code-mower 0.5.0b5__py3-none-any.whl

code_mower/calibration/__init__.py

@@ -0,0 +1,181 @@
+"""Calibration runner internals.
+
+The stable user surface is still the ``code-mower calibration`` CLI. These
+modules are intentionally small seams for corpus parsing, evidence handling,
+metrics, pilot planning, result normalization, truth matching, and
+policy/reporting code as the runner is split out of the legacy command adapter.
+"""
+
+from .corpus import load_corpus, load_json_object, parse_int
+from .evidence import (
+    KNOWN_EVIDENCE_DISPOSITIONS,
+    NON_BLOCKING_CODERABBIT_SEVERITIES,
+    USEFUL_EVIDENCE_DISPOSITIONS,
+)
+from .evidence_report import build_reviewer_evidence_report, render_evidence_text
+from .arms import DEFAULT_CLI_LANES, DEFAULT_LOCAL_LLM_PROFILES, default_arms
+from .commands import (
+    command_lane_id,
+    command_metadata_for_run,
+    local_llm_profiles_from_command,
+    materialize_command,
+    option_value,
+    parse_repo_path_map,
+    repo_path_for_item,
+    resolve_path_for_cwd,
+    reviewer_id_from_command,
+    summary_path_for_command,
+    text_from_timeout_stream,
+)
+from .context_inputs import (
+    CONTEXT_PACK_CLI_LANES,
+    changed_files_from_checkout,
+    context_pack_file_for_command,
+    render_materialized_context_pack_prompt_text,
+    repo_roots_from_path_map,
+    selected_context_pack_manifest,
+)
+from .identity import head_slug, safe_slug
+from .metrics import float_or_zero
+from .overlap import build_overlap_report, finding_key, render_overlap_text
+from .policy import (
+    MERGE_GATE_MIN_CLEAN_RUNS,
+    MERGE_GATE_MIN_FINDINGS,
+    MERGE_GATE_USEFUL_RATE,
+    SELECTIVE_USEFUL_RATE,
+    build_lane_policy_report,
+    render_policy_text,
+)
+from .planning import build_pilot_plan, render_plan_text
+from .results import (
+    AUDIT_INPUT_INSUFFICIENT_PATTERNS,
+    audit_input_insufficient_result,
+    coderabbit_blocking_findings,
+    infra_run_record,
+    local_llm_findings,
+    run_records_from_summary,
+)
+from .run_results import (
+    CALIBRATION_RUN_RESULTS_MODE,
+    CALIBRATION_RUN_RESULTS_SCHEMA,
+    apply_run_disposition_rules,
+    corpus_with_run_results,
+    csv_values,
+    load_run_results,
+    normalize_disposition,
+    run_matches_disposition_rule,
+)
+from .runner import (
+    default_code_mower_command,
+    load_summary,
+    result_command_dir,
+    run_calibration_commands,
+    utc_now_iso,
+    write_json,
+)
+from .run_status import (
+    RUN_STATUS_AUDIT_INPUT_INSUFFICIENT,
+    RUN_STATUS_BLOCKED,
+    RUN_STATUS_CATEGORY_ALIASES,
+    RUN_STATUS_INFRA_ERROR,
+    RUN_STATUS_PASS,
+    RUN_STATUS_UNKNOWN,
+    count_normalized_findings,
+    normalize_run_status_category,
+    status_from_verdict,
+)
+from .truth import (
+    TRUTH_EXPECTATION_ALIASES,
+    TRUTH_EXPECTATION_KNOWN_BLOCKED,
+    TRUTH_EXPECTATION_KNOWN_CLEAN,
+    TRUTH_EXPECTATION_UNKNOWN,
+    expected_finding_matches,
+    normalize_truth,
+    normalize_truth_expectation,
+    truth_for_item,
+)
+from .value_report import build_value_report, render_value_report_text
+
+__all__ = [
+    "KNOWN_EVIDENCE_DISPOSITIONS",
+    "AUDIT_INPUT_INSUFFICIENT_PATTERNS",
+    "CALIBRATION_RUN_RESULTS_MODE",
+    "CALIBRATION_RUN_RESULTS_SCHEMA",
+    "CONTEXT_PACK_CLI_LANES",
+    "DEFAULT_CLI_LANES",
+    "DEFAULT_LOCAL_LLM_PROFILES",
+    "MERGE_GATE_MIN_CLEAN_RUNS",
+    "MERGE_GATE_MIN_FINDINGS",
+    "MERGE_GATE_USEFUL_RATE",
+    "NON_BLOCKING_CODERABBIT_SEVERITIES",
+    "RUN_STATUS_AUDIT_INPUT_INSUFFICIENT",
+    "RUN_STATUS_BLOCKED",
+    "RUN_STATUS_CATEGORY_ALIASES",
+    "RUN_STATUS_INFRA_ERROR",
+    "RUN_STATUS_PASS",
+    "RUN_STATUS_UNKNOWN",
+    "SELECTIVE_USEFUL_RATE",
+    "TRUTH_EXPECTATION_ALIASES",
+    "TRUTH_EXPECTATION_KNOWN_BLOCKED",
+    "TRUTH_EXPECTATION_KNOWN_CLEAN",
+    "TRUTH_EXPECTATION_UNKNOWN",
+    "USEFUL_EVIDENCE_DISPOSITIONS",
+    "build_lane_policy_report",
