The Optimisers, live production estate

The estate as built

Five layers, from the systems that hold the money to the cockpits people actually open. Drag to rotate it, and switch between what runs today, what is built but not yet deployed, and where it is going.

Partly measured, as of 10 September 2026. The automation fleet, the Worker folders and the table dependencies were measured from the estate on 10 September. The cockpit payload figure and anything about the second Airtable base are still documentation only. Treat every node as a claim to verify, and see the drift list at the foot of the page for the ones already known to disagree.
Built and running

What is deployed and executing today. Amber is not a criticism, it is the honest reading: most layers work while carrying a known unproven edge.

Working Working with a known gap Not working or not proven Drag to rotate. Select any layer to open it.

The money chain

The one flow that crosses every layer, and the reason the estate exists. Five systems hold the same number. When they disagree, that disagreement is the signal, and it arrives as silence rather than as an error.

Pipedrive dealThe contract. Value is annualised, derived from recurring product lines.
Retainer PeriodThe monthly retainer, generated by the fleet.
Billing ItemThe ops ledger row carrying the amount.
Xero invoiceWhat we actually billed. Source of truth, read-only to us by design.
Revenue SummaryLeadership view and the run-rate.

Known drift

Places where the documentation and the deployed estate are already recorded as disagreeing. These are the first things to measure, because each one has produced a wrong answer before.

The Airtable proxy Worker runs an older version than the one on disk

Production is recorded at v1.1 while the working copy is v1.3. Anything relying on behaviour added after v1.1 is relying on code that is not deployed.

The SEO audit watcher is documented as a Cloudflare Worker and is not one

It is a Windows scheduled task, recorded as disabled since 13 August, reading a credential from a local file. The architecture note describing it as a Worker is wrong, and it was believed to be running.

The fleet ratchet is red on five checks

Measured 10 September: the credential sweep cannot be assessed at all, direct vendor API bypass stands at 641 against a ceiling of zero, unregistered wrangler configs at 8, notify registry drift at 2, colliding PowerShell function names at 4, and three deployed scripts carry a header that contradicts their own DRY_RUN constant.

The Invariants table is not evaluated, now measured

The table exists and is populated. Nothing reads it, so an invariant breach produces no signal at all.