Phuket experiment desk

A winning variant is not a story until someone writes the readout.

Mindagentgrid teaches app analytics through A/B test readouts for app features. We sit with product pairs who already ship flags, and we train them to stop treating a dashboard screenshot as a decision.

The island is not a metaphor. Seasonality, prepaid SIMs, and tourism weeks break naive runtimes. If your last “significant” test crossed a holiday, we will say so before we talk tuition.

Open the Variant Readout Desk
“Guardrail Grammar stopped us shipping a checkout badge that lifted tap-through and quietly raised refunds. I still think their sample-size worksheet is slower than our PM wanted, and that slowness is why we kept the badge off.”
— Pimchanok R., product lead, travel marketplace · remote seat from Chiang Mai
63 product pairs through the readout desk since 2017
11 wks flagship Variant Readout Desk, including the memo lab
14 d shortest runtime we will still call a feature test, holidays aside
2017 studio year on Thepkrasattri Road — still one room

Flagship

Variant Readout Desk

Eleven weeks on declaring MDE, designing guardrails, refusing peeking theatre, and writing a decision memo a PM can defend in a Tuesday standup. Informational tuition from THB 41,800 per seat. No checkout on this site.

Printed bar charts and a notebook on a pale desk

11 weeks · Phuket room + remote seats

Variant Readout Desk

Feature-flag hygiene, CUPED intuition without a research paper, and a scored memo. Limitation we state up front: a weak feature will still look weak after a clean test.

Read the syllabus

Shorter desks beside it

Guardrail Grammar, Sample Size Without Theatre, and the Sequential Peeking Clinic share the same faculty and the same allergy to inflated lift slides.

If you only have one quarter, take the flagship. If your last experiment was stopped early because a chart “looked done,” start with Peeking Clinic, then come back.

Compare programmes

What we actually teach

Readouts over theatre

Declare the bet

Every feature test starts with a written MDE, a primary metric, and at least one guardrail. If those three lines are missing, we do not call the work an experiment.

Write for a human

The memo is the product. Charts stay in an appendix. If a designer cannot repeat the decision in a hallway, the readout is unfinished.

Respect the island calendar

Thai holidays, tourism weeks, and prepaid top-up cycles change “active.” Phuket is where we teach because the logs we study already include that noise.

Journal

Recent notes from the desk

Open notebook with handwritten experiment notes

12 Mar 2026

Peeking is not a strategy

Why stopping a feature test on day six because the chart “looked green” is still peeking, even when the PM is impatient.

Team reviewing printed diagrams around a table

27 Jan 2026

Guardrails before the lift

Refund rate, crash-free sessions, and support tickets belong in the readout header, not in a footnote after you ship.

Laptop and papers on a wooden writing desk

8 Nov 2025

A readout a PM will sign

The one-page structure we use in class: decision, evidence, caveats, and the holdout you still owe finance.

All journal entries

Thepkrasattri Road

One room, no sales floor, invoices only.

Seats are confirmed by email after we read your current experiment log — or the honest admission that you do not have one yet. Pricing on this site is informational. Write to the desk if you want dates for the next Variant Readout Desk.

Enquire about a seat