06-reference

commoncog no learning dont close loops

Tue Apr 14 2026 20:00:00 GMT-0400 (Eastern Daylight Time) ·reference ·source: Commoncog ·by Cedric Chin

“You Aren’t Learning If You Don’t Close the Loops” — @CedricChin

Why this is in the vault

Chin’s operator-level argument for why Deming’s PDSA loop fails in practice — the Plan and Do are easy; the Study and Act get killed by distraction, shiny new ideas, and forgetting. Directly relevant to RDCO because MAC, consulting engagements, and the newsletter are all multi-month execution loops where the Study step is the bottleneck.

The core argument (paraphrased)

PDSA (Plan-Do-Study-Act, Deming’s cycle) is the scientific method applied to business execution. It’s “boring and brain-dead and common-sensical” — and almost nobody closes the loops.

The failure mode isn’t planning or doing. It’s Study and Act. Three killers:

  1. Mid-execution distraction — a shinier idea or an emergency pulls you off.
  2. End-of-loop distraction — something new pops up right as the loop closes, and you skip the retrospective.
  3. Forgetting — loops that take 3-6 months (SEO, a product bet) generate too much intervening chaos; you “don’t harvest the learnings” and the work goes to waste.

Chin’s Commoncog Burnout Guide example: went viral, he “nearly forgot about the original hypotheses” until re-reading the opening 6-pager. Hence the operative rule: “write out your hypotheses before you execute.” Otherwise you can’t tell at the end whether you succeeded by your own definition.

Interesting wrinkle on sunk cost: PDSA implies the opposite. Half-assed loops are waste — not because you should finish everything regardless of payoff, but because the cost was already paid and refusing to Study forfeits the only asset the loop produces (learning). Chin leans toward finishing every loop and eating the cost, because in practice distraction is the dominant failure.

Closing line: “the trick to applying PDSA seems to be that there’s no trick” — you just have to force your org to close the loops it starts.

Mapping against Ray Data Co

1. The consulting engagement IS a PDSA loop — and the Study step is where RDCO differentiates. A 3-6 month phData/MG-style engagement (see 2026-04-14-levie-agent-deployer-role-jd) is structurally a Deming loop: plan (scope + MAC spec), do (implement), study (measure model quality against MAC), act (hand off skills/playbook). Traditional consulting mostly skips Study — they ship and leave. RDCO’s state-ownership posture (../04-tooling/rdco-state-ownership-architecture) makes Study structural: the vault + MAC results ARE the deliverable. The engagement can’t end without the Study artifact, which solves Chin’s “forgot to harvest” problem by design. Worth stealing for the engagement template: require a written hypothesis at kickoff, reference it at closeout.

2. The Sanity Check newsletter is a multi-month execution loop at risk of exactly this failure. Each issue is a micro-loop; the content arc is a macro-loop. Easy to get distracted by viral-adjacent topics (Chin’s Burnout Guide trap) and forget the original hypothesis — e.g., “does data-quality-first positioning pull operators into the funnel?” If no written kickoff hypothesis exists per content bet, there’s no way to Study. Action: before the next arc starts, write the hypothesis down, put it in the vault, reference it at the arc’s end. This is a cheap import of Chin’s rule.

3. Related but softer — MAC itself isn’t PDSA. Worth being honest: MAC is SPC for data models (routine vs exceptional variation), not PDSA (plan-do-study-act). They’re sibling ideas from the Deming canon but not the same mechanism. Chin’s 2026-04-15-commoncog-becoming-data-driven-first-principles covers the SPC side; this essay covers the PDSA side. Don’t conflate them in external writing.