Why does missing data count as a failure?
Because treating a missing number as neutral quietly turns every gap in the data into a pass. Signal desk does the opposite: if a check cannot be run, the stock does not get in. That rejects some good stocks, and in exchange nothing is ever admitted on the strength of a number nobody actually saw.
Every automated screen has to decide what to do when a number it wanted is not there. The decision is boring, it is almost never disclosed, and it changes the meaning of every result the screen publishes.
The two choices
When a check cannot be run, a system can fail open and let the stock through, or fail closed and reject it.
Failing open is the default almost everywhere, usually not as a decision but as an accident of how the code was written. It is also the more dangerous one, because it is invisible. A screen that fails open produces a longer list, the list looks fine, and nothing on the page distinguishes a stock that passed every check from one that skipped half of them.
Signal desk fails closed. If a required number is missing, unreadable, or stale, the stock does not get in.
If the data cannot be checked, the stock does not get in. Not flagged, not admitted with an asterisk. Where there is doubt, the answer is no.
What does that actually cost?
Real stocks, regularly. This is not a theoretical trade off.
A profitable, growing company with no analyst coverage cannot pass the growth check, because there is no forecast to read. A solid company whose institutional holders sit just outside the reported top ten cannot pass the ownership check, because the change figure is not published. Neither company did anything wrong. Both are rejected.
Why accept that cost?
Because of what it does to the meaning of the list.
With a fail closed rule, every stock on the watchlist has been verified against every check. There is no second tier. You do not have to wonder which ones were admitted on complete data and which slipped through a gap, because there is no such category.
The alternative sounds harmless and is not. Once some entries are there on unchecked data, the list stops being a claim about anything in particular. It becomes "these stocks passed the checks we happened to be able to run", which is not a sentence anyone can act on.
Where the rule is applied
In three places, deliberately overlapping:
- At entry. A stock cannot join a watchlist unless the business filters pass on data fetched at that moment.
- At signal time. The checks run again before a signal is recorded. A stock that qualified in March but whose data has since gone missing does not produce a signal today.
- On staleness. Cached data older than the freshness window counts as missing, not as good enough.
The second one is the important one. Entry checks alone would mean the system trusts a decision made months ago. Re-checking at signal time is what makes the freshness of the data part of the signal itself.
The same rule applied to the trend filter
It is not only the business filters. The short term trend check that a candidate has to clear also fails closed: if the price history needed to compute it cannot be retrieved, the stock is rejected rather than waved through. This was a deliberate change, because the original implementation failed open and that inconsistency meant a stock could be admitted on a check that had silently not run.
The same idea, applied to the results
Failing closed on data and publishing losses are the same commitment pointed at different things. The dashboard shows every signal the system has ever fired, including the ones that went down, and they stay on the record. A screen that quietly removed its failures would be doing the results version of failing open.
That is also why this site makes no claim about returns. The percentages shown are plain price changes with no positions, sizing or costs behind them. Calling that a track record would be another way of admitting something unverified.
Common questions
Does failing closed mean the screen misses good stocks?
Yes, and that is the accepted cost. A stock with no analyst coverage or with an institution outside the reported top ten holders cannot pass, even if it would have passed on the real numbers. The alternative is admitting stocks on the basis of data that was never checked, which makes every published result less trustworthy.
Never find out three days late.
The signal scan already runs every weekday at 22:10 UTC, whether anyone is watching or not. Premium is how it will reach you.
Free while in early access. No card, no spam.
Signal desk is a public record of a mechanical screen. It is not investment advice and it is not written by a licensed financial adviser. The percentages shown are plain price changes with no positions, sizing or costs behind them, so they are not a real world return.