How the Sextant track record works
Every night, our system publishes a model portfolio. This page explains how we score it — and how you could check our numbers yourself.
The one-line promise
Anyone holding our published portfolios, this page, and free public closing prices can rebuild our performance chart and get the same answer. If you can't, we've broken something, and we want to know.
Two scoreboards
Did each night's call turn out to be good? Every nightly portfolio is scored on its own, as if you'd bought it that night and held it. A rolling 60-day average of these scores is our internal gate: until it stays convincingly positive — including through at least one major change of view — no real money moves. Ever.
What would one continuously-run portfolio have earned? We also track a single portfolio that follows the system night after night. Every change from one night to the next counts as trading, every trade gets charged a cost, and the result is one performance line shown against the S&P 500. This record starts on 11 August 2026 — the first night the system ran with no human touching it.
Both scoreboards use the same prices, so they can never tell different stories.
When we decide, and when we deal
Our nightly chain starts at one in the afternoon New York time and Sextant is its last stage, so each book is decided in the hour before the New York close — typically around half past three — from the last complete close, while the market is open. Its orders are for that day's close. Every published book carries both times: decided at, executed at. If a slow night ever pushed a decision past the bell, the honest fill would be the next session's close; that guard is a stated design question, and the published decision time is how you would catch it.
That close happens after the record for the night has been built, so the newest line of the record is always marked the following night. The published curve runs one session behind the book. It is the cost of marking at a price we could actually have had, and we pay it on purpose.
From 29 September 2026 there is one book per trading session. No book is made for a Saturday, a Sunday or a New York holiday, because no order could fill that day. Before that date the system ran every night and a weekend's books could only fill at Monday's close; the three books of 27, 28 and 29 September share one close and are the only ones in the record that do.
How we price things
We use official closing prices — our own data first, Yahoo Finance as backup. Both scoreboards draw from the same source. Each book is marked at the close it filled at: the first session close on or after the day it was decided.
One honest quirk: because the system decides before that close, a position has no real result on its first night. Rather than print a meaningless 0.0%, we label it "no evidence yet" and wait. Real results start accumulating from the next rebuild.
What the rules are, and when they changed
The fast lane buys a signal on the second night in a row that Radar names it, sizes it by how many clusters name it and for how many nights, and sells it on silence (a fifth quiet night) or when the price has given back one measured bad week from its best level since purchase. The slow lane holds thesis positions locked between the weekly human review. Nine tenths of the book goes to the fast lane and one tenth to the slow, fixed until the slow lane earns more on its own ninety-day record; no instrument takes more than 15% before the book-wide risk budget; how much of the book is out follows the market weather, two dials read from prices each night and published with every book.
These rules went live with the book of 27 September 2026. The six weeks before ran two earlier rule sets, and their books stand exactly as they were run, losses and all. We did not rewrite them, and we will not: a record that can be rewritten is not a record.
If the system holds bets in both directions on the same instrument, we count both — we never quietly cancel them out to make the trading look smaller.
Trading costs
Nobody trades for free, so neither does our record. Every trade is charged a cost based on how easy that instrument is to trade — from 0.03% for the most liquid ETFs to 0.10% for everything else. These are deliberately conservative estimates, not real fills, and we say so. On day one, the entire starting portfolio is charged as if bought from scratch.
When we can't score something
Sometimes an instrument can't be priced — an unusual contract, a bad data feed. We never quietly drop it. It appears in the published record with the reason stated, so you can always see what was left out and why.
Can a human interfere?
The system runs by itself. A manual-override facility exists for emergencies, but every use of it is logged with a reason and an expiry date, and the record shows both versions: what the machine did alone, and what happened with human hands on it. To date, it has never been used.
The receipts
Every nightly portfolio, every score, and every ledger update is published as a timestamped file. The chart and the table you see are built from those files and nothing else.
The bottom line
Our rules limit losses; they don't manufacture profits. If the system's judgement isn't good enough, the scoreboard will say so. That's not a bug in the process. That is the process.