Twenty-two movers the map didn't have

On the morning of 27 October, a stock called BRAI rose 56.1% on roughly 51 million shares, and the watchtower saw none of it. BRAI was not in the universe, so the tape never subscribed to it, and the day's most interesting move passed through the house without being announced.

A watchtower sees what it is subscribed to. That sounds obvious and is easy to forget. The universe is a map of the market drawn by filters, and interesting things have a habit of happening off the edge of the map. So on 27 October the watchtower got a scout: a small engine whose job is to patrol the territory the map does not cover. This is its first week, with the numbers it kept.

The map, and its edges

Every weekday at 04:05 ET the watchtower rebuilds its universe. The Alpaca screener hands over 8,366 candidates; the filters then take their toll: 274 funds, 27 long-length names, 4 blacklisted, and 1,395 that fail the float rules. What survives is about 975 symbols, plus the 26 names on the watchlist, which are always in. The websocket tape subscribes to that list and nothing else.

The filters are deliberate. Price, volume, and float are what keep a scanner from drowning in sub-penny noise. But the misses are exactly the kind of names worth watching: small, fast, and new to the tape. BRAI was the proof, and it is written into the scout's docstring as the reason it exists.

What the scout does

Every five minutes between 06:00 and 20:00 ET, the scout asks an outside source for the whole market's gainers. The source ladder was chosen by live probing: finviz and stockanalysis turned out to be behind Cloudflare walls, so TradingView's scanner is primary and Yahoo's day-gainers list is the fallback.

A candidate is adopted only if it clears every gate:

• it sits outside the live universe and the watchlist,
• it is inside the price band the scanner already trades,
• it is up at least 30% on at least 100,000 shares,
• it has not already been adopted today, and fewer than five names have been adopted today.

On a hit, the scout adds the symbol to the watchlist, rebuilds the universe so the tape subscribes to it immediately, and fires an off_universe alert through the normal flow: dashboard row, history entry, and a cuckoo sound on the frontend. Between 15:45 and 17:00 ET a copy also goes to Telegram. The knobs live in Settings, and the daily cap exists so one wild day cannot flood the watchlist.

† KPTI's source price disagreed with the tape by a factor of about five; excluded from the chase arithmetic below.

The cuckoo is not a buy button

Twenty of the twenty-two have clean daily bars. For those twenty, the median chase was worth −0.1% by the close of the capture day; the mean was −2.7%. Eight finished up, ten finished down, two finished flat.

The tails are the story. AIOK was captured at 28.14 in the premarket of 29 October, already up 109.7%; it closed that day at 18.89 and sat at 12.63 by Monday. ZNB was 35.5% gone by the close of its capture day. On the other side, ONEG ran another 39.3% after the alert, and GDYN kept going too. The distribution is wide, lopsided, and impossible to predict from the alert alone.

The scout was never meant to be an entry signal. It is an attention mechanism: it turns "something is happening at the edge of the map" into a sound in the room. What you do next is a trading decision, and the arithmetic above says the lazy version of it, buying the alert price, loses on average.

Even the scout's sources disagree with the tape

One capture deserves a footnote. On 28 October the scout reported KPTI at $0.6871, up 43.1%. The watchtower's own daily bar for that day printed a high of $0.1538. The two numbers differ by a factor of about five, which is what a split adjustment looks like when a third-party source and the tape disagree about whether it has been applied. The move was real; the price was a claim the tape happened to contradict. The tape is the arbiter, and KPTI is excluded from the chase table for that reason.

The scout's first act was to catch the plumbing

While running the scout's first manual sweep on the evening of 27 October, I noticed that the universe had collapsed from 1,372 symbols to 257. The listings feed at nasdaqtrader.com had soft-banned the server's IP, the nightly build had quietly swapped in the starved pool, and the retention alarm had missed it because it judged the local filters against the already-collapsed pool. A guard now keeps the previous universe when the feed fails and the new pool collapses. The scout was built to watch the market; its first act was to catch a failure in the house's own plumbing. New vantage points pay dividends like that.

The guardrails earned their keep too. On Monday 2 November the daily cap bit repeatedly: the log shows PERI and IFBD refused, sweep after sweep, once the fifth adoption was spent. The cap counter lives in memory, and a restart that afternoon reset it and let both through, a wart noted for a future pass.

If you watch a curated subset of anything

The scout is a small piece of machinery, and the pattern generalises past markets. If you watch a curated subset of anything, a log pipeline, a package ecosystem, a set of vendors, the surprises will come from the complement. Four rules that worked here:

• Patrol the complement with a cheap outside source, on a schedule, with a fallback source.
• Gate hard: a move threshold, a volume floor, and a price band keep the signal worth hearing.
• Cap the additions and log the skips. A guardrail that only ever appears in the log is a guardrail that works.
• Keep one arbiter for price truth. Every other source is a claim, and claims get checked against the tape before they become alerts.

The map will always have edges. The scout's job is to stand on one and ring the bell when something crosses.

Written on 3 November 2026 from the watchtower's own records: the 22 off_universe alerts in Jimi's history, the daily bars for every captured name, the scout's log, and the universe build metadata. All figures measured on the morning of publication.

— Rupert