devagent-ai 0.8.4__tar.gz → 0.8.6__tar.gz

{devagent_ai-0.8.4/devagent_ai.egg-info → devagent_ai-0.8.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: devagent-ai
-Version: 0.8.4
+Version: 0.8.6
 Summary: Evidence-driven local autonomous software engineering agent
 Author: Tom Ha
 Maintainer: Tom Ha
@@ -9,6 +9,7 @@ Project-URL: Homepage, https://github.com/tomha85/devagent
 Project-URL: Repository, https://github.com/tomha85/devagent
 Project-URL: Issues, https://github.com/tomha85/devagent/issues
 Project-URL: Changelog, https://github.com/tomha85/devagent/blob/main/CHANGELOG.md
+Project-URL: Sponsor, https://github.com/sponsors/tomha85
 Keywords: ai,agent,developer-tools,software-engineering,verification,local-first
 Classifier: Development Status :: 4 - Beta
 Classifier: Intended Audience :: Developers
@@ -23,6 +24,7 @@ License-File: LICENSE
 License-File: NOTICE
 Requires-Dist: openai>=1.40.0
 Requires-Dist: anthropic>=0.34.0
+Requires-Dist: cryptography>=42.0.0
 Requires-Dist: tomli>=2.0.0; python_version < "3.11"
 Provides-Extra: dev
 Requires-Dist: pytest>=8.0; extra == "dev"
@@ -35,11 +37,14 @@ Dynamic: license-file
 [![Python 3.10+](https://img.shields.io/badge/Python-3.10%2B-blue.svg)](https://www.python.org/)
 [![Status: Beta](https://img.shields.io/badge/status-beta-blue.svg)](#project-status)
 [![Production CI](https://github.com/tomha85/devagent/actions/workflows/ci.yml/badge.svg)](https://github.com/tomha85/devagent/actions/workflows/ci.yml)
+[![Sponsor DevAgent](https://img.shields.io/badge/Sponsor-DevAgent-EA4AAA?logo=githubsponsors&logoColor=white)](https://github.com/sponsors/tomha85)
 
 **A local, evidence-driven software engineering agent that turns a requirement into a tested, independently reviewed patch, prints a developer-grade engineering report, and only publishes a branch when the result is `VERIFIED`.**
 
 > **From requirement to evidence-backed verified branch.**
 
+> ❤️ **DevAgent is free during public beta.** If DevAgent saves you engineering time, consider [sponsoring continued development](https://github.com/sponsors/tomha85).
+
 DevAgent runs against a developer's local repository. It discovers the application, gathers source evidence, compiles explicit acceptance criteria, plans a bounded change, creates backups, implements the minimum necessary patch, runs repository-supported verification, independently reviews the final diff, and decides `VERIFIED`, `PARTIALLY_VERIFIED`, or `BLOCKED` from evidence rather than model confidence.
 
 For a normal verified run the deterministic harness prints the complete engineering report first, then commits reviewed paths and fast-forward pushes the developer's current non-protected branch. If the developer starts on `main`, `master`, or `trunk`, DevAgent creates a new safe branch instead. Runtime DevAgent never creates a pull request, merges, rebases, force-pushes, or deploys.
@@ -545,6 +550,23 @@ Remaining work is primarily **breadth and external validation**, not missing cor
 
 The project intentionally prioritizes trustworthy outcomes, reproducible evidence, and safe engineering behavior over feature count or unsupported "best agent" claims.
 
+## ❤️ Support DevAgent
+
+DevAgent is currently free to use during public beta.
+
+If DevAgent saves you engineering time or helps you deliver safer, better-verified software or PLC engineering work, you can support continued development through [GitHub Sponsors](https://github.com/sponsors/tomha85).
+
+Your sponsorship helps fund:
+
+- new engineering and PLC capabilities;
+- Siemens and Rockwell verification;
+- additional AI provider support;
+- regression and production qualification;
+- documentation and examples;
+- continued free public releases.
+
+[![Sponsor DevAgent](https://img.shields.io/badge/Sponsor-DevAgent-EA4AAA?logo=githubsponsors&logoColor=white)](https://github.com/sponsors/tomha85)
+
 ## Contributing
 
 Contributions are welcome. Read [CONTRIBUTING.md](CONTRIBUTING.md) before opening a pull request. Changes to safety, verification, reporting, acceptance semantics, providers, or bounded publication behavior should include regression evidence and must not give model-generated actions unrestricted Git publishing authority.

