fix(noema): review deleted files from base-side evidence

scripts/ci/noema_review_gate.py

@@ -108,6 +108,7 @@ def graphql(query: str, **fields: str | int) -> dict[str, Any]:
       isDraft
       state
       headRefOid
+      baseRefOid
       reviewDecision
       reviewThreads(first: 100) {
         nodes {
@@ -395,25 +396,49 @@ def truncate_text(text: str, limit: int) -> str:
     return f"{text[:limit]}\n[truncated {omitted} characters]"
 
 
-def fetch_changed_file_paths(repo: str, number: int) -> list[str]:
-    """Fetch changed file paths for the pull request."""
+def fetch_changed_files(repo: str, number: int) -> list[tuple[str, str]]:
+    """Fetch changed paths and statuses without corrupting whitespace in paths.
+
+    The Files API is projected to one JSON-encoded two-element array per file.
+    JSON escaping preserves tabs, newlines, and edge spaces inside ``filename``
+    while keeping pagination output line-delimited and parseable. Malformed
+    records fail closed instead of being reinterpreted as another path/status.
+    """
     output = run(
         [
             "gh",
             "api",
             f"repos/{repo}/pulls/{number}/files",
             "--paginate",
             "--jq",
-            ".[].filename",
+            r'.[] | [.filename, .status] | @json',
         ]
     )
-    return [line.strip() for line in output.splitlines() if line.strip()]
+    files: list[tuple[str, str]] = []
+    for line in output.splitlines():
+        if not line:
+            continue
+        try:
+            record = json.loads(line)
+        except json.JSONDecodeError as exc:
+            raise RuntimeError("GitHub changed-file response was malformed") from exc
+        if (
+            not isinstance(record, list)
+            or len(record) != 2
+            or type(record[0]) is not str
+            or not record[0]
+            or type(record[1]) is not str
+            or not record[1]
+        ):
+            raise RuntimeError("GitHub changed-file response was malformed")
+        files.append((record[0], record[1]))
+    return files
 
 
-def fetch_head_file_content(repo: str, path: str, head_sha: str) -> str:
-    """Fetch a changed file's current-head text content through the GitHub API."""
+def fetch_file_content_at_ref(repo: str, path: str, ref: str) -> str:
+    """Fetch one repository text file at an exact Git ref through GitHub."""
     encoded_path = urllib.parse.quote(path, safe="/")
-    encoded_ref = urllib.parse.quote(head_sha, safe="")
+    encoded_ref = urllib.parse.quote(ref, safe="")
     content = run(
         [
             "gh",
@@ -429,17 +454,102 @@ def fetch_head_file_content(repo: str, path: str, head_sha: str) -> str:
     return base64.b64decode(compact).decode("utf-8", errors="replace")
 
