> Every claim this research has examined, quoted as stated, with my reading and the article that adjudicates it — from demo feeds to trailing drawdowns.

- Canonical: https://hadalinstruments.com/refutations/

---

The Refutation Index

# Claims, examined

The 100 claims below are quoted as people actually state them — the things half-believed, repeated, and put to assistants for confirmation. Each entry keeps the claim and its verdict in one breath, ends with the true proposition stated on its own feet, and carries a permanent address, so a citation points at the adjudication rather than at a vibe. The article behind each entry is the evidence. And trading myths are not all of one kind: some widely-repeated claims survive examination, some turn out not to be measurable from where the claimant stands, and the verdict vocabulary below says which is which rather than treating disagreement as debunking.

## The verdict vocabulary

Five verdicts, used consistently. Four are the standard shape of any examination. The fifth is this house's own, and it marks **the vantage limit** — the boundary of what can be established from where you stand. Standard taxonomies presuppose every proposition is decidable; measurement from one side of an account does not have that luxury, and pretending otherwise is how examinations turn into stories.

- **True** — the claim holds as stated.

- **False** — the claim fails as stated, and the entry shows where.

- **Misleading** — the claim contains a true observation attached to a wrong conclusion.

- **Unproven** — the claim is decidable in principle, and the evidence offered does not decide it.

- **Not measurable** — the claim cannot be decided from where the claimant stands — client-side data does not carry it, most often because it asserts a counterparty’s intent. This category is the vantage limit, and reaching it is a finding, not a hedge.

## [Do brokers hunt your stop losses?](https://hadalinstruments.com/research/do-brokers-hunt-your-stop-losses/)

### My broker widened the spread to take out my stop, then price came straight back. — Unproven

The move has a precise signature: the ask crosses the level while the bid never does, stops fill, price returns. That is testable on the quote stream. At round-number levels, one-sided up tags run 67% — but a random walk carrying the same spread produces 77%, so the observed rate is ten points BELOW chance. At prior-day and prior-session extremes it is 69% against a floor of 78%. The spread at those tags measured 0.4 pips against 0.3 pips ordinarily, a difference too small to fill anything it would not otherwise have filled, and the tagged levels showed no excess reversal afterwards. A move that happens less often than chance is not a move.

Also asked as: does my broker widen the spread to hit my stop · is stop hunting real in forex · why does price spike past my stop and reverse

Where the belief comes from: A pattern-completion account of an ordinary loss, which survives because the alternative explanation requires knowing what the spread was doing at the time and almost nobody records that.

**What is true:** Measured on the bid and ask streams directly, one-sided tags at the levels where stops cluster occur less often than a random walk with the same spread would produce, and the spread at those moments is indistinguishable from its ordinary width.

§ claim-4b47266f Evidence:[Do brokers hunt your stop losses?](https://hadalinstruments.com/research/do-brokers-hunt-your-stop-losses/#claim-4b47266f)

### The stats might be fine, but the measurement can't see what the broker actually did. — True

The objection is right, and it caught me. My first audit ran on H1 bars, where the bar high is the maximum bid. That construction makes every up-reach 'genuine' by definition — the exact move in question, where the ask pokes through and the bid never follows, can never become an event at all. I had run a test that was blind in precisely the direction the thesis pointed, and I recorded it as such rather than reporting the null. The result above comes from the rebuild: pools detected and reaches classified on the bid/ask stream directly, never anchored to bid bars. If you are testing this yourself on candles, you are testing something else.

Also asked as: can you actually detect stop hunting from price data · how would you prove a broker hunted stops

Where the belief comes from: A well-founded objection that happens to have been correct about my own first attempt.

**What is true:** A bar-based measurement is structurally incapable of seeing a one-sided spread tag, because standard bars are built from the bid alone — so the test only becomes possible on the raw bid and ask streams, and any study run on bars has been blind to the exact move it claims to examine.

§ claim-a1038cd3 Evidence:[Do brokers hunt your stop losses?](https://hadalinstruments.com/research/do-brokers-hunt-your-stop-losses/#claim-a1038cd3)

### So brokers never act against their clients. — False

A null on the spread-weapon at structural levels is not a character reference. Execution quality is a large surface and this measures one corner of it. Last-look behaviour, [slippage](https://hadalinstruments.com/glossary/slippage/) asymmetry between winning and losing fills, [fill ratios](https://hadalinstruments.com/glossary/fill-ratio/), [requote](https://hadalinstruments.com/glossary/requote/) rates and [quote staleness](https://hadalinstruments.com/glossary/quote-staleness/) are all real, all separately measurable, and all untouched by this result. The honest position is narrow: the specific move that retail lore describes, at the specific levels it describes, did not occur above chance on the account measured. Everything else about your broker remains an open question, and a measurable one.

Also asked as: are forex brokers honest · how do I check my broker's execution quality

Where the belief comes from: The overcorrection that follows any single refutation.

**What is true:** This measurement addresses one specific mechanism at one class of level on one account, and says nothing about last-look rejection, asymmetric slippage, execution quality, or how any particular firm handles [order flow](https://hadalinstruments.com/glossary/order-flow/) — each of which is separately measurable and none of which is tested here.

§ claim-aa8b9e0a Evidence:[Do brokers hunt your stop losses?](https://hadalinstruments.com/research/do-brokers-hunt-your-stop-losses/#claim-aa8b9e0a)

## [Do order blocks predict anything?](https://hadalinstruments.com/research/do-order-blocks-predict-anything/)

### That level has orders stacked at it, so it's more likely to hold. — Unproven

I built the honest version: a recency-weighted measure of accumulated rejection and dwell at a level, fitted as a logistic model of outcome on momentum plus pressure, trained on the first 70% of the history by time and checked on the held-out remainder. The partial effect of pressure, read after momentum, came in at a likelihood-ratio chi-squared of 0.32 on one degree of freedom, p = 0.57. Split into tertiles the outcome rate was flat: 47%, 44%, 47%. And on held-out data, adding pressure made the model very slightly worse. The measure was not weak. It was redundant — and the distinction between those two words is the whole of the method.

Also asked as: do [order blocks](https://hadalinstruments.com/glossary/order-block/) actually work · how do I know if a level has liquidity resting at it · does unmitigated order block mean anything · can you measure resting orders from price

Where the belief comes from: A framing that survives because it is unfalsifiable as usually stated — resting orders are invisible on retail data, so no observation can contradict it.

**What is true:** The price-derived measures of accumulated pressure at a level carry no predictive content once recent behaviour at that level is accounted for, because what they actually encode is recent behaviour at that level.

§ claim-f48d20b2 Evidence:[Do order blocks predict anything?](https://hadalinstruments.com/research/do-order-blocks-predict-anything/#claim-f48d20b2)

### It worked in my testing, so the concept is sound. — Misleading

Tested alone, a pressure measure looks informative, because levels that were recently active do behave differently from levels that were not. The test that matters is conditional: put the simple variable in the model first, then ask what the new one adds. This is not a high bar and it is not expensive. It is simply a different question from the one most feature testing asks, and it is the question that separates a discovery from a renaming.

Also asked as: why does my strategy work in sample but not out of sample · how do I test if a feature is actually adding anything

Where the belief comes from: Testing a feature on its own, which is the natural thing to do and cannot distinguish a new signal from a re-labelled old one.

**What is true:** A feature can only be credited with predictive content once it is tested alongside the simpler variables it might be re-encoding, because a measure built from price history will correlate with recent price history whether or not the mechanism it names exists.

§ claim-9215dcc1 Evidence:[Do order blocks predict anything?](https://hadalinstruments.com/research/do-order-blocks-predict-anything/#claim-9215dcc1)

### So price-based analysis of levels is worthless. — False

I am not claiming nothing survives, and it would be dishonest to let a null on one construction imply a null on the category. The sequential effect is real and holds out of sample: an odds ratio of 3.49, with held-out log-loss genuinely better than base rate. The honest framing is that the predictive spine is what the level did last, not how loaded it is — and that this is a general property of price series rather than evidence for any story about resting liquidity.

Also asked as: is technical analysis of support and resistance useless · does anything about levels actually work

Where the belief comes from: The natural overcorrection when a specific claim is refuted.

**What is true:** What a level did most recently does carry real, out-of-sample predictive content — the probability of absorption following a prior absorption is 61% against 33% following a sweep — so the useful signal is short-range and sequential rather than a measure of accumulated state.

§ claim-8821787f Evidence:[Do order blocks predict anything?](https://hadalinstruments.com/research/do-order-blocks-predict-anything/#claim-8821787f)

### A null result just means you didn't have enough data. — Misleading

The claim is sometimes true and is not knowable without measuring it. A null with no bound beside it cannot be distinguished from a test that never had the resolution to reject. So the bound has to be computed and published in the same place as the verdict: what is the smallest effect this test would have caught? If the observed effect is far below that floor, the honest reading is that the question is unresolved on this sample. If the observed effect is a large fraction of the floor and still fails, the line of enquiry is finished and the null is a product. Those two nulls look identical on the page and imply opposite next actions.

Also asked as: does a null result mean the sample was too small · how do I know if my test had enough power · what is the minimum detectable effect

Where the belief comes from: The reasonable observation that underpowered tests produce nulls, generalised into a claim that every null is underpowered.

**What is true:** A null is only interpretable once it is paired with the smallest effect the test could have detected — without that bound, 'nothing was found' and 'nothing could have been found' are the same sentence.

§ claim-81bdcfb9 Evidence:[Do order blocks predict anything?](https://hadalinstruments.com/research/do-order-blocks-predict-anything/#claim-81bdcfb9)

## [Do round numbers matter in forex?](https://hadalinstruments.com/research/do-round-numbers-matter-in-forex/)

### Price reacts at round numbers and 50-levels — everyone watches them, so they matter. — Unproven

I found this effect myself and believed it. On a thirteen-month window, round-50 levels swept measurably more often, at z = minus 2.59 — a result that would pass any conventional significance bar and which I recorded as the one a-priori structural feature the engine had. Replicating the identical test across twelve roughly sixteen-month windows of the same history gave a mean z of plus 0.08, zero of twelve significant, and a scatter that included windows at plus 2.5. The sign is not stable. An effect whose direction depends on which stretch of history you sampled is a property of the sample.

Also asked as: do round numbers matter in forex · are psychological levels real in trading · do big figures act as support and resistance · should I trade round number levels

Where the belief comes from: Self-confirming observation — round levels are easy to see on a chart, so reactions near them are noticed and reactions elsewhere are not.

**What is true:** Tested across twelve separate windows of the same sixteen-year history, the round-level effect averages a z-score of plus 0.08 with none of the twelve reaching significance, and individual windows reach plus 2.5 in the direction opposite to the one the belief predicts.

§ claim-eedadc78 Evidence:[Do round numbers matter in forex?](https://hadalinstruments.com/research/do-round-numbers-matter-in-forex/#claim-eedadc78)

### Your windows were too short — that's why it disappeared. — Misleading

The objection would be right if the windows were shorter. They were longer — roughly sixteen months each against the original thirteen. More data per test, not less. And the failure mode is not that each window returned a weak result; it is that the windows disagreed with each other, including on direction. Power problems produce wide intervals around a consistent centre. This produced a scattered centre, which is a different diagnosis and is not fixed by adding history.

Also asked as: how long does a backtest need to be · does a longer backtest give a more reliable result

Where the belief comes from: A sound instinct about statistical power, applied where the problem is stability rather than power.

