OIC-SEMANTIC-001: bounded synthetic candidate extraction

AGENTS.md

@@ -1,8 +1,8 @@
 # Agent operating boundary
 
-This repository is a non-semantic infrastructure foundation. The semantic OIC
-code-start gate remains **BLOCKED** by `STATUS.md`. No agent may implement or
-claim document interpretation, candidate extraction, institutional admission,
+This repository includes an owner-opened, bounded semantic candidate-extraction
+slice defined by `docs/decisions/OIC-OWNER-DECISION-004.md`. No agent may claim
+general document interpretation, institutional admission,
 Open Control Envelope generation, Rego compilation, or runtime semantic
 decisions under this work order.
 
@@ -22,7 +22,9 @@ bash scripts/generate_sbom.sh
 bash scripts/wheel_smoke_test.sh
 ```
 
-Implemented CLI commands are limited to `oic validate-schema`,
+The only authorized semantic production path is
+`src/oic/candidate_extraction.py`, limited to the three synthetic Northstar
+sources. Implemented CLI commands are limited to `oic validate-schema`,
 `oic verify-bootstrap`, `oic verify-manifest`, and `oic doctor`. They verify
 infrastructure contracts only. Do not invent semantic commands, hosted APIs,
 gateways, remote context services, or MCP servers.

README.md

@@ -15,14 +15,17 @@ Target capability:
 > evidentiary standard required to authorize, explain, audit, or correct those
 > actions.
 
-**What exists today:** a tested, non-semantic Python infrastructure package for
+**What exists today:** a tested Python infrastructure package plus one bounded,
+deterministic candidate extractor for three synthetic Northstar sources. The
+infrastructure supports
 offline schema validation, historical bootstrap verification, current manifest
 verification, environment and gate diagnostics, and reproducible CI and SBOM
 checks. It is not a functioning institutional compiler.
 
-**What is blocked:** semantic implementation remains blocked pending the corpus
-provenance and ZTL/VEIP interface evidence listed in [`STATUS.md`](STATUS.md).
-Nothing in this README opens that gate or promotes the repository's maturity.
+**What is blocked:** every semantic capability beyond source → candidate
+normative unit remains blocked. Institutional admission, IR/OCE generation,
+runtime authorization, CA-3 extraction, and ZTL/VEIP runtime integration are
+not authorized. See [`STATUS.md`](STATUS.md).
 
 ## Capability being developed
 
@@ -295,17 +298,17 @@ verification, operational, and independent-review evidence.
 
 **STATUS: CURRENT**
 
-Current capabilities include non-semantic schemas and contracts,
+Current capabilities include schemas and contracts,
 manifest/integrity verification, the falsification harness, supply-chain
-controls, and the benchmark preflight specification.
+controls, the benchmark preflight specification, and the bounded synthetic
+candidate extractor.
 
-Exit requires complete governing-source rights and provenance plus sufficient
-ZTL and VEIP provisional-interface evidence to open the semantic code-start
-gate.
+The global corpus manifest remains incomplete and ZTL/VEIP runtime integration
+remains unauthorized.
 
 ### Stage 1 - Source-to-control reference
 
-**STATUS: NEXT - BLOCKED by the semantic code-start gate**
+**STATUS: FIRST BOUNDED SLICE OWNER-OPENED; REMAINDER BLOCKED**
 
 Target path:
 
@@ -372,19 +375,21 @@ conformance, and community governance.
 
 ## Current phase
 
-This repository authorizes contract-first, non-semantic infrastructure work.
-It currently implements four infrastructure CLI commands:
+This repository authorizes contract-first infrastructure work and the exact
+bounded extraction scope in `docs/decisions/OIC-OWNER-DECISION-004.md`. It
+currently implements four infrastructure CLI commands:
 
 - `oic validate-schema`
 - `oic verify-bootstrap`
 - `oic verify-manifest`
 - `oic doctor`
 
-It does not implement document interpretation, candidate extraction,
+It implements deterministic candidate extraction only for three authorized
+synthetic sources. It does not implement general document interpretation,
 institutional admission, Institutional IR production, Open Control Envelope
 generation, Rego compilation, or runtime semantic decisions.
