PyPI - evalrx - Versions diffs - 0.1.2__py3-none-any.whl - Mend

evalrx 0.1.2__py3-none-any.whl

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.

Files changed (339) hide show

evalrx/__init__.py +139 -0
evalrx/agent_assets/__init__.py +2 -0
evalrx/agent_assets/skills/README.md +28 -0
evalrx/agent_assets/skills/eval-chart-style/SKILL.md +172 -0
evalrx/agent_assets/skills/evalrx-report-ui/SKILL.md +116 -0
evalrx/agent_assets/skills/nature-figure/LICENSE +201 -0
evalrx/agent_assets/skills/nature-figure/README.md +412 -0
evalrx/agent_assets/skills/nature-figure/SKILL.md +60 -0
evalrx/agent_assets/skills/nature-figure/manifest.yaml +59 -0
evalrx/agent_assets/skills/nature-figure/references/api.md +436 -0
evalrx/agent_assets/skills/nature-figure/references/backend-selection.md +100 -0
evalrx/agent_assets/skills/nature-figure/references/chart-types.md +281 -0
evalrx/agent_assets/skills/nature-figure/references/common-patterns.md +350 -0
evalrx/agent_assets/skills/nature-figure/references/demos.md +65 -0
evalrx/agent_assets/skills/nature-figure/references/design-theory.md +439 -0
evalrx/agent_assets/skills/nature-figure/references/figure-contract.md +93 -0
evalrx/agent_assets/skills/nature-figure/references/figure-legend-conventions.md +71 -0
evalrx/agent_assets/skills/nature-figure/references/nature-2026-observations.md +112 -0
evalrx/agent_assets/skills/nature-figure/references/qa-contract.md +119 -0
evalrx/agent_assets/skills/nature-figure/references/r-template-index.md +66 -0
evalrx/agent_assets/skills/nature-figure/references/r-workflow.md +161 -0
evalrx/agent_assets/skills/nature-figure/references/tutorials.md +251 -0
evalrx/agent_assets/skills/nature-figure/static/core/contract.md +29 -0
evalrx/agent_assets/skills/nature-figure/static/core/stance.md +37 -0
evalrx/agent_assets/skills/nature-figure/static/fragments/backend/python.md +37 -0
evalrx/agent_assets/skills/nature-figure/static/fragments/backend/r.md +44 -0
evalrx/agent_assets/skills/outcome-driver-analysis/SKILL.md +213 -0
evalrx/agent_assets/skills/outcome-driver-analysis/assets/analysis_report_template.md +53 -0
evalrx/agent_assets/skills/outcome-driver-analysis/references/model_selection.md +72 -0
evalrx/agent_assets/skills/outcome-driver-analysis/scripts/explanatory_var_eda.R +130 -0
evalrx/agent_assets/skills/outcome-driver-analysis/scripts/explanatory_var_eda.py +150 -0
evalrx/agent_assets/skills/outcome-driver-analysis/scripts/fit_outcome_model.R +181 -0
evalrx/agent_assets/skills/outcome-driver-analysis/scripts/fit_outcome_model.py +186 -0
evalrx/agent_assets/skills/outcome-driver-analysis/scripts/univariate_eda.R +149 -0
evalrx/agent_assets/skills/outcome-driver-analysis/scripts/univariate_eda.py +177 -0
evalrx/agent_assets/skills.py +27 -0
evalrx/agent_runtime/__init__.py +78 -0
evalrx/agent_runtime/_docker_runner.py +89 -0
evalrx/agent_runtime/cli_runtime.py +103 -0
evalrx/agent_runtime/cli_transcript.py +138 -0
evalrx/agent_runtime/cli_types.py +68 -0
evalrx/agent_runtime/codegen/__init__.py +5 -0
evalrx/agent_runtime/codegen/runner.py +94 -0
evalrx/agent_runtime/experiment_harness.py +117 -0
evalrx/agent_runtime/factory.py +102 -0
evalrx/agent_runtime/json_shape.py +44 -0
evalrx/agent_runtime/judges/__init__.py +28 -0
evalrx/agent_runtime/judges/agy.py +179 -0
evalrx/agent_runtime/judges/autodetect.py +135 -0
evalrx/agent_runtime/judges/claude.py +159 -0
evalrx/agent_runtime/judges/codex.py +120 -0
evalrx/agent_runtime/providers/__init__.py +21 -0
evalrx/agent_runtime/providers/antigravity.py +31 -0
evalrx/agent_runtime/providers/base.py +145 -0
evalrx/agent_runtime/providers/claude_code.py +49 -0
evalrx/agent_runtime/providers/codex.py +37 -0
evalrx/agent_runtime/providers/gemini_cli.py +26 -0
evalrx/agent_runtime/providers/kimi_cli.py +27 -0
evalrx/agent_runtime/providers/opencode.py +27 -0
evalrx/agent_runtime/providers/registry.py +58 -0
evalrx/agent_runtime/sandbox.py +517 -0
evalrx/agent_runtime/skill_audit.py +143 -0
evalrx/agent_runtime/skills/__init__.py +19 -0
evalrx/agent_runtime/skills/installer.py +68 -0
evalrx/agent_runtime/skills/prompt_policy.py +86 -0
evalrx/agent_runtime/skills/resolver.py +19 -0
evalrx/analysis/__init__.py +132 -0
evalrx/analysis/adjudicate.py +154 -0
evalrx/analysis/analysis_module.py +361 -0
evalrx/analysis/api.py +171 -0
evalrx/analysis/case_studio.py +651 -0
evalrx/analysis/cli.py +114 -0
evalrx/analysis/dashboard.py +350 -0
evalrx/analysis/eval_case_matrix.py +118 -0
evalrx/analysis/eval_viz_theme.py +833 -0
evalrx/analysis/explore_run.py +333 -0
evalrx/analysis/explorer.py +1276 -0
evalrx/analysis/failure_modes.py +607 -0
evalrx/analysis/fused_pipeline.py +489 -0
evalrx/analysis/holdout.py +300 -0
evalrx/analysis/hypothesis_agent.py +230 -0
evalrx/analysis/narration.py +177 -0
evalrx/analysis/operationalize.py +442 -0
evalrx/analysis/plain_language.py +42 -0
evalrx/analysis/planner.py +283 -0
evalrx/analysis/probe_search.py +203 -0
evalrx/analysis/profile.py +268 -0
evalrx/analysis/prompts/__init__.py +0 -0
evalrx/analysis/prompts/explorer.py +417 -0