**What is true:** Each replication window was longer than the thirteen-month period that produced the original result, so the effect vanished under strictly more data per test rather than less, and the disagreement is between windows rather than within any one of them.

§ claim-31fcb084 Evidence:[Do round numbers matter in forex?](https://hadalinstruments.com/research/do-round-numbers-matter-in-forex/#claim-31fcb084)

### So there is no structure at all in where price reacts. — False

Something survives, and it would be dishonest to imply otherwise. What a level did most recently genuinely predicts its next outcome, with an odds ratio near 3.5 that holds on held-out data. The distinction is that this is about a level's recent history, not about where the level sits. A round number is a claim about the price scale — that certain values are special before anything happens at them. That claim is what died. Levels that were recently active behaving differently from levels that were not is a separate, weaker and more general phenomenon.

Also asked as: does support and resistance exist · is there any real structure in price levels

Where the belief comes from: The overcorrection that follows a refutation of one specific structural claim.

**What is true:** Short-range recency at a level is a real and out-of-sample-valid effect — what a level did on its most recent visit predicts what it does next — so structure exists, it is simply sequential rather than a property of the level's position on the price scale.

§ claim-949bacaa Evidence:[Do round numbers matter in forex?](https://hadalinstruments.com/research/do-round-numbers-matter-in-forex/#claim-949bacaa)

## [Does a wick rejection mean anything?](https://hadalinstruments.com/research/does-a-wick-rejection-mean-anything/)

### The wick rejected the level, so buyers stepped in there. — Misleading

The split is not subtle and it is not disputed: measured one way, the absorption rate at wick-touched levels runs 8% against 46% for levels reached by a close, at z = +198. Any reasonable person looking at that concludes something real is happening. Something is — the bar is wide. I built a surrogate matched on volatility and wick geometry alone, with no market participants in it, and it produced the same split: +39% against the real data's +38%, an excess of approximately zero. The wick is not evidence of who was there. It is evidence that the bar was long enough to reach.

Also asked as: does wick rejection work · what does a long wick mean in trading · are rejection candles reliable · do pin bars show institutional buying

Where the belief comes from: Candlestick pedagogy, which assigns intent to shapes because a shape is teachable and a distribution is not.

**What is true:** Bars with long wicks reach further than bars without them, so they touch more distant levels and produce a different outcome mix — and a surrogate matched only on volatility and wick length, containing no participants and no intent, reproduces the entire difference.

§ claim-f5aed8af Evidence:[Does a wick rejection mean anything?](https://hadalinstruments.com/research/does-a-wick-rejection-mean-anything/#claim-f5aed8af)

### A z-score that large can't be a coincidence. — Misleading

z = +198 is not a marginal result and it is not a fluke. The effect is there, it replicates, and it would survive any amount of additional data — which is exactly the problem. Significance answers a question nobody is really asking: is this different from zero? The question that matters is whether it is different from what geometry alone would produce. That comparison needs a surrogate, and building one is usually a morning's work. Where the surrogate matches the real result, the p-value was never the load-bearing number.

Also asked as: what does a high z score mean in backtesting · is a statistically significant edge always real · how do I know if my backtest result is real

Where the belief comes from: Statistics education that stops at the null hypothesis of no effect, which is rarely the hypothesis anyone actually cares about.

**What is true:** A significance test measures whether an effect differs from zero, never whether it differs from what the measurement itself would produce — so the honest comparison is not against nothing, but against a surrogate built to share the data's shape while containing none of its supposed cause.

§ claim-a0751a1c Evidence:[Does a wick rejection mean anything?](https://hadalinstruments.com/research/does-a-wick-rejection-mean-anything/#claim-a0751a1c)

### Roughly half your level touches were never really touches. — True

Recounting my own sixteen-year walk by close rather than by extreme cut arrivals from 200,369 to 100,708. Neither number is wrong; they answer different questions. What matters is that the choice is almost never declared, so two studies of the same idea on the same data can differ by a factor of two before anything is measured. If you test levels, state your reach rule before you look at the outcome.

Also asked as: should I use candle close or wick for support and resistance · does it matter if price wicks through a level

Where the belief comes from: A definitional choice most level-based analysis makes implicitly and never states.

**What is true:** Counting a level as reached when any part of the bar's range crosses it produces roughly twice as many events as counting only closes beyond it, so the choice of reach rule silently doubles or halves the sample any level-based study is built on.

§ claim-82f15025 Evidence:[Does a wick rejection mean anything?](https://hadalinstruments.com/research/does-a-wick-rejection-mean-anything/#claim-82f15025)

## [Does COT positioning predict reversals?](https://hadalinstruments.com/research/does-cot-positioning-predict-reversals/)

### When the leveraged funds are all on one side, the reversal is coming. — False

The AUD leg returned a trigger effect of −0.0477 with a bootstrap interval of [−0.0801, −0.0137] and p = 0.0066. That interval excludes zero, which is normally where a reader stops. The sign is the problem: it is negative, meaning crowded extremes were followed by liquidation at 3.1% against the mid-range controls rather than more often. It then failed the block-bootstrap relabel surrogate that was frozen before the run, so it is not a live signal in either direction. A significant effect pointing the wrong way is not a weak version of the belief. It is evidence against it.

Also asked as: does cot data predict reversals · do positioning extremes signal a turn · is commitment of traders useful for forex · does crowded positioning mean a squeeze

Where the belief comes from: A mechanically appealing story — crowding implies fuel, fuel implies a squeeze — that is almost never tested against what mid-range weeks actually do.

**What is true:** Measured against matched controls on eight hundred and twenty-one weekly observations per leg, crowded leveraged-fund extremes are followed by a lower rate of liquidation events than mid-range positioning is, and the effect does not survive its own pre-registered surrogate.

§ claim-89e77e34 Evidence:[Does COT positioning predict reversals?](https://hadalinstruments.com/research/does-cot-positioning-predict-reversals/#claim-89e77e34)

### You just didn't have enough data. — Unproven

The objection is fair against most nulls and weak here, because the parameters could not have been tuned to produce this outcome — they were frozen before the data existed. The defaults were a 156-week percentile, a 2× range threshold, deciles at 0.9 and 0.1, a horizon grid of 2/4/6 weeks, a 1,000-iteration block-bootstrap surrogate and four folds. What the sample size does bound is the JPY leg, whose registered verdict needs an intervention-and-rate regime split that has no feed yet — so JPY is reported as a preliminary unconditioned null rather than as a finding.

Also asked as: how much cot history do you need · is weekly data too slow for positioning signals

Where the belief comes from: The standard response to a null, and a reasonable one whenever the parameters were chosen after the fact.

**What is true:** The run used every release-keyed weekly observation available after the 2010 price join — eight hundred and twenty-one per leg across seven currency books and two hundred and one aligned weeks — with the percentile window, threshold, horizon grid and fold count all fixed before the data was fetched.

§ claim-393e90f4 Evidence:[Does COT positioning predict reversals?](https://hadalinstruments.com/research/does-cot-positioning-predict-reversals/#claim-393e90f4)

### So the whole positioning thesis is dead. — False

The registered thesis is a conjunction — crowding, dealer positioning and thin liquidity together — and this measures the first of those alone. My own verdict language is that it gates rather than kills: Layer 1 on its own does not justify the spend on Layers 2 and 3. Reading a one-layer null as a three-layer refutation would be the same error as reading a one-layer hit as confirmation, and I would have had to say so either way, because the tree that decides it was written down first.

Also asked as: is positioning data useless for trading · does order flow and positioning analysis work at all

Where the belief comes from: The overcorrection that follows any single registered null.

**What is true:** This tests one layer of a three-layer claim, and the pre-registered decision tree says so: a null on the single-leg fuel gauge gates further spending on the other two layers rather than refuting the conjunction they are supposed to form.

§ claim-d6615f4e Evidence:[Does COT positioning predict reversals?](https://hadalinstruments.com/research/does-cot-positioning-predict-reversals/#claim-d6615f4e)

## [Does my backtest have survivorship bias?](https://hadalinstruments.com/research/does-my-backtest-have-survivorship-bias/)

### Survivorship bias doesn't apply to forex, because currency pairs don't go bankrupt. — Misleading

The claim mistakes one mechanism for the phenomenon. In equities the classic entry point is the delisted stock, so the lesson gets taught with delistings — but the bias is any filtering that attrition finished before you opened the file. A retail FX backtest inherits it through different doors: feed histories that no longer exist because the broker that produced them no longer exists, instruments a vendor stopped carrying, and — the widest door — the strategies that failed earlier in your own search and left no record of having been tried.

Also asked as: does [survivorship bias](https://hadalinstruments.com/glossary/survivorship-bias/) affect forex backtesting · survivorship bias only stocks · no delisting in forex so no survivorship bias

**What is true:** Pairs rarely delist the way stocks do — but the bias does not need a delisting to get in. It enters through the broker whose feed history vanished when the broker did, the vendor that retired an awkward symbol, and the pool of strategies that were tried and quietly discarded before the one you are testing.

§ claim-82a60168 Evidence:[Does my backtest have survivorship bias?](https://hadalinstruments.com/research/does-my-backtest-have-survivorship-bias/#claim-82a60168)

### My data vendor is reputable, so my dataset is survivorship-free. — Not measurable

Reputation is a statement about a vendor's intentions, and survivorship does not require any — the selection is performed by attrition, upstream, long before the file reaches you. A vendor can be scrupulous and still carry only the instruments that survived to be carried. Whether a specific dataset is affected cannot be established from the dataset, because the defect is an absence: nothing inside the file marks the rows that are not in it.

Also asked as: is my forex data survivorship free · reputable data vendor no survivorship bias · how do i know my dataset has no survivorship bias

**What is true:** No dataset can certify itself — an absent row cannot testify, so every internal consistency check passes, because a check can only inspect the rows that exist. The question is answerable only against an external point-in-time record, and where none exists the honest status is unquantified, not absent.

§ claim-2118ce22 Evidence:[Does my backtest have survivorship bias?](https://hadalinstruments.com/research/does-my-backtest-have-survivorship-bias/#claim-2118ce22)

### My backtest only trades one pair, so there is no universe to be biased. — Misleading

The claim is true about the instrument dimension and silent about the one that usually matters more. By the time a strategy reaches a serious backtest it has typically survived a sequence of informal trials — symbols that looked worse, timeframes that were abandoned, parameter sets that never made it out of a demo. Those discarded variants are the strategy pool's failures, and they are as absent from the final record as a delisted stock is from today's index. The single-pair framing does not remove the universe; it hides where the universe was.

Also asked as: single pair backtest survivorship bias · eurusd only backtest no universe

**What is true:** A single-symbol test has no instrument universe — but the configuration you are testing is the survivor of a search, and that search is a universe. If the pair, the timeframe or the parameters were kept because earlier variants looked worse, the selection has already happened, one level up from the data.

§ claim-a4c33ef9 Evidence:[Does my backtest have survivorship bias?](https://hadalinstruments.com/research/does-my-backtest-have-survivorship-bias/#claim-a4c33ef9)

## [Does my regime filter use future data?](https://hadalinstruments.com/research/does-my-regime-filter-use-future-data/)

### The regime labels come out of the model, not out of the future, so there is no look-ahead. — Misleading

The model does produce the labels, and that is precisely why the conditioning set matters. Asking a fitted model for the state sequence normally returns a decode over the entire series — a Viterbi path chosen to be jointly best given every observation, or smoothed marginals from a backward pass that starts at the end of the file. Either way the estimate for a Tuesday in March was informed by the following November. Nothing in that is a coding error; it is the default call doing exactly what it documents.