 
-Run the safe non-semantic checks after the hash-locked installation described
+Run the checks after the hash-locked installation described
 in [`docs/operations/CI.md`](docs/operations/CI.md):
 
 ```bash
@@ -398,9 +403,9 @@ must not be normalized into success.
 
 ## First executable objective
 
-The first semantic objective remains blocked. When its prerequisites are
-authorized, a bounded set of public or synthetic procurement governing sources
-is intended to flow through:
+The first semantic objective is owner-opened only through source-anchored
+candidate normative units for the synthetic Northstar source set. Later stages
+remain blocked:
 
 ```text
 documents

STATUS.md

@@ -21,6 +21,17 @@ It does not authorize public quality, enterprise-readiness, legal-compliance, un
 | Public Lab restrictions visible before upload | PASS as specification | `LIMITATIONS.md`, `docs/architecture/LAB_RESTRICTIONS.md` |
 | Named owner for each implemented module | PASS for kickoff | `OWNERS.md` |
 
-**Semantic implementation gate:** BLOCKED until the preflight corpus manifest and ZTL/VEIP provisional-interface records are completed.
+SEMANTIC CODE-START GATE:
+OWNER-OPENED AT
+914830ceec70bde17004d2ccbbb13218ca44a89b
+
+FIRST AUTHORIZED SEMANTIC SCOPE:
+SOURCE → CANDIDATE NORMATIVE UNIT
+SYNTHETIC NORTHSTAR SOURCES ONLY
+
+Only the deterministic candidate-extraction slice is opened. Institutional
+admission, IR/OCE generation, runtime authorization and integration, CA-3
+semantic extraction, and all production or benchmark claims remain blocked or
+not established. Global manifest completeness remains `INCOMPLETE`.
 
 Infrastructure scaffolding and schema validation may proceed before that gate.

benchmarks/preflight/semantic-001/EXPECTED-CANDIDATES.json

@@ -0,0 +1,29 @@
+{
+  "classification": [
+    "OWNER-SPECIFIED PREIMPLEMENTATION EXPECTATION",
+    "NOT AN OBSERVED BENCHMARK RESULT",
+    "NOT INDEPENDENTLY ADJUDICATED"
+  ],
+  "expected_total": 14,
+  "expected_by_source": {
+    "SYN-NS-GOV-1": 11,
+    "SYN-NS-PROC-1": 1,
+    "SYN-NS-AMEND-2": 2
+  },
+  "candidates": [
+    {"label":"G1","source_id":"SYN-NS-GOV-1","unit_type":"definition","actor":"Requester","action":"means","object":"a fictional test actor","conditions":[]},
+    {"label":"G2","source_id":"SYN-NS-GOV-1","unit_type":"definition","actor":"Approver","action":"means","object":"a fictional test actor listed in the matrix","conditions":[]},
+    {"label":"G3","source_id":"SYN-NS-GOV-1","unit_type":"definition","actor":"Evidence","action":"means","object":"a deterministic record bound to the proposed action","conditions":[]},
+    {"label":"G4","source_id":"SYN-NS-GOV-1","unit_type":"condition","actor":null,"action":"TEST_PURCHASE_LOW","object":"approval","conditions":["Approver-A"]},
+    {"label":"G5","source_id":"SYN-NS-GOV-1","unit_type":"condition","actor":null,"action":"TEST_PURCHASE_HIGH","object":"approval","conditions":["Approver-A","Approver-B"]},
+    {"label":"G6","source_id":"SYN-NS-GOV-1","unit_type":"delegation","actor":"Approver-A","action":"delegate","object":"TEST_PURCHASE_LOW to Delegate-A","conditions":[]},
+    {"label":"G7","source_id":"SYN-NS-GOV-1","unit_type":"prohibition","actor":null,"action":"recognize","object":"other delegation","conditions":[]},
