Every panel a Hadal instrument produces closes with a statement of what it cannot support. Those statements are written into the method beside the estimator, not appended as legal cover, and they are the part of the output most worth reading carefully. This page collects the limits that apply across the whole catalogue. It describes intended behaviour: three instrument pages now cite a reading from my own research book, and no instrument has yet published a customer result.
Description is not intent
An instrument can measure that a feed has a hard cadence floor, that its two sides move in lockstep, or that spread widening clusters around particular events. None of that establishes that anyone decided it should. Sampling artefacts, aggregation layers, routing, network paths and ordinary platform architecture all produce measurable structure without anyone choosing it.
So the instruments describe behaviour and stop there. They do not certify a venue as honest or dishonest, they are not a regulatory finding, and they hold no view on where you should keep your account. Converting a measurement into an accusation is a step the instrument refuses to take, and one you should be slow to take yourself.
Measurement is not prediction
A distribution measured over a stated window describes that window. It does not forecast the next one. Feeds change, venues change infrastructure, and market conditions rotate. A finding is a description with a boundary attached, and the boundary is in the sample block for exactly this reason.
The same applies to backtest work in the other direction. A high assessed selection risk does not establish that a strategy will lose money, and a low one does not establish that it will make any. The assessment describes the search that produced the result, not the future the result is being used to argue for.
The path is in the measurement
Feed instruments measure the stream as it arrives at you. Your connection, your location, your platform and its own buffering sit inside that measurement, and there is no clean way to attribute the whole path to its origin. Results state this beside the finding rather than quietly assigning the difference to the venue.
An absence of evidence is not evidence of absence
A dimension that publishes nothing found says the effect was not present at a level the estimator can distinguish from noise, over that sample, at that resolution. It does not say the effect does not exist, and it does not clear anything. A wider window, a finer capture or a different session might resolve something this one could not, and the honest reading of a null is that the question stayed open at the power available.
This cuts both ways, which is why it belongs here rather than in a footnote. A null is not a certificate, and quoting one as though it were is the mirror image of quoting an under-powered estimate as though it were a finding.
No performance claims, ever
Nothing on this site or in any instrument output states, implies, or reports a return, a win rate, an edge, or a profit. This is a standing policy, not a pre-launch caution — it will still be true when the catalogue is selling. Where you would expect a performance figure, there is nothing, and the absence is deliberate.
The corollary matters as much: an instrument is not a strategy, and a good instrument does not fix bad discipline. The measurements can tell you that an execution environment is not what you assumed. They cannot make you act on it.
Not advice, not a service of trust
Hadal instruments are analytical tools for discretionary use. They provide no investment advice, carry no implied edge, and take no responsibility for what you do with a result. Risk management belongs to you; it cannot be delegated to a measurement and is not insured by one. The full statement of this posture is in the terms and is repeated at every point where a decision is made rather than buried at the foot of a page.
What the published corpus does not establish either
The same limits govern anything published here. A broker portrait in the Observatory is a profile, never a ranking, and never a composite score — because a composite hides the dimension that actually mattered to you behind an average that flattered a different reader. Portraits are presented alphabetically for that reason.
Where a measurement has not been made, the site prints NOT YET PUBLISHED rather than a placeholder figure. That is the honest state of the corpus today, across every panel, and it will change one measurement at a time.
Related
The three shapes of an empty result are described in reading a result. The refusal behaviour that stops an under-powered estimate from publishing at all is in when an instrument refuses to answer.
Documentation — page 6 of 9, in reading order