This is a method I intend to run, with a public ship/kill board from week one. It is not a Rileytech services page. Until there's a receipt, it's a thesis. Lab proves the lab first.

The argument


Every large-enough org is already running dozens of unvalidated, founder-dependent experiments — vibe-coded tools, side projects, "AI pilots" with no end date. The activity exists. The discipline doesn't.
These two plays add the discipline without killing the cheapness of building. No hype vocabulary. Everything here is a forcing function — the thing large orgs can't manufacture on their own.

Shared rules (both plays)

  1. Value is named before work starts. In writing, one number: revenue booked, hours reclaimed × rate, or a cost line that shrinks. Chosen with the sponsor, not for them. No number, no start.

  2. Evidence at 90 days, ROI at 180. Realized return in 90 days selects only for trivial problems. Ninety days proves demand (someone paid, signed, or switched). Six months banks the money.

  3. Default to death. Everything expires unless actively renewed by a named owner. Killing things is a win condition, not a failure. Track kills as visibly as ships.

  4. Adoption is the trigger. Anyone can build anything. The moment a second person depends on it, it needs an owner (a name, not a team) or an expiry date. That's the whole governance model — anything heavier becomes permission theater.

  5. Intake the person and the problem, never the artifact. The build was the discovery. Most artifacts should not survive the value bar, and that's fine.

  6. The three-camp test. If the same problem shows up in three companies from different camps, it's structural. If it doesn't, it's a local quirk. Two people per company — the pro and the person in the trench. Skip fans.

Spin-up: the lab proves the lab

No tool gets built in the first 90 days. The lab runs its own rules on itself first — same clock, same kill authority, same exits.
What has to be true at day 90:

  1. The three-camp test has been run on two or three candidate problems. At least one survives.
  2. A named design partner and a signed one-pager (problem, value number, what they're putting in). "They seemed excited" is not a sign.
  3. The kill log has entries. An empty log means nothing was tested.
    If these don't hold, kill the lab. Applying the rule to yourself first is what makes the kill authority believable.
    Candidate problems need a receipt: a specific workflow, a specific handoff, a specific thing someone does on a Tuesday — not "AI adoption" or "knowledge management."

Play 1 — The Lab (prove it or lose it)

A small crew of operators. Their job is not to have ideas — it's to make someone else's validated idea real, fast, and sellable.
Intake bar (all four): a person with the problem who keeps their day job; a named value number; passes the three-camp test; a committed buyer who said yes before a line is written.
The clock: Day 0 value and proof signed. Days 0–10: can't-exist screen (hallway yes/no on whether a viable version is categorically illegal). Days 1–30: discovery + smallest thing that touches the real workflow — idea-owner does discovery, coached. Days 30–60: in front of the buyer plus two more; watch what they stop doing. Days 60–90: evidence — someone paid, signed, or abandoned the old way while it was still available.
Only three exits: Kill (publicly, credit for the clean kill). Renew once (named blocker in a sentence, one renewal max). Graduate (owner + maintenance answer → Play 2 or sale).
Approvals constrain how, never whether. The clock does not stop for legal. "Waiting on legal" is not a named blocker.

Play 2 — The Proving Ground

Internal first, external later. Internal users are customer zero — captive, tolerant of v0.4, honest if you make them choose.
Price in what's scarce: retire something to adopt; capacity slots; sponsor pays in people; default-death renewal. Measurement is revealed preference at day 60 with the old option still open.
Graduation ladder: build for the outsider from day one → shadow council → design partners → paid pilot (real money) → external product with its own owner. If no design partner will pay anything, that's a kill or keep-internal decision — not a marketing problem.

Why I'm writing this


Because publishing the method before the receipt is the honest tease. A `/proof` services page for a program never run is the enthusiast move. This post is the lab proving the lab, on the page.
If you're reading along: the ship/kill board will be public from week one when spin-up starts. Until then, this is the plan — and the kill criteria apply to the plan itself.
Related: [Skills Over Seats](https://rileytech.net/posts/skills-over-seats-why-youre-botching-your-new-transformation-strategy-with-ai-this-time) · [The Blank Slate](https://rileytech.net/blank-slate) · [Services](https://rileytech.net/services)