Tool · finding the story in data
Don't read tea leaves. Test stories.
Paste your data. Get the story it confirms.
Every data series is about to become a story in someone’s meeting — the only question is whether the story is tested. We keep a catalog of data-story templates, each with truth conditions your data can falsify and a canonical form. Bring a series; we classify the form, fire the conditions, and hand back the story that survived — problem, why it matters, what to do — with the evidence trail attached. If nothing survives, we say so.
1 · bring data
2 · the form your data fits
the scene — pre-built from the deck; toggle to see your actual data
3 · the story — confirmed against your data
On this trajectory, you breach the benchmark
Attrition heading for the benchmark (forecast) is rising and clears the certainty cutoff — still under Industry benchmark (12.0) today, but on the naive linear trend it crosses in about 2.9 periods.
A number on the safe side of the line reads as 'fine' while the slope is the real story. Acting at the crossing is acting late — the people who make it breach are disengaging now, in the window that still looks safe.
What to do: Treat the projected crossing as the deadline: scope the intervention against the periods you have, and re-measure the slope (not just the level) each period. The calibrated forecast is the next step up.
- ✓ the trend is a sustained rise — best-fit form is Trending up (fit 91%)
- ✓ the rise clears the certainty cutoff — trend up at 97% certainty (needs up, ≥80%)
- ✓ on the naive trend, it crosses the line within 6 periods — on the naive trend, crosses Industry benchmark in ~2.9 periods (needs ≤6)
how sure is this — and what would more certainty cost?
achieved: form fit 91% · trend certainty 97% over 8 periods
- → more history — 8 periods is thin; 16–24 sharpens the trend call
- → segment the series (by team/level) — org-level trends can hide where the story lives
One read is often wrong. Each step above tightens the error bars — the full system prices these steps against what the decision is worth, so you buy exactly the certainty you need.
other candidates tested
- Top talent is quietly leaving3/3 conditions · 97%
- Inside the line — and holding1/2 conditions · 57%
- Past the benchmark — and climbing1/2 conditions · 57%
- Something changed — and it stuck0/1 conditions · 18%
How it works.The engine never invents a story at runtime — it draws from a curated corpus. Each template declares the conditions under which it’s true (a confirmed trend past a stated certainty cutoff, a material change, a specific shape). Your data either fires them or it doesn’t. The chart is the same discipline: we don’t plot whatever the line does, we ask which canonical form it best matches — flat, trending, spike, step, dip and recovery — and past what certainty the answer changes. You can always toggle back to your actual data; the form is the message, the wiggle is not.
What this is the seed of. A v1: deterministic, in your browser, nothing uploaded. The full system grows the catalog systematically, lets you set your own thresholds, matches with AI, and delivers through the report surfaces executives actually read — so what lands on the desk is narrative and data captured together.