{devagent_ai-0.8.4 → devagent_ai-0.8.6}/README.md

@@ -4,11 +4,14 @@
 [![Python 3.10+](https://img.shields.io/badge/Python-3.10%2B-blue.svg)](https://www.python.org/)
 [![Status: Beta](https://img.shields.io/badge/status-beta-blue.svg)](#project-status)
 [![Production CI](https://github.com/tomha85/devagent/actions/workflows/ci.yml/badge.svg)](https://github.com/tomha85/devagent/actions/workflows/ci.yml)
+[![Sponsor DevAgent](https://img.shields.io/badge/Sponsor-DevAgent-EA4AAA?logo=githubsponsors&logoColor=white)](https://github.com/sponsors/tomha85)
 
 **A local, evidence-driven software engineering agent that turns a requirement into a tested, independently reviewed patch, prints a developer-grade engineering report, and only publishes a branch when the result is `VERIFIED`.**
 
 > **From requirement to evidence-backed verified branch.**
 
+> ❤️ **DevAgent is free during public beta.** If DevAgent saves you engineering time, consider [sponsoring continued development](https://github.com/sponsors/tomha85).
+
 DevAgent runs against a developer's local repository. It discovers the application, gathers source evidence, compiles explicit acceptance criteria, plans a bounded change, creates backups, implements the minimum necessary patch, runs repository-supported verification, independently reviews the final diff, and decides `VERIFIED`, `PARTIALLY_VERIFIED`, or `BLOCKED` from evidence rather than model confidence.
 
 For a normal verified run the deterministic harness prints the complete engineering report first, then commits reviewed paths and fast-forward pushes the developer's current non-protected branch. If the developer starts on `main`, `master`, or `trunk`, DevAgent creates a new safe branch instead. Runtime DevAgent never creates a pull request, merges, rebases, force-pushes, or deploys.
@@ -514,6 +517,23 @@ Remaining work is primarily **breadth and external validation**, not missing cor
 
 The project intentionally prioritizes trustworthy outcomes, reproducible evidence, and safe engineering behavior over feature count or unsupported "best agent" claims.
 
+## ❤️ Support DevAgent
+
+DevAgent is currently free to use during public beta.
+
+If DevAgent saves you engineering time or helps you deliver safer, better-verified software or PLC engineering work, you can support continued development through [GitHub Sponsors](https://github.com/sponsors/tomha85).
+
+Your sponsorship helps fund:
+
+- new engineering and PLC capabilities;
+- Siemens and Rockwell verification;
+- additional AI provider support;
+- regression and production qualification;
+- documentation and examples;
+- continued free public releases.
+
+[![Sponsor DevAgent](https://img.shields.io/badge/Sponsor-DevAgent-EA4AAA?logo=githubsponsors&logoColor=white)](https://github.com/sponsors/tomha85)
+
 ## Contributing
 
 Contributions are welcome. Read [CONTRIBUTING.md](CONTRIBUTING.md) before opening a pull request. Changes to safety, verification, reporting, acceptance semantics, providers, or bounded publication behavior should include regression evidence and must not give model-generated actions unrestricted Git publishing authority.

{devagent_ai-0.8.4 → devagent_ai-0.8.6}/devagent/__init__.py

@@ -1,3 +1,3 @@
 """DevAgent: evidence-driven local software engineering automation."""
 
-__version__ = "0.8.4"
+__version__ = "0.8.6"

devagent_ai-0.8.6/devagent/__main__.py

@@ -0,0 +1,4 @@
+from devagent.entrypoint import main
+
+raise SystemExit(main())
+

devagent_ai-0.8.6/devagent/entrypoint.py

@@ -0,0 +1,22 @@
+from __future__ import annotations
+
+import sys
+from typing import Sequence
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    """Route domain subcommands without changing the existing software CLI."""
+
+    arguments = list(sys.argv[1:] if argv is None else argv)
+    if len(arguments) >= 2 and arguments[0] == "plc" and arguments[1] == "inspect":
+        from devagent.plc.inspect_cli import main as plc_inspect_main
+
+        return plc_inspect_main(arguments[2:])
+    if arguments and arguments[0] == "plc":
+        from devagent.plc.cli import main as plc_main
+
+        return plc_main(arguments[1:])
+
+    from devagent.cli import main as software_main
+
+    return software_main(arguments)

