fix: enforce TEPP request and search evidence boundaries

CHANGELOG.d/2.12.7-korean-search-boundary.md

@@ -0,0 +1,7 @@
+# 2.12.7 — Preserve Korean search corroboration boundaries
+
+## Fixed
+
+- Accept a Korean organization token followed by a grammatical particle in a
+  natural Searxng snippet while still rejecting the token inside a longer
+  unrelated Hangul word.

backend/app/analysis_run_start.py

@@ -11,7 +11,7 @@
 
 import hashlib
 import json
-from datetime import datetime, timezone
+from datetime import UTC, datetime
 from typing import Any
 from uuid import UUID
 
@@ -21,15 +21,15 @@
     AnalysisRunCreateError,
     fetch_visible_analysis_run,
 )
-from backend.app.post_eligibility import SOURCE_POST_ELIGIBILITY_SQL
 from backend.app.analysis_run_outbox import (
     latest_outbox_delivery_is_claimed,
     latest_outbox_delivery_is_delivered,
     outbox_request_digest,
 )
 from backend.app.lineage_ingestion import records_from_source_posts
+from backend.app.post_eligibility import SOURCE_POST_ELIGIBILITY_SQL
 from lineageweave.adjudication_client import AdjudicationClient
-from lineageweave.http_client import HttpClientError, post_json
+from lineageweave.http_client import post_json
 from lineageweave.lineage_persistence import lineage_edge_specs
 from lineageweave.models import Edge
 from lineageweave.tepp_client import AnalysisRunRequest, TeppClient, TeppNotAvailable
@@ -103,8 +103,8 @@ def transport(payload: dict[str, Any]) -> dict[str, Any]:
         try:
             headers = {"authorization": f"Bearer {api_key}"} if api_key.strip() else {}
             return post_json(url, payload, headers=headers, timeout=30.0)
-        except (HttpClientError, OSError, ValueError, TypeError) as exc:
-            raise TeppNotAvailable("TEPP transport unavailable") from exc
+        except Exception as exc:
+            raise TeppNotAvailable("TEPP transport request failed") from exc
 
     return TeppClient(transport=transport)
 