evalrx/analysis/prompts/failure_modes.py +33 -0
evalrx/analysis/prompts/holdout.py +27 -0
evalrx/analysis/prompts/hypothesis_agent.py +78 -0
evalrx/analysis/prompts/run_codebase.py +47 -0
evalrx/analysis/prompts/stats_agent.py +72 -0
evalrx/analysis/prompts/stats_tool_generator.py +43 -0
evalrx/analysis/result_marker.py +47 -0
evalrx/analysis/run_codebase.py +242 -0
evalrx/analysis/run_view.py +205 -0
evalrx/analysis/stage_views.py +93 -0
evalrx/analysis/stats_agent.py +944 -0
evalrx/analysis/stats_tool_agent.py +261 -0
evalrx/analysis/stats_tool_generator.py +415 -0
evalrx/analysis/stats_tools.py +1153 -0
evalrx/analysis/trajectory_records.py +193 -0
evalrx/analysis/workbench.py +431 -0
evalrx/analyzers/__init__.py +42 -0
evalrx/analyzers/agent/__init__.py +25 -0
evalrx/analyzers/agent/counterfactual.py +84 -0
evalrx/analyzers/agent/first_error_judge.py +96 -0
evalrx/analyzers/agent/ignored_obs.py +81 -0
evalrx/analyzers/agent/loop_detect.py +79 -0
evalrx/analyzers/agent/reliability.py +165 -0
evalrx/analyzers/agent/tool_shap.py +225 -0
evalrx/analyzers/agent/trajectory_rubric.py +168 -0
evalrx/analyzers/attention/__init__.py +19 -0
evalrx/analyzers/attention/relative_attn.py +610 -0
evalrx/analyzers/attention/rollout.py +73 -0
evalrx/analyzers/attention/sink.py +56 -0
evalrx/analyzers/attention/summary.py +190 -0
evalrx/analyzers/attribution/__init__.py +6 -0
evalrx/analyzers/attribution/generic_attn.py +31 -0
evalrx/analyzers/attribution/gradcam.py +30 -0
evalrx/analyzers/base.py +12 -0
evalrx/analyzers/geometry/__init__.py +6 -0
evalrx/analyzers/geometry/cka.py +70 -0
evalrx/analyzers/geometry/linear_probe.py +157 -0
evalrx/analyzers/hallucination/__init__.py +9 -0
evalrx/analyzers/hallucination/chair.py +78 -0
evalrx/analyzers/hallucination/opera.py +29 -0
evalrx/analyzers/hallucination/pope.py +119 -0
evalrx/analyzers/hallucination/selfcheck.py +155 -0
evalrx/analyzers/hallucination/vcd.py +29 -0
evalrx/analyzers/lens/__init__.py +7 -0
evalrx/analyzers/lens/layer_contrast.py +133 -0
evalrx/analyzers/lens/logit_lens.py +138 -0
evalrx/analyzers/lens/tuned_lens.py +30 -0
evalrx/analyzers/patching/__init__.py +5 -0
evalrx/analyzers/patching/causal_trace.py +30 -0
evalrx/analyzers/perturbation/__init__.py +23 -0
evalrx/analyzers/perturbation/_shapley.py +54 -0
evalrx/analyzers/perturbation/context_shap.py +174 -0
evalrx/analyzers/perturbation/cot_faithfulness.py +239 -0
evalrx/analyzers/perturbation/format_sensitivity.py +237 -0
evalrx/analyzers/perturbation/mm_shap.py +146 -0
evalrx/analyzers/perturbation/modality_ablation.py +196 -0
evalrx/analyzers/perturbation/perturbation_battery.py +274 -0
evalrx/analyzers/perturbation/prompt_contrast.py +265 -0
evalrx/analyzers/perturbation/rise.py +94 -0
evalrx/analyzers/perturbation/vl_shap.py +102 -0
evalrx/analyzers/reasoning/__init__.py +33 -0
evalrx/analyzers/reasoning/_text.py +328 -0
evalrx/analyzers/reasoning/answer_extraction_audit.py +327 -0
evalrx/analyzers/reasoning/arith_audit.py +226 -0
evalrx/analyzers/reasoning/contamination.py +214 -0
evalrx/analyzers/reasoning/knowledge_split.py +253 -0
evalrx/analyzers/reasoning/self_repair.py +246 -0
evalrx/analyzers/reasoning/step_rollout_value.py +216 -0
evalrx/analyzers/reasoning/termination_audit.py +258 -0
evalrx/analyzers/uncertainty/__init__.py +18 -0
evalrx/analyzers/uncertainty/calibration.py +174 -0
evalrx/analyzers/uncertainty/coverage_gap.py +199 -0
evalrx/analyzers/uncertainty/entropy.py +90 -0
evalrx/analyzers/uncertainty/logprob_entropy.py +69 -0
evalrx/analyzers/uncertainty/self_consistency.py +204 -0
evalrx/analyzers/uncertainty/verbalized_conf.py +64 -0
evalrx/cli.py +411 -0
evalrx/config.py +77 -0
evalrx/contract/__init__.py +179 -0
evalrx/contract/common.py +452 -0
evalrx/contract/emit.py +948 -0
evalrx/contract/export.py +237 -0
evalrx/contract/m1.py +325 -0
evalrx/contract/m2.py +317 -0
evalrx/contract/m3.py +165 -0
evalrx/contract/m4.py +130 -0
evalrx/contract/m5.py +292 -0
evalrx/contract/methodology.py +76 -0
evalrx/contract/pre_m1.py +58 -0
evalrx/contract/typescript.py +140 -0
evalrx/core/__init__.py +85 -0
evalrx/core/analyzer.py +174 -0
evalrx/core/capability.py +54 -0
evalrx/core/case.py +443 -0
evalrx/core/experiment.py +106 -0
evalrx/core/model.py +198 -0
evalrx/core/pipeline.py +42 -0
evalrx/core/registry.py +142 -0
evalrx/core/result.py +64 -0
evalrx/core/spec.py +173 -0
evalrx/core/tokentype.py +165 -0
evalrx/core/tool.py +92 -0
evalrx/datasets/__init__.py +41 -0
evalrx/datasets/base.py +68 -0
evalrx/datasets/gui_os.py +52 -0
evalrx/datasets/llm_qa.py +57 -0
evalrx/datasets/pure_qa.py +12 -0
evalrx/datasets/vlm_qa.py +695 -0
evalrx/datasets/web_search_qa.py +52 -0
evalrx/eval_agent/__init__.py +341 -0
evalrx/eval_agent/_tools.py +81 -0
evalrx/eval_agent/ab_runner.py +50 -0
evalrx/eval_agent/agentic/__init__.py +43 -0
evalrx/eval_agent/agentic/actions.py +216 -0
evalrx/eval_agent/agentic/board.py +107 -0
evalrx/eval_agent/agentic/loop.py +190 -0
evalrx/eval_agent/agentic/tools.py +538 -0
evalrx/eval_agent/checkpoint.py +57 -0
evalrx/eval_agent/cli_agent.py +59 -0
evalrx/eval_agent/cli_skills.py +5 -0