devagent_ai-0.8.6/devagent/plc/__init__.py

@@ -0,0 +1,185 @@
+"""Vendor-neutral PLC engineering and production verification foundations for DevAgent."""
+
+from devagent.plc import analysis as _analysis
+from devagent.plc.rockwell_compare_hardening import install as _install_rockwell_compare_hardening
+from devagent.plc.rockwell_alias_hardening import install as _install_rockwell_alias_hardening
+from devagent.plc.rockwell_entrypoint_hardening import install as _install_rockwell_entrypoint_hardening
+from devagent.plc.rockwell_st_v10 import install as _install_rockwell_st_v10
+
+# Install fail-closed compare, alias, controller-entrypoint, and ST guards before
+# production verification is imported. Every downstream Rockwell proof must
+# share the same canonical writer identity and executable-entrypoint model.
+_install_rockwell_compare_hardening()
+_install_rockwell_alias_hardening()
+_install_rockwell_entrypoint_hardening()
+_install_rockwell_st_v10()
+
+from devagent.plc.rockwell_compare_reachability_hardening import (
+    install as _install_rockwell_compare_reachability_hardening,
+)
+
+# The typed-compare theorem/FAT path is independent from boolean output-logic
+# normalization, so explicitly bind it to the same executable routine closure.
+_install_rockwell_compare_reachability_hardening()
+
+from devagent.plc.rockwell_requirement_hardening import install as _install_rockwell_requirement_hardening
+from devagent.plc.rockwell_v10_semantics import install as _install_rockwell_v10_semantics
+from devagent.plc.rockwell_action_requirements import install as _install_rockwell_action_requirements
+from devagent.plc.rockwell_risk_hardening import install as _install_rockwell_risk_hardening
+from devagent.plc.rockwell_closeout_gate_hardening import install as _install_rockwell_closeout_gate_hardening
+
+_install_rockwell_requirement_hardening()
+# V10 extends the already-hardened V9 theorem. It proves bounded OTL/OTU action
+# effects and may prove final retained scan state only inside the deliberately
+# narrow same-active-Main-RLL-routine ordering theorem. Wider scheduling remains
+# fail-closed until explicitly modeled.
+_install_rockwell_v10_semantics()
+# Explicit natural-language requirements may bind to deterministic MOV/COPY/
+# CLR/RES local action effects, but never to final scan/process behavior.
+_install_rockwell_action_requirements()
+_install_rockwell_risk_hardening()
+# Install the V9 support-contract guard before importing regression/production
+# modules. Those modules import safe_analysis by value, so the patched support
+# check must already be visible when safe_analysis is first loaded.
+_install_rockwell_closeout_gate_hardening()
+
+from devagent.plc.rockwell_core_review_v12 import install as _install_rockwell_core_review_v12
+
+# V12 makes the commercial engineering-review contract explicit: cause/effect,
+# unreachable logic, contradictory linear paths, and sequence-branch risks are
+# deterministic review outputs. They remain evidence-backed findings, not AI
+# guesses and not runtime PASS claims.
+_install_rockwell_core_review_v12()
+
+from devagent.plc.fat_procedure_v12 import install as _install_fat_procedure_v12
+
+# Every FAT candidate, including tests created later by requirement mapping,
+# must be an engineer-ready manual procedure. DevAgent plans the FAT; it does not
+# connect to or execute the engineer's external simulator/HIL/controller.
+_install_fat_procedure_v12()
+
+from devagent.plc.rockwell_regression_evidence_hardening import (
+    install as _install_rockwell_regression_evidence_hardening,
+    install_domain_evidence as _install_rockwell_regression_domain_evidence,
+)
+
+_install_rockwell_regression_evidence_hardening()
+
+from devagent.plc.rockwell_regression_case_compat_v12 import (
+    install as _install_rockwell_regression_case_compat_v12,
+)
+
+# FAT-plan regression is semantic and case-insensitive. Studio 5000 identifier
+# spelling changes alone must not force a retest or create a false regression.
+_install_rockwell_regression_case_compat_v12()
+
+from devagent.plc.rockwell_branch_coverage_v16 import install as _install_rockwell_branch_coverage_v16
+
+# V16 reconciles branch coverage with the full deterministic theorem set. A
+# neutral-text branch proven by the bounded data/compute action theorem counts
+# as modeled even when it is not an OTE/OTL/OTU boolean-output branch. Mixed or
+# partially understood branch grammars remain fail-closed and withheld.
+_install_rockwell_branch_coverage_v16()
+
+from devagent.plc.rockwell_system_service_install_v17 import (
+    install as _install_rockwell_system_service_v17,
+)
+
+# V17 does not widen static proof. Reachable GSV/SSV system-service logic stays
+# PARTIAL, but every such runtime-dependent gap receives an evidence-linked,
+# engineer-executed FAT procedure and a specific commissioning risk. The install
