§ 01

Data handling

Everything on this page is a dated artifact you can read. No certification marks, no badges — we hold none. Nothing here describes a control we intend to have.

Last reviewed 2026-09-04

§ 02

Artifacts

One row per artifact. The status column has four values, and none of them is a promise. Today every row is published; the other three statuses exist so an absence, when there is one, reads the same way.

Artifact	Description	Status	Dated	Retrieval
Data processing path	Uploads are stored by the backend. Anthropic assists extraction and normalization; calculations run server-side before the finished deliverable is served.	PUBLISHED	2026-09-04	On this page
Deal-content provider list	Anthropic is the configured provider for model-assisted extraction and normalization under the linked commercial and data-processing terms.	PUBLISHED	2026-09-04	On this page
Anthropic Commercial Terms	These API terms state that Anthropic may not train models on customer content; explicit opt-in is covered in service-specific terms.	PUBLISHED	2025-06-17	Anthropic terms
Anthropic model-training policy	This policy says commercial API data is not used for training unless the customer explicitly opts in or submits material as feedback.	PUBLISHED	2026-03-16	Anthropic policy
Anthropic API retention policy	Anthropic says API inputs and outputs are automatically deleted within 30 days, with stated service, safety, contract and legal exceptions.	PUBLISHED	2026-07-01	Anthropic policy
Anthropic Data Processing Addendum	This DPA sets processor duties, security measures, subprocessor terms and post-termination deletion obligations for Anthropic's API service.	PUBLISHED	2025-02-24	Anthropic DPA

Published — readable now · On request — exists, released under NDA · Not published — exists, not released · Not held — does not exist

§ 03

The two sentences that matter

The backend stores deal documents; Anthropic assists extraction and normalization; server-side calculations produce the finished deliverable.

Deal content is not used to train models — not by us, and under Anthropic's Commercial Terms, not by Anthropic.

Both sentences appear verbatim on the landing page. If they ever diverge, the dated version on this page governs.

Deal-content provider

Provider: Anthropic
Purpose: Model-assisted extraction and normalization
Training terms: The Commercial Terms say Anthropic may not train on customer content. Anthropic's policy identifies explicit customer opt-in and submitted feedback as exceptions.
Provider retention: Anthropic's commercial policy says API inputs and outputs are automatically deleted within 30 days, subject to its stated service, safety, contract and legal exceptions. This page makes no zero-data-retention claim.
Data-processing terms: Anthropic's Data Processing Addendum is incorporated into its Commercial Terms.

The property on this site is synthetic and disclosed. A synthetic property with known ground truth is the only way to measure extraction against an answer key — we know what the documents say because we wrote them. Accuracy figures, when we publish them, will come off holdouts that are never made public.

Deterministic means the kernel returns the same figure for the same inputs on every run, and emits the operand tree behind that figure alongside it. Model-assisted extraction reads the documents; the arithmetic downstream of extraction does not vary between runs.

USALI (Uniform System of Accounts for the Lodging Industry) is published by Hospitality Financial and Technology Professionals (HFTP). KeyTwelve is an independent product and is not affiliated with, endorsed by, or certified by HFTP. For official USALI guidance and definitions, see hftp.org.