evalrx/eval_agent/evolution.py +396 -0
evalrx/eval_agent/git_manager.py +215 -0
evalrx/eval_agent/hypothesis.py +172 -0
evalrx/eval_agent/label_quarantine.py +209 -0
evalrx/eval_agent/legacy.py +530 -0
evalrx/eval_agent/log_schema.py +497 -0
evalrx/eval_agent/loop.py +2159 -0
evalrx/eval_agent/loop_reports.py +116 -0
evalrx/eval_agent/model_instrumentation.py +282 -0
evalrx/eval_agent/narration.py +193 -0
evalrx/eval_agent/nl_runner.py +460 -0
evalrx/eval_agent/orchestrator.py +61 -0
evalrx/eval_agent/preregister.py +93 -0
evalrx/eval_agent/prompts/__init__.py +1 -0
evalrx/eval_agent/prompts/agentic.py +46 -0
evalrx/eval_agent/prompts/case_discovery.py +25 -0
evalrx/eval_agent/prompts/diagnosis.py +125 -0
evalrx/eval_agent/prompts/experiment_writer.py +265 -0
evalrx/eval_agent/prompts/explore_step.py +37 -0
evalrx/eval_agent/prompts/fix_agent.py +257 -0
evalrx/eval_agent/prompts/hypothesis_tester.py +15 -0
evalrx/eval_agent/prompts/nl_runner.py +38 -0
evalrx/eval_agent/prompts/probe_agent.py +25 -0
evalrx/eval_agent/prompts/probe_candidate_generator.py +14 -0
evalrx/eval_agent/prompts/probe_generator.py +35 -0
evalrx/eval_agent/prompts/whitebox_probe_generator.py +38 -0
evalrx/eval_agent/report.py +58 -0
evalrx/eval_agent/run_context.py +354 -0
evalrx/eval_agent/run_log.schema.json +1215 -0
evalrx/eval_agent/run_logger_v2.py +1764 -0
evalrx/eval_agent/run_metadata.py +208 -0
evalrx/eval_agent/stages/__init__.py +56 -0
evalrx/eval_agent/stages/case_discovery.py +293 -0
evalrx/eval_agent/stages/diagnosis.py +1017 -0
evalrx/eval_agent/stages/experiment_writer.py +1634 -0
evalrx/eval_agent/stages/fix_agent.py +3916 -0
evalrx/eval_agent/stages/fix_internals.py +499 -0
evalrx/eval_agent/stages/fix_pipeline.py +725 -0
evalrx/eval_agent/stages/fix_tiers.py +187 -0
evalrx/eval_agent/stages/fix_tools.py +1034 -0
evalrx/eval_agent/stages/hypothesis_tester.py +1014 -0
evalrx/eval_agent/stages/probe.py +439 -0
evalrx/eval_agent/stages/probe_agent.py +1079 -0
evalrx/eval_agent/stages/probe_candidate_generator.py +128 -0
evalrx/eval_agent/stages/probe_generator.py +326 -0
evalrx/eval_agent/stages/probe_search_agent.py +106 -0
evalrx/eval_agent/stages/protocol.py +112 -0
evalrx/eval_agent/stages/repair_catalog.py +273 -0
evalrx/eval_agent/stages/surgery.py +524 -0
evalrx/eval_agent/stages/whitebox_probe_generator.py +351 -0
evalrx/eval_agent/store.py +231 -0
evalrx/logging_utils.py +112 -0
evalrx/models/__init__.py +161 -0
evalrx/models/_discover.py +101 -0
evalrx/models/agent.py +380 -0
evalrx/models/backends/__init__.py +58 -0
evalrx/models/backends/api.py +169 -0
evalrx/models/backends/base.py +57 -0
evalrx/models/backends/gemini_compat.py +579 -0
evalrx/models/backends/hf_local.py +2074 -0
evalrx/models/backends/openai_compat.py +301 -0
evalrx/models/backends/vllm_offline.py +116 -0
evalrx/models/base.py +24 -0
evalrx/models/blackbox/__init__.py +4 -0
evalrx/models/blackbox/agent.py +31 -0
evalrx/models/blackbox/base.py +29 -0
evalrx/models/blackbox/gemini.py +279 -0
evalrx/models/blackbox/llm_api.py +17 -0
evalrx/models/blackbox/vlm_api.py +17 -0
evalrx/models/compose.py +66 -0
evalrx/models/inference.py +88 -0
evalrx/models/paper_methods/__init__.py +8 -0
evalrx/models/paper_methods/aad.py +53 -0
evalrx/models/paper_methods/ifcd.py +204 -0
evalrx/models/paper_methods/pai.py +164 -0
evalrx/models/paper_methods/tcd.py +202 -0
evalrx/models/paper_methods/vcd.py +45 -0
evalrx/models/paper_methods/vicrop.py +137 -0
evalrx/models/toolcodec.py +143 -0
evalrx/models/tools/__init__.py +20 -0
evalrx/models/tools/perception.py +300 -0
evalrx/models/tools/visual.py +174 -0
evalrx/models/whitebox/__init__.py +26 -0
evalrx/models/whitebox/agent.py +31 -0
evalrx/models/whitebox/base.py +24 -0
evalrx/models/whitebox/qwen.py +61 -0
evalrx/models/whitebox/qwen2_5_omni.py +29 -0
evalrx/models/whitebox/qwen2_audio.py +25 -0
evalrx/models/whitebox/qwen_omni.py +53 -0
evalrx/models/whitebox/qwen_vl.py +62 -0
evalrx/observability/__init__.py +21 -0
evalrx/observability/envelope.py +122 -0
evalrx/observability/outbox.py +111 -0
evalrx/observability/tracer.py +882 -0
evalrx/reporting/__init__.py +28 -0
evalrx/reporting/case_study.py +947 -0
evalrx/reporting/compiler.py +587 -0
evalrx/reporting/dynamic.py +1882 -0
evalrx/reporting/html_report.py +2225 -0
evalrx/reporting/langfuse_exporter.py +38 -0
evalrx/reporting/langfuse_source.py +155 -0
evalrx/reporting/model.py +151 -0
evalrx/reporting/run_events.py +184 -0
evalrx/reporting/server.py +557 -0
evalrx/reporting/stages.py +58 -0
evalrx/reporting/static_export.py +142 -0
evalrx/reporting/web_dist/index.html +146 -0
evalrx/specs.py +727 -0
evalrx/stats/__init__.py +47 -0
evalrx/stats/api.py +192 -0
evalrx/stats/bootstrap.py +86 -0
evalrx/stats/ebh.py +27 -0
evalrx/stats/evalue.py +98 -0
evalrx/stats/friedman.py +138 -0
evalrx/stats/mcnemar.py +40 -0
evalrx/stats/multiplicity.py +159 -0
evalrx/stats/subset_sampling.py +55 -0
evalrx/term_links.py +43 -0
evalrx/viz/__init__.py +7 -0
evalrx/viz/labels.py +77 -0
evalrx/viz/prompts.py +39 -0
evalrx/viz/renderer.py +590 -0
