22 Slice Regulatory Approvals

Pipeline Slice: Regulatory Approvals

Status: SHARP CORE VERTICAL / IN-PLACE PUBLICATION. Fixture-complete multi-authority capture, live FDA approval, REMS/CDx, and scoped label capture, source-current effective claims, normalization, evidence, resolution, canonical assertions, product protections, six query models, contracts, and provenance are implemented. Live compound-label approved-use decomposition, non-FDA live adapters, and approval-notification/designation feeds remain source expansions; atomic approvals-family publication remains an activation gate. Denoising state: sharp (noise -> rough -> sharp -> converged) Updated: 2026-07-25

Purpose

Model regulatory approval intelligence: source approvals, labels, designations, indication language, safety programs, companion diagnostics, product exclusivity/patent records, and structured label-derived evidence. This slice turns regulator-published artifacts into quality-passing regulatory assertions that agents can query through SQL; reviewed curation overlays apply only when the normal pipeline needs correction.

Legacy Data Gov treated approvals as a multi-regulator unification problem: FDA, EMA, KEGG/PMDA, and CDE/NMPA sources were collected into source tables, matched to canonical drugs, assembled into unified DrugApproval records, enriched with labels, and parsed into structured indications (../bioloupe-data-gov/docs/drug-approvals.md, ../bioloupe-data-gov/lib/tasks/regulatory/, ../bioloupe-data-gov/app/models/drug_approval.rb).

Current Decided Shape

Regulatory approvals produce evidence-backed approval and indication assertions, not merely a list of approved products. The slice owns source-specific regulatory identity and label evidence; drug/disease/biomarker identities are resolved through exact/cross-reference/verifier policy before publishing to q_*; curation is the reviewed override path.

flowchart LR
  FDA[FDA / openFDA / Drugs@FDA / labels]
  FDASupp[FDA REMS / companion diagnostics]
  EMA[EMA public medicines data]
  KEGG[KEGG / PMDA-derived drug source]
  CDE[CDE / NMPA source records + labels]
  Raw[Raw approval snapshots]
  LabelRaw[Immutable FDA label versions]
  SuppRaw[Raw supplement snapshots]
  Normalized[Normalized approval records]
  SuppNorm[Normalized supplement records]
  SuppLinks[Approval supplement links]
  Evidence[Approval / indication / label evidence claims]
  Resolution[Drug, disease, biomarker, organisation resolution]
  Verify[Verifier / accepted basis]
  Curation[Reviewed curation overlay]
  Canonical[Canonical approval assets]
  Query["q_drug_approvals family"]

  FDA --> Raw
  FDASupp --> SuppRaw
  EMA --> Raw
  KEGG --> Raw
  CDE --> Raw
  Raw --> Normalized --> Evidence --> Resolution --> Verify --> Canonical --> Query
  FDA --> LabelRaw --> Evidence
  SuppRaw --> SuppNorm --> SuppLinks --> Evidence
  Curation -. optional override .-> Canonical