+    {"label":"G8","source_id":"SYN-NS-GOV-1","unit_type":"evidence_duty","actor":null,"action":"proposal_carries_evidence","object":null,"conditions":[]},
+    {"label":"G9","source_id":"SYN-NS-GOV-1","unit_type":"exception","actor":null,"action":"TEST_EMERGENCY","object":"omit Approver-B","conditions":["emergency record present","Approver-A approval present"]},
+    {"label":"G10","source_id":"SYN-NS-GOV-1","unit_type":"discretion","actor":"Approver-A","action":"refer","object":"any proposal for human review","conditions":[]},
+    {"label":"G11","source_id":"SYN-NS-GOV-1","unit_type":"prohibition","actor":null,"action":"treat referral as approval","object":null,"conditions":[]},
+    {"label":"P1","source_id":"SYN-NS-PROC-1","unit_type":"condition","actor":null,"action":"TEST_PURCHASE_HIGH","object":"approval","conditions":["Approver-A"]},
+    {"label":"A1","source_id":"SYN-NS-AMEND-2","unit_type":"condition","actor":null,"action":"TEST_PURCHASE_HIGH","object":"approval","conditions":["Approver-A","Approver-B"]},
+    {"label":"A2","source_id":"SYN-NS-AMEND-2","unit_type":"prohibition","actor":null,"action":"use as current evidence","object":"SYNTHETIC-NORTHSTAR-PROCEDURE-v1","conditions":[]}
+  ]
+}

docs/decisions/OIC-OWNER-DECISION-004.md

@@ -0,0 +1,44 @@
+# OIC Owner Decision 004 — First bounded semantic implementation
+
+- Decision date: 2026-08-28
+- Authorized base: `914830ceec70bde17004d2ccbbb13218ca44a89b`
+- Scope: `BOUNDED SEMANTIC CODE START`
+- Gate disposition: `OPEN` for the scope below only
+- Authorized production path: `src/oic/candidate_extraction.py`
+- Authorized sources: `SYN-NS-GOV-1`, `SYN-NS-PROC-1`, `SYN-NS-AMEND-2`
+
+## Exact owner authorization
+
+> I authorize the OIC semantic code-start gate to OPEN on
+> Institutional-Compiler main at exact commit
+> 914830ceec70bde17004d2ccbbb13218ca44a89b.
+>
+> This authorization permits bounded semantic implementation under
+> TDD-OIC-001 v1.1 and the admitted gate constraints.
+>
+> It does not authorize institutional admission, runtime authorization,
+> OCE execution, Rego/OPA execution, ZTL runtime integration, VEIP runtime
+> integration, production claims, or benchmark claims unless separately
+> authorized.
+
+## Authorization boundary
+
+- First authorized semantic production path: `src/oic/candidate_extraction.py`
+- Authorized source IDs: `SYN-NS-GOV-1`, `SYN-NS-PROC-1`, `SYN-NS-AMEND-2`
+- CA-3 semantic extraction: `NOT AUTHORIZED`
+- Institutional admission: `NOT AUTHORIZED`
+- Institutional IR: `NOT AUTHORIZED`
+- OCE generation or execution: `NOT AUTHORIZED`
+- Rego/OPA: `NOT AUTHORIZED`
+- ZTL runtime: `NOT AUTHORIZED`
+- VEIP runtime: `NOT AUTHORIZED`
+- Production claims: `NOT AUTHORIZED`
+- Benchmark claims: `NOT AUTHORIZED`
+
+Candidate output remains extracted, uncertain, unadmitted, and without machine
+confidence. The stale procedure remains visible as superseded. Network/model/API
+dependencies and licensing or corpus-rights changes are not authorized. Global
+manifest completeness remains `INCOMPLETE`.
+
+This decision does not permit the evaluator or implementation to extend its own
+scope. Further semantic paths or sources require a separate owner decision.

