> Pay-per-test measurement on your own artifact: overfit, execution cost, dataset forensics and more — fixed menu, written report, method published first. - Canonical: https://hadalinstruments.com/assays/ --- The Assay Desk # One measurement. Run for you. The instruments are for desks that run their own measurements. An assay is for everyone else: you send the artifact — a backtest, a statement, a data file, a quote log — and what comes back is a written report on the same published method the instrument uses, with every input hashed and every limit stated. No software to install, no method to learn, no subscription. One question, answered once, with the working shown. **Intake is not open yet.** The menu and its anticipated pricing are published now, before the desk takes its first artifact — the same order this site publishes everything: terms first, so they can be read before anyone is asked to rely on them. The submission channel is built and waiting, and the payment rail is built; intake opens when the desk's own terms and the data-handling terms publish alongside it, and the [launch list](#assay-first) hears first. Nothing on this page can be bought today, and the page would rather say so here than let you find out at the bottom. ## The menu — twelve assays, anticipated pricing Every price below is an **anticipated** one-off figure, labelled exactly as the instrument prices are and awaiting the same ratification. Each assay runs the published battery of the instrument it names — the method is on that instrument's page and the family's receipt page *before* you pay, which is the point: you can read precisely what will be computed, and what will not be claimed, in advance. Why a single assay costs more than a month of the instrument it runs: the instrument is software you operate, so a month of it is a month of your own hours. An assay is that same published battery run for you, on your artifact, ending in a written report — and that delivery cost lands once rather than spreading across a subscription. Pricing the one-off below the subscription would quietly make the subscription the worse purchase, which is the shape of a trap rather than a menu. Every figure here is anticipated and awaiting ratification, and nothing on this page is purchasable yet. FIG. 01 MEASURED The menu, as a map: twelve assays running seven published batteries. Every assay is a question pointed at an instrument's existing method — none carries a private methodology of its own, which is why each one can be read in full before it is bought. Mapping the menu establishes which battery answers which question. It does not rank the assays, and it does not price them — every figure on this page remains anticipated. Source [assays.json — the register, exported and hashed](https://hadalinstruments.com/registers/assays.json) SHA-256 Artifact digest, shortened for display: `cf13f3b56c…e2a078`[Verify](https://hadalinstruments.com/registers/assays.json) Date 2026-08-30 The mapping as a table, derived from the same exported register the digest above names. Instrument | Assays carried | P9 | 4 | P1 | 2 | P3 | 2 | P10 | 1 | P18 | 1 | P2 | 1 | P8 | 1 | 01 ### Overfit Assay £299 anticipated · one-off “Is my backtest real, or did I fit it to noise?” - **You send** — Your backtest trade list or platform export (CSV or the platform’s native report), plus an honest count of how many variants you tried before this one — the number the mathematics cannot work without. - **You get** — The probability of backtest overfitting, the deflated Sharpe ratio and the effective number of independent trials, computed on the published P2 method, in a written report with every input hashed and every assumption stated. - **It does not establish** — Whether the strategy will be profitable. A backtest that survives the battery is a backtest that was not obviously fitted — it is not a forecast, and the report says so on its first page. - **Instead of buying it** — The mathematics is published, not proprietary — the cross-validation scheme behind the overfitting probability, the deflation of a performance statistic for trials and track length — and the method this assay runs is on the instrument’s page before you pay. Someone competent with the papers can implement it themselves; that road costs the writing, then the proving that what was written is right, and the discipline of recording trial counts while the search is still going, which is the part no road removes — you supply that number here too. A platform’s own backtest report is a different object: it describes the configuration that won rather than the search that selected it, so it cannot price a multiplicity it never observed. And if the strategy is still being re-fitted, the recurring instrument is the road, because every added parameter moves the trial count and a statistic conditioned on the old one is a snapshot of a search that has carried on. - **The method** — Published on [P2, the instrument this assay runs](https://hadalinstruments.com/instruments/overfit-auditor/) — read it before you buy, not after. 02 ### Execution-Cost Assay £249 anticipated · one-off “What is my execution actually costing me?” - **You send** — Your account statement or order history export covering the period you want measured — fills, timestamps, requotes if the platform records them. - **You get** — Implementation shortfall decomposed against quoted prices: spread paid, slippage by session and order type, and the distribution rather than the average — the tail is where the money went. - **It does not establish** — Whether your broker is acting against you. Cost is measurable from your side; intent is not, and a report that claimed otherwise would be selling you a story. - **Instead of buying it** — The raw material is already yours — the broker sends the statements — and the published method is on the instrument’s page before you buy. Export them into a spreadsheet and you can work out what each fill went off at against the price on your own ticket; that is real work, and it is better than not looking. Where that road stops is the reconciliation: setting each fill against the quote conditions running at the instant it executed needs a synchronised record of the feed it arrived on, and that record, not the arithmetic, is the difficult part. If you want the measurement to keep pace as statements keep arriving, that is the recurring instrument; this reads the window you send, once. - **The method** — Published on [P9, the instrument this assay runs](https://hadalinstruments.com/instruments/execution-cost-auditor/) — read it before you buy, not after. 