Source Inventory

Source	Greenfield role	Source identifiers	Legacy evidence	Web/reference evidence
FDA openFDA / Drugs@FDA	US application/product/label source	NDA/BLA/ANDA application number, submission number, SPL set id where available	regulatory:fda:*, FdaDatum, FDA label enrichment (../bioloupe-data-gov/docs/drug-approvals.md)	openFDA exposes drug endpoints including product labeling and Drugs@FDA data: https://open.fda.gov/apis/drug/
FDA Orange Book	live FDA product approval/exclusivity/patent source, not a stale legacy cache	application number, product number	legacy supplementary Orange Book records are used but local rows were stale in June 2026	FDA publishes Orange Book data files: https://www.fda.gov/drugs/drug-approvals-and-databases/orange-book-data-files
FDA Purple Book	live biological product approval source and biologic relationship evidence	BLA/application/product names	regulatory:purplebook:sync, PurplebookRecord were active in June 2026	FDA Purple Book downloads list current CSV/XLSX files: https://purplebooksearch.fda.gov/downloads
DailyMed / SPL labels	official structured label sections for indications, warnings, dosing, adverse reactions	SPL set id/version, application numbers	Fda::LabelSync, Fda::DailyMedClient, boxed warnings extraction	DailyMed provides SPL labeling resources; openFDA also exposes product labeling: https://dailymed.nlm.nih.gov/dailymed/ and https://open.fda.gov/apis/drug/label/
EMA public medicines data	EU centralized medicine authorizations and product information	EMA product number, medicine name, authorization status	regulatory:ema:download_and_extract, EmaApprovedDrug	EMA publishes human medicines data and product information downloads: https://www.ema.europa.eu/en/medicines
KEGG drug database	Japan/PMDA-linked approval/reference drug source	KEGG drug id / remark codes	KeggApprovedDrug, KEGG scraper	KEGG DRUG is a drug information resource with REST access patterns: https://www.kegg.jp/kegg/drug/
CDE / NMPA	China chemical/marketed approval source and official labels	CDE/NMPA application/source ids; Chinese source fields	regulatory:cde:collect, ChinaApprovedDrug, CDE PDF label processing (../bioloupe-data-gov/docs/drug-approvals.md)	CDE public portal/source systems: https://www.cde.org.cn/
FDA companion diagnostics / REMS	live supplementary regulatory facts linked to approval context; not approval rows	device/test ids, REMS program ids, source application numbers	companion diagnostics and REMS tasks/models were active in June 2026	FDA publishes companion diagnostic and REMS resources through official pages: https://www.fda.gov/medical-devices/in-vitro-diagnostics/list-cleared-or-approved-companion-diagnostic-devices-in-vitro-and-imaging-tools and https://www.fda.gov/drugs/drug-safety-and-availability/risk-evaluation-and-mitigation-strategies-rems
FDA approval notifications / breakthrough designations	deferred regulatory events/designations, not approval rows or safety claims	notification URLs, event dates, designation ids	notifications and breakthroughs tasks/models were active in June 2026	FDA publishes approval-notification resources through official pages: https://www.fda.gov/drugs/resources-information-approved-drugs/hematologyoncology-cancer-approvals-safety-notifications

Source Decision

Model FDA, EMA, KEGG, and CDE as first-class source families because the product intentionally unifies all four. The complete fixture graph exercises every family. FDA has live approval and supplement adapters; EMA, KEGG, and CDE remain fixture/import-ready until their exact public live contracts are implemented. That adapter maturity never changes the region-aware canonical model.

FDA labels remain part of this regulatory domain but use a separate source interface: regulatory_label_source. The label endpoint has a different configured cohort, paging limit, document-version selector, and activation scope from the bulk approval/product downloads. Its Dagster definition and fixture live under defs/sources/regulatory/; reusable mechanics remain in the deeper lib/sources/regulatory_labels/ module. This is one domain owner with two runtime interfaces, not a second SOC-owned source family.

evidence.approval_claims_effective, evidence.indication_claims_effective, and evidence.regulatory_safety_claims_effective admit claims only from each normalized record’s current immutable source artifact; supplement claims also require the current accepted supplement-to-approval link. Canonical rebuilds read these effective views, so a corrected source snapshot cannot leave stale prior-artifact approval facts active. Migration 20260725_0008 applies the same rule to an already-versioned database.

Asset Graph for This Slice

flowchart TD
  SourceSystems[[Regulatory sources]]
  Raw[sources.regulatory.*_snapshots]
  SuppRaw[sources.regulatory.fda_supplement_snapshots]
  Normalized[normalized.approval_source_records]
  LabelRaw[sources.regulatory.fda_label_snapshots]
  EmbeddedLabelNorm[normalized.label_documents]
  LabelNorm[normalized.regulatory_label_records]
  SuppNorm[normalized.regulatory_supplement_records]
  SuppLinks[resolution.approval_supplement_links]
  ApprovalClaims[evidence.approval_claims]
  IndicationClaims[evidence.indication_claims]
  SafetyClaims[evidence.regulatory_safety_claims]
  Resolution[resolution.regulatory_entities]
  Curation[[curation.effective_overrides]]
  Canonical[canonical.drug_approvals]
  Query["q_drug_approvals family"]

  SourceSystems --> Raw --> Normalized --> ApprovalClaims --> Resolution
  SourceSystems --> SuppRaw --> SuppNorm --> SuppLinks --> SafetyClaims
  Raw --> EmbeddedLabelNorm --> IndicationClaims --> Resolution
  LabelRaw --> LabelNorm
  LabelNorm --> SafetyClaims --> Resolution
  Resolution --> Canonical --> Query
  Curation -. reviewed overrides .-> Canonical