+    "build_reviewer_evidence_report",
+    "build_value_report",
+    "build_pilot_plan",
+    "command_lane_id",
+    "command_metadata_for_run",
+    "context_pack_file_for_command",
+    "count_normalized_findings",
+    "changed_files_from_checkout",
+    "audit_input_insufficient_result",
+    "coderabbit_blocking_findings",
+    "apply_run_disposition_rules",
+    "corpus_with_run_results",
+    "csv_values",
+    "default_code_mower_command",
+    "default_arms",
+    "expected_finding_matches",
+    "finding_key",
+    "float_or_zero",
+    "head_slug",
+    "infra_run_record",
+    "load_summary",
+    "load_corpus",
+    "load_json_object",
+    "load_run_results",
+    "local_llm_findings",
+    "local_llm_profiles_from_command",
+    "materialize_command",
+    "normalize_disposition",
+    "normalize_run_status_category",
+    "normalize_truth",
+    "normalize_truth_expectation",
+    "option_value",
+    "parse_int",
+    "parse_repo_path_map",
+    "build_overlap_report",
+    "render_overlap_text",
+    "render_evidence_text",
+    "render_materialized_context_pack_prompt_text",
+    "render_plan_text",
+    "render_policy_text",
+    "render_value_report_text",
+    "run_records_from_summary",
+    "run_matches_disposition_rule",
+    "repo_path_for_item",
+    "repo_roots_from_path_map",
+    "result_command_dir",
+    "resolve_path_for_cwd",
+    "reviewer_id_from_command",
+    "run_calibration_commands",
+    "safe_slug",
+    "selected_context_pack_manifest",
+    "status_from_verdict",
+    "summary_path_for_command",
+    "text_from_timeout_stream",
+    "truth_for_item",
+    "utc_now_iso",
+    "write_json",
+]

code_mower/calibration/arms.py

@@ -0,0 +1,267 @@
+"""Built-in calibration arm catalog and command-lane defaults."""
+
+from __future__ import annotations
+
+from typing import Any
+
+
+DEFAULT_LOCAL_LLM_PROFILES = (
+    "qwen3-coder-next-lmstudio",
+    "gemma4-ollama",
+)
+DEFAULT_CLI_LANES = ("gemini_cli", "antigravity_cli", "hermes_cli", "coderabbit_cli")
+
+
+def default_arms() -> list[dict[str, Any]]:
+    return [
+        {
+            "arm_id": "topology-baseline",
+            "kind": "cross-provider",
+            "reviewers": [
+                {"reviewer_id": "codex-audit", "provider": "codex", "lenses": ["base-audit"]},
+                {"reviewer_id": "claude-audit", "provider": "claude", "lenses": ["base-audit"]},
+            ],
+            "purpose": "Measure decorrelation from different model families.",
+        },
+        {
+            "arm_id": "same-provider-lenses",
+            "kind": "lens-shift",
+            "reviewers": [
+                {"reviewer_id": "claude-base-audit", "provider": "claude", "lenses": ["base-audit"]},
+                {
+                    "reviewer_id": "claude-package-runtime",
+                    "provider": "claude",
+                    "lenses": ["base-audit", "package-runtime"],
+                },
+            ],
+            "purpose": "Measure whether prompt lenses create useful disagreement.",
+        },
+        {
+            "arm_id": "same-provider-doctrine-lenses",
+            "kind": "lens-shift",
+            "reviewers": [
+                {"reviewer_id": "claude-base-audit", "provider": "claude", "lenses": ["base-audit"]},
+                {
+                    "reviewer_id": "claude-generic-programming",
+                    "provider": "claude",
+                    "lenses": ["base-audit", "generic-programming"],
+                },
+                {
+                    "reviewer_id": "claude-context-driven-quality",
+                    "provider": "claude",
+                    "lenses": ["base-audit", "context-driven-quality"],
+                },
+            ],
+            "purpose": (
+                "Measure whether the same model with materially different review "
+                "doctrine creates useful disagreement beyond the same-model noise floor."
