Does the score
hold up?
Every launch Cyclops scored before its outcome was known, grouped by score, against how it actually ended. Published as it stands, including where it falls short.
Higher score,
worse ending?
Scored before the outcome, outcome now known, since 29 October 2026.
The base rate: the dashed line below.
Scored, outcome not known yet. Not counted.
Hover or focus a bar. Across all judged launches, 0% ended badly (dashed line).
Show the numbers as a table
| Score | Judged | Ended badly | Rate | 95% interval |
|---|---|---|---|---|
| 0–9 | 8 | 0 | 0% | 0%–32% |
| 10–19 | 11 | 0 | 0% | 0.0%–26% |
| 20–29 | 22 | 0 | 0% | 0.0%–15% |
| 30–39 | 3 | 0 | 0% | 0%–56% |
| 40–49 | 0 | 0 | — | — |
| 50–59 | 0 | 0 | — | — |
| 60–69 | 0 | 0 | — | — |
| 70–79 | 0 | 0 | — | — |
| 80–89 | 0 | 0 | — | — |
| 90–100 | 0 | 0 | — | — |
A score is not a probability, so there is no diagonal to match. What a useful score shows is bars that rise from left to right, with low deciles under the dashed base rate and high ones above it. Wide whiskers mean few launches; read them as uncertain.
BY BANDEach launch counts once, with the first score Cyclops gave it, and only when that score was taken before the outcome was recorded. "Ended badly" = collapsed (-90% from peak) or creator sold out (-80% while the creator sold). Intervals are 95% Wilson intervals.
Who gets
flagged?
A first-time creator has no history for the score to read, and a repeat creator’s past launches are scoring inputs. This checks whether that tilts the score against one group: how often launches that did not end badly were still scored at or above each threshold (false positives), and how many that did end badly were caught.
A false positive is a launch scored at or above the threshold that did not end badly (still trading or no trades in three days). The rate is false positives over all launches that did not end badly, per group. First-time creators launched for the first time in the indexed Torch history; repeat creators had launched before. Creator history is itself a scoring input (devhistory, devrugs, serial), so a gap between the groups is expected and is shown, not hidden.