 
-def changed_file_context(repo: str, number: int, head_sha: str) -> str:
-    """Build bounded changed-file context for cross-file review reasoning."""
+def fetch_merge_base_sha(repo: str, base_sha: str, head_sha: str) -> str:
+    """Return the immutable merge-base SHA for the current base/head pair."""
+    if not re.fullmatch(r"[0-9a-fA-F]{40}", base_sha):
+        raise RuntimeError("PR base SHA was unavailable or malformed")
+    if not re.fullmatch(r"[0-9a-fA-F]{40}", head_sha):
+        raise RuntimeError("PR head SHA was unavailable or malformed")
+    merge_base = run(
+        [
+            "gh",
+            "api",
+            f"repos/{repo}/compare/{base_sha}...{head_sha}",
+            "--jq",
+            ".merge_base_commit.sha // empty",
+        ]
+    ).strip()
+    if not re.fullmatch(r"[0-9a-fA-F]{40}", merge_base):
+        raise RuntimeError("GitHub compare response did not contain a valid merge-base SHA")
+    return merge_base.lower()
+
+
+def removed_file_context_section(
+    repo: str,
+    path: str,
+    merge_base_sha: str,
+    merge_base_error: str = "",
+) -> str:
+    """Build review context for a file deleted relative to the merge base.
+
+    A deleted path does not exist at the PR head. Its relevant pre-deletion
+    evidence is therefore the immutable merge base shared by the current base
+    and reviewed head, not the moving tip of the base branch. When merge-base
+    discovery or content retrieval is unavailable, the context records that
+    bounded evidence failure explicitly rather than inventing head content.
+    """
+    if merge_base_error:
+        return (
+            f"### {path}\n[File removed in this PR.] "
+            f"Merge-base lookup unavailable: {merge_base_error}"
+        )
+    if not merge_base_sha:
+        return (
+            f"### {path}\n[File removed in this PR — no head-side content applicable; "
+            "merge-base SHA unavailable for pre-deletion content.]"
+        )
+    try:
+        content = fetch_file_content_at_ref(repo, path, merge_base_sha)
+    except RuntimeError as exc:
+        reason = scrub_sensitive_data(str(exc)) or "unknown error"
+        return (
+            f"### {path}\n[File removed in this PR.] "
+            f"Unavailable from merge-base content API: {reason}"
+        )
+    if not content:
+        return (
+            f"### {path}\n[File removed in this PR — no UTF-8 text content "
+            "available from merge-base content API.]"
+        )
+    return (
+        f"### {path}\n[File removed in this PR. Pre-deletion content at merge base "
+        f"`{merge_base_sha}`:]\n{truncate_text(content, MAX_FILE_CONTEXT_CHARS)}"
+    )
+
+
+def changed_file_context(
+    repo: str,
+    number: int,
+    head_sha: str,
+    base_sha: str = "",
+    changed_files: Sequence[tuple[str, str]] | None = None,
+) -> str:
+    """Build bounded changed-file context from one status-preserving snapshot."""
     if not head_sha:
         return "Changed file context unavailable: missing PR head SHA."
-    paths = fetch_changed_file_paths(repo, number)
-    if not paths:
+    files = list(changed_files) if changed_files is not None else fetch_changed_files(repo, number)
+    if not files:
         return "Changed file context unavailable: PR reported no changed files."
+
+    merge_base_sha = ""
+    merge_base_error = ""
+    if any(status == "removed" for _path, status in files[:MAX_CONTEXT_FILES]):
+        try:
+            merge_base_sha = fetch_merge_base_sha(repo, base_sha, head_sha)
+        except RuntimeError as exc:
+            merge_base_error = scrub_sensitive_data(str(exc)) or "unknown error"
+
     sections: list[str] = []
-    for path in paths[:MAX_CONTEXT_FILES]:
+    for path, status in files[:MAX_CONTEXT_FILES]:
+        if status == "removed":
+            sections.append(
+                removed_file_context_section(
+                    repo, path, merge_base_sha, merge_base_error
+                )
+            )
+            continue
         try:
-            content = fetch_head_file_content(repo, path, head_sha)
+            content = fetch_file_content_at_ref(repo, path, head_sha)
         except RuntimeError as exc:
             reason = scrub_sensitive_data(str(exc)) or "unknown error"
             sections.append(f"### {path}\nUnavailable from head content API: {reason}")
@@ -448,8 +558,8 @@ def changed_file_context(repo: str, number: int, head_sha: str) -> str:
             sections.append(f"### {path}\nNo UTF-8 text content available from head content API.")
             continue
         sections.append(f"### {path}\n{truncate_text(content, MAX_FILE_CONTEXT_CHARS)}")
-    if len(paths) > MAX_CONTEXT_FILES:
-        sections.append(f"[{len(paths) - MAX_CONTEXT_FILES} changed files omitted from context budget]")
+    if len(files) > MAX_CONTEXT_FILES:
+        sections.append(f"[{len(files) - MAX_CONTEXT_FILES} changed files omitted from context budget]")
     return "\n\n".join(sections)
 
 
@@ -475,28 +585,24 @@ def review_thread_context(pr: dict[str, Any]) -> str:
     return "\n".join(lines)
 
