sourcecode 3.1.1__tar.gz → 3.2.0__tar.gz

{sourcecode-3.1.1 → sourcecode-3.2.0}/CHANGELOG.md

@@ -1,9 +1,154 @@
 # Changelog
 
-## [Unreleased]
+## [3.2.0] — 2026-07-28
+
+**Answer coherence (M7, first four seams).** One fact, one authority (ADR-0008). Four
+questions this tool answered twice — transaction boundaries, endpoint denominators, endpoint
+security coverage, outbound integrations — now have a single authority each, and every
+consumer binds to it instead of re-deriving.
+
+**Upgrade note.** Two behaviours change for anyone gating on the output:
+
+- `impact` / `impact-chain` / `plan` / `compare` now count transaction boundaries that were
+  structurally reported as `0` on every repository. Risk scores move **up**: on a
+  200-class BroadleafCommerce sample, 42 classes go from 0 to ≥1 boundary and 11 change
+  `risk_level`, always upward. A pinned `--fail-on` threshold can now trip where it did not.
+- `impact-chain.security_surfaces[]` and `endpoints_affected[]` gain `verdict` +
+  `confidence` and no longer let the declared annotation policy stand as a coverage answer.
+  `security_policy` is kept as a deprecated alias of `declared_policy` (two minor versions),
+  with its value unchanged.
+
+New payload keys (`population`, `declared_clients`, three-part endpoint partitions) are
+additive; the deprecated aliases listed below keep their previous values.
+
+### Fixed
+
+- **`export --integrations` answered "0 LDAP" in documents whose stack block named Spring
+  LDAP.** Reported in two field evaluations two majors apart. Reproduced on a Boot app that
+  declares `spring-boot-starter-data-ldap` and configures `spring.ldap.urls` in YAML: the
+  directory is wired entirely by properties, so no LDAP client type is ever imported and the
+  construct scan is *right* to find nothing — but the report never mentioned the declaration
+  it could see, and the document as a whole said the system has no outbound integrations
+  while naming a directory client library elsewhere.
+  - New authority `integration_coordinates.declared_integration_coordinates(root)` answers
+    one question — which client libraries the build declares — from Maven/Gradle
+    coordinates with `file:line` evidence, excluding test/provided scope.
+  - `export --integrations` (and `export --c4`) now carry `declared_clients`
+    (`kind`, `coordinate`, `evidence`, `observed_in_source`), `declared_clients_count` and
+    `kinds_declared_not_observed`, and the coverage reason names the declared kind and its
+    coordinate. A declared coordinate is never counted as an integration: `count` and
+    `by_kind` are unchanged and still report observed constructs only.
+  - `retrieve integration-inventory` states the same declared-not-observed kinds as
+    observations.
+  - `detectors/java.py` reads its directory coordinates from that one list, so the stack
+    block and the integration report can never recognize different libraries for the same
+    external system (ADR-0008 R12).
+  - Measured: alfresco declares `cxf-rt-frontend-jaxws` with no SOAP construct in source —
+    previously invisible, now stated. Broadleaf's declared JMS/LDAP coordinates are all
+    marked `observed_in_source: true`; repositories that declare no client library keep an
+    empty block. The `export --integrations` and `export --c4` baseline cells drift by
+    these added keys, by design.
+
+- **`impact-chain` called endpoints "none_detected" that the same run classified as guarded
+  by a custom authorization mechanism.** Measured on keycloak: `security_posture` reports
+  406 endpoints `protected_custom` (Likely) — 324 of them carrying the declared policy
+  `none_detected`, because their gate is a bespoke annotation outside every published
+  vocabulary — while `impact-chain` printed `security_policy: none_detected` for those same
+  `endpoint_id`s. One document answered both "guarded by a custom mechanism" and "no
+  security detected" about `DELETE /admin/realms/{realm}`. The declared policy is a fact
+  about an annotation; it was never an answer about coverage.
+  - New authority `security_posture.endpoint_security_surface(cir)` returns the per-endpoint
+    answer keyed by `endpoint_id` (ADR-0008 R1). `impact-chain` binds to it and reuses the
+    projection the security audit already derived — no second derivation, memoized per CIR
+    so `pr-impact`'s per-class queries pay for it once.
+  - `security_surfaces[]` and `endpoints_affected[]` now carry `verdict` + `confidence`
+    (`security_verdict` / `security_confidence` on affected endpoints) from the four-way
+    posture vocabulary, plus `declared_policy` for the annotation actually found. On the