03 ### Prop-Breach Forensics £199 anticipated · one-off “Why did my prop evaluation actually fail?” - **You send** — The account statement of the failed (or live) evaluation and the firm’s rule set as published to you. - **You get** — A breach anatomy: the halt-distance timeline, which rule interaction actually terminated the account, and where the loss sequence diverged from the position sizing the rules implied — reconstructed, dated, and stated without comfort. - **It does not establish** — Whether the firm’s rules are fair, or whether you would have passed with different luck. It reconstructs what happened; it does not adjudicate the firm. - **Instead of buying it** — The rule set was published to you and the statement is your own record, so the reconstruction is open to anyone willing to do it by hand — distance to a limit is subtraction, and the method is on the instrument’s page before you buy. What makes it laborious is overlapping rules moving on separate clocks, reconstructed backwards across a whole account, where the rule that bound need not be the one attached to the trade that looks like the cause. The firm’s own answer, where you ask for one, names the rule it cited — that is where a reconstruction starts rather than what it produces, and no road here settles whether the rules were fair. If the account is still running, the forward-looking instrument is the road to take instead, since a halt distance is only actionable while it can still be closed. - **The method** — Published on [P3, the instrument this assay runs](https://hadalinstruments.com/instruments/prop-evaluee-risk-guardian/) — read it before you buy, not after. 04 ### Dataset Forensics Assay £399 anticipated · one-off “Can I trust the data file my research stands on?” - **You send** — The historical data file itself (any common format), and where you got it — the provenance question is part of the audit. - **You get** — A defect inventory: gaps and their pattern, weekend and holiday artefacts, look-ahead and point-in-time violations detectable from structure, timezone and rollover seams, and a content-hash manifest so the audited file is pinned to the report. - **It does not establish** — That the data is correct. Structure can prove a file defective; it cannot prove it faithful to the market — only a second independent source can, and the report names that limit. - **Instead of buying it** — The first pass is yours to make, and the method is published before you pay — on the instrument's page and in the research behind it: values held flat across a hole, gaps that land on session boundaries and a timestamp column that misbehaves are all findable in a spreadsheet by anyone willing to look. A slower route costs only a wait — put today's export away, pull the same range again later, and diff the overlapping past, where any change is an edit applied to history after the fact. Faithfulness is what no file audit reaches, this one included: settling it takes an independent recording of the venue the file claims to describe, and one vendor's export set against another's is not that — side by side they establish that the two recordings disagree, not which of them the venue printed. What is left is the part that resists hand-work, run once against one file at one version; a dataset that keeps being extended and re-merged is asking a standing question, which is what the instrument's recurring licence is for. - **The method** — Published on [P10, the instrument this assay runs](https://hadalinstruments.com/instruments/data-forensics/) — read it before you buy, not after. 05 ### Stress Replay Assay £249 anticipated · one-off “What happens to my sizing when the market breaks?” - **You send** — Your position sizing rules and current or intended exposure — the parameters, not the strategy logic. - **You get** — Your ruin and drawdown profile replayed through recorded historical shocks at your actual sizes, with the halt distances stated per scenario — the number you hold between yourself and a margin call, computed rather than felt. - **It does not establish** — The next shock. By construction the events that break books are the ones the sample did not contain, and the report carries that sentence rather than burying it. - **Instead of buying it** — You can do this by hand for a single position and a single event, and the method is published before you pay: pull the recorded path for the event, mark the position through it, and decide where the stop would truly have filled. The arithmetic was never the hard part — the honesty is, because filling at your own stop is the easy assumption and it is the one that makes the answer comfortable — and the effort scales badly once several positions have to move together in the correlation the event itself produced. A platform's own risk display is answering a neighbouring question: a stop level on a screen is a price you have asked for rather than one the market has undertaken to give you. Whatever route you take, the answer describes the book as it stood when it was computed and no other, and doing nothing leaves the discovery for the event itself, where it arrives at speed and resizing is no longer available. - **The method** — Published on [P8, the instrument this assay runs](https://hadalinstruments.com/instruments/stress-harness/) — read it before you buy, not after. 