Reserve Risk prices long-term-holder conviction against the market. We tested it inside our model, measured it, and it made the model worse — so it carries zero weight.
Price divided by the HODL Bank — the running total of coin-days accumulated by holders who have not sold. A low reading means conviction is high while the price of buying alongside it is low; a high reading means conviction is being spent into strength.
Our data: Checkonchain, full history from 2010. Banked in this repo as a frozen, write-once series with a SHA-256 prefix, like every other graded input.
| Measure | Without it | With it |
|---|---|---|
| Walk-forward geomean edge | 1.176 | 1.106 |
| Folds beating buy-and-hold | 10 of 17 | 8 of 17 |
| Full-history dollar multiple | 4.25x | 2.54x |
| Current-cycle edge | 1.41x | 1.48x |
Measured against the model as it stood on 2026-07-15. That baseline is several re-baselines old and is NOT the current published record — see the register for today's figures. Recorded at src/services/cycleProjection.ts:3053.
No — and that is what makes the result worth publishing. We measure a candidate's independence as the phi coefficient against the nupl < 0 anchor, the valuation condition the model already reads. Reserve Risk scores 0.412. Genuinely independent. For comparison, two candidates we had believed were independent scored 0.9945 and 0.9936 against the same anchor — they were the anchor, algebraically, wearing a different name.
Recorded at brain/IMPROVEMENT_PROPOSALS.md:436.
Reserve Risk is not a bad metric, and it is not a duplicate of the valuation signals we do use. Measured against the same valuation anchor it scores phi 0.412, which is the range we call usable.
It still made our model worse. Adding it at weight 0.04 cost 0.070 of walk-forward geomean edge and two of seventeen folds. Statistically distinct is not the same as additive: the model already reads the valuation axis through MVRV, and Reserve Risk arrived carrying mostly the same call at a worse time.
The one thing it improved was the current cycle, from 1.41x to 1.48x. That is exactly the improvement we distrust most. A change that helps only the most recent cycle is the signature of fitting the present, and our gate is designed to reject it however good the headline looks.
It stays wired at weight 0 and can be re-swept in one command. We did not delete it; we priced it at nothing.
Every signal we add has to survive the same gate, and most do not. Our published record is measured over 18 walk-forward folds (the year and cycle folds tile the same span, so they are not independent of each other). The figures in the table above are dated results against the model as it stood on the day of the test — the register is the only authority on where the model stands today.
Read the current call in today's briefing, or the timestamped record of every call in the call ledger.