docs/gates/OIC-SEMANTIC-CODE-START-GATE-OPEN-v0.1.json

@@ -0,0 +1,21 @@
+{
+  "gate_id": "OIC-SEMANTIC-CODE-START-GATE-OPEN-v0.1",
+  "state": "OPEN",
+  "owner_decision": "docs/decisions/OIC-OWNER-DECISION-004.md",
+  "authorization_base_sha": "914830ceec70bde17004d2ccbbb13218ca44a89b",
+  "authorized_semantic_paths": ["src/oic/candidate_extraction.py"],
+  "authorized_source_ids": ["SYN-NS-GOV-1", "SYN-NS-PROC-1", "SYN-NS-AMEND-2"],
+  "authorized_scope": "SOURCE_TO_CANDIDATE_NORMATIVE_UNIT",
+  "forbidden_capabilities": [
+    "INSTITUTIONAL_ADMISSION",
+    "INSTITUTIONAL_IR",
+    "OPEN_CONTROL_ENVELOPE",
+    "POLICY_TARGET_GENERATION",
+    "RUNTIME_AUTHORIZATION",
+    "ZTL_RUNTIME",
+    "VEIP_RUNTIME",
+    "CA_3_SEMANTIC_EXTRACTION",
+    "NETWORK_MODEL_API_DEPENDENCY"
+  ],
+  "global_manifest_status": "INCOMPLETE"
+}

docs/operations/FOUNDATION.md

@@ -265,8 +265,9 @@ enforcement artifact (none exists to produce).
 
 ## 8. Current limitations
 
-- **This is not a compiler.** No semantic implementation exists. See `LIMITATIONS.md`.
-- **The semantic implementation gate is BLOCKED** and nothing here opens it.
+- **This is not a compiler.** One bounded synthetic candidate extractor exists.
+- **The semantic code-start gate is OWNER-OPENED** only for that extractor and
+  those three synthetic sources; institutional admission remains blocked.
 - **Preflight corpus provenance is OPEN.** `SOURCE_MANIFEST.csv` has no rows.
 - **ZTL and VEIP are PROVISIONAL / NOT CONFIGURED.** Their interfaces are unfrozen
   (ADR-009, ADR-010) and no adapter, container, or call exists.

scripts/falsify_infrastructure.py

@@ -23,8 +23,8 @@
         "tests/contract/test_semantic_conformance.py::test_reject_duplicated_reason_codes",
     ),
     (
-        "semantic code-start remains blocked",
-        "tests/contract/test_claims_discipline.py::test_operator_guide_states_the_semantic_gate_is_blocked",
+        "semantic code-start scope expansion rejection",
+        "tests/contract/test_semantic_code_start_gate_closure.py::test_owner_gate_refuses_allowlist_self_extension",
     ),
 )
 

scripts/verify_code_start_gate.py

@@ -1,5 +1,5 @@
 #!/usr/bin/env python3
-"""Fail-closed verifier for the non-semantic code-start evidence package."""
+"""Fail-closed verifier for the owner-opened bounded semantic code-start gate."""
 
 from __future__ import annotations
 
@@ -38,6 +38,9 @@
         "src/oic/schemas.py",
     }
 )
+OWNER_OPEN_BASE = "914830ceec70bde17004d2ccbbb13218ca44a89b"
+OWNER_AUTHORIZED_SEMANTIC_PATHS = frozenset({"src/oic/candidate_extraction.py"})
+AUTHORIZED_SOURCE_IDS = frozenset({"SYN-NS-GOV-1", "SYN-NS-PROC-1", "SYN-NS-AMEND-2"})
 
 
 class GateEvidenceError(ValueError):
@@ -50,7 +53,7 @@ def _require(condition: bool, message: str) -> None:
 
 
 def discover_unadmitted_production_paths(root: Path) -> list[str]:
-    """Compare tracked ``src/oic`` paths with the immutable pre-gate baseline."""
