> Diagnosing a run that did not behave: refusals, disagreeing results, missing dimensions, rejected files, and what to gather before opening a support request. - Canonical: https://hadalinstruments.com/docs/troubleshooting/ --- Getting help # Troubleshooting Refusals, disagreeing runs, missing dimensions, rejected files — and what to gather before asking for help. - **Page kind** — Diagnostics - **Applies to** — All 21 instruments, all 3 delivery shapes - **Describes** — Intended behaviour — nothing has shipped - **Instrument version** — NOT YET PUBLISHED Not on this page This page diagnoses a run. It does not describe the support model itself, or the refusal behaviour it sends you to check first. [Support and FAQ](https://hadalinstruments.com/docs/support/)[When an instrument refuses](https://hadalinstruments.com/docs/insufficient-n/) This page is written against the failure modes the design anticipates, so that the first version of it exists before the first customer does. It will be rewritten from real reports once there are real reports; until then it is a specification of intended behaviour, not a record of observed faults. The sections below are an ordered sequence of checks, not an index of unrelated symptoms. Work down them in order: each is cheaper to perform than the one after it, and the first one that matches is usually the whole explanation. The last section is what to send if none of them are. ## A dimension returned a refusal instead of a number Expected behaviour, not a fault. The sample fell below the completeness bar declared for that estimator. The panel states how many observations were required and how many were present. Widen the window, check your session filter, and confirm the file is tick-level rather than aggregated. The full treatment is in [when an instrument refuses to answer](https://hadalinstruments.com/docs/insufficient-n/). ## Two runs disagree Compare the sample blocks before comparing anything else. In almost every case the two runs were computed over different windows, different session filters, or different files — and the disagreement is not a disagreement at all, it is two correct answers to two different questions. If the sample blocks match exactly, including the input hash, then the runs should agree. A genuine divergence at identical inputs is a defect worth reporting, and it is the single most valuable report anyone can send. ## A dimension is missing entirely from the output Two possibilities, and the output distinguishes them. If the dimension appears with a refusal, the case is [A dimension returned a refusal instead of a number](#a-dimension-returned-a-refusal-instead-of-a-number). If it does not appear at all, that battery does not measure it — the instrument reports what its method covers and does not manufacture a nearby proxy to fill the gap. Check the instrument’s own page for the list of dimensions it is responsible for; the [instruments index](https://hadalinstruments.com/instruments/) links all twenty-one. ## The file was rejected before the run started The input did not parse, or it parsed but could not be described. A run cannot begin until the instrument can state the sample’s boundary — symbol, venue, window, coverage — because that block travels with the result permanently. Common causes: an export with a truncated header, mixed timezones inside one file, silently concatenated symbols, or a bar export where a tick export was required. ## The result is not what you expected Check the definition before checking the market. The terms in these outputs are used strictly, and several of them — provenance, staleness, coordination, regime, [censoring](https://hadalinstruments.com/glossary/censoring/) — have looser everyday meanings that produce exactly this reaction. The [glossary](https://hadalinstruments.com/glossary/) holds the working definition each instrument actually uses, including what the term is not. If the definition is not the problem, read the limits block on the panel. A result that seems to say something dramatic usually says something narrower, and the “does not establish” line is where the narrowing is written down. See [what a result does not establish](https://hadalinstruments.com/docs/limits/). ## A published number here looks wrong Then re-derive it. Anything published on this site is intended to ship with its inputs, its method version and a content hash, precisely so that a stranger can recompute it and disagree with evidence. If your recomputation disagrees with the published one, send the recomputation. A correction with a receipt attached is published as a correction, in the open, with the original left legible beside it. ## What to gather before asking for help Support is asynchronous and text-first, so a complete first message resolves in one round trip instead of five. Include: - The instrument and the shape it runs in — on-platform, hub, or an engagement. - The full sample block, copied verbatim, including the input hash. - The method version shown on the panel. - What you expected, what you got, and which of the two you are more confident about. - The artifact file itself where the issue concerns a specific run. What not to send: account credentials, broker logins, or anything else that would let someone act on your behalf. Nobody here will ever ask for them, and a message that requests them is not from Hadal. How the rest of the process works is described in [support and FAQ](https://hadalinstruments.com/docs/support/). ## What none of this can fix yet There is no support channel open, because there is nothing to support: no instrument has shipped and no run can be performed. Every check above is a specification of how a fault is meant to be diagnosed, published before the first fault exists so that the shipped behaviour can be held against it. When the channel opens it will be stated on the [support page](https://hadalinstruments.com/docs/support/), with what it covers. Documentation — page 7 of 9 , in reading order