
|
|
51
|
+
|
|
52
|
+
Each source model contributes once to the shared level, at effort max.
|
|
53
|
+
This is the phase 1 anchor rule without weighting its level by whichever
|
|
54
|
+
candidate settings happen to be requested in a prediction batch.
|
|
55
|
+
"""
|
|
56
|
+
grouped = {}
|
|
57
|
+
for model, bench, effort, usd, _source in TB_COSTS:
|
|
58
|
+
# The source table writes numeric versions with hyphens. The belief
|
|
59
|
+
# and price table use dots outside Anthropic, just as the run reader.
|
|
60
|
+
if not model.startswith("claude-"):
|
|
61
|
+
model = re.sub(r"(\d)-(\d)", r"\1.\2", model)
|
|
62
|
+
model = canonical_model_id(model)
|
|
63
|
+
grouped.setdefault(model, []).append((bench, effort, usd))
|
|
64
|
+
tb = {}
|
|
65
|
+
for model, rows in grouped.items():
|
|
66
|
+
latest = max(r[0] for r in rows)
|
|
67
|
+
tb[model] = math.exp(float(np.mean([math.log(c) + .4 * (4 - e)
|
|
68
|
+
for b, e, c in rows if b == latest])))
|
|
69
|
+
table = load_prices()
|
|
70
|
+
prices = {}
|
|
71
|
+
for model in table.rates:
|
|
72
|
+
rate = table.rate(model)
|
|
73
|
+
if rate is not None:
|
|
74
|
+
prices[canonical_model_id(model)] = float(rate["input"] + rate["output"])
|
|
75
|
+
matched = sorted(set(tb) & set(prices))
|
|
76
|
+
level = (math.exp(float(np.mean([math.log(tb[m] / prices[m]) for m in matched]))) if matched else None)
|
|
77
|
+
return {"method": METHOD, "tb": tb, "prices": prices, "price_level": level,
|
|
78
|
+
"sigma0": .6, "nu0": 5., "widths": dict.fromkeys(GROUPS, 1.), "run_variance": {}}
|
|
79
|
+
|
|
80
|
+
|
|
81
|
+
def anchor(recipe: dict, model: str, effort: int) -> tuple[float, float]:
|
|
82
|
+
step = math.exp(.4 * (effort - 4))
|
|
83
|
+
tb = recipe["tb"].get(model)
|
|
84
|
+
price = recipe["prices"].get(model)
|
|
85
|
+
value = anchors(np.array([tb * step if tb is not None else np.nan]),
|
|
86
|
+
np.array([price * step if price is not None else np.nan]), level=recipe["price_level"])[0]
|
|
87
|
+
return float(value), .6 if tb is not None else 1.2
|
|
88
|
+
|
|
89
|
+
|
|
90
|
+
def row_terms(recipe: dict, terms) -> list[tuple]:
|
|
91
|
+
"""Map a structural cost row to its fixed centre and four Gaussian kernels.
|
|
92
|
+
|
|
93
|
+
Already mapped composite observations are accepted unchanged. Raw composite
|
|
94
|
+
observations must be mapped piece by piece before averaging their terms.
|
|
95
|
+
"""
|
|
96
|
+
if any(n == CONSTANT or n.startswith(PREFIXES) for n, _, _ in terms):
|
|
97
|
+
return list(terms)
|
|
98
|
+
desc = describe(terms)
|
|
99
|
+
if desc is None:
|
|
100
|
+
return []
|
|
101
|
+
key, model, effort = desc
|
|
102
|
+
a, sd = anchor(recipe, model, effort)
|
|
103
|
+
w = recipe["widths"]
|
|
104
|
+
features = [("clevel:store", .5 * w["level"]),
|
|
105
|
+
(f"ctransfer:{model}", sd * w["transfer"]),
|
|
106
|
+
(f"csetting:{key}", math.sqrt(.09 * w["setting"] ** 2 + 1e-9))]
|
|
107
|
+
features += [(f"cstep:{j}", .2 * w["steps"]) for j in range(1, 5) if effort >= j]
|
|
108
|
+
variance = sum(v * v for _, v in features)
|
|
109
|
+
center = math.log(a) - (variance + recipe["sigma0"] ** 2) / 2
|
|
110
|
+
return [(CONSTANT, None, center)] + [(n, None, v) for n, v in features]
|
|
111
|
+
|
|
112
|
+
|
|
113
|
+
def noise_variance(recipe: dict, terms) -> float:
|
|
114
|
+
desc = describe(terms)
|
|
115
|
+
key = desc[0] if desc is not None else next((n.split(":", 1)[1] for n, _, _ in terms
|
|
116
|
+
if n.startswith("csetting:")), None)
|
|
117
|
+
return float(recipe["run_variance"].get(key, recipe["sigma0"] ** 2))
|
|
118
|
+
|
|
119
|
+
|
|
120
|
+
def fit_recipe(hr, forest, *, eb: bool) -> dict:
|
|
121