evalrx/viz/schema.py +36 -0
evalrx/viz/style.py +134 -0
evalrx-0.1.2.dist-info/METADATA +532 -0
evalrx-0.1.2.dist-info/RECORD +339 -0
evalrx-0.1.2.dist-info/WHEEL +5 -0
evalrx-0.1.2.dist-info/entry_points.txt +3 -0
evalrx-0.1.2.dist-info/licenses/LICENSE +121 -0
evalrx-0.1.2.dist-info/top_level.txt +1 -0

evalrx/eval_agent/run_log.schema.json ADDED Viewed

@@ -0,0 +1,1215 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "https://evalrx.dev/schemas/run_log.schema.json",
+  "title": "EvalRX run event",
+  "description": "One logged event, wherever a run stores it (run.json / M<n>/log.json). Discriminated on `event`; matches schema_version 6. Permissive: extra (additive) fields are allowed.",
+  "$defs": {
+    "envelope": {
+      "type": "object",
+      "required": [
+        "event",
+        "schema_version",
+        "ts",
+        "trace_id",
+        "event_seq",
+        "stage",
+        "span_id"
+      ],
+      "properties": {
+        "event": {
+          "type": "string",
+          "enum": [
+            "run_start",
+            "probe",
+            "explore",
+            "analysis",
+            "diagnosis",
+            "surgery",
+            "experiment",
+            "fix",
+            "loop_end",
+            "tool_codegen",
+            "tool_registry",
+            "agent_decision",
+            "agent_tool",
+            "case_record",
+            "report_published",
+            "stage_skipped",
+            "model_call",
+            "diagnose_report",
+            "unrouted"
+          ]
+        },
+        "schema_version": {
+          "const": 6
+        },
+        "ts": {
+          "type": "string",
+          "format": "date-time",
+          "pattern": "^\\d{4}-\\d{2}-\\d{2}T\\d{2}:\\d{2}:\\d{2}(\\.\\d+)?(Z|[+-]\\d{2}:\\d{2})?$"
+        },
+        "trace_id": {
+          "type": "string"
+        },
+        "event_seq": {
+          "type": "integer",
+          "minimum": 1
+        },
+        "span_id": {
+          "type": "string"
+        },
+        "stage": {
+          "enum": [
+            "RUN",
+            "M1",
+            "M2",
+            "M3",
+            "M4",
+            "M5"
+          ]
+        },
+        "cycle": {
+          "type": "integer"
+        }
+      }
+    },
+    "run_start": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "run_start"
+            },
+            "evalrx_version": {
+              "type": "string"
+            },
+            "python_version": {
+              "type": "string"
+            },
+            "git_commit": {
+              "type": [
+                "string",
+                "null"
+              ]
+            },
+            "data_fingerprint": {
+              "type": "string"
+            },
+            "label_distribution": {
+              "type": "object"
+            },
+            "n_cases": {
+              "type": "integer"
+            },
+            "model": {
+              "type": "string"
+            },
+            "decision_judge": {
+              "type": "string"
+            },
+            "max_actions": {
+              "type": "integer"
+            },
+            "explorer": {
+              "type": "string"
+            }
+          },
+          "required": [
+            "event"
+          ]
+        }
+      ]
+    },
+    "probe": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "probe"
+            },
+            "analyzers": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
+            "findings": {
+              "type": "object"
+            },
+            "artifact_paths": {
+              "type": "object"
+            },
+            "selection_rationale": {
+              "type": "string"
+            },
+            "judge_io": {
+              "type": "object",
+              "properties": {
+                "prompt_path": {
+                  "type": "string"
+                },
+                "prompt_chars": {
+                  "type": "integer"
+                },
+                "raw_path": {
+                  "type": "string"
+                },
+                "raw_chars": {
+                  "type": "integer"
+                }
+              }
+            },
+            "duration_sec": {
+              "type": "number"
+            },
+            "n_model_calls": {
+              "type": "integer"
+            },
+            "model_calls_path": {
+              "type": "string"
+            }
+          },
+          "required": [
+            "event",
+            "cycle",
+            "analyzers",
+            "findings",
+            "artifact_paths"
+          ]
+        }
+      ]
+    },
+    "analysis": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "analysis"
+            },
+            "severity": {
+              "type": "string"
+            },
+            "n_findings": {
+              "type": "integer"
+            },
+            "findings": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
+            "narrative": {
+              "type": "string"
+            },