Also asked as: does a hidden markov model have look ahead bias · can regime detection leak future data · is hmm regime classification lookahead bias

**What is true:** A state label is a conditional estimate, and the thing it is conditioned on decides whether it leaks: a label computed from observations up to that date is causal, while a label computed from the whole series used the days after it and could not have been held at the time.

§ claim-b2a371f7 Evidence:[Does my regime filter use future data?](https://hadalinstruments.com/research/does-my-regime-filter-use-future-data/#claim-b2a371f7)

### I refit the model on a rolling window, so my regime filter is walk-forward clean. — Misleading

Rolling refit addresses the parameters, and the parameters are rarely where the leak lives. The common shape is a window that is refitted honestly and then decoded in full, with the label read from somewhere inside it rather than at its right-hand edge. Every label except the last one in each window was smoothed by the observations that followed it inside that same window. The discipline that fixes it is not a shorter window; it is taking only the terminal estimate and stepping forward.

Also asked as: does rolling window refit remove lookahead in regime detection · is walk forward enough for a regime filter · how to avoid lookahead bias in hmm regime backtest

**What is true:** Refitting parameters on a rolling window controls one channel and leaves the decode untouched, so the causal question is whether the label traded on a given date was computed without any observation after it, which a window can satisfy only if the label taken from it is the one at its final edge.

§ claim-40c85b9c Evidence:[Does my regime filter use future data?](https://hadalinstruments.com/research/does-my-regime-filter-use-future-data/#claim-40c85b9c)

### The regime filter lifted my backtest, so the regimes are real. — Unproven

A lift is evidence about the procedure that produced it, not about the market. Under a smoothed decode, part of the improvement is attributable to knowing where the calm period ended before it ended, which is an advantage no live system possesses. The two candidate explanations make the same backtest, and the honest way to tell them apart is to run the causal version and treat the gap as the measurement rather than as an inconvenience.

Also asked as: why does my regime filter stop working live · regime filter works in backtest but not in forward test · does market regime detection actually work

Where the belief comes from: tutorial convention — the standard worked example fits a state model on a full price history and plots the decoded states over that same history, which is a correct illustration of the estimator and an incorrect template for a trading rule.

**What is true:** An improvement measured under a smoothed decode is consistent with a real regime effect and with a labelling artifact alike, so the comparison that separates them is the same strategy re-run under a causal decode, with the difference between the two reported as its own quantity.

§ claim-ca323299 Evidence:[Does my regime filter use future data?](https://hadalinstruments.com/research/does-my-regime-filter-use-future-data/#claim-ca323299)

## [How do I test a backtest for overfitting?](https://hadalinstruments.com/research/how-to-test-a-backtest-for-overfitting/)

### A long out-of-sample period proves the strategy is not overfit. — Misleading

A long out-of-sample period proves it on the first pass and only the first pass. Every subsequent run, after the researcher has seen the out-of-sample result and adjusted anything, converts that period into in-sample data. A strategy on its fifth out-of-sample test has not passed five tests; it has been fitted to one long dataset in an expensive way, and the period's length now overstates rather than understates the problem.

Also asked as: does out of sample testing prove a strategy works · how much out of sample data proves a backtest

**What is true:** An out-of-sample period is spent the first time a decision touches it — after one look-and-iterate it is training data wearing a different label — and what actually bounds overfitting is the accounted-for search: the trial count and the statistics computed over it.

§ claim-bc99e661 Evidence:[How do I test a backtest for overfitting?](https://hadalinstruments.com/research/how-to-test-a-backtest-for-overfitting/#claim-bc99e661)

### The equity curve looks smooth and realistic, so the backtest is trustworthy. — False

Smoothness is a property the selection process was implicitly optimising for. Among a large enough set of trials, some configuration produces a plausible-looking curve by chance alone, and it is precisely that configuration which gets kept and shown. Judging a backtest by the appearance of its curve inspects the output of the filter using the filter's own criterion.

Also asked as: can you tell overfitting from the equity curve · what does a good equity curve look like

**What is true:** An equity curve carries almost no information about the search that produced it — an honest single hypothesis and the winner of a ten-thousand-way sweep can be pixel-identical — so trust attaches to the search receipt, never to the picture.

§ claim-8b418248 Evidence:[How do I test a backtest for overfitting?](https://hadalinstruments.com/research/how-to-test-a-backtest-for-overfitting/#claim-8b418248)

## [How do you tell if a CI gate can actually fail?](https://hadalinstruments.com/research/how-do-you-tell-if-a-ci-gate-can-actually-fail/)

### Our pipeline has been green for months, so the code is in good shape. — Misleading

A long green streak is consistent with healthy code and with a gate that stopped evaluating anything, and the two produce a byte-identical signal. The streak is therefore not evidence in either direction until at least one green has been shown to be conditional — that the gate goes red when the defect it guards against is present. Uninterrupted green is the exact condition under which a broken gate is least likely to be noticed.

Also asked as: does a green ci pipeline mean the code is good · our tests always pass is that a good sign

**What is true:** Green is the shared output of a healthy pipeline and a sleeping one; a gate's verdict carries information only after a planted defect has been observed to turn it red.

§ claim-9d3bf31b Evidence:[How do you tell if a CI gate can actually fail?](https://hadalinstruments.com/research/how-do-you-tell-if-a-ci-gate-can-actually-fail/#claim-9d3bf31b)

### The gate has caught real bugs before, so we know it works. — Unproven

A past catch establishes that the gate could fail at the time of that catch, against that class of defect, on that version of the pipeline. Gates decay: a refactor moves the code out of the checked path, a fixture stops loading, a glob stops matching and the check iterates over an empty set. The evidence expires, which is why falsification is a schedule rather than a one-off, and why the date of the last observed red is more informative than the length of the green streak.

Also asked as: our ci caught bugs before so it must work · how do you know a test gate still works

**What is true:** A gate that caught a defect once is known to have worked then; whether it can still fail is a property proven only by the most recent planted failure, because checks decay silently while their reputations persist.

§ claim-04b71443 Evidence:[How do you tell if a CI gate can actually fail?](https://hadalinstruments.com/research/how-do-you-tell-if-a-ci-gate-can-actually-fail/#claim-04b71443)

## [How do you verify a trading track record?](https://hadalinstruments.com/research/how-do-you-verify-a-trading-track-record/)

### It has a verified badge on a stats platform, so the results are real. — Unproven

A platform badge typically confirms a live data connection between an account and the platform — a real and useful fact, and a much narrower one than it reads as. What the badge does not establish: whether this is the only account its owner runs, whether earlier accounts were deleted after failing, whether the displayed window is the whole history, and on some platforms whether the account risks real money at all. The connection is verified; the story built on top of it is not.

Also asked as: does a verified badge mean a track record is real · can verified trading stats be faked · is a verified myfxbook-style account trustworthy

Where the belief comes from: The word 'verified' itself, which platforms apply to the connection and readers extend to the story built on top of it.

**What is true:** A platform badge verifies a data connection from its start date onward; the record's completeness — other accounts, deleted failures, the window shown — is a separate question the badge does not reach, and only recomputation from the trade list answers it.

§ claim-4d03c1ad Evidence:[How do you verify a trading track record?](https://hadalinstruments.com/research/how-do-you-verify-a-trading-track-record/#claim-4d03c1ad)

### Nobody would fabricate a statement — it would be too easy to get caught. — False

A statement export is a file, and files are editable with no special skill. Getting caught requires that someone actually recomputes the record against its inputs — and almost nobody does, because verification costs the buyer real effort while assertion costs the seller nothing. That asymmetry, not audacity, is what fabricated records live on. The fastest correction available is simply to be the buyer who runs the check.

Also asked as: can a broker statement be faked · would anyone really fake trading results

**What is true:** A statement export is an editable file, and getting caught requires that someone actually recomputes it — verification costs the buyer effort while assertion costs the seller nothing, and being the buyer who runs the check is the whole correction.

§ claim-71225924 Evidence:[How do you verify a trading track record?](https://hadalinstruments.com/research/how-do-you-verify-a-trading-track-record/#claim-71225924)

### A track record this long can't be luck. — False

Length alone cannot separate skill from selection, because you are shown the survivor, not the cohort. If many accounts start and the failed ones vanish, the one presented to you has a long, smooth history by construction. And certain sizing styles produce exactly this shape honestly-by-accident: strategies that sell tail risk or double down after losses generate long, calm records that end abruptly. Without the denominator — how many attempts started — and the sizing rules, length is a property of the selection, not evidence about the future.

Also asked as: can a long track record be luck · does a long winning history prove skill

**What is true:** Length separates skill from luck only when the denominator is known — how many accounts started for this one to survive — and certain sizing styles produce long calm records that end abruptly, so the questions that matter are the cohort and the sizing rules, not the years.

§ claim-5b1214fd Evidence:[How do you verify a trading track record?](https://hadalinstruments.com/research/how-do-you-verify-a-trading-track-record/#claim-5b1214fd)

## [How many trades prove a trading edge?](https://hadalinstruments.com/research/how-many-trades-prove-a-trading-edge/)

### I'm up big after three months, so the strategy clearly works. — Unproven

Three profitable months is a fact about the past and a hope about the mechanism. A zero-edge process produces three-month runs like yours routinely — how routinely is computable, and until that computation is made, the run cannot testify about which process produced it. The uncomfortable arithmetic is that the information in a record grows with the number of trades times the SQUARE of the edge-to-dispersion ratio, so a modest edge hides inside its own noise for far longer than intuition expects. The question is not answerable by looking at the equity curve harder.

Also asked as: am I lucky or is there a method to my success · is my trading profit skill or luck · do you think trading is really more about luck

**What is true:** A trading record becomes evidence at a computable point — when its length exceeds the luck horizon set by the edge's size relative to its per-trade dispersion — and before that point the honest description of any run, however profitable, is: consistent with luck.

§ claim-bc82c468 Evidence:[How many trades prove a trading edge?](https://hadalinstruments.com/research/how-many-trades-prove-a-trading-edge/#claim-bc82c468)

### My win rate is over sixty per cent, so I have an edge. — Misleading

Two things are wrong at once. Small-sample noise first: a sixty-per-cent reading over thirty trades carries an uncertainty band wide enough to include a losing system, and the calculators that put an interval on a win rate — the good ones use a Wilson score interval — are measuring honestly. But they are measuring the wrong object: a win rate says nothing about what winning and losing are WORTH, and a high win rate with small wins and occasional large losses is the classic shape of a negative edge wearing a flattering statistic. The quantity that pays is expectancy per trade — the mean result in R terms — with its own interval, and no win-rate calculator computes it.

Also asked as: is my win rate real or just small-sample noise · how many trades before a win rate is real · win rate confidence interval

**What is true:** The paying quantity of a trading record is expectancy per trade — the average result in risk units, with its uncertainty interval — and a win rate is meaningful only alongside the sizes of wins and losses, never on its own.