06 ### Feed Fidelity Assay £349 anticipated · one-off “Is the feed my terminal shows me behaving consistently?” - **You send** — A quote log exported from your own platform — exact export instructions for cTrader and MetaTrader are already published in the preparation guide, because the capture method bounds what the assay can claim. - **You get** — The fidelity battery’s client-side subset run on your capture: publication cadence, staleness distribution, spread regimes and their boundaries, measured against your own venue’s baseline and reported with the capture’s limits stated first. - **It does not establish** — A verdict on your broker. One exported log from one terminal measures your feed as you received it — venue-side behaviour, other account tiers and intent are all outside what this data can carry. - **Instead of buying it** — You are exporting the quote log either way, and what to do with it next is published before you pay: an inter-tick interval histogram is within reach of a spreadsheet, and the cadence floor is the dimension that surfaces most readily by hand. The rest resists hand-work — the dimensions that only exist as a distribution rather than an average, and a widening reported as a regime with a start and an end rather than an afternoon you half-remember — though no route escapes the capture underneath it: a log begun after a bad fill describes the window you already noticed, and that limit travels with the file into whatever is computed from it, this reading included. Comparing your chart against another firm's is open to anyone, and it establishes that the feeds differ, which was never in doubt, because neither is a reference for the other. Doing nothing leaves the feed as a background assumption, so a bad fill stays an anecdote rather than a sample you can read. - **The method** — Published on [P1, the instrument this assay runs](https://hadalinstruments.com/instruments/broker-feed-auditor/) — read it before you buy, not after. 07 ### Verdict Recompute £499 anticipated · one-off “Someone claims a result. Does it recompute?” - **You send** — The claimed result, the inputs it was supposedly computed from, and the stated method — whatever exists of each. - **You get** — An independent recomputation attempt with a three-way verdict: reproduces, diverges (with the divergence quantified), or cannot be recomputed from what was provided — the last being the verdict vendors would least like to exist, and the one a claim without its inputs earns. - **It does not establish** — Fraud. A result that does not recompute from the provided inputs is exactly that — no more. The report states what was missing, not why. - **Instead of buying it** — The check is published on this site before purchase, and asking costs nothing: request the inputs the result was computed from and the method that was stated, then run that method yourself and see whether you land where the claim landed. What that cannot produce is a reading that did not come from you, arrived at by a method someone else can follow to the same place. Where the claim rests on a live account, a third-party statistics page that connects to it is the adjacent route, and it establishes a real fact of narrower reach — that data flowed from an account — not that the record is whole or that the claim recomputes from it. Left alone, the claim stays what it was when it was made: something to take on trust, or to decline. - **The method** — Published on [P18, the instrument this assay runs](https://hadalinstruments.com/instruments/reproducible-verdict-kernel/) — read it before you buy, not after. 08 ### Single-Trade Investigation £149 anticipated · one-off “What actually happened to this one trade?” - **You send** — The trade’s full ticket from your platform — entry, exit, timestamps, order type, fill confirmations — the platform’s log excerpt if you have it, and the window you dispute described in your own words. - **You get** — A dated anatomy of the event window: the quoted spread and its regime at your timestamps, your fill against the prevailing quotes, slippage decomposed, and requote or rejection context where your logs carry it — reconstructed against my own independently journaled tick corpus where the instrument sits inside it, and from your artifacts alone where it does not, with the report saying which. Hashed inputs throughout, written to be handed onward — to the firm, a regulator, or your own adviser. - **It does not establish** — Whether anyone owed you a better fill. One trade against one venue window measures what happened, not why — intent is not measurable from the client side — and whether the anatomy supports a complaint is a judgement for the firm or your own adviser, made with the measurement in hand rather than without it. - **Instead of buying it** — You already hold the ticket and the journal, and reading them against your platform’s own chart will show you roughly where the quote sat when you were filled — worth doing, and you pay nobody for it. What that cannot give you is a view of the moment from outside the platform whose fill is in question; licensing a tick history that covers your instrument and aligning it to your own clock across a server timezone is a genuine route to one, and the reconciliation, not the arithmetic, is the work. Putting the trade to the firm is the other available route and it is free; what it examines is the firm’s own record of the window. Doing nothing leaves the fill where it is — something you remember, rather than something measured. - **The method** — Published on [P9, the instrument this assay runs](https://hadalinstruments.com/instruments/execution-cost-auditor/) — read it before you buy, not after. 