Implemented concrete assets

Layer	Asset key	Grain	Notes
source	sources.regulatory.approval_snapshot_store	one physical append-only multi-authority store	Single DDL owner; family capture assets depend on it and never create the table themselves.
source	sources.regulatory.fda_snapshots	one FDA application/product source snapshot	Preserve raw openFDA/Orange Book/Purple Book payload and source retrieval metadata.
source	sources.regulatory.fda_label_snapshots	one immutable FDA SPL/openFDA label version	Exact label capture with an explicit active-version selector and scoped activation metadata.
source	sources.regulatory.fda_supplement_snapshots	one FDA REMS program or CDx device-list row artifact	Preserve raw grouped REMS CSV payloads and official CDx table rows; not approval rows.
source	sources.regulatory.ema_snapshots	one EMA product row/document snapshot	Store source export version and product-info locator.
source	sources.regulatory.kegg_snapshots	one KEGG drug/source snapshot	Keep KEGG id and scraper/parser provenance.
source	sources.regulatory.cde_snapshots	one CDE API/PDF/source snapshot	Preserve Chinese source fields, extracted text, translation provenance.
normalized	normalized.approval_source_records	one source approval/product record	Source-faithful, not canonicalized.
normalized	normalized.label_documents	one embedded multi-authority label/document identity	Metadata projected from an approval snapshot; not the exact FDA section store.
normalized	normalized.regulatory_label_records	one selected FDA label document/section/study/outcome/AE record	Projects only explicitly active label versions; live compound indication sections remain source text until governed decomposition.
normalized	normalized.regulatory_supplement_records	one source-faithful REMS/CDx routing record	Carries source application numbers, drug terms, biomarker terms, and artifact locator.
evidence	evidence.approval_claims	one approval/status/date/region claim	Approval status, region, date, application number, product identity.
evidence	evidence.indication_claims	one parsed indication assertion	Disease, line/setting, biomarker, population, indication type.
evidence	evidence.regulatory_safety_claims	one label warning/REMS/boxed warning/CDx claim	Safety and diagnostic linkage facts.
resolution	resolution.approval_supplement_links	one accepted REMS/CDx-to-approval link	Exact source application-number link to normalized.approval_source_records; no fuzzy auto-link.
resolution	resolution.regulatory_entities	one candidate set per source term/subject	Drug, disease, biomarker, organisation, regulator identity.
canonical	canonical.drug_approvals	one effective approval assertion	Region/source-specific approval fact.
query	query.q_drug_approvals*	agent-facing grains below	Materialized SQL read models.

Source-capture implementation spec

Regulatory source capture must follow the source-resource standard in 12-source-first-ingestion.md. The implementation seam is one RegulatorySourceResource in lib/sources/regulatory/resource.py; the Dagster assets in defs/sources/regulatory/ remain thin.

class RegulatorySourceResource(dg.ConfigurableResource):
    fda_approval_mode: Literal["fixture", "live"] = "fixture"
    fda_supplement_mode: Literal["fixture", "live"] = "fixture"
    max_live_records: int | None = None
    source_http: SourceHttpResource

    def approval_capture(self, source_family: Literal["fda", "ema", "kegg", "cde"]) -> RegulatoryApprovalCapture: ...
    def supplement_capture(self) -> RegulatorySupplementCapture: ...

Exact FDA label capture uses a separate typed interface because its variation and activation semantics differ:

class RegulatoryLabelSourceResource(dg.ConfigurableResource):
    mode: Literal["fixture", "live", "import"] = "fixture"
    application_numbers: list[str] = []
    search_query: str = ""
    max_live_records: int | None = None
    source_http: SourceHttpResource

    def capture(self) -> RegulatoryLabelCapture: ...

Complete exact-application live capture may replace only those applications; complete free-query capture may upsert only observed label sets; bounded or incomplete capture cannot change the active selector; live capture can never replace the global inventory. Operator imports are captured immutably but do not activate automatically.

max_live_records is only a bounded live-smoke guard. Full source refreshes leave it unset. Any bounded live run must publish capture_scope=bounded_smoke and is_complete_source_capture=false so downstream users do not mistake the sample for the full regulatory corpus.

