The method, on the record.
A research tool should be understandable before it is trusted. This page sets out the inputs, the extraction steps, the review path, and the places the model can fail.
01 Data we read
Embassy works from public material: annual and quarterly reports, current reports, proxy statements, issuer announcements, historical price and volume data, and public sector classifications. We do not use paywalled research or non-public information.
- SEC filings: 10-K, 10-Q, 8-K and proxy statements
- Issuer announcements and public investor materials
- Historical price and volume series
- Public sector and industry classifications
02 What the model does
The model compares a new document with the prior record, extracts the passages that changed, marks unusual movements in the stated data, and writes a short digest. Each output keeps a path back to the source passage; no accuracy rate is promised.
- Detect changes between a filing and its prior version
- Flag unusual values and announce amendments
- Generate a dated, sourced digest
- Preserve the source reference with the output
03 What it does not do
The desk does not give target levels, buy or sell recommendations, personalised advice or market-direction forecasts. It does not manage money, execute trades or decide whether a security is suitable for you.
- No buy or sell recommendations
- No target levels or ratings
- No personalised advice
- No market-direction forecasts
04 Validation
A sample of outputs is checked by a person against the underlying filing for source fidelity and obvious extraction errors. You can report an error from the Contact page; corrected outputs keep a new version and a dated note.
- Sampled human review against the source
- Versioned corrections
- An error-report path that reaches the desk
- No claim that every output has been reviewed
05 Known limitations
The record can be messy. Accounting-standard changes, one-time items, restatements and low-liquidity names are common sources of error. A delayed feed can make a change look newer than it is.
- Accounting-policy changes and one-time items
- Restatements and amended filings
- Trading halts and low-liquidity securities
- Source and market-data delays
- Extraction errors in tables and footnotes
06 No performance claims
We do not publish performance figures, outcome statistics or comparisons against a benchmark. A sourced change record is not a track record, and we will not present it as one.
- No performance figures
- No outcome statistics
- No benchmark comparison
- No claim that the method produces a particular outcome