+    28 endpoints reachable from `RealmAdminResource`, all 28 move `none_detected` →
+    `protected_custom` (Likely), matching `spring-audit` exactly.
+  - An endpoint the authority cannot speak about — or a repository with no posture
+    projection — is `coverage_unknown` with the reason stated, never a manufactured "none"
+    (ADR-0008 R9). On spring-petclinic all 17 endpoints read `coverage_unknown` in both
+    commands instead of `none_detected` in one and `coverage_unknown` in the other.
+  - `security_policy` is kept as a deprecated alias of `declared_policy` for two minor
+    versions; its value is unchanged.
+
+- **The security posture reported three counts under names that all read as "handlers", and
+  they did not reconcile.** The field report put `custom_authorization_mechanisms` at
+  713/936, `gate_coverage` at 698 and 2635, and a rollup of 939 + 2635 = 3574 — the only
+  figure matching `endpoints.total`. None was arithmetically wrong: they counted three
+  different populations, and no key said which. Reproduced exactly on keycloak, where the
+  old payload showed `gate_bearing: 337` beside `not_carrying_gate: 270` against a stated
+  total of 676 — a partition that visibly loses 69 endpoints, because the first number
+  counts handler *methods* and the second counts *endpoints*.
+  - One authority, `security_posture.endpoint_population`, now supplies every denominator,
+    and the posture payload carries a `population` block stating both units once
+    (ADR-0008 R1).
+  - `gate_coverage` gains an explicit three-part endpoint partition —
+    `endpoints_carrying_gate` + `endpoints_standard_guarded` +
+    `endpoints_not_carrying_gate` = `endpoints_total` — counted in the single loop that
+    visits each endpoint once, so the parts sum by construction. On keycloak:
+    406 + 0 + 270 = 676. Handler-method figures keep their own names
+    (`handler_methods_total`, `handler_methods_carrying_gate`) and are never part of that
+    sum.
+  - Custom-gate detection states the unit its percentage used: the denominator is handler
+    method *declarations* (overloads share an FQN, so keycloak has 551 distinct handler
+    methods across 601 declarations), and `endpoints_total` / `handler_methods_total` are
+    printed beside it so no reader has to infer which one a coverage figure meant.
+  - `total_controller_handlers`, `gate_bearing`, `not_carrying_gate` and `handler_total`
+    are kept as deprecated aliases with their existing values for two minor versions. No
+    consumer breaks; the numbers they held did not change.
+
+- **`impact` counted zero transaction boundaries on every repository — corrected, and the
+  two commands that answer this question now answer it once.** The field report was that
+  `impact` reported `transactional_boundaries_count: 0` for a symbol while `retrieve
+  transactions-reaching` reported 47 for the same one. The cause was worse than a
+  disagreement: the blast radius derived boundaries from a call-graph node
+  `role == "transaction_boundary"` that no producer ever assigns — 0 of 28 662 nodes carry
+  it on BroadleafCommerce — so the figure was **structurally always zero**, and the
+  `n_txn` term never contributed to `risk_score` or to the `n_txn >= 2` escalation rule
+  behind `risk_level`.
+  - Both consumers now ask one authority, `spring_semantic.boundaries_declared_within`
+    over the transaction-boundary index (ADR-0008 R1), which follows Spring's own
+    inheritance in both directions: a method carries its class's boundary, a class carries
+    the boundaries of the methods it declares, and asking about one method never sweeps in
+    a sibling's.
+  - Two named facts, one authority (ADR-0008 R3): `transactional_boundaries_touched` is
+    the target plus everything that calls it; the new `transactional_boundaries_reaching`
+    is the callers-only subset — what would run this change inside a transaction — and is
+    the number `retrieve transactions-reaching` reports, now by construction rather than
+    by coincidence.
+  - **Figures move, upward, because a real cost was being counted as nothing.** Measured
+    on BroadleafCommerce (161 declared boundaries): `CatalogServiceImpl` 0 → 10 boundaries
+    and `risk_score` 76.73 → 77.54; `OrderServiceImpl` 0 → 32 and 73.83 → 76.86;
+    `CustomerServiceImpl` 0 → 32 and 66.10 → 71.01. Across a 200-class sample, 42 classes
+    now report at least one boundary and 11 change `risk_level` — every one of them
+    upward. A pinned `--fail-on` threshold may therefore trip where it did not before;
+    that is the correction, not a regression.
 
