Triage monitors regulatory filings across every holding in the Militia Long/Short Equity ETF and ranks them by how much they matter to the fund. It pulls filings from seven countries, scores each one on a 0-to-10 scale, generates a plain-English summary, and surfaces the ones that need attention.
The goal is to catch material regulatory disclosures before they move the stock. A routine annual report from a 0.3% position gets buried. An unexpected current report from a 5% long gets flagged immediately.
Connectors run on a schedule and pull filings from each source. Each connector handles authentication, rate limits, and format differences for its jurisdiction.
Each filing receives a Composite Materiality Score (CMS) from 0 to 10. The CMS combines five sub-scores, each measuring a different dimension of how much the filing matters to the fund.
| Sub-Score | Weight | What It Measures |
|---|---|---|
| PWI (Position Weight Impact) | 25% | How large the holding is in the fund. A 5% position scores higher than a 0.2% position because a price move hits the fund harder. Computed directly from portfolio weights using a log formula, no AI involved. |
| FTS (Filing Type Severity) | 15% | How important this type of filing typically is. An 8-K current report starts with a higher base than a routine 10-K annual report. Base scores come from a filing type taxonomy table, then AI adjusts based on actual content. |
| CSF (Content Surprise Factor) | 25% | How much the filing departs from what the market expected. A restatement of prior earnings or an unexpected CEO departure scores high. A routine quarterly update scores low. AI-scored by reading the filing text. |
| CFI (Cash Flow Impact) | 25% | Estimated effect on the company's future cash flows. A major contract win, revenue warning, or restructuring charge scores high. A change in auditor or board committee composition scores low. AI-scored. |
| DA (Directional Alignment) | 10% | Whether the news cuts for or against the fund's position. Bad news on a long position (or good news on a short) scores highest, because that is the scenario where the fund needs to act. AI-scored. |
After scoring, Claude Sonnet reads the filing text along with the scorer's reasoning and the fund's position context (weight, direction, holding name). It produces structured output with several fields: a plain-English summary, a rationale explaining why the filing matters to the fund, key numbers extracted from the filing, risk flags, entities mentioned, and financial changes.
For non-English filings (Japanese, Spanish, German, French, Chinese), the summarizer also produces a side-by-side translation. The scorer's sub-score explanations are passed into the summarizer prompt so the summary synthesizes real analysis rather than inventing reasoning from scratch.
| Source | Jurisdiction | Filing Types |
|---|---|---|
| EDGAR | United States | 10-K, 10-Q, 8-K, DEF 14A, SC 13D/G, Form 4, S-1 |
| TDnet | Japan | Earnings, material facts, corporate actions, forecasts |
| EDINET | Japan | Securities reports (yuho), quarterly reports, amendments |
| SIX | Switzerland | Ad-hoc announcements, annual/interim reports |
| CNMV | Spain | Hechos relevantes, periodic financial reports |
| BMV | Mexico | Relevant events, financial statements |
| AMF | France | Regulated information, periodic reports |
| HKEx | Hong Kong | Announcements, circulars, financial reports |