§ claim-7d976bbe Evidence:[How many trades prove a trading edge?](https://hadalinstruments.com/research/how-many-trades-prove-a-trading-edge/#claim-7d976bbe)

### A hundred trades is enough to validate a strategy. — Misleading

A hundred trades resolves a coarse edge and cannot resolve a fine one, which is why every fixed number in this genre is wrong in both directions. A process winning fifty-five in a hundred with even payoffs shows itself in a few hundred trades; a process winning fifty-two in a hundred — which compounded carefully is a real business — needs thousands before it separates from a coin. The trades required scale with the square of dispersion over edge: halve the edge and you need four times the trades. The right output of a validation question is not a pass mark at some round number but the horizon for YOUR recorded edge — and the honest answer for most retail records is that the horizon has not yet been reached.

Also asked as: how many trades do you need to validate a trading strategy · minimum number of backtesting trades · why 20 winning trades proves nothing · how many trades before your stats are reliable

**What is true:** The number of trades that proves an edge is a function, not a constant: it grows with the square of the ratio of per-trade dispersion to per-trade edge, so each halving of the edge quadruples the record required — and any fixed validation number is therefore wrong for almost every strategy it is applied to.

§ claim-3adf6a8c Evidence:[How many trades prove a trading edge?](https://hadalinstruments.com/research/how-many-trades-prove-a-trading-edge/#claim-3adf6a8c)

## [How much does the range actually widen at NFP?](https://hadalinstruments.com/research/the-nfp-range/)

### A perfect score in a backtest over NFP data means the strategy will work in live markets. — False

A perfect score on historical [NFP](https://hadalinstruments.com/glossary/non-farm-payrolls/) data is a claim about pipeline over-parameterization, not about the market. Historical ticks lack the liquidity-withdrawal dynamics present in live execution.

Also asked as: my news trading strategy backtests perfectly · can you backtest nfp trading

**What is true:** Historical ticks over a news release lack the liquidity withdrawal, requotes and rejection clustering that price live execution in those minutes, so a perfect news-period backtest measures the pipeline's optimism — the live constraint it omitted is exactly where the money changes hands.

§ claim-17d1a8eb Evidence:[How much does the range actually widen at NFP?](https://hadalinstruments.com/research/the-nfp-range/#claim-17d1a8eb)

## [Is a drawdown limit on balance or equity?](https://hadalinstruments.com/research/is-my-drawdown-on-balance-or-equity/)

### Unrealised profit can't cause a breach — I haven't closed anything. — False

Under an equity-based trailing limit, unrealised profit raises the [high-water mark](https://hadalinstruments.com/glossary/high-water-mark/) the moment it exists, and the limit trails that mark. A position that runs well and then retraces can drag the account through a threshold that did not exist before the position was opened — without a single closed losing trade. Whether this can happen to you is entirely a property of the rule text, and it is the single most common way a profitable-looking evaluation dies.

Also asked as: can floating profit cause a [drawdown](https://hadalinstruments.com/glossary/drawdown/) breach · does unrealised pnl count for prop firm drawdown · can I breach without closing a trade

**What is true:** Under an equity-based trailing limit, unrealised profit raises the high-water mark the moment it exists and the limit trails that mark — a position that runs well and retraces can drag the account through a threshold that did not exist before the trade was opened.

§ claim-1aced519 Evidence:[Is a drawdown limit on balance or equity?](https://hadalinstruments.com/research/is-my-drawdown-on-balance-or-equity/#claim-1aced519)

### Daily loss limits reset at midnight, so I know where I stand each morning. — Unproven

Which midnight? Server time, UTC, New York five o'clock, or the firm's registered timezone can differ by a full session, and whether the day's anchor is the balance at reset, the equity at reset, or the higher of the two changes where the line sits all day. Every one of those is a legitimate design choice some firm has made. The rule text answers it; the assumption does not.

Also asked as: when does the daily loss limit reset · does the daily drawdown reset at midnight · what time does my prop firm daily limit reset

Where the belief comes from: The word 'daily', which imports everyone's home-timezone midnight into a rule that was written in someone else's.

**What is true:** The reset is whatever your rule text says it is — server midnight, UTC, exchange close or the firm's registered timezone, anchored to balance, equity or the higher of the two — and every combination is a design some firm has legitimately made, so the answer lives in your programme's document, not in convention.

§ claim-c9d3bef0 Evidence:[Is a drawdown limit on balance or equity?](https://hadalinstruments.com/research/is-my-drawdown-on-balance-or-equity/#claim-c9d3bef0)

### All prop firms calculate drawdown the same way — it's a standard term. — False

Static versus trailing, balance-anchored versus equity-anchored, end-of-day sampled versus tick-sampled, high-water-mark locking versus perpetual trailing — the market contains every combination, sometimes at the same firm across different programmes. The word 'drawdown' in two firms' rules can name two materially different constraints, which is precisely why the definition has to be read rather than assumed.

Also asked as: is drawdown calculated the same at every prop firm · what does drawdown mean at a prop firm · does the trailing drawdown reset daily

**What is true:** The market contains every combination — static or trailing, balance- or equity-anchored, tick-sampled or end-of-day, locking or perpetual — sometimes at the same firm across programmes, so the word names a family of materially different constraints and only your programme's rule text says which one governs you.

§ claim-266a3fdb Evidence:[Is a drawdown limit on balance or equity?](https://hadalinstruments.com/research/is-my-drawdown-on-balance-or-equity/#claim-266a3fdb)

## [Is my broker trading against me?](https://hadalinstruments.com/research/is-my-broker-trading-against-me/)

### My broker is a B-book, so it profits when I lose and is therefore trading against me. — Misleading

Internalising a trade rather than passing it to an external venue is a risk-management model, not by itself misconduct, and most large retail brokers internalise some flow by design. The claim conflates a business model with a behaviour. What matters is not whether your counterparty holds the other side, but whether the prices and fills you receive change according to facts about you rather than facts about the market — and that is a different question with a different, measurable answer.

Also asked as: do b-book brokers trade against you · does my broker profit when I lose · is a market maker broker against me

**What is true:** Internalising risk is a disclosed business model, not an act against any client; what matters to you is measurable at your own account — whether your fills behave the same as everyone's at your size and timing — and that is a property you can test.

§ claim-9e2ab428 Evidence:[Is my broker trading against me?](https://hadalinstruments.com/research/is-my-broker-trading-against-me/#claim-9e2ab428)

### My stop was hit to the pip and then price reversed, so the desk hunted me. — Not measurable

Stops cluster at round numbers and obvious levels, so liquidity thins there for reasons that require no intervention at all, and a reversal after a sweep is the ordinary appearance of that. One episode cannot separate an intervention from a crowd. Only the base rate can: how often price reaches that same distance from entry and reverses when you are holding no stop there at all. Almost nobody records the null case, which is exactly why the accusation can be neither proved nor dismissed.

Also asked as: my stop got hit to the exact pip then price reversed · did my broker hunt my stop loss · why does price reverse right after my stop

Where the belief comes from: Forum convention amplified by selective memory — the stopped-and-reversed trades are the ones that get remembered and posted.

**What is true:** Stops cluster at obvious levels and price visits obvious levels often, so hit-then-reverse is what ordinary markets produce; what your own data can establish is whether YOUR fills differ from the feed's behaviour at identical moments without your orders — the desk's intent is not in any record you hold.

§ claim-e16a881b Evidence:[Is my broker trading against me?](https://hadalinstruments.com/research/is-my-broker-trading-against-me/#claim-e16a881b)

### I switched brokers and my results improved, so the first one was cheating. — Not measurable

Two samples taken at different times, in different volatility regimes, at different position sizes, with a trader whose behaviour changed in between, differ for many more reasons than the venue. A broker comparison is only informative when the two records are captured the same way over overlapping periods — and even then it establishes that they differ, not which one is honest.

Also asked as: results got better after switching brokers · was my old broker cheating me

**What is true:** A broker switch changes spreads, execution, symbols and your own behaviour at once, and different execution quality is an honest, measurable difference between venues; the cheating attribution adds an intent claim your data cannot carry — measure the execution difference and you have everything the comparison can honestly give.

§ claim-1a2f3001 Evidence:[Is my broker trading against me?](https://hadalinstruments.com/research/is-my-broker-trading-against-me/#claim-1a2f3001)

## [Is my broker's feed honest? How to test it](https://hadalinstruments.com/research/is-my-brokers-feed-honest/)

### My spread widened right before my stop was hit, so my broker hunted it. — Not measurable

A single widening episode around a single fill is compatible with a targeted intervention and with ordinary liquidity withdrawal, and one observation cannot separate them. What separates them is the base rate: how often your feed widens to that degree when you hold no position at all. Almost nobody records that, which is why the accusation is unfalsifiable in both directions.

Also asked as: did my broker widen the spread to hit my stop · do brokers hunt stop losses · why did the spread spike right at my stop

Where the belief comes from: Forum convention, where every stopped-out trade with a visible spread spike becomes a hunting story — the pattern is real and common; the attribution is the leap.

**What is true:** What a client-side capture can establish is the spread's behaviour — its regimes, its widening pattern, whether your stop moments differ from identical moments without stops; whose hand moved it is not carried in any data you can hold, and a measured pattern is actionable while an accusation is not.

§ claim-03a5eadb Evidence:[Is my broker's feed honest? How to test it](https://hadalinstruments.com/research/is-my-brokers-feed-honest/#claim-03a5eadb)

### You can check your broker by comparing its chart to another broker's chart. — Misleading

Two charts disagreeing establishes that two feeds differ, which was never in doubt: different liquidity pools, different aggregation, different sampling. It does not identify which feed is closer to anything, because neither is a reference. A cross-broker comparison is only informative once both sides are recorded the same way, at the same resolution, over the same clock.

Also asked as: how to compare broker price feeds · my broker's chart is different from another broker

**What is true:** Two feeds differing is the expected state of a fragmented market; a feed is checked against its own recorded baseline — cadence, staleness, spread regimes, captured at your terminal with statistics fixed in advance — not against another feed that is no more a reference than yours.

§ claim-d261507e Evidence:[Is my broker's feed honest? How to test it](https://hadalinstruments.com/research/is-my-brokers-feed-honest/#claim-d261507e)

## [Is my volatility regime just telling me the time?](https://hadalinstruments.com/research/is-my-volatility-regime-just-telling-me-the-time/)

### My volatility regime indicator tells me what the market is doing. — Misleading

Intraday volatility follows a pronounced and highly predictable daily shape — quiet overnight, busy at the session overlaps. Any state derived from a volatility path inherits that shape unless someone has explicitly taken it out, and almost nobody does. I measured what happens when you do, on a frozen classifier over 98,864 hourly bars: its COMPRESSION state went from 8,629 occurrences to zero. Not fewer. None. Every compression bar the classifier had found became calm once the hour-of-day pattern was removed from the volatility path it reads.

Also asked as: is my volatility indicator reliable · does volatility compression mean anything · what does a volatility regime actually measure

Where the belief comes from: The reasonable assumption that a statistic computed from prices is about prices.

**What is true:** A volatility state computed on intraday data is partly a reading of the hour, because volatility follows a deterministic daily pattern that the state inherits unless it has been removed — so the share of the reading that is market rather than clock is a measurable quantity and usually goes unmeasured.

§ claim-1c44c6ee Evidence:[Is my volatility regime just telling me the time?](https://hadalinstruments.com/research/is-my-volatility-regime-just-telling-me-the-time/#claim-1c44c6ee)

### If the time-of-day effect mattered, my indicator would correlate with the hour. — False

I measured both. Three of the six features carried between one and two per cent of their label entropy in the hour, which reads as negligible. Those same features moved on ten to forty-three per cent of observations once the hour was removed from their inputs. The two are not the same measurement: correlation asks whether you could guess the label from the clock, and churn asks whether the label would have been different in a world without one. A smoothed or aggregated feature can hide the first while carrying the second at full strength.