 ### Added
 
+- **ADR-0008 "Answer Coherence" — one fact, one authority (normative, no behaviour change
+  yet).** Two independent field evaluations, on releases two majors apart, scored the
+  product identically and named the same ceiling: not a missing capability, but the same
+  question answered twice with different numbers (`impact` reporting 0 transactional
+  boundaries where `retrieve` reports 47; `compare` ranking on caller arrays truncated at
+  30 while the exact magnitudes sit in `stats`; `has_tests` contradicting `analysis_gaps`
+  inside one document). The ADR fixes the rule the codebase had never stated: a fact has
+  exactly one authority, consumers transport it, quantities never come from the length of a
+  rendered collection, and a surface that cannot resolve emits `unknown` — never `0`. It
+  also changes the unit of remediation from *defect* to *fact*: a coherence fix is complete
+  only when the authority is corrected, every consumer is rebound, and a cross-command
+  parity assertion is added to the release battery.
+  - `docs/architecture/adr/0008-answer-coherence.md` — the rule, with the code-verified seam
+    inventory in Appendix A.
+  - `docs/DEFECT-LEDGER.md` — every field defect, its class, and the release that closed it,
+    published rather than tracked privately.
+  - `.planning/FIELD-EVAL-SYNTHESIS-2026-07-28.md` and
+    `.planning/ROADMAP-STRATEGIC-2026-07-28.md` — the evidence and the resulting direction
+    (M7 Coherence → M8 Payload economy → M9 Core & surface → M10 Enrichment interop →
+    M11 Compliance substrate).
+
 - **`migrate-check --blast-radius` states the regression scope behind each blocker.** The
   finding list says what must change; a migration plan is built around what has to be
   re-tested when that change lands. Each affected product file now resolves to the types

{sourcecode-3.1.1 → sourcecode-3.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sourcecode
-Version: 3.1.1
+Version: 3.2.0
 Summary: Persistent structural context and ultra-fast repeated analysis for AI coding agents
 License-File: LICENSE
 Keywords: agents,ai,codebase,context,developer-tools,llm

{sourcecode-3.1.1 → sourcecode-3.2.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "sourcecode"
-version = "3.1.1"
+version = "3.2.0"
 description = "Persistent structural context and ultra-fast repeated analysis for AI coding agents"
 readme = "README.md"
 requires-python = ">=3.9"

{sourcecode-3.1.1 → sourcecode-3.2.0}/src/sourcecode/__init__.py

@@ -4,4 +4,4 @@ ASK Engine is the product. ``ask`` is the canonical CLI command; ``sourcecode``
 the legacy compatibility alias and the Python/PyPI package name. See
 docs/PRODUCT_IDENTITY.md (normative)."""
 
-__version__ = "3.1.1"
+__version__ = "3.2.0"

{sourcecode-3.1.1 → sourcecode-3.2.0}/src/sourcecode/cli.py

@@ -4084,7 +4084,11 @@ def impact_cmd(
       - direct_callers     — classes that directly call or depend on the target
       - indirect_callers   — transitive callers (BFS, bounded by --depth)
       - endpoints_affected — HTTP endpoints that transitively depend on the target
-      - transactional_boundaries_touched — @Transactional classes in the call chain
+      - transactional_boundaries_touched — @Transactional declarations on the target
+                             and on everything that calls it
+      - transactional_boundaries_reaching — the callers-only subset: what would run
+                             this change inside a transaction (the figure
+                             `ask retrieve transactions-reaching` reports)
       - risk_score / risk_level — quantified change risk
 