File/module	Required role
lib/sources/regulatory/models.py	typed approval/supplement capture records, source locator/value objects, source-mode metadata
lib/sources/regulatory/adapters.py	fixture and live adapters for openFDA/Drugs@FDA, Orange Book, Purple Book, REMS, CDx, later EMA/KEGG/CDE
lib/sources/regulatory/resource.py	source mode, max_live_records, base URLs, nested source_http, retry/error classification
lib/sources/regulatory/repository.py	DDL/upsert into sources_regulatory.approvals_snapshots and sources_regulatory.supplement_snapshots, source-artifact registration
defs/sources/regulatory/approvals.py	Dagster asset declarations only: keys, groups, pools, resource calls, repository writes, metadata
defs/sources/regulatory/supplements.py	same as approvals; no env flag branch and no direct low-level HTTP dependency
defs/sources/regulatory/labels.py	thin FDA label asset declaration; owns no HTTP, parsing, or activation logic
lib/sources/regulatory_labels/	label-specific model, fixture/live/import adapters, active-version policy, and immutable repository

The source assets use pool="source_fda" for FDA live calls and pool="postgres" when bulk database writes are the bottleneck. Default test and dev materializations use fixtures. Live smoke tests mock source_http by default and only hit the real internet under an explicit opt-in run.

Rejected alternatives:

Alternative	Rejection
Keep BIOLOUPE_LIVE_SOURCE_FDA_ENABLED / BIOLOUPE_LIVE_SOURCE_FDA_SUPPLEMENTS_ENABLED in asset bodies	The asset body becomes the variation seam, source mode is hidden from run config, and tests couple to branching instead of contracts.
Put openFDA download logic directly in approvals.py	Makes an already large Dagster definition file own HTTP, parsing, persistence, and graph metadata.
Separate fda_source, ema_source, kegg_source, cde_source resources immediately	Prematurely expands the approval interface. One approval resource is deeper until authority-specific auth/runtime constraints force a split.
Put FDA labels inside RegulatorySourceResource only to minimize resource count	Resource count is not the design goal. Label API, cohort, paging, version selection, and activation vary together and justify one small stable interface.

Slice-Local Domain Model

Regulatory approvals stress regionality, source authority, label versioning, and indication decomposition.

Entity	Grain	Canonical key recommendation	Notes
Regulatory source record	one source-published product/application row	reg_source:{source}:{source_record_id}	Source identity remains stable even if canonical links change.
Regulatory supplement artifact	one REMS program or CDx device-list row artifact	reg_supplement:{source}:{artifact_kind}:{source_record_id}	Routes official supplemental evidence to approvals without becoming an approval row.
Approval assertion	one product/drug approval in one region/source context	approval:{source}:{source_record_identity}:{region}:{status_or_date_hash}	Source-record identity is product-grained where the source is product-grained; application number alone is not unique.
Label document/version	one official label/source document version	label:{source}:{document_id}:{version_or_seen_at}	Label text is evidence, not the canonical indication itself.
Indication assertion	one structured use from a label/source record	indication:{approval_key}:{disease_context}:{line_setting_biomarker_hash}	A single label can yield many indications.
Regulatory designation	one designation/program fact	designation:{source}:{designation_type}:{subject_id}	Breakthrough, orphan, PRIME, accelerated, conditional, etc.
Regulatory product protection record	one source-scoped product + patent/exclusivity record	product_protection:{source}:{application}:{product}:{kind}:{source_record_id}	Patent and regulatory exclusivity are distinct. An expiry date is source context, not a universal loss-of-exclusivity date.
Safety program / warning	one REMS/boxed-warning/safety claim	safety:{source}:{program_or_warning_id}	Safety evidence may be approval-wide or indication-specific.
Companion diagnostic link	one drug/biomarker/test/indication relation	cdx:{source}:{test_or_device_id}:{drug_or_indication}	Requires drug + biomarker/test resolution.
Applicant / holder / partner role	one organisation role on an approval	approval_org_role:{approval_key}:{role}:{org_term_hash}	Ownership/partner semantics may be time-sensitive.

Invariants

• Approval facts are region- and source-scoped. A US FDA approval is not equivalent to EMA/CDE/KEGG approval.
• Drug approval is not the same grain as indication. One approval/label can contain multiple indications; one indication can require multiple biomarker/disease/treatment-line constraints.
• Label-derived evidence must retain the label/source document version and text locator.
• Translated CDE label claims must retain original Chinese text, translation version, and source PDF/API locator.
• Innovation type (original, generic, biosimilar, etc.) is a derived claim with source-specific mapping evidence, not a universal raw field.
• Patent and regulatory-exclusivity records retain source, region, application/product linkage, protection kind, source code/identifier, expiry date, use code where present, and source release. They do not collapse into one global loss_of_exclusivity fact.
• REMS and CDx source artifacts link to approvals only through accepted resolution.approval_supplement_links rows. Exact source application number is the first implemented basis; approval notifications and breakthrough designations remain separate event/designation work.