+            "conclusion": {
+              "type": "string"
+            },
+            "evidence_chain": {
+              "type": "array"
+            },
+            "stats_tool_results": {
+              "type": [
+                "object",
+                "array"
+              ]
+            },
+            "stats_plan": {
+              "type": [
+                "object",
+                "array"
+              ]
+            },
+            "stats_results": {
+              "type": [
+                "object",
+                "array"
+              ]
+            },
+            "corrected_rejections": {
+              "type": [
+                "object",
+                "array"
+              ]
+            },
+            "visualizations": {
+              "type": "array"
+            },
+            "figures": {
+              "type": "array"
+            },
+            "judge_io": {
+              "type": "object",
+              "properties": {
+                "prompt_path": {
+                  "type": "string"
+                },
+                "prompt_chars": {
+                  "type": "integer"
+                },
+                "raw_path": {
+                  "type": "string"
+                },
+                "raw_chars": {
+                  "type": "integer"
+                }
+              }
+            },
+            "duration_sec": {
+              "type": "number"
+            }
+          },
+          "required": [
+            "event",
+            "cycle",
+            "severity",
+            "n_findings",
+            "findings",
+            "narrative"
+          ]
+        }
+      ]
+    },
+    "explore": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "explore"
+            },
+            "ok": {
+              "type": "boolean"
+            },
+            "n_observations": {
+              "type": "integer"
+            },
+            "n_charts": {
+              "type": "integer"
+            },
+            "n_charts_rendered": {
+              "type": "integer"
+            },
+            "n_tables": {
+              "type": "integer"
+            },
+            "n_candidate_signals": {
+              "type": "integer"
+            },
+            "n_hypotheses": {
+              "type": "integer"
+            },
+            "adjudication": {
+              "type": "object"
+            },
+            "observations": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
+            "caveats": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
+            "figures": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
+            "attempts": {
+              "type": "integer"
+            },
+            "error": {
+              "type": "string"
+            },
+            "out_dir": {
+              "type": "string"
+            },
+            "report_path": {
+              "type": "string"
+            },
+            "duration_sec": {
+              "type": "number"
+            }
+          },
+          "required": [
+            "event",
+            "cycle",
+            "ok",
+            "n_observations",
+            "n_charts",
+            "n_tables"
+          ]
+        }
+      ]
+    },
+    "diagnosis": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "diagnosis"
+            },
+            "model_name": {
+              "type": [
+                "string",
+                "null"
+              ]
+            },
+            "n_hypotheses": {
+              "type": "integer"
+            },
+            "hypotheses": {
+              "type": "array",
+              "items": {
+                "type": "object",
+                "properties": {
+                  "statement": {
+                    "type": "string"
+                  },
+                  "plain_statement": {
+                    "type": [
+                      "string",
+                      "null"
+                    ]
+                  },
+                  "failure_mode": {
+                    "type": [
+                      "string",
+                      "null"
+                    ]
+                  },
+                  "status": {
+                    "type": [
+                      "string",
+                      "null"
+                    ]
+                  },
+                  "test_design": {
+                    "type": [
+                      "string",
+                      "null"
+                    ]
+                  }
+                }
+              }
+            },
+            "raw_judge_output": {
+              "type": [
+                "string",
+                "null"
+              ]
+            },
+            "referenced_charts": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
+            "explore_context_used": {
+              "type": "boolean"
+            },
+            "explore_figures": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
+            "failure_modes_used": {
+              "type": "boolean"
+            },
+            "judge_io": {
+              "type": "object",
