QS RESEARCH · FORWARD TEST · PRE-REGISTERED
FST Live forward test: 40 sessions, pre-registered
The FST Live agent is trading a paper options book every market day for 40 sessions under rules fixed in advance. Every session is scored after the close and published here, win or lose, with a daily check that the rules were not changed.
Paper trading: simulated fills, not real money. Results are hypothetical and are not investment advice.
Last updated 2026-11-21 01:40 UTC through Nov 20, 2026.
Cumulative net P&L, whole show
Paper fills, after the assumed commission. The horizontal axis is the session number in the 40-session window.
Every session
| Session | Window | Entries / floor | Closes | Gross | Net | Cumulative net | Main / sleeve net | Checks |
|---|---|---|---|---|---|---|---|---|
| Nov 12, 2026 | void (window 1) | 7 | 7 | +$38.00 | +$22.40 | not counted | — | — |
| Nov 13, 2026 | 1 / 40 | 5 / 3 | 5 | +$67.00 | +$57.90 | +$57.90 | +$57.90 / $0.00 | PASS |
| Nov 16, 2026 | 2 / 40 | 4 / 3 | 4 | +$219.00 | +$208.60 | +$57.90 | +$208.60 / $0.00 | FAIL |
| Nov 17, 2026 | 3 / 40 | 3 / 3 | 3 | +$233.00 | +$229.10 | +$57.90 | +$92.70 / +$136.40 | FAIL |
| Nov 18, 2026 | 4 / 40 | 7 / 3 | 7 | +$663.00 | +$644.80 | +$57.90 | +$557.10 / +$87.70 | FAIL |
| Nov 19, 2026 | 5 / 40 | 3 / 3 | 3 | +$130.00 | +$126.10 | +$57.90 | $0.00 / +$126.10 | FAIL |
| Nov 20, 2026 | 6 / 40 | 3 / 3 | 3 | +$57.00 | +$51.80 | — | +$169.10 / −$117.30 | FAIL |
A day's numbers are the closes of trades entered that session, net of $0.65 per contract per side. Entries are counted against the floor of 3 a session (1 on an early close).
Daily attestation
FAIL Session 6 (Nov 20, 2026): failed check(s): build, deploy, files, build_secret, function, stamps, window.
Sessions with a failed check:
- Session 2 (Nov 16, 2026): settings, stamps, window
- Session 3 (Nov 17, 2026): function, settings, stamps, window
- Session 4 (Nov 18, 2026): build, deploy, files, build_secret, function, settings, stamps, window
- Session 5 (Nov 19, 2026): build, deploy, files, build_secret, function, settings, stamps, window
- Session 6 (Nov 20, 2026): build, deploy, files, build_secret, function, stamps, window
After every close the record checks that the agent ran the configuration fixed before the test: the same code build, the same code files, the same settings and the same risk limits, with no redeploy and no manual change. A failed check is published on the day it happens. Each daily line also carries the hash of the line before it, so a past day cannot be rewritten without breaking the chain.
pinned build 7e624893 · start record sha256 fdbebaf5b707b6dc… · chain head b8cce962cc04d8ff… after 7 lines
What is being tested
FST Live is QuantSignals' trading agent running a paper options book in public, every market day, with no human trading decision. This test asks one question: can an AI agent that trades options live every market day make money after costs? The answer is read once, after session 40, not day by day. Until then every number here is interim, not a verdict.
The rules were written down and approved on Nov 10, 2026, before any session of the test was traded, and they cannot be changed during the window. Any human change to the agent's code, settings or risk limits voids the window and restarts the count. That has already happened once: window 1 was voided on its first day, Nov 12, 2026, when a code change reached the agent during the session. It stays on this page, labelled void. Window 2 started Nov 13, 2026. Read the pre-registration summary.
Pre-registration summary
Approved on Nov 10, 2026, before the first trade of the test. Five dated amendments followed, all before window 2's first session: the agent's build and settings were read back and pinned, the 3-entries floor was confirmed as part of the pass bar, the sleeve's exits were set to 10% and its review floor to −$600, and the restart after window 1 was voided raised the confidence level. Nothing can change during the window.
- Window: 40 consecutive NYSE sessions, Nov 13, 2026 to Jan 12, 2027. Early-close days count. A missed session stays in the window as inactive.
- Two books, judged apart and never netted for a verdict. The main book is the edge claim: the mean net P&L per closed trade. The activity sleeve (one SPY contract at fixed checkpoints when the day is short of entries, with a 10% stop and a 10% target) keeps the show trading every day; it is never called an edge and must lose no more than $1,200 over the window.
- Costs: $0.65 per contract per side, charged on every contract. A harsher cost case is reported beside it.
- Pass bar (main book): at least 60 closed trades, mean net P&L above $0 with its session-bootstrap lower bound above $0 (97.5%, since this is the second window started), drawdown of the whole show no more than $1,500, at least one entry on all 40 sessions and at least 3 on 36 of them.
- Verdicts: PASS (on paper, every bar met). NOT PROVEN: the mean is positive but its lower bound is not above $0, there are fewer than 60 closes, or only the 3-entries floor is missed. FAIL: the mean net is $0 or less, the drawdown bar is broken, or fewer than 36 of 40 sessions were active. NOT MEASURED: the window did not finish by the kill date, or the data rule was not met.
- Early FAIL: if the whole show's net drawdown passes $1,500 on any close, the test fails that day and it is published that day. Nothing else is decided before the end.
- Kill date: Feb 15, 2027. A window not completed and read by then closes as NOT MEASURED.
- The show never stops on any result: it trades every session the market is open.
Watch the agent trade during market hours, with its reasoning on screen, at FST Live.
Watch FST Live