Here lie our ideas
Every agency claims it tests. Here's ours, with headstones: concepts we designed, believed in, shipped, and killed, each with the data that ended it. We publish the funerals because an 83% kill rate is either a boast or a confession, and honestly it's both. Names of client games withheld; causes of death exact.
⚰ Mar 2026 · 5 days
The fake slider
"Promised a swipe, delivered regret."
CTR 3.1%, IPM 1.2, D1 of its cohort 11 points under control. Bait clicks, tourist installs. Never again, until someone new joins and suggests it.
⚰ Apr 2026 · 6 days
Cinematic intro, 40 seconds
"Beautiful. Unwatched past second four."
Hook rate 12% vs bench median 31%. The game deserved a trailer; the auction wanted a hook. Both things are true.
⚰ Feb 2026 · 8 days
Timer-pressure puzzle hook
"Great CTR, brought the wrong crowd."
Clicks +38% over control, cohort D1 −7 points. Recruited adrenaline, sold zen. The mismatch was the whole autopsy.
⚰ May 2026 · 4 days
CEO-says-sorry UGC
"Apologized for a game nobody was mad at."
The 'our game was broken, we fixed it' format needs a famous game. IPM 1.8. Confusion is not intrigue.
⚰ Jun 2026 · 7 days
Hyper-real kitchen ASMR
"Ten million views on organic. Two installs."
Watch time superb, install intent absent. A reminder that virality is a currency some registers don't accept.
⚰ May 2026 · 6 days
The graph-forward sleep ad
"Showed the dashboard. Sold the homework."
Cost per trial 41% above the phone-face-down winner. Full story in the sleep-app post-mortem.
⚰ Jan 2026 · 9 days
Playable with the real tutorial
"Faithful to the game, fatal to the funnel."
Completion 22% vs 61% for the compressed build. A playable is a wink, not a syllabus. Revision two lives on happily.
⚰ Mar 2026 · 3 days
Winter-theme reskin, in March
"Seasonal, in the wrong season."
Shipped late from the backlog. IPM fine for one weekend of northern-hemisphere cognitive dissonance, then cratered. Calendar-aware naming conventions exist now.
⚰ Apr 2026 · 1 day
The influencer voice clone
"Died of a policy check, mercifully fast."
AI voice too close to a real creator's. Killed pre-launch by our own review, which is the cheapest possible funeral. The bench's only proud death.
⚰ Jun 2026 · 6 days
Choice hook 17-B
"The winner's twin, minus the winning."
Identical to 17-A except the stranger sorts correctly. Nobody corrects a correct stranger. IPM half of its sibling; the control condition earned its keep.
Why we bury in public
Three reasons. Discipline: a kill decision you must write an epitaph for is a kill decision you actually make, and our median time from launch to funeral is six days because of it. Memory: the bench's worst habit would be re-testing old corpses, and a graveyard is cheaper than déjà vu. And honesty with clients: the winners in our client stories make sense only next to the nineteen graves behind each one. The market's direction makes the discipline non-optional, with top-quartile campaigns now refreshing creative weekly; a fuller method write-up lives in the testing framework guide.
We'll dig graves for you too
The bench ships 20 to 40 concepts a month and buries most of them. What survives is why clients stay.
Start a run