+            ),
+        },
+        {
+            "arm_id": "same-provider-risk-ops-lenses",
+            "kind": "lens-shift",
+            "reviewers": [
+                {"reviewer_id": "claude-base-audit", "provider": "claude", "lenses": ["base-audit"]},
+                {
+                    "reviewer_id": "claude-security-threat-model",
+                    "provider": "claude",
+                    "lenses": ["base-audit", "security-threat-model"],
+                },
+                {
+                    "reviewer_id": "claude-operability",
+                    "provider": "claude",
+                    "lenses": ["base-audit", "operability"],
+                },
+            ],
+            "purpose": (
+                "Measure whether security and operability lenses catch useful "
+                "production-risk findings beyond the base audit and same-model noise floor."
+            ),
+        },
+        {
+            "arm_id": "gemini-risk-ops-lens-fanout",
+            "kind": "executable-lens-fanout",
+            "reviewers": [
+                {
+                    "reviewer_id": "gemini-base-audit",
+                    "provider": "local-cli",
+                    "lane_id": "gemini_cli",
+                    "lenses": ["base-audit"],
+                },
+                {
+                    "reviewer_id": "gemini-security-threat-model",
+                    "provider": "local-cli",
+                    "lane_id": "gemini_cli",
+                    "lenses": ["base-audit", "security-threat-model"],
+                },
+                {
+                    "reviewer_id": "gemini-operability",
+                    "provider": "local-cli",
+                    "lane_id": "gemini_cli",
+                    "lenses": ["base-audit", "operability"],
+                },
+            ],
+            "purpose": (
+                "Execute Gemini CLI against the same head with base, security, "
+                "and operability lenses so lens evidence can be collected without "
+                "manual command fan-out."
+            ),
+            "requires_explicit_arm": True,
+        },
+        {
+            "arm_id": "gemini-doctrine-lens-fanout",
+            "kind": "executable-lens-fanout",
+            "reviewers": [
+                {
+                    "reviewer_id": "gemini-base-audit",
+                    "provider": "local-cli",
+                    "lane_id": "gemini_cli",
+                    "lenses": ["base-audit"],
+                },
+                {
+                    "reviewer_id": "gemini-generic-programming",
+                    "provider": "local-cli",
+                    "lane_id": "gemini_cli",
+                    "lenses": ["base-audit", "generic-programming"],
+                },
+                {
+                    "reviewer_id": "gemini-context-driven-quality",
+                    "provider": "local-cli",
+                    "lane_id": "gemini_cli",
+                    "lenses": ["base-audit", "context-driven-quality"],
+                },
+            ],
+            "purpose": (
+                "Execute Gemini CLI against the same head with base, generic "
+                "programming, and context-driven quality lenses so same-model "
+                "doctrine shifts can be measured with real lane outputs."
