Startup valuation methods as auditable code: Berkus, Scorecard, Risk Factor Summation, VC Method, First Chicago and market multiples.
Four well-named tools with strong descriptions (50-200 chars each) and clear purposes. All tools have input schemas with type definitions and parameter descriptions. Tool names follow verb_noun pattern (list_, check_, value_, explain_). Descriptions are LLM-optimized and include WHEN to use each tool. However, output schemas are not formally documented in the source code, only inferred from docstrings. Parameter descriptions are present but some lack explicit constraints (e.g., 'currency' has no enum or format spec). No error handling guidance visible in tool definitions. Security considerations (input validation, error messages) not evident in the provided code.
Given whatever is known about a company, report which valuation methods can already run, which are blocked and on what, and which missing field would unlock the most methods. Use this to decide what to ask the user for next, instead of guessing at values.
Run a single valuation method and return its full derivation: every step of the arithmetic, the assumptions made on the company's behalf, what the figure does not account for, and citations for the method. Use this when writing up a valuation someone will have to defend.
List every valuation method available, what each suits, what it requires, and the exact company-data format the other tools accept. Call this first so you fill in field names correctly rather than guessing.
Run every applicable valuation method and return what each concluded, with a median and a range. Methods that cannot run are reported rather than silently skipped. Report the range and the individual methods; the median alone hides the disagreement that matters.
Output schemas not formally documented. Tool descriptions mention what is returned (e.g., 'median and range', 'full derivation') but JSON Schema output types are not visible in the source code. LLMs cannot plan downstream operations without knowing response structure.
Parameter 'currency' lacks enum constraint or format specification. Description says 'Currency label (default: USD)' but does not specify valid values (ISO 4217 codes? Free-form strings?). LLMs may hallucinate invalid currency codes.
No error handling guidance in tool definitions. If a company object is missing required fields, the tool description does not explain what error the LLM should expect or how to recover. check_valuation_readiness partially addresses this by reporting blocked methods, but value_company and explain_valuation do not.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
Company object parameter is loosely typed as 'object' with optional nested fields. No formal schema constraints (required vs optional, type validation, nested field types). LLMs cannot validate input before calling the tool.
No pagination or result-limiting guidance. If value_company returns many valuation methods, the response could be large. No mention of limits, pagination, or how to handle large result sets.