09 ### Outage-Window Report £199 anticipated · one-off “What did the platform outage actually cost me?” - **You send** — Your platform journal and log files covering the window — they record connection failures, rejected actions and their timestamps — plus your account statement showing the positions held through it, and the outage as you experienced it, in your own words. - **You get** — A timeline of the lockout built from your own logs — first failed action, every rejected attempt, restoration — laid against the reference price path from my journaled corpus where the instrument sits inside it, with the achievable exit stated as a range under stated assumptions, never a single number. Hashed inputs, written to be handed onward. - **It does not establish** — What you would have done. A counterfactual is bounded, not known — the report states the range the reference market offered, not the exit you would have taken, and whether the window’s cost is anyone’s liability is a question for the firm or your adviser, with the measurement in hand. - **Instead of buying it** — The journal is already on your own machine, and it names every failed action and every rejection with a timestamp; assembling those into a timeline is careful reading of a log file, and anyone prepared to do it can. The part that resists is the price path across the window, because the history your terminal downloads afterwards comes from the same venue whose platform was unavailable to you, and a series from anywhere else has to be sourced and aligned to your clock before it says anything. The platform’s own status notice is the remaining free route; it describes the platform’s condition rather than the positions you were holding through it. Left alone, terminal logs rotate on their own schedule, and the window ends up described by what you remember rather than by what your own files recorded. - **The method** — Published on [P1, the instrument this assay runs](https://hadalinstruments.com/instruments/broker-feed-auditor/) — read it before you buy, not after. 10 ### Payout-Denial Recompute £199 anticipated · one-off “Does the arithmetic behind my denied payout actually hold?” - **You send** — Your full account statement or platform export, the firm’s rule text as published to you — consistency rules, drawdown definitions, profit-split terms — and the firm’s stated reason for the denial, verbatim. - **You get** — The cited arithmetic recomputed from your own record: profit-day percentages, drawdown paths on both balance and equity basis, split calculations — and where the denial cites conduct windows, the cited trades tested against the published calendar: opened when, relative to which window, by how much. Every figure derived, every assumption stated. - **It does not establish** — Whether the firm must pay. A recomputation shows whether the stated condition was met under the stated rules; contract interpretation and enforcement belong to the firm, a regulator where one exists, or your own adviser — armed with arithmetic instead of assertion. - **Instead of buying it** — The inputs are already in your hands: your own statement, the firm’s rule text as published to you, and the reason it gave. The arithmetic itself is not hard — a competent person rebuilds it in a spreadsheet — and how to work out which drawdown definition governs an account is published on this site before anything is bought. The firm’s own appeal channel, where it offers one, costs nothing to use and returns, at most, the firm’s reading of its own rules; what it cannot return is the same derivation made from outside the disagreement, stated the same way whichever direction it comes out. Who decides whether you are owed is unchanged either way — the firm, a regulator where one exists, or your own adviser. - **The method** — Published on [P3, the instrument this assay runs](https://hadalinstruments.com/instruments/prop-evaluee-risk-guardian/) — read it before you buy, not after. 11 ### Execution-Chain Attribution £249 anticipated · one-off “Was it the EA, the connection, or the broker?” - **You send** — Your platform’s journal and expert logs for the event — they carry order-send timestamps, requotes, rejections and disconnects — the account statement rows for the trades in question, and the EA’s own log if it writes one. The strategy’s logic stays yours; the assay reads timing and fills, never code. - **You get** — The loss event decomposed along the chain: what the EA requested and when, what the connection did between request and server, and what the fill did against the reference corpus at that instant — three segments, each measured, with the divergence located where the logs actually put it. - **It does not establish** — Fault. The chain attribution shows where the divergence entered, not who owes you for it — a wide fill can be an honest market, a slow hop can be your own VPS, and the report distinguishes what the logs prove from what they merely suggest. - **Instead of buying it** — The logs are already on your own machine, and their sequence is readable: what the EA sent, what came back, what filled. Someone comfortable in a journal file can reconstruct the request side unaided — what went out, and what the connection did with it — and the method is published before purchase rather than after. The limit is the last link: placing a fill against what the market was doing at that instant needs a record of that instant kept independently of the account being examined, and your own terminal’s logs are a record of what reached you rather than something to check it against. Moving the VPS or the account and watching whether it recurs is the other route — it costs what the move costs, answers by elimination rather than by measurement, and speaks to the next event rather than this one. - **The method** — Published on [P9, the instrument this assay runs](https://hadalinstruments.com/instruments/execution-cost-auditor/) — read it before you buy, not after. 