/bottom-up-market-size

name: bottom-up-market-size
description: >
  Size a market bottom-up from an Economic Mind market report and stress-test that number against
  independent sizing approaches - the headline size, the transparent per-company build (each player's
  revenue, share-in-market, in-market revenue, confidence and justification carried verbatim from the
  report), an explicit read on how much to trust the headline and why, and a football-field comparison
  of the bottom-up figure against 2-3 credible cross-checks. Use whenever the user wants to "size the X
  market bottom-up", "how big is X and how sure are we", "bottom-up market size", "build / show the
  market-size buildup", "triangulate the size of X", or "benchmark our market size against other
  estimates". The bottom-up figures come ONLY from the Economic Mind report, unchanged; cross-checks come
  from a credible-source research pass. The deliverable is data (Markdown + JSON); presentation is left to
  the client.

Bottom-up market size

Size a market the way a careful analyst would: take Economic Mind's **bottom-up build** as the spine, show every company that rolls up into it, say plainly how much the headline can be trusted and why, then put that number on a **football field** next to a few independent estimates so the reader can judge it. The deliverable is **data** - a Markdown report plus a JSON sidecar - not a styled document; this skill is about the research and the numbers, and layout is the client's job.

The whole value is **credibility under scrutiny**: a market size is only as good as the build beneath it and the cross-checks beside it. So nothing is hand-waved - the build is carried verbatim from the report, the confidence is reasoned from the report's own stamps, and every cross-check names a credible source.

Two source channels, never mixed

1. **The bottom-up build - ONLY the Economic Mind report, unchanged.** The headline size, the per-company revenue, share-in-market, in-market revenue, confidence ratings, justifications, the scope boundary and the vintage all come **straight from the report**. Do **not** recompute, re-estimate, re-rank, or "improve" any of them, and do not pull a player's revenue from elsewhere. You are presenting the report's build faithfully, not rebuilding it. (The only arithmetic permitted on it is summing the in-market revenues to show the build-to-total and reconciling that to the report's stated headline - shown transparently, never overriding the report.) 2. **The cross-checks - a credible-source research pass.** The alternative sizing approaches on the football field are **not** in the report; you construct 2-3 of them from independent credible sources (see the source rules). These sit **beside** the bottom-up for comparison and **never alter it**. This is the one place the web is used, and only for the cross-checks.

Keeping these apart is what makes the exercise honest: the anchor stays pristine and the triangulation is clearly a separate, independently-sourced opinion.

Work to files first, then read them back

Context can be compressed mid-task, and a market build has a lot of cells. So **immediately after each tool call returns, write its raw output to a file** before the next call - the report content, then each cross-check as you research it. Build the deliverable by reading the files back, not from recall. Create one working directory at the start, e.g. `em-size/`.

Source rules for the cross-checks (strict)

The football field is only as credible as its weakest source, so the bar is high. Use **only**: (a) primary filings & registries (Proff, Brønnøysund, Allabolag, S&P Capital IQ, Creditsafe, audited annual reports); (b) official statistics (Eurostat, SSB, SCB, national statistics offices); (c) regulators & standards bodies (IMO, USCG, national building authorities); (d) listed-company reports and reputable financial data; (e) named industry associations.

**Never** use market-research aggregator / press-release "market-size" mills - MarketsandMarkets, Grand View, IMARC, Verified Market Reports, SNS Insider, Precedence, FactMR, Custom Market Insights, SkyQuest, Fortune Business Insights, Mordor, Spherical and the like. Their figures are unreliable and scope-inconsistent. When web-searching, pass these as `blocked_domains`. Build cross-checks from **first principles on credible bases** (e.g. official installation statistics × a fire-protection share; a listed group's disclosed segment revenue grossed for the tail; installed base × annual value). If a number exists only from a mill, do not use it - say no credible independent estimate was available and lean on the build.

**Restate scope.** Any external estimate almost certainly uses a different market definition. Before it can sit on the football field, restate it to the **report's in-scope definition** (geography, product boundary, revenue basis) and say how you did so in its comment. Never fabricate; if you cannot scope-adjust an estimate honestly, drop it.

Workflow

Phase 1 - Resolve the market & pull the build

1. `mkdir -p em-size`. 2. `search_markets("<the user's market phrase>")` (limit ~5). If the top hit's `score` is ≥1.5× the second's (or there is one hit), use it and announce the choice. Otherwise show the top 3 (Market, Geography, Size, Relevance) and ask the user to pick. If nothing matches, say the market is not covered and stop - do not improvise a size. 3. `get_market(slug)` for the catalog, then `get_market(slug, "bottom-up-supply-side")` (use the catalog's actual element name) to get the definition, headline size, and the per-company build. Tables come back as markdown - read them, don't guess. 4. **Immediately write** the raw element to `em-size/report-bottomup.md`, and write `em-size/build.csv` - one row per company, every field carried verbatim from the report (leave blank only where the report itself is blank):

   rank,company,hq,tota