A first run is deliberately unexciting. You give an instrument a bounded sample, it states what that sample can and cannot support, and it returns a description with the sample attached. This page sets out that sequence so the first real run holds no surprises. It describes intended behaviour; no instrument has shipped and no measurement from one has been published.
Step one: declare the sample
Before anything is computed, the instrument records what it was given: instrument or symbol, venue, date range, session hours, row count, and the hash of the file it read. This block travels with every result forever. It exists so that a number you show someone in six months can be traced back to the exact bytes that produced it.
Nothing about this step is optional or hidden. If two runs disagree, the sample declaration is the first place to look, and it is usually the whole explanation.
Step two: the completeness bar
Each battery has a stated minimum before it will estimate anything — a floor on observations, on covered sessions, or on events of the relevant kind. The floor is set in the method, ahead of the run, not chosen after seeing the data. Where the sample clears the bar, the estimate publishes. Where it does not, the instrument prints the refusal and the shortfall instead of a number.
That is the behaviour most new users find unfamiliar, so it has its own page: when an instrument refuses to answer.
Step three: the estimate, with its uncertainty
What comes back is a distribution far more often than a single figure. A spread regime has a start, an end and a context. Quote staleness is an age distribution across ticks, not an average with the interesting part averaged away. An overfitting assessment reports a probability, not a verdict.
Every panel carries three things: the estimate, the sample it rests on, and an explicit statement of what it does not establish. The third is not a disclaimer bolted on at the end — it is written into the method beside the estimator itself, and it is the part worth reading twice. See what a result does not establish.
Step four: the artifact
A run is intended to produce a file you keep: the inputs, the method version, the estimates, and a content hash over the whole thing. Recomputation is the reason. A published Hadal number is supposed to be one a stranger can re-derive from the same inputs and the same stated method, and that is only possible if the run leaves a durable record. The methodology page describes the standard the published measurements are held to.
What you provide, per suite
Whatever the suite, what you provide is recorded before anything is computed.
- Feed & Execution — a tick capture from your own feed, or your own fill records. What comes back describes publication cadence, two-sided coordination, per-side staleness, spread regimes, and freeze behaviour, each with its sample.
- Backtest Honesty — a strategy, its parameter search, and the data it was fitted on. What comes back is an assessment of how much of the result could be selection rather than signal, plus the provenance defects found in the data itself.
- Desk Discipline — account and rule parameters. What comes back describes distance to breach and the behaviour of your own risk envelope, never a recommendation about position size.
- The Honesty Stack, Premium Add-on, The Terminal and Integrity Services each state their own inputs on their suite pages. Read the “does not establish” block on the instrument page before the first run, not after it.
What a first run will never ask you for
No step in the sequence above requires a broker login, an account password, an API key with trading permissions, or any other credential that would let something act on your behalf. On-platform instruments read what the platform already shows them. Hub instruments read the file you hand over and nothing else. If a message or a screen ever asks you for trading access in order to run a measurement, it is not from Hadal.
When this becomes real
No instrument is available and no measurement pipeline is live, so no first run can be performed today. This page exists now, before customer one, because a self-service centre written after launch is written under pressure — and because publishing the intended sequence in advance is the only way it can be held against me later.
Documentation — page 2 of 9, in reading order