Infinite Bookkeeping
The umbrellaThe broader thesis that the next 10 years of business intelligence is a separation of quantitative ledgers from qualitative ledgers, both keyed to the same business reality, both AI-queryable.
Imagine how much you could get done if you had proper AI working for you at its full potential.
A published framework . v1.0 . CC-BY 4.0Infinite Entry: a published framework for AI-era business intelligence.
A conservation-law foundation for measuring what changed in a business, why it changed, and how much each driver contributed. Published as an open specification under the Infinite Bookkeeping umbrella.
The premise
Most business intelligence was built for humans to pivot. AI tools need something different. Infinite Entry specifies what that something is.
The definition
What Infinite Entry is.
One umbrella, six distinct names. Infinite Entry sits in a specific slot, and that slot matters.
Infinite Entry is the published quantitative specification for AI-era business intelligence, sitting under the broader Infinite Bookkeeping umbrella.
It defines how a business's metrics decompose into the drivers that moved them, with the constraint that the parts sum to the whole. The decomposition is mathematically rigorous, model-agnostic, and designed for AI tools to query at any time grain and any dimension level. It is published as an open standard so that any team, any tool, and any model can interoperate against the same contract.
The Infinite Bookkeeping umbrella covers both the quantitative side (this spec) and the qualitative side (Infinite Brain, the formal version of which is deferred to a later release). The canonical open-source reference implementation is the Infinite Brain OS repository. The commercial productized data layer is StarMynd ABI. The company that maintains the standard and ships the reference implementation is StarMynd.
The naming framework
Six names. One umbrella. Distinct slots.
The standard, the implementation, the productized layer, and the company are all separate concerns. Each layer has its own name on purpose.
Infinite Entry
The quantitative specificationThis page. The published, open, citable framework for how AI-era business intelligence decomposes metric movements into driver contributions, subject to a conservation law.
You are hereInfinite Brain
Qualitative specification (deferred)The companion qualitative spec covering decisions, notes, agent definitions, and knowledge-graph structure. The formal v1.0 of this spec is deferred. The implementation pattern is already in the open-source repo.
Infinite Brain OS
The canonical open-source productThe MIT-licensed reference implementation on GitHub. The architecture pattern made visible in code. Forkable, runnable, self-hostable. The receipts that the spec is implementable.
StarMynd
The company that maintains the standardThe company that authors and maintains Infinite Entry, ships the open-source Infinite Brain OS, runs the productized ABI data layer, and offers the AI Architect engagement on top.
ABI
The productized commercial data layerStarMynd ABI delivers Infinite Entry compliant business data into a client's isolated GCP project. The unsexy data engineering rendered as a recurring product so client teams do not have to staff for it.
The math
The conservation law.
Infinite Entry is built on a single mathematical constraint. When a business metric changes between two periods, the sum of the driver contributions must equal the total change. No driver attribution is allowed to leak, drift, or double-count. The decomposition is exact.
This constraint is what makes the framework AI-queryable across any time grain and any dimensional cut. A weekly view sums to the monthly view. A channel view sums to the total. A cohort cut respects the same conservation rule as the aggregate. The math holds whether the question is asked by a human, by Claude, by Codex, or by an autonomous agent.
The framework borrows its rigor from the long-running tradition of paired-entry accounting in business and from index decomposition analysis in industrial economics. Both traditions require that totals reconcile to their parts. The formal proof of the conservation property, applied to AI-era marketing decomposition, is published as part of the v1.0 specification.
Read the conservation law proof
The methods
The two decomposition methods cited in the spec.
Infinite Entry does not invent new math. It selects, names, and applies the two best-fit methods from index decomposition analysis and cooperative game theory.
L
LMDI: Log Mean Divisia Index
Used when drivers combine multiplicatively. Channel rate times channel volume, conversion rate times traffic, retention rate times subscribers.
The Log Mean Divisia Index method decomposes a multiplicative change into additive log contributions with zero residual. Every driver gets a well-defined slice of the change, and the slices sum to the total change exactly. The method is the workhorse of energy-economics decomposition for the last three decades and applies cleanly to subscription-DTC, performance-marketing, and pricing metrics.
Ang, B. W. (2005). The LMDI approach to decomposition analysis: A practical guide.S
Shapley value attribution
Used when drivers interact non-trivially. Mixed-effect lifts where channel A's increase materially changes channel B's response.
The Shapley value is the cooperative game theory solution to fair attribution among interacting players. Every driver receives the average marginal contribution it makes across every possible ordering of the other drivers. The sum equals the total change, exactly. The cost is computational. Infinite Entry specifies when the LMDI shortcut is sufficient and when the full Shapley computation is required.
Shapley, L. S. (1953). A value for n-person games. Contributions to the Theory of Games.
The pattern
Why publish the standard openly.
A company can ship a great product. A company can also publish the open standard that defines the product category. The second is the longer-running play. We have seen it three times in the last five years.
Anthropic
+
MCP
Anthropic ships Claude as a commercial product. Anthropic also publishes the Model Context Protocol as an open standard that any tool, any agent, any model provider can implement. The standard outlives any one model release.
Vercel
+
Next.js
Vercel ships a hosting product. Vercel also stewards Next.js as the open framework that defines a category of React applications. The framework runs on Vercel best, but the framework is the standard.
Stripe
+
OpenPayments
Stripe ships a payments product. Stripe also co-publishes open payment standards (and earlier, the Open API Initiative) so the broader category of programmable money has a contract every participant can read.
StarMynd + Infinite Entry
StarMynd ships AI Architect, StarMynd ABI, and the open-source Infinite Brain OS. Infinite Entry is the open standard underneath all three. Anyone can read it, implement it, cite it, and run it. Our reference implementation is one way to satisfy the spec, not the only way.
Standards leadership underway
Citation recruitment is open.
We are actively recruiting external citations from named voices in AI engineering, data engineering, and marketing analytics. Frameworks become standards when other practitioners adopt them, name them in their own work, and extend them. If you are an operator, an engineer, or a researcher who finds Infinite Entry useful in your stack, the citation channel is open. Email, GitHub issue, or LinkedIn DM all work.
v1.0 publishes June 1, 2026. The reference implementation in the Infinite Brain OS repository publishes June 10, 2026. Citations gathered in 2026 land in the v1.1 acknowledgments section.
Read the v1.0 specification.
The full formal specification including the conservation law proof, the LMDI and Shapley application notes, the dimensional model, and the cited prior art. Open access. CC-BY 4.0.
If the spec URL is not yet live in your preview window, the standalone spec page lands on June 1, 2026 from a separate workstream. The conservation law page is internal to that workstream as well.
Read the spec. Or get the launch update.
The v1.0 specification lands June 1, 2026. The open-source reference implementation in Infinite Brain OS lands June 10, 2026. Subscribe for both.
Spec drop notifications
Get notified when v1.0 ships, when v1.1 ships, and when new acknowledgments land.
One email per release. No social proof claims. No funnels. Just the spec drops and the cited extensions.