     \b
@@ -4714,7 +4718,10 @@ def export_cmd(
     ),
     integrations: bool = typer.Option(
         False, "--integrations",
-        help="Detect outbound integrations (HTTP/LDAP/JMS clients) with file:line evidence.",
+        help=(
+            "Detect outbound integrations (HTTP/LDAP/JMS clients) with file:line "
+            "evidence, plus the client libraries the build declares."
+        ),
     ),
     c4: bool = typer.Option(
         False, "--c4",
@@ -4742,6 +4749,11 @@ def export_cmd(
                      rolled up from class-level relation edges.
     --integrations   Outbound integrations (RestTemplate/WebClient/Feign/LDAP/JMS)
                      with file:line evidence — external-system dependency arrows.
+                     Also lists the client libraries the BUILD declares
+                     (declared_clients) and which of those no source construct
+                     uses — a config-wired client has no construct to find, so a
+                     count of 0 is never left standing alone against a client
+                     library named in the stack block.
     --c4             Unified architecture document mapped onto the open C4 model
                      (context/containers/components/code) + an API surface and a
                      per-directory content-hash manifest for incremental consumers.
@@ -4792,9 +4804,12 @@ def export_cmd(
         graph = ir.get("graph", {})
         _ep_data = extract_java_endpoints(root)
         endpoints = _ep_data.get("endpoints", [])
+        from sourcecode.integration_coordinates import declared_integration_coordinates
         _integrations = SemanticIntegrationEngine(
             ContextGraph.build(file_list, root)
-        ).as_report(len(file_list))
+        ).as_report(
+            len(file_list), declared=declared_integration_coordinates(root),
+        )
         data = _build_c4_export(
             root,
             file_list,
@@ -4831,9 +4846,12 @@ def export_cmd(
     if integrations:
         from sourcecode.context_graph import ContextGraph
         from sourcecode.semantic_integration_engine import SemanticIntegrationEngine
+        from sourcecode.integration_coordinates import declared_integration_coordinates
         data["integrations"] = SemanticIntegrationEngine(
             ContextGraph.build(file_list, root)
-        ).as_report(len(file_list))
+        ).as_report(
+            len(file_list), declared=declared_integration_coordinates(root),
+        )
 
     _prog.finish()
     output = _serialize_dict(data, format)
@@ -5684,19 +5702,35 @@ def _render_gate_coverage_section(result: "SpringAuditResult") -> list[str]:  #
     gc = (result.security_posture or {}).get("gate_coverage")
     if not gc:
         return []
-    not_covered = gc.get("not_carrying_gate", 0)
+    not_covered = gc.get("endpoints_not_carrying_gate", gc.get("not_carrying_gate", 0))
     gates = ", ".join(f"`{g}`" for g in gc.get("gate_annotations", [])) or "the detected gate"
-    total = gc.get("total_controller_handlers", 0)
+    # One unit, named: these are endpoints, not handler methods. The two counts differ
+    # (one method can serve several mappings) and must never be summed together.
+    total = gc.get("endpoints_total", gc.get("total_controller_handlers", 0))
     lines: list[str] = ["", "---", ""]
     if not_covered == 0:
-        lines.append(f"✅ **Gate coverage** — all {total} controller handlers carry {gates}.")
+        _gated = gc.get("endpoints_carrying_gate", total)
+        _std = gc.get("endpoints_standard_guarded", 0)
+        if _std:
+            lines.append(
+                f"✅ **Gate coverage** — all {total} endpoints are guarded: {_gated} carry "
+                f"{gates}, {_std} rely on a standard guard."
+            )
+        else:
+            lines.append(f"✅ **Gate coverage** — all {total} endpoints carry {gates}.")
         return lines
 
     covered = gc.get("possibly_filter_covered", 0)
+    _standard = gc.get("endpoints_standard_guarded", 0)
     lines.append(
-        f"🔓 **Gate coverage** — {not_covered} of {total} controller handlers do not carry "
+        f"🔓 **Gate coverage** — {not_covered} of {total} endpoints do not carry "
         f"{gates}."
     )
+    if _standard:
+        lines.append(
+            f"_{gc.get('endpoints_carrying_gate', 0)} carry the gate, {_standard} rely on a "
+            f"standard guard, {not_covered} on neither — the three sum to {total}._"
+        )
     if gc.get("reconstructed_filter_patterns"):
         lines.append(
             f"_{covered} of those match a reconstructed servlet filter pattern "
@@ -6369,6 +6403,16 @@ def migrate_check_cmd(
       ask migrate-check . --output migration.json
       ask migrate-check . --snapshot --ref sprint-12   persist a readiness point
       ask migrate-check . --trend                      days-remaining over time
+      ask migrate-check . --blast-radius               endpoints behind each blocker
+
+    \b
+    Blast radius (--blast-radius):
+      Ranks the affected product files by the HTTP endpoints whose call path runs
+      through them — the re-test plan, ordered by regression scope. Reach is never
+      build breakage: a javax→jakarta finding fails compilation whether or not an
+      endpoint reaches it. A file whose types the IR could not resolve is reported
+      unresolved (reach unknown, never zero), and a repository with no endpoints in
+      the IR gets a null share with the reason stated instead of a 0.0.
 
