You cannot committee-score the entire market every minute. MarketXED scanners shrink the tape into a universe of names that are allowed to be debated. Filters typically include price, volume, and listing quality. The job of a scanner is eligibility, not a buy list. Nothing here is financial advice.
What a universe is for
A universe is the set of tickers the agents are allowed to waste compute on. If the screen is all already-extended gainers, the committee will keep arguing about late continuation. If the screen is too broad, you get noise and PDT churn. Save a filter that matches how you actually trade: cash-account slow turnover is a different universe than a margin account hunting RTH momentum.
- Liquidity: can you exit without becoming the print?
- Price band: avoid names where a 1.5 percent stop is a rounding error or a halt magnet.
- Session: regular hours only for alerts; pre-market scans are context, not a text.
Market data may come from public Yahoo-style quotes and bars when that gate is on, or from the broker/data path the box is configured to use. If a quote source is killed, the honest UI is “no scan,” not a stale leaderboard. Do not write your process as if one vendor is eternal.
Scanners versus the committee
The scanner nominates. The committee votes. Wyckoff and sentiment explain texture. Mixing those jobs is how people turn a gainer screen into a personality. If a name fails the universe rules, it should not appear as an actionable pick even if a model loves the story.
Custom mode (larger names, sector ETFs) is a different prior than a small-cap gainer list. The learning loop’s universe prior exists because those two worlds do not share a base rate. Recalibrate expectations when you change the screen; do not carry last week’s hit rate into a new filter.
A sane daily routine
Set the filter before the open. Do not retune mid-morning to chase whatever is already on the tape. When a name prints, read playbook risk and account constraints first. If the universe keeps producing the same failure cluster (late markup, news spike, thin float), tighten the screen instead of loosening conviction gates.
Save one or two lists, not twelve. Repeated use is the feature: a stable universe is what makes calibration and agent weights mean something. A new screen every day is a new experiment with n = 0.
What scanners will not do
They will not find shorts if the execution path cannot short. They will not waive pattern-day-trader counts. They will not replace a time stop. They will not stay accurate if the quote feed is off. Use them as a fence around attention. The trade, if any, still has to clear the committee, the playbook, and your broker rules.
When to throw the list out
Throw the list out when the same failure tag repeats (late markup, news gap, halt risk) and the filter did not change. That is the screen selecting the error. Tighten price, volume, or exclude the theme. Do not lower the conviction gate to “find trades.” A smaller universe with two clean labels beats a fifty-name circus.
Also throw it out when the quote path is off. A cached leaderboard is worse than an empty one. Rebuild after the feed is healthy. If you switch from a gainer list to megacaps, restart your mental base rate; the learning prior will catch up only after new settled picks exist in that mode.
Worked example: one filter, one week
Monday you save a liquid, mid-price RTH list. By Thursday the same failure tag (late markup) has printed four times. Do not add sectors to “find something.” Tighten the list or sit in cash. Friday you switch to megacaps for a calmer tape. Reset your expected hit rate to the new prior. Last week’s gainer scoreboard does not travel with you.
Write the filter in one line and tape it to the week: price band, minimum volume, RTH only. If a name fails the line, it is not a special situation. It is out. The committee never sees it. That boredom is the feature. Universes that feel exciting are usually last hour’s gainers wearing a new shirt.