+# happens before production imports capture analysis/risk functions by value.
+_install_rockwell_system_service_v17()
+
+from devagent.plc.semantic_coverage_report import install as _install_semantic_coverage_report
+
+# Ensure any later CLI import of render_production_report receives the
+# reachability-aware semantic coverage augmentation.
+_install_semantic_coverage_report()
+
+from devagent.plc.rockwell_system_service_report_install_v17 import (
+    install as _install_rockwell_system_service_report_v17,
+)
+
+# Keep the public report renderer identity stable while adding an explicit V17
+# system-service runtime boundary inside the semantic coverage section.
+_install_rockwell_system_service_report_v17()
+
+from devagent.plc.four_contract_v13 import install as _install_four_contract_v13
+
+# V13 makes the commercial four-core contract explicit and testable:
+# engineering analysis, logic/risk review, five-area optimization recommendations,
+# and engineer-ready FAT planning. All optimization output is advisory only.
+_install_four_contract_v13()
+
+from devagent.plc.professional_report_install_v14 import install as _install_professional_report_v14
+
+# V14 adds a customer-facing executive layer while preserving every detailed
+# engineering/evidence section and the established public renderer identity.
+# This is presentation-only: it cannot change deterministic verdicts/readiness.
+_install_professional_report_v14()
+
+from devagent.plc.report_clarity_install_v16 import install as _install_report_clarity_v16
+
+# V16 clarifies project-only review and coverage terminology without changing
+# deterministic engineering, risk, FAT, or release-readiness decisions.
+_install_report_clarity_v16()
+
+from devagent.plc.agent_harness_install_v15 import install as _install_agent_harness_v15
+
+# V15 applies modern agent orchestration only to the probabilistic assistance
+# layer: bounded propose/critic/revise graphs, deterministic evidence guards, and
+# per-run trace capture. The PLC proof/risk/readiness core stays authoritative.
+_install_agent_harness_v15()
+
+from devagent.plc.siemens_integration_v1 import install as _install_siemens_integration_v1
+
+# Siemens V1 is a vendor branch over the already-qualified production contract.
+# It accepts offline TIA Portal Openness/XML and generated-source exports, proves
+# only bounded top-level SCL semantics, and leaves control-flow/LAD/FBD/GRAPH/STL
+# or protected behavior PARTIAL/OPAQUE. Qualified Rockwell dispatch still enters
+# the exact guarded Rockwell analyzer above.
+_install_siemens_integration_v1()
+
+from devagent.plc.siemens_scl_control_flow_v2 import install as _install_siemens_scl_control_flow_v2
+
+# Siemens V2 adds a deliberately bounded deterministic theorem for complete,
+# single-level IF/ELSIF/ELSE Boolean assignment chains. Missing/incomplete branch
+# assignments, nesting, CASE/loops, calls, cyclic/self-references, and unsupported
+# expressions stay fail-closed and require engineer-executed FAT evidence.
+_install_siemens_scl_control_flow_v2()
+
+from devagent.plc.siemens_scl_control_flow_hardening_v2 import (
+    install as _install_siemens_scl_control_flow_hardening_v2,
+)
+
+# A syntactically complete inner IF is still nested control flow. Do not allow an
+# inner chain to become independently FULL when its enclosing region is outside
+# the V2 single-level theorem.
+_install_siemens_scl_control_flow_hardening_v2()
+
+from devagent.plc.production_v5 import run_production_verification_v5
+from devagent.plc.siemens_v5_bindings_v1 import install as _install_siemens_v5_bindings_v1
+
+# Agent Harness V15 loads Production V5 before Siemens integration. Refresh only
+# V5's by-value shared production/evidence/review bindings after Siemens installs.
+# Siemens then reaches the vendor dispatcher while Rockwell remains behind the
+# exact qualified Rockwell functions selected by those vendor-aware wrappers.
+_install_siemens_v5_bindings_v1()
+
+# Production is now loaded; bind the stage-14 baseline evidence augmentation.
+_install_rockwell_regression_domain_evidence()
+from devagent.plc.safe_analysis import analyze_rockwell_l5x
+from devagent.plc.plc_dispatch import analyze_plc_project
+from devagent.plc.siemens_cli_install_v1 import install as _install_siemens_cli_v1
+
+# CLI parsing is loaded only after production_v5 exists, avoiding circular
+# imports while exposing both vendor input contracts through `devagent plc`.
+_install_siemens_cli_v1()
+
+# Keep existing imports from ``devagent.plc.analysis`` on the guarded Rockwell
+# public path. The base module remains reusable by ``safe_analysis`` without
+# recursive calls; vendor-neutral callers should use analyze_plc_project.
+_analysis.analyze_rockwell_l5x = analyze_rockwell_l5x
+
+__all__ = ["analyze_plc_project", "analyze_rockwell_l5x", "run_production_verification_v5"]