+              "properties": {
+                "prompt_path": {
+                  "type": "string"
+                },
+                "prompt_chars": {
+                  "type": "integer"
+                },
+                "raw_path": {
+                  "type": "string"
+                },
+                "raw_chars": {
+                  "type": "integer"
+                }
+              }
+            },
+            "duration_sec": {
+              "type": "number"
+            }
+          },
+          "required": [
+            "event",
+            "cycle",
+            "model_name",
+            "n_hypotheses",
+            "hypotheses",
+            "raw_judge_output"
+          ]
+        }
+      ]
+    },
+    "surgery": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "surgery"
+            },
+            "module": {
+              "type": "string"
+            },
+            "hypothesis": {
+              "type": "string"
+            },
+            "failure_mode": {
+              "type": [
+                "string",
+                "null"
+              ]
+            },
+            "status": {
+              "type": "string"
+            },
+            "fixed": {
+              "type": "boolean"
+            },
+            "confidence_score": {
+              "type": [
+                "number",
+                "null"
+              ]
+            },
+            "evidence": {
+              "type": [
+                "object",
+                "null"
+              ]
+            },
+            "n_refocused_cases": {
+              "type": [
+                "integer",
+                "null"
+              ]
+            },
+            "duration_sec": {
+              "type": "number"
+            },
+            "judge_io": {
+              "type": "object",
+              "properties": {
+                "prompt_path": {
+                  "type": "string"
+                },
+                "prompt_chars": {
+                  "type": "integer"
+                },
+                "raw_path": {
+                  "type": "string"
+                },
+                "raw_chars": {
+                  "type": "integer"
+                }
+              }
+            }
+          },
+          "required": [
+            "event",
+            "cycle",
+            "module",
+            "hypothesis",
+            "failure_mode",
+            "status",
+            "fixed"
+          ]
+        }
+      ]
+    },
+    "experiment": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "experiment"
+            },
+            "module": {
+              "type": "string"
+            },
+            "hypothesis": {
+              "type": "string"
+            },
+            "failure_mode": {
+              "type": [
+                "string",
+                "null"
+              ]
+            },
+            "status": {
+              "type": [
+                "string",
+                "null"
+              ]
+            },
+            "fixed": {
+              "type": "boolean"
+            },
+            "provider": {
+              "type": [
+                "string",
+                "null"
+              ]
+            },
+            "metrics": {
+              "type": [
+                "object",
+                "null"
+              ]
+            },
+            "returncode": {
+              "type": [
+                "integer",
+                "null"
+              ]
+            },
+            "timed_out": {
+              "type": [
+                "boolean",
+                "null"
+              ]
+            },
+            "llm_calls": {
+              "type": [
+                "integer",
+                "null"
+              ]
+            },
+            "code_paths": {
+              "type": "object"
+            },
+            "output_paths": {
+              "type": "object"
+            },
+            "workspace_snapshot": {
+              "type": [
+                "object",
+                "null"
+              ]
+            },
+            "record": {
+              "type": "string"
+            }
+          },
+          "required": [
+            "event",
+            "cycle",
+            "module",
+            "hypothesis",
+            "failure_mode",
+            "status",
+            "fixed"
+          ]
+        }
+      ]
+    },
+    "fix": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "fix"
+            },
+            "module": {
+              "type": "string"
+            },
+            "record": {
+              "type": "string"
+            }
+          },
+          "required": [
+            "event",
+            "cycle",
+            "module"
+          ]
+        }
+      ]
+    },
+    "loop_end": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "loop_end"
+            },
+            "cycles": {
+              "type": "integer"
+            },
+            "tokens_used": {
+              "type": [
+                "integer",
+                "null"
+              ]