     \b
     Readiness time series:
@@ -6587,7 +6631,8 @@ def impact_chain_cmd(
       - indirect_callers    — transitive callers (BFS up to --depth hops)
       - endpoints_affected  — HTTP endpoints reachable through the call chain
       - transaction_boundary — @Transactional semantics on the target (if any)
-      - security_surfaces   — per-endpoint security policy + SEC findings
+      - security_surfaces   — per-endpoint security verdict + confidence
+                              (security_posture authority) + declared policy + SEC findings
       - impact_findings     — TX/SEC audit findings touching the call chain
       - risk_level          — critical | high | medium | low
 

{sourcecode-3.1.1 → sourcecode-3.2.0}/src/sourcecode/detectors/java.py

@@ -15,6 +15,7 @@ from sourcecode.detectors.parsers import (
     substitute_maven_properties as _resolve_maven_property,
     unique_strings,
 )
+from sourcecode.integration_coordinates import coordinates_for_kind
 from sourcecode.schema import FrameworkDetection
 from sourcecode.tree_utils import flatten_file_tree
 
@@ -488,11 +489,15 @@ class JavaDetector(AbstractDetector):
         if "spring-boot-starter-data-jpa" in text or "spring-data-jpa" in text:
             frameworks.append(FrameworkDetection(name="Spring Data JPA", source=source))
         # The Boot starter names neither of the module coordinates, so a Boot app
-        # wiring LDAP the standard way declared no LDAP framework at all.
-        if (
-            "spring-ldap-core" in text
-            or "spring-security-ldap" in text
-            or "spring-boot-starter-data-ldap" in text
+        # wiring LDAP the standard way declared no LDAP framework at all. The coordinate
+        # list lives in `integration_coordinates` — the same set `export --integrations`
+        # reads, so this stack block and that report can never recognize different
+        # libraries for the same external system (ADR-0008 R12).
+        # Only the Spring-family coordinates may carry the "Spring LDAP" label — the
+        # authority also knows non-Spring directory clients, and naming the wrong library
+        # would trade one incoherence for another.
+        if any(
+            c in text for c in coordinates_for_kind("ldap") if c.startswith("spring-")
         ):
             frameworks.append(FrameworkDetection(name="Spring LDAP", source=source))
         if "spring-aspects" in text or "spring-aop" in text:

