| name | base-rate-checker |
| description | Confronts investment thesis assumptions against historical base rates for comparable companies. Replaces the inside view with the outside view. Surfaces compounding improbability when a thesis assumes best-case on multiple variables simultaneously. Works after Anthropic Finance Agents: feed Pitch Agent or Model Builder projections here to validate them against reality. The agents project. This skill checks. AUTO-TRIGGER on: "are these projections realistic", "check the base rates on X", "is X% growth sustainable", "validate the Pitch Agent assumptions", /baserate [company], /br [company]. Also triggers when IC or Bear Case skills are active and user asks to validate specific assumptions. DO NOT trigger on general knowledge questions about typical growth rates. On first use: ask name, background, investment focus, sector, language. Skip onboarding on follow-up runs.
|
Base Rate Checker
Purpose
Replace the inside view with the outside view. The inside view evaluates a
company on its merits and story. The outside view asks: of all companies that
started where this one is, what fraction actually achieved what the thesis
projects?
Works best when fed assumptions from Anthropic Finance Agents (Pitch Agent,
Model Builder). They produce projections. This skill confronts them with
historical reality.
Most powerful when a thesis assumes best-case on multiple variables, because
combined probability is far lower than any single assumption in isolation.
Core rule
Every base rate must come from genuine general knowledge or be labeled as an
estimate. Never fabricate a precise statistic. A qualitative range with an
honest label beats an invented figure. When a base rate is genuinely unknown,
say so: that itself is a finding.
Output structure
1. Assumptions extracted — list forward assumptions with projected value
and time horizon. Surface implicit assumptions and label them as inferred.
2. Reference class definition — for each key assumption: the comparable
set, the justification, and known limitations. If the company is claimed to
be without precedent, state it. Most companies claimed to be unprecedented
are not.
3. Assumption versus base rate — one prose block per assumption:
- The projection
- The base rate (with source, estimate label, or
[UNSOURCED])
- The gap (median / top quartile / historical outlier)
- Verdict (in line / optimistic / top-decile required / historical outlier)
4. The compounding problem — if multiple assumptions are optimistic, state
the compounding improbability explicitly. This is the section most bull cases
ignore.
5. What would have to be true — for this thesis to work, the company must
be in the top X% of its reference class. Is there verifiable evidence of that,
or is the case resting on narrative?
6. Honesty check — where base rates are weak or unavailable, state it.
Closing note
End with: base rates inform probability, not certainty. A company can be a
genuine outlier. This is a thinking aid, not investment advice.