Slice-Local Entity Resolution

Subject kind	Source term examples	Candidate target types	Publishable basis	High-impact?
approval_drug_link	brand name, active ingredient, product name, CDE English fields	drug spine resolved_drug_key / brand product / ingredient concept	exact synonym, source cross-reference, context-verified, verifier-passed fuzzy/LLM, reviewed curation override	yes
approval_indication_disease_link	label disease language, CDE translated disease terms	canonical Disease / NCIt disease concept	exact/NCIt, context-verified, verifier-passed semantic/LLM, reviewed curation override	yes
approval_biomarker_link	PD-L1, HER2, EGFR mutation, companion diagnostic biomarker	canonical Biomarker / gene/protein/variant	exact/cross-reference/context-verified/reviewed curation override	yes
approval_organisation_link	applicant, MAH, sponsor, license holder, partner	canonical Organisation	exact legal name/alias, source cross-reference, reviewed curation override	medium/high
companion_diagnostic_link	test/device/biomarker/drug pair	diagnostic test/device + biomarker + drug	FDA exact device list, source cross-reference, reviewed curation override	high
approval_trial_link	label clinical-study reference, NCT in label	canonical Trial	exact registry id or verifier or reviewed curation override	medium/high

No approval-drug or indication-disease fuzzy match publishes without deterministic basis, verifier acceptance, or reviewed curation override. Legacy ActiveAdmin locks and source drug_id matches become explicit resolution receipts or curation overlays, not mutable columns.

Implemented approval-drug resolution is ingredient-first, product-name fallback. The resolver first tries the source active-ingredient/generic drug_term against the drug spine’s resolution.entity_names; if that does not produce a resolved key, it tries the source product_name/brand against the same spine. This uses source-derived canonical.drug_brands populated from approval/product evidence, so branded approvals like KEYTRUDA can resolve to drug:pembrolizumab without reading legacy brand_drugs, minting approval-local drug identity, or requiring NCIt/RxNorm authority enrichment to finish first. Authority-vocabulary external ids improve q_drugs; they are not a prerequisite for resolving a regulator’s own product row.

Evidence Claim Schemas

approval_claim

Field	Meaning
claim_id	stable id from source snapshot + field path + extractor version
source_family	FDA / EMA / KEGG / CDE / supplementary FDA source
source_record_key	source-scoped source id
application_number	NDA/BLA/ANDA/etc when present
product_name	source product/brand name
region	USA/EU/Japan/China/etc
approval_status	approved, withdrawn, pending, conditional, etc. source-faithful then normalized
approval_date	source date + precision
claim_value	structured JSON for source-specific fields
source_locator	API/download/PDF row locator

Supplement evidence claims reuse evidence.regulatory_safety_claims after the supplement artifact has an accepted approval_supplement_link. REMS emits safety_type='rems'; CDx emits safety_type='companion_diagnostic'. The source_record_key remains the linked approval source record so existing canonical/query assets materialize downstream, while source_artifact_key and source_locator point to the REMS/CDx artifact, not the approval snapshot.

indication_claim

Field	Meaning
approval_key	parent approval/source record
label_key	source label/document version
source_artifact_key	immutable label/API/PDF/source snapshot key
source_locator	SPL section, label section, API row, PDF page/span, or source field locator
raw_indication_text	exact label/source text span
disease_context_term	raw disease phrase
line_or_setting	treatment line, adjuvant/metastatic/maintenance/etc
biomarker_requirements	raw and structured biomarker constraints
population_constraints	age, prior therapy, disease subtype, geography, etc
indication_type	treatment / diagnosis / prevention / mitigation / unknown
resolution_subject_keys	disease/biomarker/drug/org subjects

Curation Override Subjects

Subject kind	Why overrideable	Example curation decision
approval_drug_link	Wrong drug match corrupts approvals, SOC, forecasting, and competitive landscape.	FDA application X resolves to Drug Y, not same-name unrelated brand.
approval_indication_assertion	Labels often contain compound, nested indication language.	Split one label paragraph into two disease-specific indications.
approval_indication_disease_link	Disease/subtype matching determines downstream SOC and market context.	“HER2-positive gastric cancer” resolves to disease + biomarker context.
innovation_type_assertion	Source-specific codes are not semantically uniform.	CDE category 5.1 maps to original for this source context.
approval_organisation_role	Applicant/holder/partner impacts ownership analytics.	Label applicant is legal holder, not commercial owner.
companion_diagnostic_link	CDx claims combine test, biomarker, drug, and indication.	FDA CDx row links test A to biomarker B and drug C for indication D.
product_protection_link	Application/product mismatches corrupt exclusivity context and commercial summaries.	Repoint an Orange Book patent/exclusivity row to the source product it actually covers.

