The Stop List for Free Inference and Free Hosts

Description and use

Teams established a stop list that links free dependencies to risk signals and pre‑agreed actions. Using a JSON ledger, linting, and a detach drill, they created a repeatable process for deciding when to abandon free inference services.

Problem

Teams adopt free inference endpoints or servers without clear criteria for when to abandon them, leading to hidden risks beyond cost.

Outcome

Teams gain a repeatable, low‑overhead process to evaluate and exit free inference dependencies, reducing hidden operational risk.

Workflow

• Define a capability ledger JSON per job listing endpoint, capabilities, latency, fallback, and owner.
• Lint the ledger with a CI pre‑merge check that flags red‑line capabilities such as "mutates_durable_state".
• Run a detach drill by pointing provider base URLs to a port that refuses connections and compare job behavior to the normal run.
• If the job fails or behaves unexpectedly, decide to add a second writer or mark the job as an accepted exception.
• Maintain an exit‑criteria table to record agreed actions for each risk signal.
• Use the stop list to allow on‑call engineers to execute the exit decision at any time without committee.

Evidence

A stop list fixes this. It is a short document, dated, that pairs each free dependency with an observable signal and a pre‑agreed action.

Source

Running it against the sample above exits 1 with one line naming the durable write.

Source

Exit criteria are worthless if nobody can measure the exit. The drill runs one command twice: once normally, once with every provider base URL pointed at a port that refuses connections.

Source