+            },
+            "timings_sec": {
+              "type": "object"
+            },
+            "total_duration_sec": {
+              "type": "number"
+            },
+            "resolved": {
+              "type": "boolean"
+            },
+            "n_hypotheses": {
+              "type": "integer"
+            },
+            "final_hypotheses": {
+              "type": "array",
+              "items": {
+                "type": "object",
+                "properties": {
+                  "statement": {
+                    "type": "string"
+                  },
+                  "plain_statement": {
+                    "type": [
+                      "string",
+                      "null"
+                    ]
+                  },
+                  "failure_mode": {
+                    "type": [
+                      "string",
+                      "null"
+                    ]
+                  },
+                  "status": {
+                    "type": [
+                      "string",
+                      "null"
+                    ]
+                  },
+                  "test_design": {
+                    "type": [
+                      "string",
+                      "null"
+                    ]
+                  }
+                }
+              }
+            },
+            "stopped_by": {
+              "type": [
+                "string",
+                "null"
+              ]
+            },
+            "n_verified": {
+              "type": "integer"
+            },
+            "verified_hypotheses": {
+              "type": "array"
+            }
+          },
+          "required": [
+            "event",
+            "cycles"
+          ]
+        }
+      ]
+    },
+    "tool_codegen": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "tool_codegen"
+            },
+            "module": {
+              "type": "string"
+            },
+            "tool_name": {
+              "type": "string"
+            },
+            "need": {
+              "type": "string"
+            },
+            "source": {
+              "type": "string"
+            },
+            "ok": {
+              "type": "boolean"
+            },
+            "error": {
+              "type": [
+                "string",
+                "null"
+              ]
+            },
+            "code_paths": {
+              "type": "object"
+            }
+          },
+          "required": [
+            "event",
+            "cycle",
+            "module",
+            "tool_name",
+            "need",
+            "source",
+            "ok"
+          ]
+        }
+      ]
+    },
+    "tool_registry": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "tool_registry"
+            },
+            "module": {
+              "type": "string"
+            },
+            "n_tools": {
+              "type": "integer"
+            },
+            "tools": {
+              "type": "array"
+            }
+          },
+          "required": [
+            "event",
+            "cycle",
+            "module",
+            "n_tools",
+            "tools"
+          ]
+        }
+      ]
+    },
+    "agent_decision": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "agent_decision"
+            },
+            "step": {
+              "type": "integer"
+            },
+            "action": {
+              "type": "string"
+            },
+            "params": {
+              "type": "object"
+            },
+            "rationale": {
+              "type": "string"
+            },
+            "valid": {
+              "type": "boolean"
+            },
+            "repair_attempts": {
+              "type": "integer"
+            },
+            "fallback_used": {
+              "type": "boolean"
+            },
+            "judge_io": {
+              "type": "object",
+              "properties": {
+                "prompt_path": {
+                  "type": "string"
+                },
+                "prompt_chars": {
+                  "type": "integer"
+                },
+                "raw_path": {
+                  "type": "string"
+                },
+                "raw_chars": {
+                  "type": "integer"
+                }
+              }
+            },
+            "duration_sec": {
+              "type": "number"
+            }
+          },
+          "required": [
+            "event",
+            "step",
+            "action",
+            "valid"
+          ]
+        }
+      ]
+    },
+    "agent_tool": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "agent_tool"
+            },
+            "step": {
+              "type": "integer"
+            },
+            "tool": {
+              "type": "string"
+            },
+            "ok": {
+              "type": "boolean"
+            },
+            "summary": {
+              "type": "string"
+            },
+            "error": {
+              "type": [
+                "string",
+                "null"
+              ]
+            },