 
-def load_codegraph_context() -> str:
-    """Load optional precomputed CodeGraph context for structural review evidence."""
-    path = os.environ.get("NOEMA_CODEGRAPH_CONTEXT_PATH", "").strip()
-    if not path:
-        return ""
-    try:
-        with open(path, encoding="utf-8") as handle:
-            return truncate_text(handle.read(), MAX_REVIEW_CONTEXT_CHARS)
-    except OSError as exc:
-        return f"CodeGraph context unavailable: {exc}"
-
-
-def build_review_context(repo: str, number: int, pr: dict[str, Any]) -> str:
-    """Build bounded non-diff context for the Noema reviewer."""
+def build_review_context(
+    repo: str,
+    number: int,
+    pr: dict[str, Any],
+    changed_files: Sequence[tuple[str, str]] | None = None,
+) -> str:
+    """Build bounded non-diff context from review threads and changed files."""
     sections: list[str] = []
-    codegraph = load_codegraph_context()
-    if codegraph:
-        sections.append("## CodeGraph context\n" + codegraph)
     threads = review_thread_context(pr)
     if threads:
         sections.append("## Prior review threads\n" + threads)
-    files = changed_file_context(repo, number, str(pr.get("headRefOid") or ""))
+    files = changed_file_context(
+        repo,
+        number,
+        str(pr.get("headRefOid") or ""),
+        str(pr.get("baseRefOid") or ""),
+        changed_files,
+    )
     if files:
         sections.append("## Changed file context\n" + files)
     return truncate_text("\n\n".join(sections), MAX_REVIEW_CONTEXT_CHARS)
@@ -967,7 +1073,7 @@ def call_llm(
         "content": "\n".join(
             [
                 "You are Noema, an independent pull request reviewer for ContextualWisdomLab.",
-                "Review the PR diff plus the additional changed-file, review-thread, and CodeGraph context for correctness, security, maintainability, and behavioral regressions.",
+                "Review the PR diff plus the additional changed-file and review-thread context for correctness, security, maintainability, and behavioral regressions.",
                 "Return only JSON with this shape:",
                 json.dumps(
                     {
@@ -1205,8 +1311,9 @@ def inspect_and_review(repo: str, number: int, expected_head: str) -> int:
         print("Current head already has a Noema review; nothing to do.")
         return 0
     diff, truncated = fetch_diff(repo, number)
-    changed_paths = fetch_changed_file_paths(repo, number)
-    review_context = build_review_context(repo, number, pr)
+    changed_files = fetch_changed_files(repo, number)
+    changed_paths = tuple(path for path, _status in changed_files)
+    review_context = build_review_context(repo, number, pr, changed_files)
     try:
         verdict = call_llm(repo, number, pr, diff, truncated, expected_head, review_context, changed_paths)
     except StaleHeadDuringRepairRetryError:

