PROTOTYPE2026
Holdout
An incrementality testing harness: geo splits, budget holdouts and lift readouts standardized into a repeatable protocol — so 'did it actually work?' becomes a routine question, not a research project.
Labs · Experiments
Where we test what
we believe
Internal research, open source tooling and prototypes. Most experiments fail; the ones that don't become infrastructure. Both outcomes are the point.
RESEARCH2026
Hookrate
A creative intelligence study: thousands of short-form ad openings scored against retention curves to map which hook mechanics survive the first three seconds — by category, format and platform.
PROTOTYPE2026
Answerable
Tracking brand visibility inside AI answers: which assistants cite you, for which queries, with which sources — share of voice for the post-ten-blue-links era.
OPEN SOURCE2025
Cleanroom
A tracking-audit toolkit: crawls a site, diffs the measured events against the tracking plan and flags duplicates, gaps and consent leaks before they corrupt a quarter of decisions.
CONCEPT2025
Slopgate
A quality gate for AI-assisted content: scoring drafts for genericity, unverifiable claims and template smell — the reviewer that keeps volume from becoming noise.
ARCHIVED2024
Payback
A cohort-economics simulator used to stress-test growth plans: how CAC, margin and repeat rate interact at 3× spend — and where 'working' channels quietly become unaffordable.
Lab work is deliberately unglamorous: hypotheses, harnesses, measurements. When an experiment graduates, it appears in our services. When it doesn't, it appears here anyway — negative results are results.