KAN › The method · lesson 6 of 7

Escalate a consequential choice

Most choices are yours to make inside the authority you were given. A consequential one goes to its owner, and SDMR is the shape that makes it quick to decide well.

Teaches from kan/6w-sdmr.md, “Construct and escalate an SDMR decision”, and kan/mcp/README.md, “Send the right record”.

When to escalate

Escalate through the Operator's channel “when confidence is below 70%, FIB is 13 or higher, or material ambiguity remains in requirements, scope, address or ownership after available evidence is checked.” Existing authorization stays valid for the choice it covers: “an answered question does not need another approval.” And “Continue independent work while dependent work waits.”

The six parts, in order

1. Question: one sentence naming the choice and its consequence.
2. Steelman: the strongest argument for change, linked to evidence.
3. Steward: the strongest existing guarantee, and the cost of changing it.
4. Options: two to four real alternatives, each with its weakest point.
5. Recommendation: a separate choice with a confidence percentage and the principal uncertainty.
6. Falsifying check: the observation or violating input that would refute the recommendation.

“Avoid options engineered to make a preferred answer inevitable. A format-complete decision with weak evidence remains weak.”

Try it

In the EchoEdu delivery

This course's first checkpoint (KAN CRS0RWxHsb2J1s) carried three choices. Two, the page shell and the archive notice, came in a compact form of SDMR: the options with their weakest points, a recommendation with its confidence, and for the shell a falsifying check. The remaining one, the course's slug, was a plain question to the owner of the file it lives in. The shell recommendation, plain HTML pages at 75%, named what would refute it: the design epic requiring lesson bodies in a source other than HTML. The answer (KAN CRS0RWxJg2IS8G) checked that falsifier against the Operator's own design ruling before accepting the recommendation, mixed two of the options for the notice, and settled the slug by giving the edit to the file's owner.

Half an hour later the Operator ruled again, and both accepted choices fell: the course would be built on Mercury UI and Effector rather than plain HTML, and the old course would be retired rather than archived (KAN CRS0RWzLyQCdWa). That is the method working, not failing: “Existing authorization remains valid for the choice it covers”, and a new ruling covers a new choice. The decision record names what it supersedes, so nobody follows the old one.

When the Operator rules, the ruling travels with their words. A decision record needs to, quote (their words, verbatim) and quote_source. The reframe that started this course was recorded that way: “Reframe agile-agent-workflow to be the KAN course. Flow, integration, MCP, seats”.

Check yourself

1. Your confidence in a routine choice is 85% and the change is reversible. Do you escalate?
2. Why does the steward argue for what exists, and what does the chosen option keep from it?
3. What makes a falsifying check useful rather than decorative?

Previous: Deliver through PTE · Next: Good, bad, test and blue

Teaches ecd/kan at 2d1b5c8e, 25 October 2026.