12 ### Copy-Trade Divergence £199 anticipated · one-off “Why are my copy-trading results worse than the signal?” - **You send** — Your account statement and journal from the copying account, and the address of the provider’s public record page — the published history your subscription copies. Nothing from inside the provider is needed; the comparison is your fills against their public record. - **You get** — Per-trade divergence decomposed: entry and exit timing lag, spread and slippage differences, and sizing effects — measured for every copied trade in the window, with each component’s contribution stated separately, so the gap has an anatomy instead of a feeling. - **It does not establish** — Whether the provider’s record is genuine — that is the Verdict Recompute’s question — or whether copying could ever match a published record exactly: some divergence is structural, and the report separates that floor from the part your setup adds. - **Instead of buying it** — The records are already available to you: your own account’s statement and journal, and the public page your subscription copies. Lining them up trade by trade is spreadsheet work, the method is published rather than held back, and a competent person can do it unaided. What a spreadsheet returns is one lump difference per trade, and a lump does not tell you which part of it any copier would have paid anyway — the separation is the labour, not the subtraction. Cancelling is the other route and costs nothing: it ends the question instead of answering it, and the share of the gap that was structural is a property of copying rather than of the subscription you left. - **The method** — Published on [P9, the instrument this assay runs](https://hadalinstruments.com/instruments/execution-cost-auditor/) — read it before you buy, not after. FIG. 02 MEASURED Every item on the menu publishes what it will not tell you. Twelve assays, twelve stated non-claims — the full wording is on each card above, at the point of sale, and that wording is the authoritative version. Stating a non-claim establishes the report's boundary. It does not narrow what the report positively measures. Source [assays.json — the register, exported and hashed](https://hadalinstruments.com/registers/assays.json) SHA-256 Artifact digest, shortened for display: `cf13f3b56c…e2a078`[Verify](https://hadalinstruments.com/registers/assays.json) Date 2026-08-30 here would have added one landmark per placement, which on the homepage means a landmark list that is mostly disclaimer. --> Standing policy - **No promise of profit, ever.** Hadal makes no performance claims and carries no implied edge. Past measurements describe instrument behaviour — never future returns. - **You own risk management.** Hadal cannot control it and does not insure it. Good tools do not fix bad discipline — and this site says so. - **Analytical tools, for discretionary use.** Nothing here is investment advice or a recommendation to trade. Every decision, and every outcome, is yours. The same statement stands in the footer of every page.[Terms](https://hadalinstruments.com/terms/) [Privacy](https://hadalinstruments.com/privacy/) ## How an assay runs - You send the artifact in the assay's stated format — [the preparation guide](https://hadalinstruments.com/docs/assay-preparation/) gives exact export instructions per platform, published before intake so you can check what you hold against what an assay needs. Off-format submissions come back with a note, not a surcharge. - The artifact is content-hashed on receipt — the report is pinned to exactly what you sent, so neither side can later dispute what was measured. - I run the instrument's published battery on the engine. An assay is run by a person on real machinery; it is not an app, and this page will never pretend otherwise. - You receive the written report: the figures, the intervals, the assumptions, the limits — and the hash manifest that lets a competent third party check what was done. - Questions about the report go through a [ticket](https://hadalinstruments.com/support/), like everything else here. Turnaround is targeted at five business days from a valid submission — a target, not a guarantee, for the reasons the [support page](https://hadalinstruments.com/support/) states: a promised window nobody audits is marketing, and this desk would rather publish its measured turnaround distribution once one exists. Data handling terms — retention, deletion, and the submission channel itself — publish in the same act that opens intake, alongside the [privacy policy](https://hadalinstruments.com/privacy/) they extend. An assay desk that took your trading data before publishing what happens to it would have the sequence backwards. ## Off the menu The menu is the product. Anything off it — a custom metric, a modified method, a bespoke report format — is not a dearer assay; it is a configured engagement with a stated floor, because custom work is priced to be rare. The menu itself was derived from the instrument catalogue and from published research into the dispute-investigation gap, not from a demand history I do not have. The floors and the two-slot cap on configured engagements are published on the [licensing page](https://hadalinstruments.com/licensing/), in the doors register, where they have been since before this desk existed. ## Hear when intake opens The launch list is the one channel that works today, and assay intake will be announced on it first — with final prices, the submission channel, and the data-handling terms in the same message. here would have added one landmark per placement, which on the homepage means a landmark list that is mostly disclaimer. --> Standing policy - **No promise of profit, ever.** Hadal makes no performance claims and carries no implied edge. Past measurements describe instrument behaviour — never future returns. - **You own risk management.** Hadal cannot control it and does not insure it. Good tools do not fix bad discipline — and this site says so. - **Analytical tools, for discretionary use.** Nothing here is investment advice or a recommendation to trade. Every decision, and every outcome, is yours. The same statement stands in the footer of every page.[Terms](https://hadalinstruments.com/terms/) [Privacy](https://hadalinstruments.com/privacy/)