Write an assignment ledger
Document eligibility, exposure, and leftover client state before anyone opens a metric dashboard.
Flagship program
Signal Lab: read the flag before you celebrate the lift.
An eight-week studio residency for product, analytics, and mobile engineering leads who already run experiments and still leave the readout meeting with three competing stories.
What you will be able to do
Name cache interference
Identify TTL and key-shape problems that make treatment users look like control, or the reverse.
Hold a limitation-first readout
Run a meeting where the first slide is what the test cannot claim, including sample-size and window issues.
Protect guardrails
Place crash-free sessions, latency, and refund rate beside conversion instead of below the fold.
Modules
Eight weeks, one artefact a week.
Week 1 — Exposure or theatre
Rebuild your assignment story from logs, not from the flag console screenshot. If exposure cannot be proven, the residency pauses here on purpose.
Week 2 — Metric windows and late events
When the clock starts on iOS and Android, how ATT consent changes the population, and what to do with SDK retries that arrive on day three.
Week 3 — Cache, CDN, and node lies
The module the studio is named for. Disk caches, WebView persistence, and keys that ignore flag version.
Week 4 — Holdouts that actually hold
Contamination through shared devices, family accounts, and “internal users” who were never excluded.
Week 5 — Guardrails as first-class outcomes
Crash-free sessions, p95 latency, refunds, and ticket volume. A green primary KPI with a red guardrail is an unfinished decision.
Week 6 — Release windows for Thai calendars
Store review lags, staged rollouts, Songkran freezes, and OEM update weeks. Taught against your real ship dates.
Week 7 — Readout rehearsal
Each team presents one live or recent flag. The room is allowed to stop the presenter if the limitation is missing.
Week 8 — Ledger handover
A written grammar your organisation can keep after we leave, plus six months of alumni ledger hour.
Instructor
Apinya Sutham
Studio director. Former instrumentation lead at a Bangkok super-app. Apinya designed Signal Lab after three “onboarding wins” collapsed when a CDN TTL was shortened. She still teaches week 3 herself.
Informational fee
Atelier Residency is listed at THB 186,000 for a team of four. Additional seats are billed separately. This page does not take payment. Confirm availability and invoicing with the desk.
Questions we actually get
Do we need a specific analytics vendor?
No. Signal Lab has been taught against Amplitude, Mixpanel, BigQuery exports, and a homegrown event bus. We care whether exposure is logged at user level, not which logo is on the dashboard.
Can we join without an engineer in the room?
You can, and it usually goes poorly in week 3. Cache and client state questions need someone who can open the app. We ask for one engineering observer even if they only attend those sessions.
What is a real limitation of Signal Lab?
We cannot repair missing historical exposure logs. If your last two quarters of flags have no trustworthy assignment data, we can teach the grammar going forward, but we will not reconstruct fake certainty about tests you already shipped. Teams who need a forensic rebuild of old experiments should not expect that from this residency.
Is there a public cohort or only private ateliers?
Signal Lab runs as Atelier Residency for a single organisation. Cohort Grammar is the public bench. Mixing unrelated companies into week 3 cache reviews has never been useful.
From recent residencies
“Week 3 ended our checkout flag. Painful. Correct.”
Growth lead · Bangkok fintech
Release-window week finally put iOS and Android on one calendar. I still think eight weeks is long for a team that already “knows experiments,” but the limitation-first readout is now our default.
Siriwan Phol · marketplace experiment owner