+            "duration_sec": {
+              "type": "number"
+            }
+          },
+          "required": [
+            "event",
+            "step",
+            "tool",
+            "ok"
+          ]
+        }
+      ]
+    },
+    "case_record": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "case_record"
+            },
+            "case_id": {
+              "type": "string"
+            },
+            "case": {
+              "type": "object"
+            },
+            "media_paths": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            }
+          },
+          "required": [
+            "event",
+            "case_id",
+            "case",
+            "media_paths"
+          ]
+        }
+      ]
+    },
+    "report_published": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "report_published"
+            },
+            "report_schema_version": {
+              "type": "integer"
+            },
+            "catalog_version": {
+              "type": "string"
+            },
+            "json_render_version": {
+              "type": "string"
+            },
+            "source_event_seq": {
+              "type": "integer"
+            },
+            "sha256": {
+              "type": "string"
+            },
+            "generated_by": {
+              "type": "object"
+            },
+            "report_paths": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            }
+          },
+          "required": [
+            "event",
+            "report_schema_version",
+            "catalog_version",
+            "sha256",
+            "generated_by"
+          ]
+        }
+      ]
+    },
+    "stage_skipped": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "stage_skipped"
+            },
+            "stage": {
+              "type": "string"
+            },
+            "reason_code": {
+              "type": "string"
+            },
+            "detail": {
+              "type": "string"
+            }
+          },
+          "required": [
+            "event",
+            "stage",
+            "cycle",
+            "reason_code"
+          ]
+        }
+      ]
+    },
+    "model_call": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "model_call"
+            },
+            "duration_sec": {
+              "type": "number"
+            },
+            "error": {
+              "type": [
+                "string",
+                "null"
+              ]
+            }
+          },
+          "required": [
+            "event",
+            "cycle",
+            "inputs",
+            "output"
+          ],
+          "anyOf": [
+            {
+              "required": [
+                "analyzer",
+                "method",
+                "call_index"
+              ]
+            },
+            {
+              "required": [
+                "role",
+                "operation"
+              ]
+            }
+          ]
+        }
+      ]
+    },
+    "diagnose_report": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "diagnose_report"
+            },
+            "hypotheses": {
+              "type": "array"
+            },
+            "m4_results": {
+              "type": "array"
+            },
+            "discovery": {
+              "type": "array"
+            }
+          },
+          "required": [
+            "event",
+            "cycles",
+            "hypotheses",
+            "m4_results",
+            "discovery"
+          ]
+        }
+      ]
+    },
+    "unrouted": {
+      "allOf": [
+        {
+          "$ref": "#/$defs/envelope"
+        },
+        {
+          "type": "object",
+          "properties": {
+            "event": {
+              "const": "unrouted"
+            },
+            "key": {
+              "type": "string"
+            },
+            "tag": {
+              "type": [
+                "string",
+                "null"
+              ]
+            }
+          },
+          "required": [
+            "event",
+            "key",
+            "tag"
+          ]
+        }
+      ]
+    }
+  },
+  "oneOf": [
+    {
+      "$ref": "#/$defs/run_start"
+    },
+    {
+      "$ref": "#/$defs/probe"
+    },
+    {
+      "$ref": "#/$defs/analysis"
+    },
+    {
+      "$ref": "#/$defs/explore"
+    },
+    {
+      "$ref": "#/$defs/diagnosis"
+    },
+    {
+      "$ref": "#/$defs/surgery"
+    },
+    {
+      "$ref": "#/$defs/experiment"
+    },
+    {
+      "$ref": "#/$defs/fix"
+    },
+    {
+      "$ref": "#/$defs/loop_end"
+    },
+    {
+      "$ref": "#/$defs/tool_codegen"
+    },
+    {
+      "$ref": "#/$defs/tool_registry"
+    },
+    {
+      "$ref": "#/$defs/agent_decision"
+    },
+    {
+      "$ref": "#/$defs/agent_tool"
+    },
+    {
+      "$ref": "#/$defs/case_record"
+    },
+    {
+      "$ref": "#/$defs/report_published"
+    },
+    {
+      "$ref": "#/$defs/stage_skipped"
+    },
+    {
+      "$ref": "#/$defs/model_call"
+    },
+    {
+      "$ref": "#/$defs/diagnose_report"
+    },
+    {
+      "$ref": "#/$defs/unrouted"
+    }
+  ]
+}