fix(v2 output_assembly): final-pass schema:url self-loop sweep

Empirical batch 4 (post-b3101f8) showed Bug P self-loops STILL
materialising on non-EPFL Persons across 4 separate runs (pyffs 1/1,
detect-libc 3/8, gimie 3/10, pandoc 14/31). The pattern: zero
self-loops on EPFL-affiliated repos, all of them on github-only
contributors.

Root cause: the LLM person agent emits `payload["id"] = "<urn-or-name>"`
and `payload["schema:url"] = "https://github.com/<login>"` separately;
my agent-level guard from 97285c1 ran AT THE AGENT, where those two
fields didn't match yet. Downstream canonicalisation then resolved
`id` to the github URL — and the self-loop materialised AFTER all the
agent-level guards had already run.

Fix at `output_assembly`, which is the first stage where every entity
is in its final canonical form (post-id-resolution, pre-jsonld-build).
`_drop_person_url_self_loops()` walks Person entities, unwraps both
the bare-string and `{"@id": ...}` shapes of `schema:url`, and pops
the field when the value equals the resolved `id`. Applied to both
the root entity and `related_entities`. Mutates in place — the
upstream payloads have already been deep-copied at this point.

Companion: regenerated `tests/v2/golden/extract/*.json` because the
octocat user/Person fixture had a self-loop the sweep now drops.

Author: caviri

src/v2/pipeline/stages/output_assembly.py

@@ -230,6 +230,20 @@ def assemble_output(  # noqa: C901, PLR0912
     related_entities, dedup_warnings = _merge_duplicate_entities(related_entities)
     warnings.extend(dedup_warnings)
 
+    # Person `schema:url` self-loop drop — final canonical-form sweep.
+    # The LLM person agent's own guard runs BEFORE canonicalization
+    # has rewritten `id`, so when the LLM emits
+    # `"id": "<urn>", "schema:url": "https://github.com/X"` the
+    # equality check fails at agent time. Downstream stages then
+    # resolve `id` to the github URL and the self-loop materialises in
+    # the final graph. Re-check here, after every id has been
+    # canonicalised, and drop the redundant url. Production audit
+    # observed 14 self-loops in `gabyx/pandoc` alone after the
+    # agent-level fixes shipped.
+    _drop_person_url_self_loops(related_entities)
+    if isinstance(root_entity, dict):
+        _drop_person_url_self_loops([root_entity])
+
     # Article ↔ Repo linkback. Production audit (Bug V) found
     # `schema:citation` literally `null` in 441/441 repos, including
     # the 19 repos that successfully extracted ``schema:ScholarlyArticle``
@@ -256,6 +270,38 @@ def assemble_output(  # noqa: C901, PLR0912
     )
 
 
+def _drop_person_url_self_loops(entities: list[dict[str, Any]]) -> None:
+    """Mutate Persons in-place to drop `schema:url` self-loops.
+
+    A self-loop is `schema:url == id` (in either the bare-string or
+    `{"@id": ...}` form). Mutates the list in place since this stage
+    already deep-copies upstream payloads.
+    """
+    for entity in entities:
+        if not isinstance(entity, dict):
+            continue
+        type_value = entity.get("type") or entity.get("@type")
+        if isinstance(type_value, list):
+            is_person = any("schema:Person" in str(t) for t in type_value)
+        else:
+            is_person = isinstance(type_value, str) and "schema:Person" in type_value
+        if not is_person:
+            continue
+        pid = entity.get("id") or entity.get("@id")
+        if not isinstance(pid, str):
+            continue
+        url = entity.get("schema:url")
+        target: str | None = None
+        if isinstance(url, str):
+            target = url.strip()
+        elif isinstance(url, dict):
+            inner = url.get("@id") or url.get("id")
+            if isinstance(inner, str):
+                target = inner.strip()
+        if target is not None and target == pid.strip():
+            entity.pop("schema:url", None)
+
+
 def _link_articles_to_root_repo(
     root_entity: Any,
     related_entities: list[dict[str, Any]],

tests/v2/golden/extract/repo_github_com_owner_repo.json

@@ -114,9 +114,6 @@
         "@id": "https://github.com/octocat",
         "@type": "schema:Person",
         "schema:name": "The Octocat",
-        "schema:url": {
-          "@id": "https://github.com/octocat"
-        },
         "pulse:githubUsername": "octocat",
         "pulse:orcidIdentifier": null,
         "pulse:infosciencePersonIdentifier": null,
@@ -178,7 +175,7 @@
   "warnings": [],
   "stats": {
     "entities_count": 3,
-    "triples_count": 21,
+    "triples_count": 20,
     "stages_completed": [
       "context_gather",
       "repo_agent",

tests/v2/golden/extract/user_github_com_username.json

@@ -14,7 +14,6 @@
       },
       "idSource": "pulse:githubUsername",
       "schema:name": "The Octocat",
-      "schema:url": "https://github.com/octocat",
       "pulse:githubUsername": "octocat",
       "pulse:orcidIdentifier": null,
       "pulse:infosciencePersonIdentifier": null,
@@ -37,7 +36,6 @@
           },
           "idSource": "pulse:githubUsername",
           "schema:name": "The Octocat",
-          "schema:url": "https://github.com/octocat",
           "pulse:githubUsername": "octocat",
           "pulse:orcidIdentifier": null,
           "pulse:infosciencePersonIdentifier": null,