Thin Channel
How rarely external truth can arrive before calibration collapses
Submitted by
—
Static demo · no live model
Corridor — any finite schedule
0.205–0.220
Brier on hold-out
Never — no signal
0.298
collapses to raw 3B
Cliff direction
Binary
channel = volume, not frequency
How often does the system get to check reality?
Move the lever from “every day” to “never” and watch the Brier curve
Current check interval
day
every 1 day · 1316 outcomes seen
Brier on hold-out · lower is better
0.205
Collapse — back to raw 3B
(b) base-rate only0.210
(a) raw 3B only0.298
improvement vs raw 3B−0.093
The lever — reveal schedule
← more often · less often →
Daily contact
One annual debrief
Never
(a) raw 3B only
0.298
useless as a forecaster
(b) topic base-rate only
0.189
no model needed
(c) mix (a + b)
0.189
c − b = −0.0001
All 11 schedules — verbatim from the run (click a row to move the lever)
| Frequency | Period | Revealed | (c) mix | (b) base only | (a) raw 3B |
|---|