Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.cf0.ai/llms.txt

Use this file to discover all available pages before exploring further.

cf0’s posture on accuracy is defense in depth. No single layer is asked to be perfect. The model orchestrates; math runs in code; outputs are schema-validated; missing inputs surface as missing, not invented.

The model doesn’t do math

Financial calculations — DCF, WACC, multiples, sensitivities, monte carlo — route through deterministic code templates, not AI inference. The model frames the question and reads the result; the result itself comes from a tested code path. cf0’s numbers are reproducible: same question, same inputs, same answer.

Schema-validated output

Every model output that reaches storage or the UI passes through schema validation. Free-form prose that doesn’t match a known structure doesn’t get persisted. Hallucinated fields are caught before they propagate.

Isolated execution sandbox

Any code-execution path runs in an isolated, per-user sandbox. The sandbox cannot reach the wider cf0 infrastructure or other users’ workspaces.

Refuse to fabricate

When a required input is missing, cf0 says so rather than inventing it. From an actual generated report:
“EV/Sales requires enterprise value; since debt/cash and shares outstanding are not provided here, EV/Sales is not computed. PEG requires forward growth; not provided, so PEG is omitted.”

Methodology gates

Before producing a DCF or valuation, cf0 pauses on the assumptions that swing the answer — WACC, terminal growth, forward revenue growth — and asks the analyst to accept or override. See Run a DCF.

Regression gate

cf0 ships against an internal evaluation harness of analyst-style research questions. Releases that regress on the eval bank don’t ship.

Honest about limits

LLMs can still hallucinate at the prose layer. The guardrails above keep the math sound, schemas tight, and sources cited. They don’t eliminate every wrong sentence — they make wrong sentences findable, citable, and challengeable. See Citations and audit trail for how every claim resolves to its source.
Last modified on May 19, 2026