Track record
Predictions, logged in public
Every Monday this page records which companies the index flags — in advance. 90 days later, each call is scored against real headcount data. Hits and misses both stay on the page, forever.
Why this page exists: backtested on 2024–2026 (2,574 company-months), companies flagged by the index's decline signal went on to shed ≥5% of staff in the following quarter 3.4× more often than the cohort base rate. Backtests persuade; a public, timestamped forward record proves. Full method + caveats on the methodology page.
Cohorts logged
1
Predictions pending
20
Resolved
0
Hit rate
—
first cohort resolves ~135 days after logging
Week of 2026-07-13logged in advance · resolves +135d
Decline signal (backtested ×3.4)
Published Cut board · top 10
The rules (fixed in advance)
- Two series are logged every Monday, automatically: the companies in the worst ~20% of Cuts by trailing-90-day headcount change (the exact signal the backtest validated), and the published Cut board's top 10.
- Each entry resolves ~135 days later — a 90-day outcome window plus the data source's publication lag — against the same ruler every time: hit = headcount fell ≥5% from the logged baseline.
- Entries are never edited after logging. Resolution fills in the outcome; nothing else changes. Misses stay on the page.
Honest limits
The backtest is within our tracked cohort (companies selected in 2026, with hindsight), on monthly data as revised by our data source today, with overlapping outcome windows. That's exactly why this page exists: the forward record has none of those outs. What gets logged here was written down before the outcome existed.