@@ -119,12 +119,12 @@ def tepp_run_request(
     """Build TEPP's published request from the frozen run, never a theta."""
     cutoff = knowledge_cutoff
     if cutoff.tzinfo is None:
-        cutoff = cutoff.replace(tzinfo=timezone.utc)
+        cutoff = cutoff.replace(tzinfo=UTC)
     return AnalysisRunRequest(
         idempotency_key=idempotency_key,
         tenant_workspace_id=str(corporate_entity_id),
         snapshot_id=snapshot_sha256,
-        knowledge_cutoff=cutoff.astimezone(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ"),
+        knowledge_cutoff=cutoff.astimezone(UTC).strftime("%Y-%m-%dT%H:%M:%SZ"),
         model_contract_version=_TEPP_MODEL_CONTRACT,
         output_profile=_TEPP_OUTPUT_PROFILE,
     )
@@ -610,7 +610,7 @@ async def deliver_queued_analysis_run(
         return await _visible_or_404(
             conn, analysis_run_id, account_id, affiliated_entity_ids
         )
-    now = datetime.now(timezone.utc)
+    now = datetime.now(UTC)
     try:
         if not latest_outbox_delivery_is_claimed(latest):
             await _append_outbox_delivery(
@@ -636,9 +636,8 @@ async def deliver_queued_analysis_run(
                 affiliated_entity_ids=affiliated_entity_ids,
                 adjudication_client=adjudication_client,
             )
-        finished = datetime.now(timezone.utc)
-        if finished < now:
-            finished = now
+        finished = datetime.now(UTC)
+        finished = max(finished, now)
         await _append_outbox_delivery(
             conn,
             analysis_run_id,
@@ -699,7 +698,7 @@ async def _deliver_lineage_reconstruction(
     adjudication_client: AdjudicationClient | None = None,
 ) -> None:
     """Persist ThreadWeave parent choices for the frozen bag."""
-    now = datetime.now(timezone.utc)
+    now = datetime.now(UTC)
     member_rows = await _snapshot_member_posts(
         conn,
         locked["analysis_source_snapshot_id"],
@@ -715,9 +714,8 @@ async def _deliver_lineage_reconstruction(
         )
     edges = lineage_edge_specs(records_from_source_posts(rows), llm=adjudication_client)
     digest = reconstruction_result_digest(edges)
-    finished = datetime.now(timezone.utc)
-    if finished < now:
-        finished = now
+    finished = datetime.now(UTC)
+    finished = max(finished, now)
     await conn.execute(
         """
         insert into analysis_run_reconstruction
@@ -759,25 +757,23 @@ async def _deliver_tepp_measurement(
     tepp_client: TeppClient,
 ) -> None:
     """Submit the frozen snapshot through ``tepp_client``. Never persist a theta."""
-    now = datetime.now(timezone.utc)
+    now = datetime.now(UTC)
     request = tepp_run_request(
         idempotency_key=str(locked["idempotency_key"]),
         snapshot_sha256=str(locked["snapshot_sha256"]),
         knowledge_cutoff=locked["knowledge_cutoff"],
         corporate_entity_id=str(locked["corporate_entity_id"]),
     )
     status_code, failure_code, envelope = _tepp_submission(tepp_client, request)
-    if status_code == _SUCCEEDED and envelope is not None:
-        if not await _persist_tepp_result(
-            conn,
-            analysis_run_id=analysis_run_id,
-            envelope=envelope,
-        ):
-            status_code = _FAILED
-            failure_code = "tepp_result_not_persisted"
-    finished = datetime.now(timezone.utc)
-    if finished < now:
-        finished = now
+    if status_code == _SUCCEEDED and envelope is not None and not await _persist_tepp_result(
+        conn,
+        analysis_run_id=analysis_run_id,
+        envelope=envelope,
+    ):
+        status_code = _FAILED
+        failure_code = "tepp_result_not_persisted"
+    finished = datetime.now(UTC)
+    finished = max(finished, now)
     await _append_status(
         conn,
         analysis_run_id,

docs/adr/0005-relation-verification-agent.md

@@ -59,6 +59,12 @@ keeps the channel unavailable (never fabricates a verification result)
 when `SEARXNG_BASE_URL` is unset, same discipline as every other
 pluggable client in this repo.
 
+Evidence token selection excludes generic fixture descriptors such as
+`fictitious`, `nonexistent`, `placeholder`, `sample`, `example`, and `demo`.
+Those words can appear in unrelated search results and are not organization
+identity. This keeps synthetic demo data out of the corroboration signal while
+retaining distinctive organization tokens and cited URLs.
+
 Persistence: `post_counterparty_entity` gains
 `verification_status_code` (`common_lookup_value` category
 `relation_verification_status`: `verify_pending` / `verify_corroborated`

docs/adr/0022-authorized-tepp-start.md

@@ -37,6 +37,8 @@ authorized transaction:
 4. submits through `TeppClient`. An empty `TEPP_TRANSPORT_URL` keeps the
    default unavailable transport. A set URL POSTs the published wire
    payload through the http(s)-only helper. File URLs stay unavailable;
+   `AnalysisRunRequest` rejects non-v1 versions and blank/non-text required
+   fields before the transport is called, matching TEPP's v1 JSON Schema;
 5. appends Failed / `tepp_not_available` when the transport is missing
    or refused, or Failed / `tepp_result_not_persisted` when TEPP accepts
    an envelope this product cannot store yet.
@@ -93,3 +95,7 @@ World Wide Web Consortium. https://www.w3.org/TR/prov-dm/
 
 World Wide Web Consortium. (2013). *PROV-O: The PROV ontology* (W3C
 Recommendation). https://www.w3.org/TR/prov-o/
+
+ContextualWisdomLab. (2026). *TEPP API and modular integration contract*
+([Computer software documentation]). GitHub.
+https://github.com/ContextualWisdomLab/TEPP/blob/main/docs/API_CONTRACT.md