|
+
recipe = source_prior()
|
|
122
|
+
cells = {}
|
|
123
|
+
for terms, y, weight, source in zip(hr.rows, hr.y, hr.w, hr.sources):
|
|
124
|
+
if source != "user":
|
|
125
|
+
continue
|
|
126
|
+
desc = describe(terms)
|
|
127
|
+
if desc is not None:
|
|
128
|
+
cells.setdefault(desc[0], [desc, terms, []])[2].append((float(y), float(weight)))
|
|
129
|
+
keys = sorted(cells)
|
|
130
|
+
summary = {"converged": True, "runs": 0, "widths": recipe["widths"]}
|
|
131
|
+
if keys:
|
|
132
|
+
descs = [cells[k][0] for k in keys]
|
|
133
|
+
model = np.array([d[1] for d in descs])
|
|
134
|
+
efforts = np.array([d[2] for d in descs])
|
|
135
|
+
values = [anchor(recipe, d[1], d[2]) for d in descs]
|
|
136
|
+
sd = np.array([v[1] for v in values])
|
|
137
|
+
E = (efforts[:, None] >= np.arange(1, 5)).astype(float)
|
|
138
|
+
components = {"level": np.full((len(keys), len(keys)), .25),
|
|
139
|
+
"transfer": (model[:, None] == model[None, :]) * np.outer(sd, sd),
|
|
140
|
+
"setting": np.eye(len(keys)) * .09, "steps": E @ E.T * .04}
|
|
141
|
+
cm = FrontierCost(np.array([v[0] for v in values]), components=components)
|
|
142
|
+
for i, key in enumerate(keys):
|
|
143
|
+
for y, weight in cells[key][2]:
|
|
144
|
+
cm.nc[i] += weight
|
|
145
|
+
cm.sx[i] += weight * y
|
|
146
|
+
cm.sxx[i] += weight * y * y
|
|
147
|
+
cm.csum[i] += weight * math.exp(y)
|
|
148
|
+
cm.fit()
|
|
149
|
+
if eb:
|
|
150
|
+
summary = cm.refit()
|
|
151
|
+
recipe["widths"] = dict(zip(GROUPS, cm.widths.tolist()))
|
|
152
|
+
recipe["run_variance"] = dict(zip(keys, cm.sig2.tolist()))
|
|
153
|
+
# Fit unit Gaussian coefficients after FrontierCost has fitted their widths.
|
|
154
|
+
# A deterministic centre is stored as coefficient 1 with negligible variance,
|
|
155
|
+
# so the existing sparse fit reader and its exact rank-one updates still apply.
|
|
156
|
+
mapped = [row_terms(recipe, cells[k][1]) for k in keys]
|
|
157
|
+
nodes = [CONSTANT] + sorted({n for row in mapped for n, _, _ in row if n != CONSTANT})
|
|
158
|
+
index = {n: i for i, n in enumerate(nodes)}
|
|
159
|
+
X = np.zeros((len(keys), len(nodes)))
|
|
160
|
+
for i, row in enumerate(mapped):
|
|
161
|
+
for n, _, value in row:
|
|
162
|
+
X[i, index[n]] += value
|
|
163
|
+
Z = X[:, 1:]
|
|
164
|
+
precision = np.eye(len(nodes))
|
|
165
|
+
precision[0, 0] = 1e18
|
|
166
|
+
mean = np.zeros(len(nodes))
|
|
167
|
+
mean[0] = 1.
|
|
168
|
+
if keys:
|
|
169
|
+
xbar, _ = cm._statistics()
|
|
170
|
+
weights = cm.nc / cm.sig2
|
|
171
|
+
precision[1:, 1:] += Z.T @ (weights[:, None] * Z)
|
|
172
|
+
mean[1:] = np.linalg.solve(precision[1:, 1:], Z.T @ (weights * (xbar - X[:, 0])))
|
|
173
|
+
for node in nodes:
|
|
174
|
+
forest.add(node, None)
|
|
175
|
+
support = {n: [len({run for row, run, source in zip(hr.rows, hr.runs, hr.sources)
|
|
176
|
+
if source == "user" and any(t[0] == n for t in row_terms(recipe, row))}), 0]
|
|
177
|
+
for n in nodes}
|
|
178
|
+
res = HeadFit("gaussian", mean, np.array(list(recipe["widths"].values())), sigma=recipe["sigma0"],
|
|
179
|
+
log_evidence=summary.get("log_evidence", 0.), precision=sparse.csr_matrix(precision), info=summary)
|
|
180
|
+
return {"name": "cost", "engine": "gaussian", "kind": "cost", "node_ids": nodes, "fit": res,
|
|
181
|
+
"phi": recipe["widths"], "center": 0., "scale": 1., "baseline": 0., "support": support,
|
|
182
|
+
"n_rows": sum(len(cells[k][2]) for k in keys), "n_runs": len(set(r for r, s in zip(hr.runs, hr.sources)
|
|
183
|
+
if s == "user")),
|
|
184
|
+
"factors": ["fixed source-only TB/price anchors; FrontierCost covariance and observed setting means"],
|
|
185
|
+
"recipe": recipe}
|