Four Positions, Zero Fills, and One Very Suspicious Taser Deal

The swing bot held four positions today — IWM, SLV, VLUE, XLK — and did exactly nothing with them. No buys, no sells, zero fills, zero dollars traded. Net result: up $2. Sometimes the strategy is just not blinking, and the blink is the mistake.

The active desk had a rougher Thursday. Down $24, no open positions. Codex — the other robot we track on its own separate paper account — came back with a clean block: no actionable setup. On a short pre-holiday session heading into a long weekend, both AIs sat on their hands. The codex real account ended up $0.38 on 35 actual closed trades for the year. Small positive, honest number.

Under the hood, the strategy roster had a messy day: 8 strategies pruned, 9 promoted, one immediately hit the roster cap and got trimmed right back out. One strategy — SMH rsi_meanrev — got pruned AND promoted in the same session cycle. The evaluation window caught it trending down, then caught it on the bounce. That's not a signal. That's the window being too narrow relative to the noise. We flagged it as a bug worth fixing: strategies should need two consecutive bad windows before getting cut, not one, so they don't oscillate in and out on the same afternoon.

Federal disclosures surfaced this week: back in February, a sitting president purchased between $1 million and $5 million of Axon stock — the company that makes the TASER 10 and controls roughly 90% of the U.S. Taser market. Fourteen days later, ICE published a $220 million contract solicitation with specs so specific — 45-foot range, 10 deployable probes — that independent experts said the specs were written for Axon's product and would exclude every competitor. We're documenting it. The timing is the whole story.

Something completely different and genuinely interesting: 20 to 25 genomics stocks just broke out to 52-week highs simultaneously this week. Not one moonshot name dragging up the rest — a coordinated sector breakout. TWST up 100%, Maravai up 98%, NeoGenomics up 77%, Illumina — which has been dead money for years — somehow gained 40%. The median name in the group is up 33% over three months while the Nasdaq managed 4%. When an entire sector moves together like that, it's usually rotation money arriving in size, not coincidence.

And then there's this out of China: Z.ai released an open-source AI model this week that benchmarks at par with the best Western frontier models. The part that matters — it was trained entirely on Huawei chips. Not Nvidia. Huawei, the company semiconductor export controls were specifically designed to cripple. The CEO says a Claude-tier model is months away. If frontier AI can now be trained on sanctioned hardware, the entire strategic premise behind chip export controls just got considerably more complicated.

A thread in Reddit's algo trading community this week had someone sharing a strategy with a Sharpe ratio of 1.52 — which sounds strong. He tested 'several hundred' algorithms to find it. The most upvoted response said something worth writing down: 'A Sharpe of 1.5 means something very different if it came from the third idea than if it came from the three-hundredth.'

This is called the Deflated Sharpe problem. The more strategies you run through a backtesting machine, the higher the odds that your 'winner' looks good purely by accident — the same way flipping a coin a thousand times and keeping the best streak doesn't prove the coin is magic. The math to correct for this exists. You take the total number of strategies you tested, factor that into your surviving result, and adjust the significance score downward accordingly. A strategy with a 1.15 out-of-sample Sharpe out of 50 candidates is a very different animal than the same number out of 500. We have 13 live strategies right now. They survived some number of total candidates that we haven't precisely tracked. That's a gap. From here forward, every strategy that gets promoted will log how many candidates were tested before it was selected. That number goes into the significance math. If we can't count the denominator, the numerator doesn't mean what we think it does.

Combined paper balance: $1,978 across two accounts, funded at $2,000. Swing desk: $1,002, up $2. Active desk: $976, down $24. Today's combined: -$3.

The live edge read after 7 trading days: we're up 0.19%, SPY is up 1.57%. Alpha gap: -1.39 percentage points. Our live Sharpe clocks at 2.48 versus SPY's 5.85 over that window. Seven days is noise — it's barely enough data to warm up the model, let alone validate it. The backtested out-of-sample picture is the more honest read: roster median Sharpe 0.95, momentum sleeve at 1.02 versus SPY's historical 0.74. That's where the actual signal lives. The live number will need months to say anything real. We're documenting it week by week so there's a record — wins, losses, the dumb ones, all of it.