+            ),
+            "requires_explicit_arm": True,
+        },
+        {
+            "arm_id": "antigravity-risk-ops-lens-fanout",
+            "kind": "executable-lens-fanout",
+            "reviewers": [
+                {
+                    "reviewer_id": "antigravity-base-audit",
+                    "provider": "local-cli",
+                    "lane_id": "antigravity_cli",
+                    "lenses": ["base-audit"],
+                },
+                {
+                    "reviewer_id": "antigravity-security-threat-model",
+                    "provider": "local-cli",
+                    "lane_id": "antigravity_cli",
+                    "lenses": ["base-audit", "security-threat-model"],
+                },
+                {
+                    "reviewer_id": "antigravity-operability",
+                    "provider": "local-cli",
+                    "lane_id": "antigravity_cli",
+                    "lenses": ["base-audit", "operability"],
+                },
+            ],
+            "purpose": (
+                "Execute Antigravity CLI against the same head with base, "
+                "security, and operability lenses so the forward Google CLI "
+                "lane can be calibrated separately from Gemini compatibility."
+            ),
+            "requires_explicit_arm": True,
+        },
+        {
+            "arm_id": "antigravity-doctrine-lens-fanout",
+            "kind": "executable-lens-fanout",
+            "reviewers": [
+                {
+                    "reviewer_id": "antigravity-base-audit",
+                    "provider": "local-cli",
+                    "lane_id": "antigravity_cli",
+                    "lenses": ["base-audit"],
+                },
+                {
+                    "reviewer_id": "antigravity-generic-programming",
+                    "provider": "local-cli",
+                    "lane_id": "antigravity_cli",
+                    "lenses": ["base-audit", "generic-programming"],
+                },
+                {
+                    "reviewer_id": "antigravity-context-driven-quality",
+                    "provider": "local-cli",
+                    "lane_id": "antigravity_cli",
+                    "lenses": ["base-audit", "context-driven-quality"],
+                },
+            ],
+            "purpose": (
+                "Execute Antigravity CLI against the same head with base, "
+                "generic programming, and context-driven quality lenses so "
+                "same-provider doctrine shifts can be measured on the forward "
+                "Google CLI lane."
+            ),
+            "requires_explicit_arm": True,
+        },
+        {
+            "arm_id": "hermes-doctrine-lens-fanout",
+            "kind": "executable-lens-fanout",
+            "reviewers": [
+                {
+                    "reviewer_id": "hermes-base-audit",
+                    "provider": "local-cli",
+                    "lane_id": "hermes_cli",
+                    "lenses": ["base-audit"],
+                },
+                {
+                    "reviewer_id": "hermes-generic-programming",
+                    "provider": "local-cli",
+                    "lane_id": "hermes_cli",
+                    "lenses": ["base-audit", "generic-programming"],
+                },
+                {
+                    "reviewer_id": "hermes-context-driven-quality",
+                    "provider": "local-cli",
+                    "lane_id": "hermes_cli",
+                    "lenses": ["base-audit", "context-driven-quality"],
+                },
+            ],
+            "purpose": (
+                "Execute Hermes CLI against the same head with base, generic "
+                "programming, and context-driven quality lenses so its one-shot "
+                "agent profile can be compared with Gemini/Claude doctrine "
+                "evidence without granting merge authority."
+            ),
+            "requires_explicit_arm": True,
+        },
+        {
+            "arm_id": "same-provider-control",
+            "kind": "noise-floor",
+            "reviewers": [
+                {"reviewer_id": "claude-base-audit-a", "provider": "claude", "lenses": ["base-audit"]},
+                {"reviewer_id": "claude-base-audit-b", "provider": "claude", "lenses": ["base-audit"]},
+            ],
+            "purpose": "Measure repeated-run variance for the same model and same lens.",
+        },
+        {
+            "arm_id": "local-cli-models",
+            "kind": "informational-bakeoff",
+            "reviewers": [
+                *(
+                    {
+                        "reviewer_id": profile_id,
+                        "provider": "local-llm",
+                        "profile_id": profile_id,
+                        "lenses": ["base-audit"],
+                    }
+                    for profile_id in DEFAULT_LOCAL_LLM_PROFILES
+                ),
+                *(
+                    {
+                        "reviewer_id": lane_id,
+                        "provider": "local-cli",
+                        "lane_id": lane_id,
+                        "lenses": ["base-audit"],
+                    }
+                    for lane_id in DEFAULT_CLI_LANES
+                ),
+            ],
+            "purpose": "Compare cheap informational reviewers before promotion.",
+        },
+    ]