fix: normalize off-list enum values in company profile schemas

GeographicExposure.exposure_types and BusinessLine.trend rejected any
model output outside their narrow Literal set, raising a validation
error that discarded the entire Company Overview for the run (caught
by synthesis_node's try/except, but silently — confirmed on a live
AAPL run right after the previous two fixes let the model correctly
produce values like "regulatory" and "geopolitical").

Both fields now normalize via a bounded synonym table (mirroring
RiskItem._normalize_category) before validation: known synonyms map
to the canonical value, unknown ones fall back safely (trend ->
not_disclosed) or are dropped with a stderr log (exposure_types).
"geopolitical" is added as a legitimate new category, not just an
alias. The synthesis prompt now spells out both closed enumerations
explicitly, matching the existing pattern for risks_categorized.category.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

agent/company_profile_schemas.py

@@ -1,13 +1,56 @@
 """Structured payload for the Company Overview research surface."""
 from __future__ import annotations
 
+import sys
 from typing import Any, Literal, Optional
 
-from pydantic import BaseModel, ConfigDict, Field
+from pydantic import BaseModel, ConfigDict, Field, field_validator
 
 from agent.schemas import SourcedFact
 
 
+_CANONICAL_TRENDS = {"growing", "stable", "declining", "mixed", "not_disclosed"}
+
+_TREND_ALIASES: dict[str, str] = {
+    "increasing": "growing",
+    "expanding": "growing",
+    "accelerating": "growing",
+    "rising": "growing",
+    "flat": "stable",
+    "steady": "stable",
+    "unchanged": "stable",
+    "decreasing": "declining",
+    "shrinking": "declining",
+    "contracting": "declining",
+    "falling": "declining",
+    "varied": "mixed",
+    "uneven": "mixed",
+}
+
+_CANONICAL_EXPOSURE_TYPES = {
+    "revenue",
+    "operations",
+    "supply_chain",
+    "regulation",
+    "geopolitical",
+}
+
+_EXPOSURE_TYPE_ALIASES: dict[str, str] = {
+    "regulatory": "regulation",
+    "legal": "regulation",
+    "compliance": "regulation",
+    "sales": "revenue",
+    "manufacturing": "operations",
+    "production": "operations",
+    "sourcing": "supply_chain",
+    "suppliers": "supply_chain",
+    "political": "geopolitical",
+    "geo_political": "geopolitical",
+    "trade": "geopolitical",
+    "tariffs": "geopolitical",
+}
+
+
 class CompanyIdentity(BaseModel):
     model_config = ConfigDict(extra="ignore")
 
@@ -30,18 +73,60 @@ class BusinessLine(BaseModel):
     revenue_share_pct: Optional[float] = Field(default=None, ge=0, le=100)
     share_period: Optional[str] = None
 
+    @field_validator("trend", mode="before")
+    @classmethod
+    def _normalize_trend(cls, v: object) -> str:
+        if isinstance(v, str):
+            normalized = "_".join(v.strip().lower().replace("-", " ").split())
+            if normalized in _CANONICAL_TRENDS:
+                return normalized
+            if normalized in _TREND_ALIASES:
+                return _TREND_ALIASES[normalized]
+        print(
+            f"[business-trend] coercing unknown trend '{v}' to not_disclosed",
+            file=sys.stderr,
+        )
+        return "not_disclosed"
+
 
 class GeographicExposure(BaseModel):
     model_config = ConfigDict(extra="ignore")
 
     name: str
     exposure_types: list[
-        Literal["revenue", "operations", "supply_chain", "regulation"]
+        Literal[
+            "revenue", "operations", "supply_chain", "regulation", "geopolitical"
+        ]
     ] = Field(default_factory=list)
     description: SourcedFact
     revenue_share_pct: Optional[float] = Field(default=None, ge=0, le=100)
     period: Optional[str] = None
 
+    @field_validator("exposure_types", mode="before")
+    @classmethod
+    def _normalize_exposure_types(cls, v: object) -> object:
+        if not isinstance(v, list):
+            return v
+        normalized_items = []
+        seen = set()
+        for item in v:
+            if not isinstance(item, str):
+                normalized_items.append(item)
+                continue
+            normalized = "_".join(item.strip().lower().replace("-", " ").split())
+            if normalized in _EXPOSURE_TYPE_ALIASES:
+                normalized = _EXPOSURE_TYPE_ALIASES[normalized]
+            elif normalized not in _CANONICAL_EXPOSURE_TYPES:
+                print(
+                    f"[exposure-type] dropping unknown exposure_type '{item}'",
+                    file=sys.stderr,
+                )
+                continue
+            if normalized not in seen:
+                normalized_items.append(normalized)
+                seen.add(normalized)
+        return normalized_items
+
 