sourcecode-3.2.0/src/sourcecode/integration_coordinates.py

@@ -0,0 +1,175 @@
+"""integration_coordinates.py — the declared-client signal for outbound integrations.
+
+THE authority for one question: *which outbound-integration client libraries does this
+repository's build declare?* (ADR-0008 R1). It is a different question from "which client
+constructs does the source contain" — that one belongs to
+:mod:`sourcecode.semantic_integration_engine` — and the two answers only conflict when a
+document reports one and stays silent about the other.
+
+That is exactly what the field reported, twice, two majors apart: `export --integrations`
+answered `count: 0` on a repository whose stack block listed **Spring LDAP**. Reproduced on
+a Boot app that declares `spring-boot-starter-data-ldap` and configures
+`spring.ldap.urls` in YAML: the directory is wired entirely by properties, so no LDAP
+client type is ever imported and the construct scan is *right* to find nothing — but the
+document as a whole then said the system has no outbound integrations while its own stack
+block said it talks to a directory.
+
+A declared coordinate is **evidence of a declared client, never an observed integration**
+(the same discipline that stopped a `org.mybatis` coordinate from being reported as MyBatis
+usage in `detectors/java.py`). This module answers what the build declares; the engine
+answers what the code does; the payload states both and says which is which.
+
+VAI (see `security_posture` §0.5): the table below lists **published open-source artifact
+coordinates** — the identity of a library, matched literally, the same basis as the engine's
+published client-type table. No proprietary, client or convention name appears here, and no
+predicate branches on a repository's own naming.
+"""
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Iterable, Optional
+
+# Build-manifest artifact coordinate → the outbound-integration kind it declares.
+# Kinds are exactly the engine's kinds, so declared and observed answers are comparable.
+# Only coordinates whose sole purpose is to talk to an external system are listed: a
+# driver-or-anything coordinate that a repository might carry for unrelated reasons has no
+# place here, because a declaration nobody can act on is noise, not evidence.
+_COORDINATE_KIND: "dict[str, str]" = {
+    # Directory (LDAP)
+    "spring-ldap-core": "ldap",
+    "spring-security-ldap": "ldap",
+    "spring-boot-starter-data-ldap": "ldap",
+    "spring-boot-starter-ldap": "ldap",
+    "unboundid-ldapsdk": "ldap",
+    # Kafka
+    "spring-kafka": "kafka",
+    "kafka-clients": "kafka",
+    "spring-cloud-stream-binder-kafka": "kafka",
+    # RabbitMQ / AMQP
+    "spring-boot-starter-amqp": "rabbitmq",
+    "spring-rabbit": "rabbitmq",
+    "amqp-client": "rabbitmq",
+    # JMS brokers
+    "spring-boot-starter-activemq": "jms",
+    "spring-boot-starter-artemis": "jms",
+    "activemq-client": "jms",
+    "activemq-broker": "jms",
+    "artemis-jms-client": "jms",
+    "spring-jms": "jms",
+    # Redis
+    "spring-boot-starter-data-redis": "redis",
+    "spring-data-redis": "redis",
+    "lettuce-core": "redis",
+    "jedis": "redis",
+    # Elasticsearch
+    "spring-boot-starter-data-elasticsearch": "elasticsearch",
+    "elasticsearch-rest-high-level-client": "elasticsearch",
+    "elasticsearch-java": "elasticsearch",
+    # Mail (SMTP)
+    "spring-boot-starter-mail": "smtp",
+    # Declarative HTTP clients
+    "spring-cloud-starter-openfeign": "http",
+    "feign-core": "http",
+    "okhttp": "http",
+    # SOAP
+    "spring-ws-core": "soap",
+    "spring-boot-starter-web-services": "soap",
+    "cxf-rt-frontend-jaxws": "soap",
+}
+
+# Build manifests scanned. Text-level, like the framework-coordinate scan in
+# `detectors/java.py` — a build file is configuration, not source.
+_MANIFEST_NAMES: frozenset[str] = frozenset({
+    "pom.xml", "build.gradle", "build.gradle.kts",
+})
+
+# Directories that never hold a production build manifest.
+_SKIP_DIRS: frozenset[str] = frozenset({
+    ".git", ".idea", "target", "build", "node_modules", ".sourcecode-cache",
+    ".gradle", "out", "dist",
+})
+
+# A repository with thousands of modules is real (OFBiz, Broadleaf); an unbounded walk is
+# not. The cap is stated in the payload when it bites, never applied silently.
+_MAX_MANIFESTS = 400
+
+_TEST_SCOPE_WINDOW = 4          # lines after a Maven coordinate that may carry its <scope>
+_TEST_SCOPE = re.compile(r"<scope>\s*(test|provided)\s*</scope>", re.IGNORECASE)
+_GRADLE_TEST_CONF = re.compile(r"^\s*(test|androidTest)[A-Za-z]*\s*[(\s'\"]", re.IGNORECASE)
+
+
+@dataclass(frozen=True)
+class DeclaredClient:
+    """A client library the build declares, with the line that declares it."""
+
+    kind: str
+    coordinate: str
+    evidence: str               # "relpath:line" — same shape as an Integration's evidence
+
+    def to_dict(self) -> dict:
+        return {"kind": self.kind, "coordinate": self.coordinate, "evidence": self.evidence}
+
+
+def coordinates_for_kind(kind: str) -> tuple[str, ...]:
+    """Published coordinates that declare ``kind``. One home for the list, so a consumer
+    that recognizes a library recognizes the same set this authority does."""
+    return tuple(sorted(c for c, k in _COORDINATE_KIND.items() if k == kind))
+
+
+def _manifest_paths(root: Path) -> list[Path]:
+    found: list[Path] = []
+    for name in sorted(_MANIFEST_NAMES):
+        for path in sorted(root.rglob(name)):
+            if len(found) >= _MAX_MANIFESTS:
+                return found
+            if any(part in _SKIP_DIRS for part in path.parts):
+                continue
+            if path.is_file():
+                found.append(path)
+    return found
+
+
+def _is_test_scoped(lines: list[str], idx: int, manifest: str) -> bool:
+    """True when the declaration is a test/provided dependency — a build-time coordinate
+    is not a statement that the running system talks to that external party."""
+    if manifest == "pom.xml":
+        for line in lines[idx: idx + _TEST_SCOPE_WINDOW + 1]:
+            if _TEST_SCOPE.search(line):
+                return True
+        return False
+    return bool(_GRADLE_TEST_CONF.match(lines[idx]))
+
+
+def declared_integration_coordinates(
+    root: Optional[Path], *, manifests: Optional[Iterable[Path]] = None,
+) -> list[DeclaredClient]:
+    """Client libraries the build declares, deterministically ordered, deduplicated by
+    (kind, coordinate). Never raises: an unreadable manifest contributes nothing.
+    """
+    if root is None:
+        return []
+    root = Path(root)
+    paths = list(manifests) if manifests is not None else _manifest_paths(root)
+
+    seen: dict[tuple[str, str], DeclaredClient] = {}
+    for path in paths:
+        try:
+            lines = path.read_text(encoding="utf-8", errors="ignore").splitlines()
+        except OSError:
+            continue
+        try:
+            rel = str(path.relative_to(root)).replace("\\", "/")
+        except ValueError:
+            rel = path.name
+        for idx, line in enumerate(lines):
+            for coordinate, kind in _COORDINATE_KIND.items():
+                if coordinate not in line:
+                    continue
+                if _is_test_scoped(lines, idx, path.name):
+                    continue
+                key = (kind, coordinate)
+                if key not in seen:
+                    seen[key] = DeclaredClient(kind, coordinate, f"{rel}:{idx + 1}")
+    return [seen[k] for k in sorted(seen)]

