Agentic, verified-finance recipes — built and operated inside the Snickerdoodle contract.
By Nik Bear Brown and Humanitarians AI · Published by Humanitarians AI, a 501(c)(3) nonprofit organization.
Mycroft is both a book and a working agentic repository. It rests on one claim: AI made finance execution cheap, but it did not make finance judgment cheap — and the work of a serious finance function is now mostly the second thing. A model will write you a clean variance note in seconds; knowing whether it is true, whether it ties to a source, and whether someone can defend it to a controller or an audit committee got no cheaper at all. This book is about the gap between fluent finance output and finished finance work, and about building pipelines that hold that gap open instead of papering over it.
The book teaches a vocabulary you will use in every chapter — the fluency trap, the finance work stack, the phase gate where AI assistance stops and accountable human work begins, provenance beats polish, the two customers every recipe serves, and the doctrine that friction protects judgment. It then puts that vocabulary to work across ten finance recipes — variance, reconciliation, liquidity, close flux, budget challenge, control evidence, AP/AR aging, cash forecast, audit-evidence binder, and revenue-contract exceptions — each showing exactly where the machine prepares and where the human decides.
It is written for anyone whose finance work gets reviewed — analyst, accountant, controller, treasury, FP&A, audit and controls — and for the engineers building tools for them. It is not a course in accounting, not a manual for the framework that governs the repo, and not a promise that AI can run your close. It is a book about using AI to make finance judgment more available, not less necessary.
The repository is governed by Snickerdoodle, an agent-operating constitution
that treats a project as a contract between human judgment and AI execution. Start
with _MANIFEST.md for the map, SNICKERDOODLE.md for the rules, and DOMAIN.md
for what is runnable today.
Front Matter
Foundations
- The Fluency Trap
- The Reallocation Principle
- The Verified Finance Data Contract
- Two Customers
- Verifying Finance Evidence
The Recipes 6. Monthly Variance Pack 7. Subledger-to-GL Reconciliation Triage 8. Daily Cash Position and Liquidity Watch 9. Close Flux Analysis and Balance-Sheet Review 10. Budget-Request Normalizer and Challenge Pack 11. Control-Evidence Completeness Checker 12. AP/AR Exception and Aging Workbench 13. Cash Forecast Variance Explainer 14. PBC Request Tracker and Audit-Evidence Binder 15. Revenue Contract and Billing Exception Triage
Synthesis 16. The Build and the Honest Run
End Matter
- Fundamental Themes
- Appendix: Best Practices for Agentic Book Repos
- Back Matter — acknowledgments, author, references, glossary
These are Kindle / online editions designed for integration with Medhavy — मेधावी, Sanskrit for "intelligent" — an AI-powered intelligent-textbook system that turns these chapters into adaptive practice: hints, worked examples, quizzes, and feedback loops. Learn more at https://www.medhavy.com/.
Copyright © 2026 Nik Bear Brown and Humanitarians AI. Published by Humanitarians AI, a 501(c)(3) nonprofit organization.