Every study we run, published with its real numbers, including the losses, because a loss that becomes a rule is worth more than a win. Each card names the venues it applies to.
Lighter · Robinhood chain
24.0h paper run · Aug 19-20, 2026 · COIN, ORCL (control), AAPL, SUI, NEAR, BTC · $100 clips
Method (the shape, not the secret): The same simulated quoting engine that validated ORCL, pointed at the next candidates, including two crypto books with real width and real flow, to answer whether the edge is a stock-token thing or a wide-book thing. Deploy rule unchanged: capture AND after-fill drift both ≥ 0 across the full day.
What came of it: COIN failed, and it failed clearly. It is not a second ORCL. SUI (−0.56 / −11.26) and NEAR failed the same way, and NEAR is the textbook case: the best capture in the run at +4.34bps, wiped out by −9.35bps of after-fill drift. Fills that look brilliant and then get run over are the trap this rule exists to catch. AAPL missed by a hair (+0.71 / −0.24). The genuine surprise is BTC (the TIGHTEST book in the run at 1.11bps) passing both tests on 171 fills/hr, which cuts against our own 'wide niche books are the edge' assumption and is the finding we most want to re-test. And the control did not repeat: ORCL's capture more than tripled to +1.72bps, but its after-fill drift went NEGATIVE (−0.91 vs +0.18 in the 44-hour study), so by our own rule ORCL does not clear the bar on this run. One run is not a verdict on ORCL, but it is exactly why the Edge-Health Fleet exists.
🔬 Still watchingin plain English
Do not quote COIN, SUI or NEAR. The tape said no, plainly. BTC is the one that earned a second look, and even ORCL, the market we already ride, slipped under our own bar on this run. Nothing here changes what you can ride today; it changes what we test next.
every venue · around the clock
continuous · an always-on worker probing every book we trade, every few minutes
Method (the shape, not the secret): Each book's spread and flow, checked against its own ~7-day average. A verdict only flips after three consecutive checks agree, and then it posts to the LIVE EDGE HEALTH board above and into the in-game notices.
What came of it: Live now. Edges die when professionals arrive and compress the spread (gold's mainnet twin is the proof: same asset, half the spread). This fleet exists so that day is a notice, not a surprise.
🛡 Already protecting youin plain English
You do not have to watch for the day an edge dies. This does it for you. When a road starts compressing, a notice reaches you in the game before you ride into it.
See the live board ↑Lighter · Robinhood chain
44h continuous · Aug 16-18, 2026 · $100 clips, simulated against the live tape
Method (the shape, not the secret): Post-only quotes resting on both sides of a fair price, small fixed clips, hard inventory caps, and stock-tokens quoted ONLY while their real market sleeps. Every simulated fill was checked twice: what it captured at the moment of the fill, and where price drifted afterward (the test that catches poisoned fills).
What came of it: One market passed both tests across the full run: ORCL. Two more (MSFT, BABA) earned on capture but their fills kept getting run over afterwards, so they stay benched. The winner deployed into the game the same day as 📗ORCL, so you can ride the exact configuration the study validated.
✅ Ride thisin plain English
This is the study behind the ORCL setup you can ride right now: the exact configuration that passed, not a version of it.
Ride ORCL on Lighter →every venue
measured across live sessions + a continuous public-tape recorder
calm weekend (0.68bps vol)
US open · Monday ramp (3.6bps)
Method (the shape, not the secret): A 30-second volatility estimate computed from the live tape, anchored against the real cost of full farming sessions in each regime.
What came of it: Volume does not have one price. The HOUR sets it, with a ~20× spread between the cheapest and most hostile windows. This ladder became the MARKET STORM: the game measures the weather live, names the hour's price, shelters you in the worst of it, and asks before letting you ride through.
🛡 Already protecting youin plain English
The hour you drive matters more than almost anything else you choose: up to about 20× more. The game now names the hour’s price for you and shelters you in the worst of it, so you do not have to memorise a clock.
Lighter · Robinhood chain (majors)
906 simulated legs · 2 days of recorded tape · 4 exit styles on identical entries
fast-scratch vs patient exits
Method (the shape, not the secret): The same entries replayed under four exit rules, from 'cut at −3bps, bank +2 instantly' to 'cut at −10, trail winners', each scored per volatility regime.
What came of it: All four styles landed within a few dollars per million of each other in every regime. The temptation to tune exits after a bad hour is a trap; the hour itself is what decides the cost. This study is why the engine's exits stay measured and boring.
🛡 Already protecting youin plain English
Taking profit early is not the free win it feels like. The engine already exits the way the test favoured, so there is nothing for you to tune.
Lighter · Robinhood chain
one evening · Aug 19, 2026 · real fills, real money
Method (the shape, not the secret): The validated study configuration, armed live by a real rider at $100 clips.
What came of it: The captures worked exactly as studied, but the stops escalated straight into the US premarket (−$0.23 → −$0.44 → −$0.61 → −$0.72), which the paper's averages had smoothed over. Within hours the finding shipped as a wider gate: stock-tokens now also sit out premarket. An honest loss, converted into protection for every future rider.
🛡 Already protecting youin plain English
A real loss of ours became a rule that protects you: stock-token roads now sit out the US premarket automatically. You cannot ride into that particular hole any more.
Lighter · Robinhood chain × Hyperliquid
24h in · $500 per leg · live funding + live books · still running
cheapest two-venue volume
Method (the shape, not the secret): Long one venue, short the other, same size, price risk cancels and the pair earns the funding gap between the two books. The run accrues real funding minute by minute, tracks how far the losing leg drifts (the margin a live engine must manage), and cycles a simulated pair to price two-venue volume.
What came of it: Interim, and already useful. LIT leads at 50.4% APR with HYPE behind it at 15.3%, and BTC prices two-venue volume at $33 per $1M, cheaper than most single-venue taker churn we have measured. The warning is just as clear: CASHCAT's basis blew out to 202bps, costing $2,966 per $1M with a −$84.67 worst leg, which is exactly the margin event a hedged engine has to survive rather than average away. Several pairs earned nothing at all because their books never crossed. Final verdict when the run completes; hedged pairs only ship with leg protection sized off these worst-leg numbers.
🔬 Still watchingin plain English
Not ready to ride yet. Hedged pairs stay off until the run finishes and leg protection is sized off the worst drawdowns we measured, including the pair that blew out to 202bps.
Lighter · Robinhood chain
continuous · 300+ MB of tape and growing
Method (the shape, not the secret): Every public print on the watched books, recorded around the clock.
What came of it: The raw material every study above is built from, and the ground truth the game's ✓ counters reconcile against.
🛡 Already protecting youin plain English
This is the receipt drawer. Every number on this page traces back to it, and it is what the game’s ✓ counters are checked against, so nothing here is a claim we cannot show.