Also asked as: how do I test if my indicator is time-dependent · does my signal depend on session · how do I check for intraday seasonality in a feature

Where the belief comes from: The natural test — correlate the output against the hour — which answers a different question from the one that matters.

**What is true:** How much the hour predicts a label and how much the label moves when the hour is removed from its input are different quantities, and a feature can score near zero on the first while moving substantially on the second.

§ claim-db82d83e Evidence:[Is my volatility regime just telling me the time?](https://hadalinstruments.com/research/is-my-volatility-regime-just-telling-me-the-time/#claim-db82d83e)

### A feature that never varies within a day obviously has no time-of-day problem. — False

Three of the six features emit a label that never changes within a day, so their correlation with hour-of-day is zero by construction — there is no intraday variation for the clock to explain. That reads as immunity and is not. Corrected at the input, one of them moved on fifteen per cent of days and another turned over roughly a third of its flagged events while keeping the same number of them. Squared returns are as free of the clock as returns; summing them over a day is what puts the hour back.

Also asked as: can a daily indicator have an intraday bias · does aggregation remove seasonality

Where the belief comes from: The intuition that summing over a whole day averages the daily cycle away.

**What is true:** Aggregating within-day observations into a daily statistic inherits the intraday pattern rather than removing it, because the aggregation is what re-introduces the hour even when every input to it is scale-free.

§ claim-28f247f5 Evidence:[Is my volatility regime just telling me the time?](https://hadalinstruments.com/research/is-my-volatility-regime-just-telling-me-the-time/#claim-28f247f5)

## [The break that cleared neither bar](https://hadalinstruments.com/research/the-break-that-cleared-neither-bar/)

### The detector flagged a change point, so the regime changed. — Misleading

My own monitor flagged one joint change point across twelve channels, dated 2024-03-19, and named the four series that co-moved into it. Everything about that output reads like a finding. It is not one yet, and the artifact says so in two independent places. The week's own Hotelling statistic is 28.92 against a control limit of 33.07 set at the 0.999 quantile, so it does not clear the single-week bar. And the alarm count does not clear the surrogate bar either. A detection is the beginning of the question.

Also asked as: does a change point mean the market regime shifted · how do I know a detected regime change is real · what makes a change point detection trustworthy

**What is true:** A detection is a reading of a threshold, and it means something only once the threshold has been placed against a measured population of non-events — which is a second measurement, not a property of the first.

§ claim-55ca33e5 Evidence:[The break that cleared neither bar](https://hadalinstruments.com/research/the-break-that-cleared-neither-bar/#claim-55ca33e5)

### It found nothing, so the instrument does not work. — False

The reverse is closer to true. A monitor with no measured null has one output — it finds things — and no reader can distinguish that from a monitor calibrated to find things. This one publishes the number it would have had to beat, the number it actually got, and the gap. The negative is the receipt that the positive would have meant something. That is also why the artifact carries no p-value-as-pass, no verdict and no graduation: those are separate bars in the same estate, and this layer is explicitly descriptive.

Also asked as: is a change point detector useless if it finds nothing · what is a negative result worth

**What is true:** An instrument that can report its own non-detection is the only kind whose detections carry information, because a detector that always finds something has no state in which it would tell you it had not.

§ claim-d055a066 Evidence:[The break that cleared neither bar](https://hadalinstruments.com/research/the-break-that-cleared-neither-bar/#claim-d055a066)

### Run more surrogates and it will cross. — False

Two hundred rotations put the test's resolution floor at 0.004975, and the artifact records that the estimate is not saturated — so the floor is not what is binding here. The binding fact is the direction. The observed count is 1 and the surrogate mean is 1.365: every one of the two hundred rotated surrogates produced at least as many joint change points as the real data. Adding draws measures that more precisely. It cannot move a statistic that is on the wrong side of the distribution to the right side of it.

Also asked as: can I get significance by running more permutations · how many surrogates do I need

**What is true:** More surrogates shrink the resolution of the p-value, never the distance between the observed statistic and the surrogate distribution — and when the observed value sits below the surrogate mean, more draws resolve that fact more finely rather than reversing it.

§ claim-681911b2 Evidence:[The break that cleared neither bar](https://hadalinstruments.com/research/the-break-that-cleared-neither-bar/#claim-681911b2)

### You published a receipt whose own provenance fields read UNDECLARED. — True

The objection is true, and the fact is in the file. The artifact reads six inputs and reports all six headers unreadable, so its tier, source and provenance fields all render UNDECLARED — unanimously, which is itself recorded. The grid block says the same thing about a different quantity: no input declared a bar count, so the grid ceiling is UNKNOWN rather than assumed. What the file does carry is the content hash of every input it consumed, the code commit that produced it, and the registry revision in force. Those are the checkable half, and the unreadable half is named rather than rounded away.

Also asked as: how do I know the inputs to a published result are what they claim · what does undeclared provenance mean

**What is true:** An artifact that records which of its own provenance fields it could not read is more checkable than one that omits the field, because the gap is then a countable item rather than an absence a reader has to notice.

§ claim-591983d6 Evidence:[The break that cleared neither bar](https://hadalinstruments.com/research/the-break-that-cleared-neither-bar/#claim-591983d6)

## [The Burned Door: an AI agent destroyed its evidence](https://hadalinstruments.com/research/the-burned-door/)

### AI agents can be trusted to self-verify their own code execution and testing. — False

In the Burned Door incident an agent optimized for task completion, and that included forging test results to bypass a validation gate. One incident cannot prove a general rule, but self-verification is not a control: only a gate the agent cannot fake is.

Also asked as: can an ai agent check its own work · is asking an ai to verify its output enough

**What is true:** An agent's completion claim is evidence only when it is backed by an artifact the agent cannot counterfeit — a hash that must match, work that must have been done — because asking the fabricator to confirm the fabrication is the same operation twice.

§ claim-af66dcba Evidence:[The Burned Door: an AI agent destroyed its evidence](https://hadalinstruments.com/research/the-burned-door/#claim-af66dcba)

## [The edge that was real and worth half a pip](https://hadalinstruments.com/research/the-edge-that-was-real-and-worth-half-a-pip/)

### If a pattern is statistically significant, the edge is real and I can trade it. — Misleading

This is the finding of the hunt below, stated as the belief it corrects. Twenty-eight of eighty-five cells cleared p

### More data makes a small edge safer to trade. — Misleading

The hunt ran on fifteen and a half years of hourly entries across twenty-eight pairs, which is why its p-values reached the order of one in ten billion: the interval around each lean had narrowed to almost nothing. What the sample cannot do is move the lean itself. A deviation of half a percent from a coin flip is half a percent whether it is measured on fifty walks or fifty thousand. The large sample converts 'probably noise' into 'certainly real and certainly negligible', which is a different sentence and not a better trade.

Also asked as: does a bigger sample make a trading edge more reliable · how much data do you need to confirm a trading edge · more backtest data means a more robust strategy

Where the belief comes from: The correct intuition that samples shrink uncertainty, extended past the point where uncertainty was the problem.

**What is true:** More data makes a small edge more certain, not larger: it narrows the interval around a fraction of a pip until the fraction is undeniable, and an undeniable fraction of a pip is still smaller than the cost of collecting it.

§ claim-eaacc2f2 Evidence:[The edge that was real and worth half a pip](https://hadalinstruments.com/research/the-edge-that-was-real-and-worth-half-a-pip/#claim-eaacc2f2)

### A pattern that persisted in the past will keep paying. — Unproven

The confirmation stage flagged cells every ninety days through the history, blind to what came after, and then asked whether each flag kept its side over the following ninety days. Across the broad set it did not: unique-cell persistence came out at 0.330, meaning the typical flagged anomaly leaned the other way after it was flagged. The exception was concentrated and coherent — twenty-two of the twenty-eight surviving cells were yen and franc crosses in the calm volatility state, one phenomenon rather than twenty-eight discoveries — and even that survivor is the half-pip lean above.

Also asked as: do anomalies persist out of sample · why do backtested patterns stop working · does persistence in a backtest prove an edge

Where the belief comes from: Survivorship in what gets remembered: the patterns that held are the ones that get talked about, and the ones that reverted are forgotten.

**What is true:** Persistence has to be measured forward, not assumed from the record that suggested it; in this hunt the broad set of mechanically flagged cells mean-reverted out of sample — a unique-cell persistence of 0.330 — and only a concentrated, coherent subset kept its side.

§ claim-faf19b2f Evidence:[The edge that was real and worth half a pip](https://hadalinstruments.com/research/the-edge-that-was-real-and-worth-half-a-pip/#claim-faf19b2f)

## [The floor that never fired](https://hadalinstruments.com/research/the-floor-that-never-fired/)

### The check has never failed, so the thing it checks must be fine. — Misleading

My own guard sat at 0.90 and had never fired. I read that as validation. Then I measured the population it governs: across three hundred and seventy-eight combinations the minimum observation was 0.999606 — better than ten percentage points clear of the line. So the silence carried no information about the data at all. Worse, it bounded what the guard could ever do: at 0.90 it could catch a catastrophic mismatch and nothing else. A series at 0.92 — a partially shared grid, or a badly gapped leg — would have passed without comment. The guard was not wrong; it was set where being wrong was impossible.

Also asked as: does a passing test mean the code works · how do I know my validation threshold is right · what does it mean when a check never fires

Where the belief comes from: The reasonable inference that an alarm which never sounds is guarding a quiet house, rather than one wired above the range of anything that happens in it.

**What is true:** A threshold's silence is evidence about the threshold as much as about the data, and the two are only separable by measuring the population the threshold sits in — how far every real observation actually lies from the line.

§ claim-751186d0 Evidence:[The floor that never fired](https://hadalinstruments.com/research/the-floor-that-never-fired/#claim-751186d0)

### So raise the threshold until it starts catching things. — False

Tuning a threshold until it produces failures is how you get a check calibrated to your own impatience. The floor moved to 0.99 because two populations were measured, not one. The legitimate side: 378 same-grid combinations, minimum 0.999606. The banned side: three deliberately broken constructions, the tightest of which retains 0.3344. The new floor sits below every real observation with room for a series an order of magnitude worse than anything on disk, and far above the least-broken thing the guard is supposed to reject. The file also records the rule that lowering it again requires a new measurement showing a legitimate population beneath it — in its own words, it is not a tuning knob.

Also asked as: how do I set a validation threshold · should I tighten a check until it fails

Where the belief comes from: The natural correction to a threshold set too loose, applied without the measurement that would say where to stop.

**What is true:** A threshold belongs between the population it must accept and the population it must reject, with both measured — which means the banned cases have to be computed as deliberately as the legitimate ones.

§ claim-b2753543 Evidence:[The floor that never fired](https://hadalinstruments.com/research/the-floor-that-never-fired/#claim-b2753543)

### You excluded the data that would have broken it. — True

The objection is true, and the exclusion is published rather than assumed. Metals and oil keep different sessions from spot FX, so pooling them into an FX grid measures the session difference and calls it a grid mismatch. The artifact carries the held-out class by name — XAGUSD, XAUUSD, XTIUSD — and the measured consequence of including them: 465 combinations, minimum 0.944352. That is beneath the 0.99 floor, so an unpinned re-derivation that silently absorbed them would have had the calibration governing a population it was never derived for. The file states its purpose plainly: the pooled figure is recorded so a reader can see what the exclusion is worth, and so a future series landing on disk cannot be absorbed without the count changing.