 class StrategicChange(BaseModel):
     model_config = ConfigDict(extra="ignore")

agent/prompts.py

@@ -491,8 +491,8 @@ Required JSON structure:
 - risks_categorized: 3-6 items; category must be exactly one of: Regulatory, Operational, Competitive, Financial, Macro, Demand, Geopolitical — do not invent new buckets
 - management_commentary: 3-5 items (prefer MD&A sources; use transcript for tone/Q&A color not in filings)
 - guidance_history: up to 4 items, most recent first (cover the last 4 quarterly periods; one entry will typically be from an annual 10-K)
-- company_profile.business_lines: 3-5 items
-- company_profile.geographic_exposures: at most 8 items
+- company_profile.business_lines: 3-5 items; trend must be exactly one of: growing, stable, declining, mixed, not_disclosed — do not invent new labels
+- company_profile.geographic_exposures: at most 8 items; each exposure_types value must be exactly one of: revenue, operations, supply_chain, regulation, geopolitical — do not invent new labels (e.g. 'regulatory' is invalid, use 'regulation')
 - company_profile.strategic_changes: at most 3 items
 - company_profile.attention_themes: exactly 3 items
 - company_profile.watch_variables: 3-4 items

tests/test_company_profile.py

@@ -16,7 +16,8 @@ from agent.company_profile import (
     source_fingerprint,
 )
 from agent.evidence import evidence_envelope, make_evidence_record
-from agent.company_profile_schemas import CompanyProfile
+from agent.company_profile_schemas import CompanyProfile, CompanyProfileSection
+from agent.prompts import SYNTHESIS_STRUCTURED_PROMPT
 from analysis.company_attention import attach_attention_stats, cluster_questions
 from analytics.company_market import (
     _parse_published_date,
@@ -52,6 +53,94 @@ def _fact(status="VERIFIED"):
     }
 
 
+def test_geographic_exposure_normalizes_synonym_and_new_exposure_types():
+    section = CompanyProfileSection.model_validate(
+        {
+            "geographic_exposures": [
+                {
+                    "name": "Americas",
+                    "exposure_types": ["revenue"],
+                    "description": _fact(),
+                },
+                {
+                    "name": "Europe",
+                    "exposure_types": ["revenue", "regulatory"],
+                    "description": _fact(),
+                },
+                {
+                    "name": "Greater China",
+                    "exposure_types": ["operations", "geopolitical"],
+                    "description": _fact(),
+                },
+            ]
+        }
+    )
+
+    assert section.geographic_exposures[1].exposure_types == [
+        "revenue",
+        "regulation",
+    ]
+    assert section.geographic_exposures[2].exposure_types == [
+        "operations",
+        "geopolitical",
+    ]
+
+
+def test_geographic_exposure_drops_unknown_exposure_type():
+    section = CompanyProfileSection.model_validate(
+        {
+            "geographic_exposures": [
+                {
+                    "name": "Americas",
+                    "exposure_types": ["revenue", "interplanetary"],
+                    "description": _fact(),
+                },
+                {
+                    "name": "Asia",
+                    "exposure_types": ["Supply Chain"],
+                    "description": _fact(),
+                },
+            ]
+        }
+    )
+
+    assert section.geographic_exposures[0].exposure_types == ["revenue"]
+    assert section.geographic_exposures[1].exposure_types == ["supply_chain"]
+
+
+def test_business_line_trend_normalizes_synonyms_and_unknowns():
+    section = CompanyProfileSection.model_validate(
+        {
+            "business_lines": [
+                {"name": "Services", "description": _fact(), "trend": "Increasing"},
+                {"name": "Devices", "description": _fact(), "trend": "flat"},
+                {
+                    "name": "Other",
+                    "description": _fact(),
+                    "trend": "something_else",
+                },
+            ]
+        }
+    )
+
+    assert [line.trend for line in section.business_lines] == [
+        "growing",
+        "stable",
+        "not_disclosed",
+    ]
+
+
+def test_synthesis_prompt_enumerates_profile_literals():
+    assert (
+        "revenue, operations, supply_chain, regulation, geopolitical"
+        in SYNTHESIS_STRUCTURED_PROMPT
+    )
+    assert (
+        "growing, stable, declining, mixed, not_disclosed"
+        in SYNTHESIS_STRUCTURED_PROMPT
+    )
+
+
 def _metric_row(
     period,
     report_date,