+    """Compare tracked paths with the immutable baseline plus owner allowlist."""
     git = shutil.which("git")
     if git is None:
         raise GateEvidenceError("cannot enumerate tracked src/oic production paths")
@@ -63,8 +66,9 @@ def discover_unadmitted_production_paths(root: Path) -> list[str]:
     if result.returncode != 0:
         raise GateEvidenceError("cannot enumerate tracked src/oic production paths")
     current = {line for line in result.stdout.splitlines() if line}
-    added = sorted(current - ADMITTED_SRC_OIC_PATHS)
-    missing = sorted(ADMITTED_SRC_OIC_PATHS - current)
+    admitted = ADMITTED_SRC_OIC_PATHS | OWNER_AUTHORIZED_SEMANTIC_PATHS
+    added = sorted(current - admitted)
+    missing = sorted(admitted - current)
     return added + [f"MISSING:{path}" for path in missing]
 
 
@@ -74,6 +78,7 @@ def validate_evidence(
     profile: dict[str, Any],
     veip: dict[str, Any],
     gate_text: str,
+    gate_record: dict[str, Any],
     source_bytes: dict[str, bytes],
     active_text: str,
     semantic_paths: list[str] | None = None,
@@ -168,10 +173,23 @@ def validate_evidence(
         "GLOBAL REPOSITORY COMPLETENESS" in normalized_gate and "INCOMPLETE" in normalized_gate,
         "global incompleteness not explicit",
     )
+    _require("READY FOR SEPARATE EXACT-HEAD REVIEW" in normalized_gate, "gate language invalid")
     _require(
-        "READY FOR SEPARATE EXACT-HEAD REVIEW\nNOT OPEN" in normalized_gate, "gate language invalid"
+        gate_record.get("gate_id") == "OIC-SEMANTIC-CODE-START-GATE-OPEN-v0.1",
+        "wrong owner gate identity",
     )
-    _require(not semantic_paths, "semantic implementation appeared before gate opening")
+    _require(gate_record.get("state") == "OPEN", "owner gate is not OPEN")
+    _require(gate_record.get("authorization_base_sha") == OWNER_OPEN_BASE, "wrong owner base SHA")
+    _require(
+        set(gate_record.get("authorized_semantic_paths", [])) == OWNER_AUTHORIZED_SEMANTIC_PATHS,
+        "gate production allowlist mismatch",
+    )
+    _require(
+        set(gate_record.get("authorized_source_ids", [])) == AUTHORIZED_SOURCE_IDS,
+        "gate source allowlist mismatch",
+    )
+    _require(gate_record.get("global_manifest_status") == "INCOMPLETE", "gate escalates manifest")
+    _require(not semantic_paths, "production path exceeds owner authorization")
 
 
 def load_and_validate(root: Path) -> None:
@@ -186,6 +204,7 @@ def load(path: str) -> dict[str, Any]:
     gate_text = (root / "docs/gates/OIC-SEMANTIC-CODE-START-GATE-CLOSURE-v0.1.md").read_text(
         encoding="utf-8"
     )
+    gate_record = load("docs/gates/OIC-SEMANTIC-CODE-START-GATE-OPEN-v0.1.json")
     active_text = "\n".join(
         (root / path).read_text(encoding="utf-8")
         for path in (
@@ -204,6 +223,7 @@ def load(path: str) -> dict[str, Any]:
         profile,
         veip,
         gate_text,
+        gate_record,
         source_bytes,
         active_text,
         semantic_paths=semantic_paths,
@@ -216,4 +236,4 @@ def load(path: str) -> dict[str, Any]:
     except (GateEvidenceError, OSError, KeyError, json.JSONDecodeError) as exc:
         print(f"FAIL semantic code-start gate evidence: {exc}")
         sys.exit(1)
-    print("PASS semantic code-start prerequisite evidence; gate remains NOT OPEN")
+    print("PASS semantic code-start gate is owner-opened for the exact bounded path and sources")