devagent_ai-0.8.6/devagent/plc/agent_harness_install_v15.py

@@ -0,0 +1,27 @@
+from __future__ import annotations
+
+from typing import Any, Callable
+
+from devagent.plc.agent_harness_v15 import begin_run_trace, end_run_trace
+from devagent.plc import production_v5
+
+_ORIGINAL: Callable[..., Any] = production_v5.run_production_verification_v5
+_INSTALLED = False
+
+
+def _run_with_harness_trace(*args: Any, **kwargs: Any):
+    trace, token = begin_run_trace()
+    try:
+        result = _ORIGINAL(*args, **kwargs)
+        result.ai_harness_trace = list(trace)
+        return result
+    finally:
+        end_run_trace(token)
+
+
+def install() -> None:
+    global _INSTALLED
+    if _INSTALLED:
+        return
+    production_v5.run_production_verification_v5 = _run_with_harness_trace
+    _INSTALLED = True

devagent_ai-0.8.6/devagent/plc/agent_harness_v15.py

@@ -0,0 +1,208 @@
+from __future__ import annotations
+
+from contextvars import ContextVar, Token
+from typing import Any
+
+from devagent.providers import ModelProvider
+
+# The PLC agent harness is deliberately small and explicit. It is not part of the
+# deterministic PLC proof engine. It only governs AI review candidates before
+# they are admitted as AI_CANDIDATE observations.
+MAX_REVIEW_ITERATIONS = 2
+_ACTIVE_TRACE: ContextVar[list[dict[str, Any]] | None] = ContextVar(
+    "devagent_plc_agent_harness_trace",
+    default=None,
+)
+
+REVIEW_CRITIC_SCHEMA: dict[str, Any] = {
+    "type": "object",
+    "additionalProperties": False,
+    "required": ["decisions"],
+    "properties": {
+        "decisions": {
+            "type": "array",
+            "maxItems": 24,
+            "items": {
+                "type": "object",
+                "additionalProperties": False,
+                "required": ["finding_id", "decision", "reason", "supported_evidence_ids"],
+                "properties": {
+                    "finding_id": {"type": "string", "minLength": 1},
+                    "decision": {"type": "string", "enum": ["ACCEPT", "REVISE", "REJECT"]},
+                    "reason": {"type": "string", "minLength": 1},
+                    "supported_evidence_ids": {
+                        "type": "array",
+                        "maxItems": 8,
+                        "items": {"type": "string", "minLength": 1},
+                    },
+                },
+            },
+        }
+    },
+}
+
+REQUIREMENT_CRITIC_SCHEMA: dict[str, Any] = {
+    "type": "object",
+    "additionalProperties": False,
+    "required": ["decisions"],
+    "properties": {
+        "decisions": {
+            "type": "array",
+            "maxItems": 50,
+            "items": {
+                "type": "object",
+                "additionalProperties": False,
+                "required": ["requirement_id", "decision", "reason", "supported_evidence_ids"],
+                "properties": {
+                    "requirement_id": {"type": "string", "minLength": 1},
+                    "decision": {"type": "string", "enum": ["ACCEPT", "REJECT"]},
+                    "reason": {"type": "string", "minLength": 1},
+                    "supported_evidence_ids": {
+                        "type": "array",
+                        "maxItems": 8,
+                        "items": {"type": "string", "minLength": 1},
+                    },
+                },
+            },
+        }
+    },
+}
+
+
+def begin_run_trace() -> tuple[list[dict[str, Any]], Token[list[dict[str, Any]] | None]]:
+    sink: list[dict[str, Any]] = []
+    return sink, _ACTIVE_TRACE.set(sink)
+
+
+def end_run_trace(token: Token[list[dict[str, Any]] | None]) -> None:
+    _ACTIVE_TRACE.reset(token)
+
+
+def trace(
+    sink: list[dict[str, Any]] | None,
+    *,
+    graph: str,
+    node: str,
+    iteration: int,
+    outcome: str,
+    detail: str,
+    counts: dict[str, int] | None = None,
+) -> None:
+    target = sink if sink is not None else _ACTIVE_TRACE.get()
+    if target is None:
+        return
+    event: dict[str, Any] = {
+        "graph": graph,
+        "node": node,
+        "iteration": iteration,
+        "outcome": outcome,
+        "detail": detail,
