What Exists Under The Platform
The intelligence engine contains a large private source archive and extracted document warehouse. It should power public source packs and premium research access, not be dumped raw into the index. The public library is the curated front of the archive; the full archive remains gated and quality-controlled.
Source Archive Scale
| Asset | Current Scale | Best Use |
|---|---|---|
| Raw source archive | 4.1GB, 78 source folders, 4,251 files | Evidence moat and premium document library |
| Documents table | 14,226 documents | Search, source packs, document profiles, gated exports |
| Extracted text | About 315M text characters | Search, summarization, fact extraction, topic expansion |
| Financial references | 57,251 extracted figures | Review queue, data products, premium financial intelligence |
| Tables | 45,186 extracted tables in the integrated parser output | Datasets, charts, source-pack evidence, data room products |
| Validation records | 97 cross-validated claims | High-confidence public evidence and premium dossiers |
Source Categories
Publication Rule
Source material becomes public only when it supports a page with clear corridor relevance, clean metadata, valid dates, usable source attribution, no empty modules, and no low-confidence financial noise. Everything else stays in the engine until reviewed.
Evidence Products
| Product | Public Surface | Gated Surface |
|---|---|---|
| Source packs | Source lists attached to public pages | Document-level packets with provenance, extraction notes, and citation trails |
| Entity dossiers | Company, country, DFI, mine, and operator profiles | Relationship evidence, source counts, risks, documents, and reviewed facts |
| Deal evidence | Clean deal pages and finance explainers | Review queues, extracted figures, confidence notes, and source conflict checks |
| Validation records | High-confidence claims inside editorial pages | Cross-source verification tables and analyst review status |
View Terminal Source Profiles Open Document Library Editorial Methodology