{sourcecode-3.1.1 → sourcecode-3.2.0}/src/sourcecode/mcp/registry.py

@@ -781,7 +781,9 @@ Blast-radius analysis: who calls a class and what breaks if it changes? Java onl
 
 Maps to: ask impact <target> <repo_path> [--depth <depth>]
 Returns: direct_callers, indirect_callers, endpoints_affected,
-         transactional_boundaries_touched, risk_score, risk_level, stats.
+         transactional_boundaries_touched (target + callers),
+         transactional_boundaries_reaching (callers only), risk_score,
+         risk_level, stats.
 
 Use this when:
 - Planning a refactor: understand the full call chain before changing a class

{sourcecode-3.1.1 → sourcecode-3.2.0}/src/sourcecode/mcp/server.py

@@ -890,7 +890,8 @@ def get_impact_chain(repo_path: str = ".", symbol: str = "", depth: int = 4) ->
     Returns: ImpactChainResult with schema_version, symbol, resolution,
              direct_callers, indirect_callers, endpoints_affected,
              transaction_boundary (propagation/isolation/read_only),
-             security_surfaces (per-endpoint policy + finding IDs),
+             security_surfaces (per-endpoint verdict/confidence from the
+             security_posture authority + declared policy + finding IDs),
              impact_findings (TX-001..005 + SEC-001..003 findings in call chain),
              analysis_warnings, risk_level, confidence, metadata.
 
@@ -1320,7 +1321,9 @@ def get_impact_context(repo_path: str = ".", target: str = "", depth: int = 4) -
 
     Maps to: ask impact <target> <repo_path> [--depth <depth>]
     Returns: direct_callers, indirect_callers, endpoints_affected,
-             transactional_boundaries_touched, risk_score, risk_level, stats.
+             transactional_boundaries_touched (target + callers),
+             transactional_boundaries_reaching (callers only), risk_score,
+             risk_level, stats.
 