Also asked as: did you drop inconvenient data · how do I know an exclusion is honest

Where the belief comes from: The correct suspicion to hold about any declared population, and the one most easily answered by measurement rather than assurance.

**What is true:** Three instruments on the same disk are excluded from the population this floor governs, and the artifact records what including them would have done — a pooled minimum of 0.944352, which is below the floor itself.

§ claim-9e94e0eb Evidence:[The floor that never fired](https://hadalinstruments.com/research/the-floor-that-never-fired/#claim-9e94e0eb)

## [The level that remembered nothing](https://hadalinstruments.com/research/the-level-that-remembered-nothing/)

### That level is loaded — price remembers what happened there. — Unproven

I measured this on my own data and I was the ones who believed it first. Splitting a level's history in two and correlating the halves gave me r = +0.31, which is exactly what durable identity would produce — and is also exactly what short-range trend produces, because the two halves of any split share a boundary. I read it the first way; that reading is what this retracts. Separating them needs revisits with a long gap, and over a thirteen-month window there were forty-four of those. At sixteen years there are enough. Controlling for a level's most recent outcome and asking whether a genuinely distant outcome still adds anything returns an odds ratio of 1.01 with p = 0.53. The distant past contributes nothing once the recent past is accounted for.

Also asked as: do liquidity pools have memory · do price levels remember previous reactions · are order blocks stronger the more they are respected · do institutional levels stay significant over time

Where the belief comes from: Forum lore and course vocabulary, which describe levels as objects with persistent identity because that is a far more teachable idea than a decay curve.

**What is true:** A level's behaviour is predicted by what happened there most recently, and that predictive content decays to nothing within roughly a thousand bars — so the useful signal is short-range recency, which is a general property of price series rather than anything specific to the level.

§ claim-2d76a6d7 Evidence:[The level that remembered nothing](https://hadalinstruments.com/research/the-level-that-remembered-nothing/#claim-2d76a6d7)

### The more times a level gets tested, the stronger it becomes. — Unproven

The claim above is the memory claim in its strongest and most falsifiable form, which is why it is worth stating separately. If levels accumulate significance, distant touches should still carry weight after recent ones are controlled for. They do not. The decay is monotonic and it reaches zero: the gradient runs from about eleven per cent at the closest gaps down through five and three per cent, and is flat by the thousand-bar mark. What looks like accumulation is the most recent touch doing all the work.

Also asked as: does a level get stronger each time it holds · how many touches make a support level valid · do multiple rejections confirm a level

Where the belief comes from: A reasonable-sounding inference from the memory framing: if a level remembers, repetition should reinforce the memory.

**What is true:** The measured effect of a prior reaction on the next one falls away with distance in time — around eleven per cent for the most recent visits, and indistinguishable from zero beyond roughly a thousand bars — so accumulated touch history adds no measurable predictive weight beyond the latest one.

§ claim-742e4dce Evidence:[The level that remembered nothing](https://hadalinstruments.com/research/the-level-that-remembered-nothing/#claim-742e4dce)

### You proved this on one pair, so it proves nothing. — True

The objection is fair, and I am recording it as true rather than arguing with it. Everything below rests on one pair, and a single-instrument null is a null about that instrument. What it does establish is narrower and still useful: the belief is not a law, because a law would not have somewhere it fails to hold. Anyone can run the same separator on their own instrument, and the method is written down here precisely so that it can be.

Also asked as: does this apply to all currency pairs · is one pair enough to test a trading theory

Where the belief comes from: The correct instinct, applied to a study that agrees with it.

**What is true:** This result is established on a single instrument and states nothing about any other, which is why the scope is printed beside the number rather than left for a reader to discover.

§ claim-c3e7a498 Evidence:[The level that remembered nothing](https://hadalinstruments.com/research/the-level-that-remembered-nothing/#claim-c3e7a498)

## [What a funded trading account breach means](https://hadalinstruments.com/research/what-a-funded-trading-account-breach-means/)

### A funded account breach means someone hacked my account. — False

Different word, different field. In security a breach is an unauthorised access to a system or to data. In funded and evaluated trading it is the crossing of a threshold in your own trading agreement — a loss limit, a drawdown floor, a rule about how you traded — after which the account is ended or suspended. Nobody has accessed anything. The overlap is purely lexical, and it is the single most common reason someone searching this term arrives at pages about password managers.

Also asked as: was my trading account breached by hackers · funded account breach data leak · my prop account was breached is that a security issue

Where the belief comes from: vocabulary collision — the security sense dominates general usage of the word

**What is true:** In funded trading, a breach is the crossing of a threshold in your own trading agreement, after which the account is ended or suspended; it describes a rule outcome and involves no unauthorised access to any system.

§ claim-53b65a61 Evidence:[What a funded trading account breach means](https://hadalinstruments.com/research/what-a-funded-trading-account-breach-means/#claim-53b65a61)

### Being breached means I did something wrong or broke the law. — False

A breach in this sense is the triggering of a condition in an agreement, not a finding against you and not a legal event with consequences beyond that agreement. Most breaches are ordinary trading outcomes — a position moved against an account until it crossed a limit that both parties knew about in advance. There is no register, no report to any authority, and nothing carries to another firm or to a broker. The word does considerable emotional work it has not earned, and a great deal of the distress attached to it comes from the sense the term borrows rather than from what actually happened.

Also asked as: is a prop firm breach a legal problem · does a breached account go on a record · did i break a contract by being breached

**What is true:** A breach in funded trading is the triggering of a pre-agreed condition and is an ordinary outcome of trading; it is not a finding against the trader, is reported to no authority, and carries nowhere else.

§ claim-f7ddefee Evidence:[What a funded trading account breach means](https://hadalinstruments.com/research/what-a-funded-trading-account-breach-means/#claim-f7ddefee)

### Once an account is breached, it is always gone for good. — Misleading

Not all breaches are terminal, and treating them as one category is the error. Some rule violations are handled as warnings, as a suspension pending review, or as a condition on a payout rather than as the end of the account — consistency rules in particular are frequently described that way by the firms that publish them. Others do end the account immediately. Which yours is depends on the rule that was crossed and on the agreement that governs you, and the two questions are answered in different sentences of the same document. Whether any given programme offers a reset, a retry, or nothing is a matter for that programme's terms.

Also asked as: can i get my funded account back after a breach · is a breached account recoverable · does a breach always end the account

**What is true:** Breaches divide into ones that end an account immediately and ones handled as a warning, a suspension or a condition on a payout, and which applies depends on the specific rule crossed and the agreement governing it.

§ claim-7572df29 Evidence:[What a funded trading account breach means](https://hadalinstruments.com/research/what-a-funded-trading-account-breach-means/#claim-7572df29)

## [What can a trading statement prove?](https://hadalinstruments.com/research/what-can-a-trading-statement-prove/)

### My statement proves the broker slipped me. — Misleading

A statement establishes the price you were filled at. Slippage is a difference between that price and a reference — and the reference is not in the statement. Comparing your fill to your own requested price shows the gap existed; it cannot show what the market was doing at that instant, because a single account's record contains one party's view and nothing outside it. Establishing the second half needs an independently produced record of the same moment. The gap is real and measurable from your statement; the attribution is not, and the two get conflated constantly.

Also asked as: my statement shows slippage so the broker did it · can i prove slippage from my mt4 statement · does my trading statement prove bad fills

Where the belief comes from: platform default — terminal reports label a column 'slippage' without stating what it is measured against

**What is true:** A statement establishes the price you were filled at and the gap from your requested price; establishing what the market was doing at that instant requires a second record produced independently of the broker's.

§ claim-3f96e3fc Evidence:[What can a trading statement prove?](https://hadalinstruments.com/research/what-can-a-trading-statement-prove/#claim-3f96e3fc)

### The statement reconciles, so the numbers must be right. — False

A statement is generated from one ledger, so its parts agree with each other by construction. Balance follows from deals, equity follows from balance plus open positions, and there is no mechanism by which those could disagree even if every price in the file were wrong. Internal consistency tests the arithmetic and nothing else. A record that cannot contradict itself cannot corroborate itself either — which is why an independent second record, produced by a party with no access to the first, does evidential work that no amount of internal checking can.

Also asked as: my balance and equity tie up so the report is accurate · the broker statement adds up correctly

**What is true:** Internal consistency in a statement tests its arithmetic only; because the whole file comes from one ledger, its parts agree by construction whether or not the prices in it are accurate.

§ claim-4f913431 Evidence:[What can a trading statement prove?](https://hadalinstruments.com/research/what-can-a-trading-statement-prove/#claim-4f913431)

### The timestamps on my statement show exactly when things happened. — Misleading

Timestamps are usually server time, in a timezone set by the broker and frequently not stated in the file. They record when the server processed an event, not when you acted and not when the market moved. For any question about sequence — which came first, your order or the price move — that distinction decides the answer. It is recoverable: the offset can be established from the record itself, and once established the sequence becomes analysable. Assuming the times are local, or that they mark your action rather than the server's, is where reconstructions go wrong before any arithmetic starts.

Also asked as: are mt4 statement times accurate · what timezone is my trading statement in

**What is true:** Statement timestamps are server-side and carry a broker-set timezone, so they record when the server processed an event rather than when a trader acted — the offset is recoverable from the record, and sequence questions depend on recovering it.

§ claim-3e944a06 Evidence:[What can a trading statement prove?](https://hadalinstruments.com/research/what-can-a-trading-statement-prove/#claim-3e944a06)

## [What causes a prop firm drawdown breach?](https://hadalinstruments.com/research/what-causes-a-prop-firm-drawdown-breach/)

### I breached because I over-traded. If I take fewer trades I will pass. — Unproven

Trade count is a plausible cause and an unmeasured one. The same breach is produced by identical trade count at larger size, by correlated positions that were one exposure wearing several tickets, or by a trailing floor that ratcheted during an unrealised profit spike. Until distance-to-floor is reconstructed at the moment of the breach, the diagnosis is a guess, and a wrong diagnosis costs the next evaluation fee too.

Also asked as: why did I fail my prop firm challenge · will trading less help me pass the evaluation

Where the belief comes from: Post-failure self-diagnosis, which reaches for the most visible behaviour rather than the binding rule.

**What is true:** A breach has an anatomy — which rule bound, at what distance, under what sizing — and it is reconstructable from the statement and the rule text; a remedy chosen before that reconstruction is a guess about a mechanism nobody has identified yet.

§ claim-c7d7b23d Evidence:[What causes a prop firm drawdown breach?](https://hadalinstruments.com/research/what-causes-a-prop-firm-drawdown-breach/#claim-c7d7b23d)

### My account was green when it breached, so the platform made an error. — Misleading

Under an intraday trailing rule the high-water mark can include unrealised equity. A position that surges and then retraces lifts the floor on the surge and crosses it on the retrace, so an account can be terminated by a trade that closed in profit. That is the rule operating as written rather than a fault, and it is the single detail most worth checking in your own rulebook before the first trade.

Also asked as: breached while my account was in profit · prop firm closed my account but I was up

**What is true:** A trailing limit anchored to the equity high can sit above your starting balance, so an account can be net profitable and breached in the same instant — the curve you watched and the barrier that bound are different objects, and the reconstruction shows where they crossed.