Canonical Assets

Canonical asset	Grain	Built from
canonical.drug_approvals	one effective approval assertion	approval claims + drug/org resolution/verifier receipts; curation overlay when present
canonical.approval_indications	one structured indication	label/source text + disease/biomarker resolution/verifier receipts; curation overlay when present
canonical.regulatory_designations	one designation/program fact	breakthrough/orphan/conditional/etc source claims
canonical.companion_diagnostic_links	one drug-biomarker-test relation	FDA CDx evidence + entity resolution
canonical.regulatory_product_protections	one source-scoped product patent/exclusivity record	versioned Orange Book or equivalent regulator files + product/approval resolution

normalized.label_documents is the embedded multi-authority document-identity registry produced from approval snapshots. Exact FDA documents and sections are separately owned by sources_regulatory.fda_label_* and normalized.regulatory_label_*; this prevents approval rows from pretending to be complete label capture. Warning, boxed-warning, and REMS claims remain evidence-layer facts in evidence.regulatory_safety_claims and are surfaced through q_label_evidence; promote them to canonical.regulatory_safety_claims only when a reviewed safety-claim overlay exists. Approval organisation roles are deferred to the organisation-intelligence slice instead of modelled as a ghost canonical table here.

q_drug_approvals Family

Query model	Grain	Purpose
q_drug_approvals	one effective approval	region-aware approval search by drug, source, date, status, innovation type
q_approval_indications	one indication assertion	disease/biomarker/treatment-line indication discovery
q_label_evidence	one label/REMS/warning evidence span	trace label-derived and supplement-derived claims to source text
q_regulatory_designations	one designation/program fact	orphan/breakthrough/accelerated/conditional/PRIME/etc tracking
q_companion_diagnostics	one drug-biomarker-test relation	CDx intelligence for biomarker-linked therapies
q_regulatory_product_protections	one regulator product + protection record	inspect patent/exclusivity context and source expiry dates without pretending one date is universal LOE
q_provenance receipts, optionally exposed through q_regulatory_provenance	one or more receipts per provenance-required row/fact	global source snapshots + extraction runs + evidence claims + resolution/verifier receipts + curation overlays when applicable; q_regulatory_provenance may exist only as a convenience view over q_provenance

q_regulatory_product_protections requires regulatory_product_protection_key, source/region, application and product identifiers, resolved approval/drug/brand keys, protection_kind (patent or regulatory_exclusivity), source code/identifier, patent/use code when applicable, expiry date and precision, source release, assertion/supersession state, and provenance_key. The source record is authoritative for what the regulator file reports; legal validity, litigation outcomes, pediatric extensions, and market-entry predictions require separate evidence. 35-slice-commercial-intelligence.md may project selected protection context into a product summary but does not own or duplicate these records.

Downstream key contract

resolved_approval_key(s) used by downstream slices (regulatory-event mentions in 24-slice-news-intelligence.md, approval-backed evidence in 25-slice-standard-of-care.md and forecasting readiness in 28-slice-forecasting-tpp.md) is defined here as a stable key into q_drug_approvals — the same identity as the approval_key domain key (approval:{source}:{source_record_identity}:{region}:{status_or_date_hash}). It stays region- and source-scoped: a downstream resolved_approval_key is never a global “approved” flag, and downstream slices reference it rather than minting approval identity. Drug, disease, biomarker, and organisation identities inside this slice resolve into their owner spines (resolved_drug_key → 29, resolved_disease_key/resolved_biomarker_keys → 30, resolved_org_key → 26).

q_approval_indications comparative fact contract

q_approval_indications is the regulatory-label side of the shared therapeutic-fact model in 20-pipeline-families.md. It is an approved-use assertion from a regulator/source artifact, not a clinical outcome row. FDA labeling guidance treats the INDICATIONS AND USAGE section as the place to state approved uses and clinically significant limitations; DailyMed/openFDA expose structured label artifacts, while EMA/CDE/KEGG need source-family-specific locators and translations where applicable (FDA labeling guidance: https://www.fda.gov/media/114443/download; DailyMed SPL resources: https://dailymed.nlm.nih.gov/dailymed/; openFDA label endpoint: https://open.fda.gov/apis/drug/label/).

