---
name: footnote
description: Forensic linguistic analysis of SEC corporate disclosure filings. Use when the user asks about how a company's language has changed in filings, how its rhetoric compares to peers, what's "buried" in MD&A or Risk Factors, or wants citation-anchored evidence about corporate disclosure tone.
---

# The Buried Footnote — MCP usage skill

This skill describes how to work with the `footnote` MCP server (https://theburiedfootnote.com/mcp). The server provides deterministic linguistic-signal analysis of US 10-K/10-Q and Canadian 40-F filings, with every claim citing a specific accession number on SEC EDGAR.

## When to invoke this skill

- The user asks about a public company's filings, MD&A, Risk Factors, hedging language, certainty markers, disclosure tone, or year-over-year language drift
- The user wants to compare two companies' disclosure language
- The user wants the weekly editorial dossier for an industry
- The user mentions "forensic linguistics", "disclaimer density", "hedging", or similar disclosure-analysis vocabulary

## Tool inventory

**13 tools — 5 free, 8 paid.**

| Tool | Tier | Use for |
|---|---|---|
| `list_industries` | free | **Always call first** to discover valid industry slugs and ticker rosters |
| `weekly_report` | free | Narrative markdown dossier for an industry-week, with citations |
| `recent_material_events` | free | Typed event stream (bullets) for a single ticker |
| `check_filing_drift` | free | Signal-level summary of a ticker's most-recent filing |
| `forensic_findings` | free | Forensic-accounting findings for ONE issuer: auditor's-report history (going concern, material weakness, opinion type, CAMs, tenure) + deterministic forensic signals (risk-factor drift, metric disappearance, non-GAAP expansion, related-party, balance-sheet, segment, rev-rec). Each item cites EDGAR. |
| `industry_signal_snapshot` | **paid** | Per-section raw signal grid (the full numeric matrix) for an industry's latest week |
| `compare_disclosure` | **paid** | Side-by-side signal comparison of two tickers, one section |
| `find_similar_disclosures` | **paid** | Embedding-based nearest-neighbour search |
| `filing_context` | **paid** | Full drift detail + bullets + metadata for one filing |
| `archive_search` | **paid** | Keyword search over the bullet archive |
| `request_brief` | **paid** | Request a forensic Brief on one issuer — Footnote's proprietary in-depth analysis applied to the deterministic forensic evidence → an adjudicated, citation-grounded brief. Optional `thesis` = red-team mode. Async + metered (consumes one monthly brief); returns a `brief_id`. |
| `get_brief` | **paid** | Poll a brief by `brief_id`: status, and when complete the adjudicated, citation-grounded brief + evidence count. Owner-only. |
| `bilingual_drift` | **paid** | EN/FR drift on Quebec issuer filings (Phase 2, stub) |

Anonymous free tier: 20 calls/day per IP. Paid: 2000/day, Authorization: Bearer <token>.

## Recommended workflows

### "What's interesting in [industry] this week?"
1. `list_industries` → confirm the slug
2. `weekly_report(industry: <slug>)` → returns the narrative dossier as markdown
3. Quote relevant passages back to the user with the citations intact (every claim has a `[section · signal · ticker = value]` trace)

### "How is [TICKER] hedging compared to peers?"
1. `list_industries` → find which industry the ticker belongs to
2. `weekly_report(industry: <slug>)` → the narrative dossier already positions each covered issuer against its cohort (free)
3. For the raw per-section numbers behind that positioning, `industry_signal_snapshot(industry: <slug>)` returns the full cohort grid — **paid**; compare the ticker's `tentative_markers`, `modal_verb_ratio`, `certainty_balance` in `item_1a` (Risk Factors) and `item_7` (MD&A) against peers

### "What did [TICKER] file recently?"
1. `recent_material_events(ticker: <T>)` → typed event bullets with confidence scores
2. `check_filing_drift(ticker: <T>)` → if the user wants signal-level detail

### "Is anything buried in [TICKER]'s filings? / What are the red flags?"
1. `forensic_findings(ticker: <T>)` → the auditor's-report history and forensic signals for that one issuer
2. Surface the high-severity items first (going concern, material weakness, non-clean opinion, sharp risk-factor rewrites, disappeared metrics), each with its EDGAR citation
3. These are **descriptive classifications of the disclosure**, not a verdict on the company — present them as "what the filings show," and let the user form the thesis (long or short)

### "Compare TSLA and AAPL on Risk Factors language"
- This is a paid tool (`compare_disclosure`). The free path: `weekly_report(industry: us_tech_megacap)` for the narrative cohort positioning, or `check_filing_drift` on each ticker for its section-level summary. Point the user to https://theburiedfootnote.com/subscribe for the side-by-side grid (`compare_disclosure` / `industry_signal_snapshot`).

### "Get a Brief on [TICKER]" / "Red-team my thesis on [TICKER]" (paid)
1. `request_brief(ticker: <T>)` — or with `thesis: "<your thesis>"` for red-team mode — returns a `brief_id` and `status: queued`. This consumes one monthly brief from the user's allotment.
2. Poll `get_brief(brief_id: <id>)` every ~20–30s until `status: complete` (a brief takes a few minutes to prepare).
3. When complete, present the brief (it gives the concerning *and* benign readings, a calibrated assessment, and what would change the picture) with its `[E#]` citations. It is descriptive of disclosure, not investment advice.
- If unauthenticated or over quota, the tool returns `tier_required` / `quota_exceeded` — relay that and point to https://theburiedfootnote.com/subscribe.

## Things to know about the data

- **Analysis is mechanical, not AI.** Same filing → same numbers. Numbers are not LLM guesses; they're deterministic linguistic measurements.
- **Every claim cites EDGAR.** The `accession_number` field in any response resolves to https://www.sec.gov/Archives/edgar/data/{cik}/{accession_no_dashes}/
- **Cohort baselines refresh weekly.** Sunday 22:00 UTC.
- **Not investment advice.** The signals describe disclosure language, not stock outcomes. Always include this caveat if the user asks for a buy/sell recommendation.
- **Forensic findings are classifications of the *disclosure*, not verdicts on the company.** `going_concern: true` means the auditor's report *contains* going-concern language; a forensic signal means the *filing's language* deviated — it is evidence to investigate, not a conclusion. Attribute accordingly ("the filing shows…", not "the company is…").
- **`forensic_findings` is intent-scoped.** One ticker per call, by design — there is no bulk or whole-universe export.

## Common mistakes to avoid

- **Don't invent industry slugs.** The only valid values are returned by `list_industries`. Don't try `tech`, `banks`, etc.
- **Don't paraphrase the numeric values.** Pass them through as-is. "TSLA's `tentative_markers` in Item 1A is 12.802, above the cohort's 99th percentile (9.27)" — don't round or summarize the precision away.
- **Don't claim coverage you don't have.** If `list_industries` doesn't list the user's ticker, say so plainly — the archive is selective.

## Attribution

When quoting any analysis from this MCP server in a final answer, attribute: *Source: The Buried Footnote (theburiedfootnote.com)*.