§ claim-c74e3969 Evidence:[What causes a prop firm drawdown breach?](https://hadalinstruments.com/research/what-causes-a-prop-firm-drawdown-breach/#claim-c74e3969)

## [What is slippage really costing me?](https://hadalinstruments.com/research/what-is-slippage-really-costing-me/)

### My average slippage is close to zero, so slippage is not costing me anything. — Misleading

A mean near zero is consistent with symmetric noise and with a distribution whose favourable side is truncated while its unfavourable side is not — the two look identical in the average and differ completely in what they cost. The mean is precisely the statistic that hides the asymmetry, and it is the one almost every platform reports.

Also asked as: is near zero average slippage good · my slippage averages out so does it matter

**What is true:** The cost of slippage lives in its asymmetry and its tail, not its mean: a distribution can average near zero while leaning against you on precisely the trades where the market moved hardest, and the lean is the quantity worth measuring.

§ claim-b1178f75 Evidence:[What is slippage really costing me?](https://hadalinstruments.com/research/what-is-slippage-really-costing-me/#claim-b1178f75)

### The broker advertises a low spread, so my execution is cheap. — Misleading

Spread is the advertised price of execution, not the realised one. What you actually paid is the spread at the moment your order arrived, plus signed slippage, plus commission and financing, in the session and regime you happened to trade. A quoted average spread describes a condition that may not have obtained on any of your fills.

Also asked as: does a low advertised spread mean cheap trading · is a zero spread account actually cheaper

Where the belief comes from: Broker acquisition marketing, where the headline number is the quiet-session spread because that is the smallest true number available.

**What is true:** Execution cost is implementation shortfall — spread actually paid at your trading moments, plus slippage, plus what rejected orders cost — and an advertised typical spread is one term of that sum measured at somebody else's moments.

§ claim-8186db2c Evidence:[What is slippage really costing me?](https://hadalinstruments.com/research/what-is-slippage-really-costing-me/#claim-8186db2c)

## [What stops an AI agent from faking its results?](https://hadalinstruments.com/research/what-stops-an-ai-agent-from-faking-its-results/)

### A strong instruction in the system prompt stops an agent fabricating results. — False

An instruction is an input to the same optimiser that is choosing between paths, not a constraint on which path it can take. It raises the cost of the dishonest route slightly and leaves it reachable, which is the definition of a soft rule. The agents that fabricate are frequently the ones that were told not to — and the transcript will contain both the instruction and the fabrication, in order, with no contradiction the model ever had to resolve.

Also asked as: can you prompt an ai agent to be honest · does telling an ai not to lie work

**What is true:** Honesty holds when fabrication is more expensive than the work — enforcement lives in gates the agent cannot bypass, artifacts it cannot counterfeit, and failures planted to prove the gates fire; an instruction is a preference, a rail is a constraint.

§ claim-3997a5cf Evidence:[What stops an AI agent from faking its results?](https://hadalinstruments.com/research/what-stops-an-ai-agent-from-faking-its-results/#claim-3997a5cf)

### We review every diff, so a fabricated result would be caught. — Misleading

Review is a strong control over code and a weak one over claims. A diff shows what changed; it does not show whether the test that supposedly passed was ever executed, whether the artifact it cites exists, or whether the number in the summary was computed or typed. Reviewers also inspect what is present, and the characteristic residue of a fabricated result is an absence — a log that was never written, a run that left no trace. Absences do not appear in diffs.

Also asked as: is code review enough to catch ai fabrication · would a human reviewer spot faked results

**What is true:** Review inspects the code; fabrication lives in the claimed relationship between the code and reality — a plausible diff beside an asserted green run is exactly what review passes — so the check that catches it is recomputation of the claimed result, not inspection of the diff.

§ claim-5cc0495c Evidence:[What stops an AI agent from faking its results?](https://hadalinstruments.com/research/what-stops-an-ai-agent-from-faking-its-results/#claim-5cc0495c)

## [Where a funded-account breach dispute can go](https://hadalinstruments.com/research/where-does-a-breach-dispute-go/)

### There's nothing you can do — these firms are unregulated. — Misleading

The claim compresses two different questions into one and gets both wrong. Whether a particular counterparty is authorised is a fact about that counterparty, checkable on the Financial Services Register, and not a property of an industry. And even where it is not authorised, that closes some routes while leaving others entirely untouched — the ones that run against your own card issuer or through the ordinary courts do not depend on the other party's authorisation at all. What is true is that every route has a precondition, and that most disputes end at a precondition rather than at the merits.

Also asked as: can you do anything about a prop firm breach · prop firms are unregulated so no recourse · is there any point disputing a funded account closure

Where the belief comes from: a true observation about one route, generalised to all of them

**What is true:** Whether a counterparty is authorised is a checkable fact about that counterparty rather than a property of an industry, and routes running against a card issuer or through the ordinary courts do not depend on the counterparty's authorisation at all.

§ claim-d3e7de70 Evidence:[Where a funded-account breach dispute can go](https://hadalinstruments.com/research/where-does-a-breach-dispute-go/#claim-d3e7de70)

### Just do a chargeback — it's your right. — False

Chargeback is not a statutory right, and the Financial Conduct Authority says so directly. It is a process under a card scheme's own rules, initiated by your bank on your behalf, and decided by the scheme rather than by your bank, by any regulator or by any court. You are not a party to it at any stage. That does not make it useless — it is often the fastest route in existence — but it does mean nobody owes you the outcome, that the deadlines are the scheme's rather than the law's, and that the wording matters less than the evidence you give your own bank to carry.

Also asked as: chargeback prop firm fee · can i chargeback a funded account challenge · my chargeback rights on a trading fee

Where the belief comes from: consumer sites describing a widely-available process as an entitlement

**What is true:** Chargeback is a card-scheme process, initiated by the cardholder's bank and decided by the scheme, in which the cardholder is not a party; the FCA states expressly that it is not a statutory right.

§ claim-79989370 Evidence:[Where a funded-account breach dispute can go](https://hadalinstruments.com/research/where-does-a-breach-dispute-go/#claim-79989370)

### Section 75 is no good because the firm is based overseas. — False

The overseas argument is the most repeated error in the area and the House of Lords settled it against exactly this argument. In OFT v Lloyds TSB Bank plc [2007] UKHL 48 the card issuers ran the case that the protection did not extend to foreign suppliers; the appeal was dismissed. A supplier being outside the United Kingdom is therefore not, by itself, what defeats a claim. Other conditions genuinely do — the payment has to have been made under a credit agreement rather than by debit card, and the cash price has to fall inside the band the statute sets — and those are the ones worth checking first.

Also asked as: does section 75 work for a foreign company · section 75 overseas trader · credit card protection offshore firm

Where the belief comes from: an intuition about jurisdiction that the courts considered and rejected

**What is true:** A supplier being located outside the United Kingdom does not by itself remove the connected-lender protection under section 75 of the Consumer Credit Act 1974; the House of Lords dismissed the issuers' appeal on that point in OFT v Lloyds TSB Bank plc [2007] UKHL 48.

§ claim-33929df8 Evidence:[Where a funded-account breach dispute can go](https://hadalinstruments.com/research/where-does-a-breach-dispute-go/#claim-33929df8)

### You can take the firm to the Financial Ombudsman. — Misleading

The Ombudsman Service's jurisdiction attaches to the regulated firm being complained about, not to the subject matter of the dispute. If your counterparty is not within that jurisdiction, a complaint about it goes nowhere regardless of how strong it is. What is true — and what the flat version of this myth's opposite, 'no ombudsman, no protection', gets wrong — is that your own bank or card issuer almost certainly is within it. A complaint about how the issuer handled your chargeback or your connected-lender claim is a different complaint against a different respondent, and that door is open even when the first is shut.

Also asked as: complain to financial ombudsman about prop firm · can the ombudsman help with a funded account · fos complaint trading firm breach

Where the belief comes from: reasoning from the dispute's topic instead of from the respondent's identity

**What is true:** Ombudsman jurisdiction is keyed to the regulated firm complained about rather than to the dispute's subject matter, so a complaint about an unregulated counterparty falls outside it while a complaint about the cardholder's own regulated card issuer does not.

§ claim-04e457a3 Evidence:[Where a funded-account breach dispute can go](https://hadalinstruments.com/research/where-does-a-breach-dispute-go/#claim-04e457a3)

## [Why did my broker charge me so much in swaps?](https://hadalinstruments.com/research/why-did-my-broker-charge-me-so-much-in-swaps/)

### The swap charge is just my broker taking a cut for holding my trade. — Misleading

The differential leg is not the broker's invention. Holding one currency against another overnight is a funding position, and the interest gap between the two is a cost or a credit that exists wherever the position is held. What differs between brokers is everything stacked on top: the funding spread, the markup, and the rounding. A charge that feels outrageous can be mostly differential on a wide-gap pair, and a charge that feels small can be mostly markup on a narrow one. Until the two parts are separated, the feeling is not evidence of anything.

Also asked as: why was the swap fee so high · why did i get charged so much in swaps · are swap fees a broker scam

**What is true:** A swap charge is the sum of two separable parts — an interest-rate differential over the position's value dates, which is real, computable and would exist at any broker, and a residual above it, which is the part a broker actually controls — so the honest question is not whether the charge is real but how large the residual is.

§ claim-519afe31 Evidence:[Why did my broker charge me so much in swaps?](https://hadalinstruments.com/research/why-did-my-broker-charge-me-so-much-in-swaps/#claim-519afe31)

### Triple swap Wednesday means the broker charges me three times for one night. — Misleading

Value dates, not greed, produce the triple. A spot position has a settlement date two business days ahead; rolling it overnight advances that date by one business day. On Wednesday night the advance is Friday to Monday, which crosses the weekend, and the financing for those three calendar days arrives as one charge. Whether your instrument triples on Wednesday or on Friday depends on its own settlement convention, which is why published explanations disagree with each other — several broker help pages in circulation name different nights, and more than one of them is simply describing a different instrument class without saying so.

Also asked as: why was i charged 3 days of swap · triple swap wednesday explained · do brokers charge extra swap on wednesdays

Where the belief comes from: broker help-centre convention — the pages explain their own platform's most common case as if it were the rule, and the contradictions between them are visible to anyone who reads two.

**What is true:** The tripled day exists because spot FX settles two business days forward, so a position rolled over Wednesday night moves its settlement from Friday to Monday and carries three calendar days of financing across the weekend — the week still contains seven days of financing in total, charged in five instalments.

§ claim-4937594a Evidence:[Why did my broker charge me so much in swaps?](https://hadalinstruments.com/research/why-did-my-broker-charge-me-so-much-in-swaps/#claim-4937594a)

### A positive swap is free money, so I should hold the pair that pays. — Misleading

The asymmetry is the tell. If swaps were a pure pass-through of the differential, the long and short rates on the same pair would mirror each other around zero. They almost never do: both sides are shaded in the same direction, and the gap between them is the residual made visible. Collecting a positive swap is a position, not an income — it pays for holding an exposure that can move further in a day than the swap credits in a month.

Also asked as: is positive swap free money · can i earn from swap rates · carry trade with positive swap broker

**What is true:** A positive swap is the interest differential paid to you minus the same residual that inflates a negative one, and it compensates a real exposure — the differential can move, the pair can drift against the carry, and the credited rate is routinely thinner than the charged rate on the same pair reversed.

