Methodology
Diers Capital publishes real dollars, not backtests. This page states the mechanism exactly as it is coded — nothing here is a performance claim.
1. Composite ranks
A scanner reads SEC-filing signals (going-concern language, short-interest tilt, accession-number cadence) into a ranked composite. This narrows the universe; it does not vote and it does not execute.
2. Committee drops
Five seats vote keep or drop on each candidate the composite surfaces. 4 of 5 must vote keep before anything else happens. This is the only approval rule in the system — there is no override, no manual approval path, and no seat carries a veto or a tie-break beyond this count. Every drop publishes on the log exactly like every keep.
3. Executor fills
An approved thesis executes automatically in the Agentic account — a real taxable brokerage account, not a paper account — and the fill publishes on a one-day (T+1) delay.
Position-size ladder
Position size is set by account NAV, read directly from
services/portfolio/sizing.py. Every band below is compared to that code in
CI — this table cannot drift from the number that actually sizes an order.
| NAV band | Positions held | Max single position |
|---|---|---|
| $0–$25,000 | 15–18 | 8% |
| $25,000–$100,000 | 18–24 | 6% |
| $100,000–$200,000 | 22–28 | 5% |
| $200,000+ | 28–32 | 4% |
Each band is stated from its lower boundary; a NAV exactly on a boundary takes the higher tier.
Canary window
The canary runs 28 Sep–2 Oct 2026. During the canary, the first trade is capped at $1, each subsequent trade is randomised $1–$25, and the cumulative buy cap is $200 — dollar caps, not percentage targets, so a small mistake stays small while the pipeline is proven out.
What we do not report
We do not compare results against any index or benchmark. Results are reported as the account's own value, in dollars, with dates — nothing is called a win or a loss against a yardstick.