Contract field	Required meaning
Purpose	Answer: “For what patient population/context is this product approved or regulator-described, from which label/source, and with what constraints?”
Grain	One structured indication assertion for one approval/product/source label context + disease/population/line/biomarker slice.
Stable row key	approval_indication:{source_family}:{approval_key}:{label_key}:{indication_hash}.
Required columns	approval_indication_key, approval_key, label_key, source_family, region, regulator, application_number, product_name, resolved_drug_key, disease_term, resolved_disease_key, biomarker_terms, resolved_biomarker_keys, line_or_setting, population_constraints, indication_type, approval_status, approval_date, raw_indication_text, source_artifact_key, source_locator, translation_version, source_language, assertion_state, provenance_key.
Search affordances	filters on region/source/regulator, drug, disease, biomarker, line/setting, approval status/date, indication type; full text/trigram on raw indication text and product names.
Provenance rule	provenance_grain=bundle for every indication row; field receipts for raw_indication_text, approval_status, approval_date, line_or_setting, population_constraints, and translated fields.
Refresh/materialization	current runtime is an in-place plain view after label-document extraction, indication decomposition, disease/biomarker/drug resolution, verifier gates, and applicable curation-review gates; it is not last-known-good protected until the approvals family adopts atomic generation publication.
Blocking checks	schema_contract, provenance_coverage, source_artifact_immutability, source_locator, document_version, source_authority, resolution_basis, llm_verifier_required, curation_review_state when override applies.

Allowed fact_path examples:

approval_indication:approval=fda-nda-123:label=spl-abc:disease=nsclc:attr=label_text
approval_indication:approval=fda-nda-123:label=spl-abc:disease=nsclc:attr=approval_status
approval_indication:approval=cde-456:label=pdf-789:disease=gastric_cancer:attr=translation_version

This row can support SOC assertions, forecasting readiness, and competitive-landscape queries, but it remains source/regulator-scoped. It must not become a global drug is approved flag without region, source, indication, and provenance.

Current boundary: q_approval_indications materializes structured source fields from the approval-source path. The exact FDA label adapter captures live indications_and_usage sections but does not yet decompose compound section text into atomic approved-use claims. Those rows fail closed before downstream review candidates. Implementing and evaluating that governed regulatory extraction is a live activation blocker, not evidence that every raw section is already queryable as a structured indication. Per ADR-0011, even a successfully decomposed approved use remains distinct from standard of care.

q_approval_indications column contract seed

Column	Type	Nullability	Semantic type	Provenance
approval_indication_key	text	not null	query row key	row
approval_key	text	not null	approval assertion key	row
label_key	text	nullable	label artifact/version key	bundle when label-derived
source_family	text	not null	source family	row
region	text	not null	geography/regulatory region	row
regulator	text	nullable	regulator	row
application_number	text	nullable	source application id	bundle when present
product_name	text	not null	source product name	bundle
resolved_drug_key	text	nullable	resolved drug key	field when present
disease_term	text	nullable	source disease phrase	bundle when present
resolved_disease_key	text	nullable	resolved disease key	field when present
biomarker_terms	text[]	nullable	source biomarker phrases	bundle when present
resolved_biomarker_keys	text[]	nullable	resolved biomarker keys	field when present
line_or_setting	text	nullable	treatment line/setting	field when present
population_constraints	jsonb	nullable	patient population constraints	field when present
indication_type	text	not null	treatment/diagnosis/prevention/etc	bundle
approval_status	text	not null	regulatory status	field
approval_date	date	nullable	approval/effective date	field when present
raw_indication_text	text	nullable	label/source text	field when present
source_artifact_key	text	not null	immutable source artifact	bundle
source_locator	jsonb	not null	source locator	bundle
translation_version	text	nullable	translation version	field for translated sources
source_language	text	nullable	language code	row
source_authority_tier	text	not null	authority tier	row
assertion_state	text	not null	effective assertion state; curation overlay if applicable	row
provenance_key	uuid	not null	provenance bundle	bundle

Catalog notes: region, source_family, and indication_type should be controlled vocabularies. raw_indication_text is nullable only for non-label source records that express indication as structured fields; in that case source_locator must identify the source fields exactly.

q_approval_indications vocab/search/example seed

Legacy data-shape pressure from local datalake confirms that regulatory values should be normalized but source-preserving: drug_approvals.region is currently concentrated in USA, CN, EU, and JP; source_type includes FDA, China/CDE, EMA, KEGG, and Purple Book-derived records; approval_status includes drug_approved, drug_discontinued, drug_approval_unknown, drug_pending, and drug_rejected; indications.indication_type includes treatment, prevention, diagnosis, mitigation, and symptom_relief (public.drug_approvals, public.indications, sampled from local datalake).

