1. Authoritative Federal Data Sources & Explicit Scope Boundaries
Every entity dossier, cohort rate, and shared 510(k) predicate lineage link in Predicate Ledger is compiled from public-domain U.S. Food and Drug Administration bulk archives without synthetic interpolation:
- CDRH 510(k) Premarket Notification Archives (
accessdata.fda.gov/premarket/ftparea/): Five historical and current pipe-delimited ASCII files (pmn7680,pmn8185,pmn8690,pmn9195, andpmn96cur) unified into 176,625 deduplicated clearances, repairing 4,088 Latin-1 character lines and 575 unescaped quote rows. - CDRH Product Classification (
foiclass.zip): 7,098 three-letter FDA product codes supplying clinical advisory panel assignments, device classes (I, II, III), implantable flags (27,147 clearances), and life-sustaining flags (5,737 clearances). - openFDA Device Enforcement & CDRH Recall Archives (
download.open.fda.gov): 59,703 CDRH recall rows and 40,450 enforcement rows deduplicated into 59,705 unique recall events (3,677 Class I, 54,988 Class II, 1,040 Class III). - External FDA Databases Excluded From Entity Counts: Standalone Premarket Approval (PMA) clinical trial dossiers, GUDID unique device identifier catalogs, and raw individual adverse event reports in FDA MAUDE (21 CFR Part 803) remain separate external FDA databases and are never conflated with verified 510(k) clearance or recall counts.
2. Deterministic K-Number Linkage & Broadcast Exclusion
OpenFDA's pre-joined recall endpoint populates a synthetic k_numbers list covering every historical clearance in a three-letter product code whenever an underlying recall notice omits explicit K-numbers. Predicate Ledger cross-examines explicit K-number citations against raw CDRH recall fields, excluding 19,984,540 synthetic broadcast joins and repairing 255 malformed text K-number strings to isolate 48,892 verified clearance-to-recall links across 13,125 cleared devices.
| Pipeline Audit Metric | Verified Count | Validation Rule |
|---|---|---|
| Deduplicated 510(k) Premarket Submissions | 176,625 | Unique canonical K-number key across 5 CDRH archives |
| Latin-1 Encoding & Unescaped Quote Repairs | 4,088 / 575 | Preserves legacy 1976 to 1995 applicant rows without truncation |
| FOICLASS Product Codes Matched | 7,098 | Repairs 71 lowercase recall code entries |
| Corporate M&A Parent Lineages Consolidated | 40 groups (46,481 510(k)s) | Maps 1,087 historical subsidiary strings to parent owners |
| Synthetic openFDA Broadcast Links Stripped | 19,984,540 | Rejects product-code-wide synthetic joins lacking explicit K-numbers |
| Verified 510(k)-to-Recall Linkages | 48,892 | Links 38,915 recall events to 13,125 distinct 510(k)s |
3. Automated Publication Quality Gate Verification
Before any publication snapshot is promoted to production, the reconciliation pipeline executes deterministic invariant checks verifying primary-key uniqueness, non-negative durations and counts, recall-to-clearance bounds, geographic bounds, and minimum denominator enforcement:
| Quality Gate Check | Rule Identifier | Observed Violations | Expected Violations | Status |
|---|---|---|---|---|
| 510(k) Primary Key Unique | 510k_primary_key_unique | 0 | 0 | PASS |
| Recall Primary Key Unique | recall_primary_key_unique | 0 | 0 | PASS |
| Review Duration Nonnegative | review_duration_nonnegative | 0 | 0 | PASS |
| Entity Primary Key Unique | entity_primary_key_unique | 0 | 0 | PASS |
| Nonnegative Entity Counts | nonnegative_entity_counts | 0 | 0 | PASS |
| Signal Not Exceeding Clearances | signal_not_exceeding_clearances | 0 | 0 | PASS |
| California Not Exceeding National | california_not_exceeding_national | 0 | 0 | PASS |
| Denominator Support Enforced | denominator_support_enforced | 0 | 0 | PASS |
4. Statistical Denominators, Historical Recall Spans, and Interpretive Boundaries
Percentage rankings (recalled_device_pct) require at least 20 cleared 510(k) submissions in the cohort so small denominators cannot produce misleading 50% or 100% outlier ratios. Cumulative recall figures span fifty years of federal records (1976 to 2026) and combine closed historical notices with any currently open field actions; always inspect individual recall statuses in the Device & Recall Explorer or live FDA Enforcement Reports before assessing current clinical status. Patients should never stop using or alter a prescribed medical device without consulting their physician.
5. Public-Data Architecture & Cross-Agency Reference Notes
Predicate Ledger shares deterministic reconciliation rules and SHA-256 provenance verification with sibling Meru AI healthcare and regulatory data platforms:
- Meru AI for Healthcare Practice: MedTech, Clinical, and Regulatory AI
- Healthcare Price Transparency & Pharmaceutical Patent-Cliff Crosswalk Guide
- Custom MedTech Regulatory Data Integration & Research Partnerships
- Cross-Agency Entity Resolution Across Federal Regulatory Archives
- Citation-Grade Public-Data Architecture and SHA-256 Provenance
- Meru AI Public-Data Platforms Overview