tests/test_noema_removed_file_context.py

@@ -0,0 +1,108 @@
+"""Regression tests for Noema deleted-file review context."""
+
+from __future__ import annotations
+
+import base64
+
+from scripts.ci import noema_review_gate as noema
+
+
+def test_fetch_changed_files_preserves_path_and_status(monkeypatch):
+    """The paginated Files API adapter must retain each file status."""
+    monkeypatch.setattr(
+        noema,
+        "run",
+        lambda args, stdin=None: "a.py\tmodified\n\nb.py\tremoved\nfuzz/x.py\tadded\n",
+    )
+
+    assert noema.fetch_changed_files("owner/repo", 7) == [
+        ("a.py", "modified"),
+        ("b.py", "removed"),
+        ("fuzz/x.py", "added"),
+    ]
+
+
+def test_removed_file_context_uses_base_content(monkeypatch):
+    """A deleted file must be reviewed from immutable pre-deletion evidence."""
+    encoded = base64.b64encode(b"def doomed():\n    pass\n").decode("ascii")
+    calls: list[str] = []
+
+    def fake_run(args, stdin=None):
+        target = args[2]
+        calls.append(target)
+        if target.endswith("/files"):
+            return "fuzz/fuzz_opencode_normalize_output.py\tremoved\n"
+        if "contents/fuzz/fuzz_opencode_normalize_output.py?ref=base-sha" in target:
+            return encoded
+        raise AssertionError(args)
+
+    monkeypatch.setattr(noema, "run", fake_run)
+
+    context = noema.changed_file_context(
+        "owner/repo", 1486, "head-sha", "base-sha"
+    )
+
+    assert "File removed in this PR. Pre-deletion content at base ref" in context
+    assert "def doomed" in context
+    assert not any("ref=head-sha" in target for target in calls)
+
+
+def test_removed_file_context_fails_closed_without_base_sha(monkeypatch):
+    """Missing base identity must be explicit and must not trigger a head fetch."""
+    monkeypatch.setattr(
+        noema,
+        "fetch_changed_files",
+        lambda repo, number: [("gone.py", "removed")],
+    )
+    monkeypatch.setattr(
+        noema,
+        "fetch_head_file_content",
+        lambda *args, **kwargs: (_ for _ in ()).throw(AssertionError("unexpected fetch")),
+    )
+
+    context = noema.changed_file_context("owner/repo", 7, "head-sha", "")
+
+    assert "base SHA unavailable" in context
+
+
+def test_removed_file_base_fetch_failure_is_distinct_from_head_failure(monkeypatch):
+    """A base-side API failure must remain typed as base evidence failure."""
+    monkeypatch.setattr(
+        noema,
+        "fetch_changed_files",
+        lambda repo, number: [("gone.py", "removed")],
+    )
+
+    def fail_fetch(repo, path, ref):
+        raise RuntimeError("HTTP 502: token ***")
+
+    monkeypatch.setattr(noema, "fetch_head_file_content", fail_fetch)
+
+    context = noema.changed_file_context(
+        "owner/repo", 7, "head-sha", "base-sha"
+    )
+
+    assert "Unavailable from base content API" in context
+    assert "Unavailable from head content API" not in context
+
+
+def test_build_review_context_passes_live_base_ref(monkeypatch):
+    """The GraphQL base identity must reach changed-file context construction."""
+    observed: list[tuple[str, int, str, str]] = []
+    monkeypatch.setattr(noema, "review_thread_context", lambda pr: "")
+    monkeypatch.setattr(noema, "load_codegraph_context", lambda: "")
+
+    def fake_context(repo, number, head_sha, base_sha=""):
+        observed.append((repo, number, head_sha, base_sha))
+        return "files"
+
+    monkeypatch.setattr(noema, "changed_file_context", fake_context)
+
+    result = noema.build_review_context(
+        "owner/repo",
+        7,
+        {"headRefOid": "head-sha", "baseRefOid": "base-sha"},
+    )
+
+    assert observed == [("owner/repo", 7, "head-sha", "base-sha")]
+    assert "## Changed file context\nfiles" in result

tests/test_repository_branch_coverage_javascript_and_noema.py

@@ -176,9 +176,7 @@ def test_noema_review_context_includes_locations_bodies_and_all_sections(
     assert "src/runtime.py:7" in rendered
     assert "reviewer: Fix this" in rendered
 
-    monkeypatch.setattr(noema, "load_codegraph_context", lambda: "graph")
     monkeypatch.setattr(noema, "changed_file_context", lambda *_args: "files")
     context = noema.build_review_context("owner/repo", 1, pr)
-    assert "CodeGraph context" in context
     assert "Prior review threads" in context
     assert "Changed file context" in context

tests/test_repository_branch_coverage_reporting_edges.py

@@ -129,7 +129,6 @@ def test_noema_small_diff_and_empty_context_branches(
     rendered_context = noema.review_thread_context(pr)
     assert rendered_context == "- Thread open at src/runtime.py:\n  - reviewer: note"
 
-    monkeypatch.setattr(noema, "load_codegraph_context", lambda: "")
     monkeypatch.setattr(noema, "review_thread_context", lambda _pr: "")
     monkeypatch.setattr(noema, "changed_file_context", lambda *_args: "")
     assert noema.build_review_context("owner/repo", 1, pr) == ""