# Comps Desk

> A comparable company analysis that runs in the browser, followed by a paid review that judges the
> peer set, picks the anchor multiple, calls every peer and writes the comps page for an investment
> committee. Every number the review writes is checked against the table. https://comps-desk.skillsafe.ai/

Comps Desk is a web app on SkillSafe derived from the agent skill @anthropics/comps-analysis
(anthropics/financial-services-plugins, Apache-2.0). It runs on gpt-terra and is metered per review;
the comps themselves are free and need no account.

## What the free comps compute

- Input: a peer table pasted or dropped as CSV, tab-separated text or a Markdown table, with loose
  header matching (company, ticker, share price, diluted shares, market cap, debt, cash, net debt,
  revenue, prior-year revenue, growth, gross profit, EBITDA, net income, free cash flow), plus the
  target's revenue, EBITDA, net income, FCF, growth, net debt, shares and price. Up to 25 peers.
- Per peer: market cap (price x diluted shares unless given), enterprise value (market cap + net
  debt), gross, EBITDA, net and FCF margins, Rule of 40 (growth + FCF margin), EV / Revenue,
  EV / EBITDA, P / E, FCF yield and PEG. Multiples on a zero or negative denominator are NM.
- Statistics: maximum, 75th percentile, median, 25th percentile and minimum (Excel QUARTILE.INC) over
  the comparable metrics only - growth, margins and multiples, never size metrics - excluding NM
  values and any peer the user unticks.
- Checks and flags: too few peers, no multiple with three values, a target far from the peers' size,
  many NM values, IQR outliers (also more than 20% from the median), a wide multiple range, missing EV inputs, duplicates, the target in
  its own peer set, margin order, multiples outside the usual ranges, size dispersion, multiples that
  ignore growth, market cap mismatches, and sector notes.
- The target's implied enterprise value, equity value and value per share at the 25th, median and
  75th percentile EV / Revenue, EV / EBITDA and P / E, the range across methods, and where the target
  ranks inside the peer set.
- Exports: the comps as Markdown, a values CSV, and an Excel CSV in which every derived cell and every
  statistic is a live formula.
- Changing the peer set: untick a peer (or apply the review's exclusions) and the page shows what moved
  since the full set - peers counted, anchor median, implied EV, equity and per-share ranges, and flags -
  with one button to put the peers back. A table over 25 rows is cut on whole rows and every export
  says so, in the file and in its name.

## What the review returns

One JSON object: `verdict` (sound, usable_with_caveats, unreliable), `headline`, `anchor` (metric and
reason), `peer_calls` (ticker, keep / question / exclude, reason - one per peer), `metric_readings`,
`valuation_read`, `flag_responses` (one per flag), `data_requests`, `methodology_note`, `ic_summary`,
`summary`. The page offers the review's exclusions as a one-click recompute of the statistics. The review downloads as
Markdown with the comps tables appended, or as a values CSV with a "Review call" column per peer; a
review whose table has since changed is marked as describing the earlier figures.

## Limits

It does not fetch prices or filings, know what any company does, or adjust for one-off items, leases,
minority interests, calendarisation or currency. The data are the user's to source and date. It is
analysis, not investment advice.

## Links

- App: https://comps-desk.skillsafe.ai/
- API tutorial: https://comps-desk.skillsafe.ai/api.html
- Source skill: https://skillsafe.ai/skill/@anthropics/comps-analysis