CIM drafting, review and tokenised delivery automation for Anthropic

What it calls

This is the endpoint this build calls directly, with runtime values replaced by a placeholder.

• 8 parallel model calls via `generateDoc`

What it reads and writes

Every read and every write this build makes is printed below, as the build sends it, with runtime values replaced by a placeholder.

• tables `cim_drafts` and `cim_documents`

Safety checks

An automation that writes into a system of record has to refuse some cases. These are the checks this build makes before it acts.

delivery REFUSES if the newest version was written by the model, so a human must save an edit first

forged tokens rejected and token kinds are not interchangeable

a 30-day HMAC on the view link

a forward-only status ladder

a `requireEditor` guard

Schedule and settings

These are the tuned constants from the delivered build. They are included because they are the part nobody can guess in advance, and they are the part a second implementation gets wrong.

• `CIM_PAYMENT_ENABLED=0`

Where it runs

It runs as Vercel plus a Cloudflare Worker.

It also touches Cloudflare D1, Stripe and Resend.

What this record is

Data Jockey is a services firm, so this page is an implementation record. There is no product to buy here and no price on this page.

No customer is named on it either. Data Jockey publishes no customer list, and the calls and checks above are the part a reader can check.

Questions we hear

Is cim drafting, review and tokenised delivery automation for Anthropic a product or a project?

It is a project. The queries on this page are specific to how one company had its Anthropic set up: which fields carried the human identifier, which documents already existed, and where a human had to stay in the loop. A product cannot know any of that in advance, which is why the integration is the work.

Did this run in production?

Yes. It runs in production today as Vercel plus a Cloudflare Worker, and this page is read out of the code that runs.

What happens when the automation is not sure?

It stops. This build carries 5 distinct kinds of check, and every one of them exists to hand the case to a person. Automating the confident cases and routing the rest is what makes an integration safe to leave running.

Can Data Jockey build this for us?

Book a call and we will scope it against the systems you already run. Data Jockey publishes no prices, so there is no figure on this page, and what a build costs depends on what your setup turns out to need.

Where these facts come from

Every call, table, check and setting above comes from the delivery record of this build, and each identifier in it was checked against the file that implements it by tools/capture-proof-of-work.mjs. The capture is committed at data/raw/dj-proof-of-work-capture-2026-09-07.json and records a sha256 of each source file. Generated 2026-09-07. No figure here was rounded, averaged or estimated.