     Use this when:
     - Planning a refactor: understand the full call chain before changing a class

{sourcecode-3.1.1 → sourcecode-3.2.0}/src/sourcecode/repository_ir.py

@@ -26,6 +26,10 @@ from typing import Any, Iterable, Optional
 from sourcecode.caller_metrics import CALLER_METRIC_RECONCILIATION
 from sourcecode.endpoint_metrics import ENDPOINT_SURFACE_RECONCILIATION
 from sourcecode.fqn_utils import normalize_owner_fqn as _normalize_owner_fqn
+from sourcecode.spring_semantic import (
+    boundaries_declared_within as _boundaries_declared_within,
+    build_tx_index_from_ir as _build_tx_index_from_ir,
+)
 from sourcecode.path_filters import (
     is_test_path as _is_test_path,
     is_test_fixture_module_path as _is_test_fixture_module_path,
@@ -6803,6 +6807,7 @@ def compute_blast_radius(
             "security_surface_affected": [],
             "cross_module_impact": [],
             "transactional_boundaries_touched": [],
+            "transactional_boundaries_reaching": [],
             "risk_score": 0.0,
             "risk_level": "unknown",
             "confidence_score": 0.0,
@@ -6828,6 +6833,7 @@ def compute_blast_radius(
             "security_surface_affected": [],
             "cross_module_impact": [],
             "transactional_boundaries_touched": [],
+            "transactional_boundaries_reaching": [],
             "risk_score": 0.0,
             "risk_level": "unknown",
             "confidence_score": 0.0,
@@ -7134,23 +7140,19 @@ def compute_blast_radius(
         })
     cross_module_impact = cross_module_impact[:10]
 
-    # ── 7. Transactional boundaries touched ───────────────────────────────────
-    txn_nodes: list[str] = []
-    for node_dict in graph_nodes:
-        fqn = node_dict.get("fqn") or ""
-        role = node_dict.get("role") or ""
-        symbol_kind = node_dict.get("symbol_kind") or ""
-        if role == "transaction_boundary" or "Transactional" in (node_dict.get("canonical_name") or ""):
-            if fqn in affected_classes or _enclosing_class(fqn) in affected_classes:
-                txn_nodes.append(fqn)
-        elif symbol_kind == "method" and fqn in affected_classes:
-            enc = _enclosing_class(fqn)
-            for n2 in graph_nodes:
-                if n2.get("fqn") == enc and n2.get("role") == "transaction_boundary":
-                    txn_nodes.append(fqn)
-                    break
-
-    txn_nodes = sorted(set(txn_nodes))
+    # ── 7. Transactional boundaries ───────────────────────────────────────────
+    # Both figures come from the one transaction-boundary authority (ADR-0008 R1);
+    # they differ only in which symbols are asked about:
+    #   touched   — the target and everything that calls it (what this change sits in)
+    #   reaching  — the callers alone (what would run this change inside a transaction)
+    # This used to be derived here from a graph node `role == "transaction_boundary"`
+    # that no producer ever assigns, so `impact` reported 0 boundaries on repos where
+    # `retrieve transactions-reaching` reported dozens, and `n_txn` never contributed
+    # to the risk score or its escalation rules.
+    _tx_index = _build_tx_index_from_ir(ir)
+    _caller_symbols = set(direct_callers) | set(indirect_callers)
+    txn_nodes = sorted({b.symbol for b in _boundaries_declared_within(_tx_index, affected_classes)})
+    txn_reaching = sorted({b.symbol for b in _boundaries_declared_within(_tx_index, _caller_symbols)})
 
     # ── 8. Risk score ─────────────────────────────────────────────────────────
     n_direct   = len(direct_callers)
@@ -7288,6 +7290,11 @@ def compute_blast_radius(
         "security_surface_affected": security_surface_affected,
         "cross_module_impact": cross_module_impact,
         "transactional_boundaries_touched": txn_nodes,
+        # The callers-only subset: the boundaries that would run this change inside a
+        # transaction. Same authority, narrower question — and the figure
+        # `retrieve transactions-reaching` answers, so the two commands agree by
+        # construction rather than by coincidence.
+        "transactional_boundaries_reaching": txn_reaching,
         "depth_reached": _effective_depth,  # actual BFS depth used, not the requested max
         "bfs_truncated": _bfs_truncated,
         "stats": {
@@ -7297,6 +7304,7 @@ def compute_blast_radius(
             "indirect_callers_sampled": _indirect_sampled,
             "endpoints_affected_count": n_ep,
             "transactional_boundaries_count": n_txn,
+            "transactional_boundaries_reaching_count": len(txn_reaching),
             "mappers_affected_count": n_mappers,
             "modules_affected_count": n_modules,
             "security_surface_count": n_sec,