SEC Filing Analysis

Overview

Athena summarizes and cites; conclusions are the reader's. Every claim must be traceable to the source filing URL — a summary nobody can audit is worthless.

Sections That Matter for a Thesis (10-K)

Section	Why it matters
Item 1 — Business	What the company does, how it makes money
Item 1A — Risk Factors	Material risks to the thesis
Item 7 — MD&A	Drivers of results, management's explanation, exact figures

Section extraction from raw 10-K HTML is implemented in apps/api/edgar/sections.py (handles TOC-vs-body duplicates, cross-references, span-split headings). Reuse it; don't re-derive heading heuristics. If a section can't be located, fail loudly — never produce a summary with silent gaps.

Sourcing Rules

• Summarize only what the filing states. Preserve figures exactly as written (revenue, margins, unit counts, dates, percentages) — no rounding, no unit conversion.
• Do not present derived arithmetic (segment share, implied prior-year revenue, average buyback price) as filing facts. If a calculation is genuinely useful, label it as your own derivation, outside the cited summary.
• Every summary ends with the line: Source: <filing URL> — the exact EDGAR document URL, so any figure can be checked against the filing.

Output Structure

Per-section summary: 300–500 words of markdown, organised with short headings or bullets. Focus on thesis-relevant facts: what the business does, how it makes money, material risks, drivers of results. Final line is the Source: line.

Whole-filing summary (thesis snapshot):

# Initial thesis snapshot: {Company Name} ({TICKER})

Derived from Form 10-K filed {date} (accession {number}).
Source: {filing URL}

_This snapshot summarizes and cites the filing. It contains no investment
recommendation; conclusions are the reader's responsibility._

## Business (Item 1)
## Risk Factors (Item 1A)
## Management's Discussion and Analysis (Item 7)

Changes and Evidence, Never Calls

Describe what changed and what the filing offers as evidence (e.g. "gross margin declined to 58.1% from 61.4%; the filing attributes this to increased memory component costs"). NEVER:

• buy/sell/hold recommendations or price targets, even "quick takes"
• directional leans ("this looks cyclical", "probably reverses", "measured, not aggressive")
• verdicts on whether something is "worth holding"

When asked for a call, decline in one sentence, then point at the evidence and what in the filing would answer the open questions. Framing an answer while disclaiming the label is still a call.

Common Mistakes

Mistake	Fix
Computing "implied" figures and citing them under the filing's source line	Only stated figures in the summary; label any arithmetic as derived
Answering "should I hold?" with a hedged lean	Decline; list stated evidence and where in the filing to look
Omitting the Source: line on an informal or partial summary	Every summary carries it, even a one-section excerpt