+    }
+    if counts:
+        event["counts"] = dict(sorted(counts.items()))
+    target.append(event)
+
+
+def critique_review_candidates(
+    provider: ModelProvider,
+    *,
+    candidates: list[dict[str, Any]],
+    evidence: list[dict[str, Any]],
+    iteration: int,
+    trace_sink: list[dict[str, Any]] | None,
+) -> dict[str, dict[str, Any]]:
+    if not candidates:
+        return {}
+    trace(
+        trace_sink,
+        graph="PLC_ENGINEERING_REVIEW",
+        node="CRITIC",
+        iteration=iteration,
+        outcome="START",
+        detail="Evaluate AI review candidates for evidence support, boundedness, and unsupported runtime/safety claims.",
+        counts={"candidates": len(candidates)},
+    )
+    response = provider.request(
+        role="plc_engineering_critic",
+        payload={
+            "instruction": (
+                "Act as an independent PLC review critic. Evaluate only the supplied candidate findings and evidence. "
+                "ACCEPT only when the finding is directly supported by its cited evidence and does not claim runtime behavior, safety certification, PASS, or release readiness. "
+                "Use REVISE when the core concern is supported but wording/claims overreach. REJECT unsupported findings. "
+                "supported_evidence_ids must be a subset of evidence actually supporting the candidate."
+            ),
+            "candidates": candidates,
+            "evidence": evidence,
+        },
+        schema=REVIEW_CRITIC_SCHEMA,
+    )
+    decisions = {
+        str(item["finding_id"]): item
+        for item in response.get("decisions", [])
+        if isinstance(item, dict) and item.get("finding_id")
+    }
+    trace(
+        trace_sink,
+        graph="PLC_ENGINEERING_REVIEW",
+        node="CRITIC",
+        iteration=iteration,
+        outcome="COMPLETE",
+        detail="Independent critic completed candidate evaluation.",
+        counts={
+            "accepted": sum(1 for item in decisions.values() if item.get("decision") == "ACCEPT"),
+            "revise": sum(1 for item in decisions.values() if item.get("decision") == "REVISE"),
+            "rejected": sum(1 for item in decisions.values() if item.get("decision") == "REJECT"),
+        },
+    )
+    return decisions
+
+
+def critique_requirement_mappings(
+    provider: ModelProvider,
+    *,
+    mappings: list[dict[str, Any]],
+    evidence: list[dict[str, Any]],
+    iteration: int,
+    trace_sink: list[dict[str, Any]] | None,
+) -> dict[str, dict[str, Any]]:
+    if not mappings:
+        return {}
+    trace(
+        trace_sink,
+        graph="PLC_REQUIREMENT_MAPPING",
+        node="CRITIC",
+        iteration=iteration,
+        outcome="START",
+        detail="Evaluate AI requirement trace candidates without permitting verification promotion.",
+        counts={"mappings": len(mappings)},
+    )
+    response = provider.request(
+        role="plc_requirement_critic",
+        payload={
+            "instruction": (
+                "Evaluate only whether each proposed requirement-to-evidence trace candidate is plausibly supported by the supplied evidence. "
+                "Never declare VERIFIED, PASS, SAFE, compliant, or release-ready. ACCEPT only grounded traceability candidates; otherwise REJECT."
+            ),
+            "mappings": mappings,
+            "evidence": evidence,
+        },
+        schema=REQUIREMENT_CRITIC_SCHEMA,
+    )
+    decisions = {
+        str(item["requirement_id"]): item
+        for item in response.get("decisions", [])
+        if isinstance(item, dict) and item.get("requirement_id")
+    }
+    trace(
+        trace_sink,
+        graph="PLC_REQUIREMENT_MAPPING",
+        node="CRITIC",
+        iteration=iteration,
+        outcome="COMPLETE",
+        detail="Independent requirement critic completed mapping evaluation.",
+        counts={
+            "accepted": sum(1 for item in decisions.values() if item.get("decision") == "ACCEPT"),
+            "rejected": sum(1 for item in decisions.values() if item.get("decision") == "REJECT"),
+        },
+    )
+    return decisions