Controlled vocabularies:

Field	Controlled values / normalization
region	USA, EU, CN, JP, with room for future ISO/region expansion.
source_family	fda, dailymed, openfda, ema, cde, nmpa, kegg, purple_book, orange_book, curation_overlay.
approval_status	approved, discontinued, withdrawn, pending, rejected, unknown, plus source-specific raw value retained in evidence.
indication_type	treatment, diagnosis, prevention, mitigation, symptom_relief, unknown.
innovation_type if exposed on sibling approval model	original, generic, non_original, biosimilar, biosimilar_interchangeable, unknown.
source_authority_tier	regulator_label, regulator_approval_database, regulator_supplementary, third_party_regulatory_projection, curation_overlay.

Search/index contract:

Search need	Required index/affordance
Approval/indication lookup by drug/region	btree on (resolved_drug_key, region, approval_status).
Disease/biomarker indication discovery	btree on resolved_disease_key; GIN on resolved_biomarker_keys.
Label/source text search	generated tsvector and trigram index over raw_indication_text, product_name, and disease_term.
Temporal/regulatory filtering	btree on approval_date, source_family, regulator.
Provenance audit	btree on source_artifact_key; hash/btree on provenance_key; btree on source_authority_tier.
Translation QA	partial index where source_language <> 'en' or translation_version is not null.

Example row shape:

approval_indication_key: approval_indication:fda:nda-123456:spl-abc:nsclc-pdl1
approval_key: approval:fda:nda-123456:usa:approved
label_key: label:dailymed:spl-abc:v2026-01-01
source_family: fda
region: USA
regulator: FDA
application_number: NDA123456
product_name: Examplemab
resolved_drug_key: drug:examplemab
resolved_disease_key: disease:nsclc
biomarker_terms: [PD-L1 TPS >= 50%]
line_or_setting: first_line_metastatic
indication_type: treatment
approval_status: approved
raw_indication_text: indicated for first-line treatment of metastatic...
source_authority_tier: regulator_label
provenance_key: <uuid>

Pathological examples the implementation must handle:

Case	Why it is dangerous	Required behavior
Label indication has no explicit disease phrase but structured source fields imply one	Some source records expose indication via fields rather than a clean label text span.	Permit raw_indication_text is null only when source_locator identifies exact source fields and the indication decomposition is provenance-complete.
CDE/NMPA translated indication lacks original Chinese text	Translation can change disease, line, and biomarker meaning.	Block accepted query publication until original text, translation version, and source artifact locator are present.
Publication result is tempting to treat as approval	A publication can support efficacy evidence but is not a regulator approval source.	Never publish q_approval_indications from publication-only authority; keep it in publication evidence/SOC support surfaces.
Same drug has different approval status across regions	A global approved flag erases regulatory reality.	Keep rows region/source scoped; agents must filter region, source_family, and approval_status.

Checks and Quality Gates

Blocking:

• every approval assertion has source family, source record key, region, and provenance bundle;
• every published approval-drug link has accepted basis;
• every resolved_drug_key, resolved_disease_key, resolved_biomarker_keys, and resolved_org_key value resolves to an active row in its owner q_* model (referential_integrity gate in 08-contracts-and-quality.md);
• every indication row has raw source text or source field locator;
• translated CDE-derived claims retain original text + translation provenance;
• every product protection row resolves to an exact source application/product record, preserves protection kind and source release, and does not publish a derived universal LOE date;
• schema of q_drug_approvals family matches generated q_catalog;
• no high-impact unresolved conflict publishes as accepted indication/drug truth.

Warning:

• source freshness drift by regulator;
• sudden count change by source/status/region;
• unmapped source submission class or innovation code;
• label parser section drift;
• unusually high unresolved indication/disease terms.

Source Expansion Boundary

• Add sampled, contract-tested live adapters for EMA, KEGG/PMDA, and CDE/NMPA without changing the source/evidence/query grains.
• Add FDA approval-notification and breakthrough-designation capture as event/designation sources, not approval rows.
• Expand label evidence from exact locators and bounded excerpts to fuller section capture only when storage, licensing, and agent retrieval needs justify it.
• Extend source-specific innovation mappings as evidence claims; never infer one universal innovation class from incompatible regulator codes.