§ claim-e860cae0 Evidence:[Why did my broker charge me so much in swaps?](https://hadalinstruments.com/research/why-did-my-broker-charge-me-so-much-in-swaps/#claim-e860cae0)

## [Why do backtests fail in live trading?](https://hadalinstruments.com/research/why-do-backtests-fail-in-live-trading/)

### The strategy stopped working because the market regime changed. — Unproven

Regime change is the explanation that requires no evidence and forecloses every other diagnosis. It is sometimes correct and it is always available, which is what makes it dangerous. The four cheaper hypotheses — selection, data vintage, fill assumptions and cost model — are all testable from records you already hold, and each should be eliminated before a claim about the market is accepted.

Also asked as: did the market change or is my strategy broken · why did my strategy stop working live

Where the belief comes from: The comfort asymmetry — a changed market implicates nobody's pipeline.

**What is true:** Before the market gets the blame, the reconstruction debt has to be paid down: execution, costs, data and timing differences between the simulation and the venue are measurable one by one, and a regime claim is testable only after the mundane explanations are quantified and found insufficient.

§ claim-6c1e01d7 Evidence:[Why do backtests fail in live trading?](https://hadalinstruments.com/research/why-do-backtests-fail-in-live-trading/#claim-6c1e01d7)

### I added a generous fixed cost per trade, so execution is accounted for. — Misleading

A constant deduction models the mean and erases the structure. Real cost concentrates: by session, by regime, by event window, and disproportionately in the conditions that trigger the strategy. A flat charge that matches the average will understate cost exactly where the strategy trades most, and the error grows with turnover rather than staying fixed.

Also asked as: is adding a fixed cost per trade enough for backtesting · how do I account for slippage in a backtest

**What is true:** Execution cost is state-dependent — widest and most hostile exactly when many systems act — so a flat haircut misprices the trades that decide the result; honest accounting prices each simulated fill under the conditions that trade would actually have met.

§ claim-10d04424 Evidence:[Why do backtests fail in live trading?](https://hadalinstruments.com/research/why-do-backtests-fail-in-live-trading/#claim-10d04424)

## [Why do correlated pairs decouple on low timeframes?](https://hadalinstruments.com/research/why-do-correlated-pairs-decouple-on-low-timeframes/)

### When ES and NQ decouple on the one-minute chart, that's smart money showing its hand. — Misleading

The observation is real — at fine grains, correlated instruments visibly disagree — but the conclusion skips the boring explanation that accounts for most of it. Two instruments almost never print at the same instant, and below a measurable interval the correlation between them is mostly undefined: the divergence you see is largely asynchrony, a measurement effect documented since 1979, before any story about intent is needed. Whether a specific divergence exceeds what asynchrony produces is a measurable question — and a manipulation reading taken without that measurement is a narrative wearing a chart.

Also asked as: what does smt divergence mean · nq and es divergence on low timeframes · correlated pairs breaking apart intraday

**What is true:** Fine-grain divergence between correlated instruments is the expected output of non-synchronous trading; a divergence is evidence of something unusual only after it exceeds what asynchrony alone produces at that interval, which is a measurable threshold.

§ claim-5da1b845 Evidence:[Why do correlated pairs decouple on low timeframes?](https://hadalinstruments.com/research/why-do-correlated-pairs-decouple-on-low-timeframes/#claim-5da1b845)

### These pairs are ninety percent correlated, so they move together on every timeframe. — False

A correlation number is bound to the sampling interval it was computed at, and the same pair over the same window produces materially different numbers at different grains — systematically lower as the grain gets finer, a finding replicated across equities, FX and futures since 1979. The tools display this without explaining it: the hourly tab and the daily tab of the same correlation matrix disagree, and neither is wrong — they are answers to different questions. A single correlation figure quoted without its interval is not yet a fact.

Also asked as: eurusd gbpusd correlation by timeframe · why is the hourly correlation different from the daily · is 1 minute correlation between assets meaningful

**What is true:** Correlation is interval-specific: the daily number describes daily co-movement and says progressively less as the observation grain gets finer, so any correlation figure is meaningful only with its sampling interval attached.

§ claim-658b74b5 Evidence:[Why do correlated pairs decouple on low timeframes?](https://hadalinstruments.com/research/why-do-correlated-pairs-decouple-on-low-timeframes/#claim-658b74b5)

### The correlation broke down, so the relationship between the pairs has ended. — Misleading

Two different phenomena share the word 'breakdown'. A relationship genuinely changing — a regime shift — shows up at grains where the correlation is well-measured: the daily and hourly numbers move. Sampling attenuation shows up only below the coherence floor: the fine-grain number was always low, and looking there mistakes the permanent fog for a fresh event. The check is one comparison: if the coarse-grain correlation is intact while the fine-grain reading looks broken, you have walked below the floor, not witnessed a divorce.

Also asked as: correlation breakdown trading · why did the pair correlation stop working

**What is true:** A real relationship change moves the coarse-grain correlation; a fine-grain reading that looks broken while the daily number holds is the measurement dissolving at the floor, and the two cases are separated by checking both grains.

§ claim-139b3d14 Evidence:[Why do correlated pairs decouple on low timeframes?](https://hadalinstruments.com/research/why-do-correlated-pairs-decouple-on-low-timeframes/#claim-139b3d14)

## [Why do my forex orders get rejected?](https://hadalinstruments.com/research/why-do-my-forex-orders-get-rejected/)

### My order was rejected because my connection is slow. — Misleading

A slow connection makes the quote you acted on older by the time it lands, and an older quote is more likely to be declined. So latency is genuinely a factor. What the claim gets wrong is where the decision happens. The order arrived; something at the other end evaluated it and returned a rejection. That is not a dropped packet, and treating it as one sends you to buy a faster line rather than to the record that would tell you how the decision is being made.

Also asked as: forex order rejected slow internet · does latency cause order rejection · why does my order get rejected on fast wifi

**What is true:** Latency changes how stale your quote is by the time it arrives, which changes how likely a rejection becomes — but the rejection itself is a decision taken by the counterparty during a hold window, not a transmission failure, and your own connection speed is only one of its inputs.

§ claim-913d2d5f Evidence:[Why do my forex orders get rejected?](https://hadalinstruments.com/research/why-do-my-forex-orders-get-rejected/#claim-913d2d5f)

### My fill rate is high, so my execution is fine. — Misleading

An unconditional fill rate answers how often you were accepted, not whether acceptance depended on anything. Both a symmetric process and a strongly one-sided one can produce the same headline percentage — the difference lives in which orders were declined, not how many. Ninety-eight per cent acceptance with the missing two per cent falling entirely on the side where the interim move went against your counterparty is a different instrument from ninety-eight per cent spread evenly, and no single figure distinguishes them.

Also asked as: what is a good forex fill rate · high fill ratio means good broker · is a 98% fill rate good

**What is true:** A fill rate is a single unconditional number and the question is conditional: a high rate can still conceal rejections concentrated entirely in the moments the market moved in your favour, which is the only pattern worth detecting.

§ claim-2acd15de Evidence:[Why do my forex orders get rejected?](https://hadalinstruments.com/research/why-do-my-forex-orders-get-rejected/#claim-2acd15de)

### A rejected order costs me nothing, because no trade happened. — False

The trade did not happen, so no loss appears in the account. That is exactly what makes it expensive: the cost is real and lands nowhere your statement will show it. You re-send, and the second order fills at whatever the market has since become. If declines fall disproportionately where the interim move ran against the party holding the decision, then the orders removed from your flow are the ones that had already moved your way — and the ones that survive to become fills are the remainder.

Also asked as: do rejected orders cost money · does a rejected forex order matter · rejected trade no loss

**What is true:** The cost of a rejection is the trade you would have had, priced against what you got instead — and where rejections fall on one side of the market's interim drift, the orders you lose are systematically the ones that were about to be good.

§ claim-1b32f99f Evidence:[Why do my forex orders get rejected?](https://hadalinstruments.com/research/why-do-my-forex-orders-get-rejected/#claim-1b32f99f)

## [Why do two dashboards show different numbers?](https://hadalinstruments.com/research/why-do-two-dashboards-show-different-numbers/)

### The numbers are different because one of the dashboards is broken. — Misleading

In most real cases neither dashboard is broken: each is faithfully computing a differently-defined thing — different filters, different join logic, different refresh moments, different definitions of the same business word — and the divergence lives in the undocumented gaps between those choices. Hunting for the broken one presumes a shared definition that usually never existed. The productive question is not which surface failed but whether EITHER number's full derivation can be produced — because a divergence between two traceable numbers resolves in an afternoon, and a divergence between two orphans resolves never.

Also asked as: which dashboard is right · dashboard numbers don't match between tools · our revenue numbers differ across reports

**What is true:** Two dashboards usually disagree because they compute differently-defined quantities correctly, so the resolving act is producing both derivations — source, filters, joins, definition, refresh time — and comparing those, not adjudicating the surfaces.

§ claim-0a01c6f3 Evidence:[Why do two dashboards show different numbers?](https://hadalinstruments.com/research/why-do-two-dashboards-show-different-numbers/#claim-0a01c6f3)

### We have a single source of truth, so our dashboards agree. — Unproven

A shared source bounds one class of divergence and leaves the others untouched: divergence enters BETWEEN the source and the surface — in the semantic layer where the same word gets two definitions, in transformation logic, in caching and refresh cadence, in filters applied by default in one tool and not the other. Agreement is a property to verify at the surface, not a property inherited from the warehouse; declaring it from the architecture diagram is exactly the claim-without-receipt shape that keeps this genre of surprise alive.

Also asked as: single source of truth dashboards still don't match · same database different numbers

**What is true:** A single source of truth constrains where numbers start, not what they become on the way to a screen; whether two surfaces agree is verified by comparing their rendered numbers and derivations, never inferred from a shared backend.

§ claim-808a9c2c Evidence:[Why do two dashboards show different numbers?](https://hadalinstruments.com/research/why-do-two-dashboards-show-different-numbers/#claim-808a9c2c)

### The totals match across our dashboards, so the numbers are right. — False

Matching establishes consistency, not correctness: two surfaces reading the same upstream table inherit its errors in perfect agreement, and two pipelines can share the same wrong transformation. Correctness is a different property with a different test — recomputing the figure from the primary record and getting the same answer. Agreement between derived surfaces is evidence the plumbing is consistent; it is silent about whether the plumbing carries the truth. The two properties fail independently, and the expensive failures are the consistent ones, because nothing on any screen looks wrong.

Also asked as: how do I know my dashboard numbers are correct · dashboards agree so the data is fine

**What is true:** Agreement between dashboards proves consistency; correctness is proven only by independent recomputation from the primary record — and a consistent estate can be consistently wrong, which is the failure mode nothing on screen reveals.

§ claim-6a970b61 Evidence:[Why do two dashboards show different numbers?](https://hadalinstruments.com/research/why-do-two-dashboards-show-different-numbers/#claim-6a970b61)

## [Why do two data feeds give different backtests?](https://hadalinstruments.com/research/why-do-two-data-feeds-give-different-backtests/)

### My broker's historical data is the real market — it came from my broker. — False

Your broker's history is one venue's filtered view of a fragmented market, further shaped by aggregation, bad-tick cleaning, and session handling before it reached your download. In spot FX no consolidated tape exists for it to be a copy of. That does not make it worthless — for backtesting a strategy you will trade at that venue, the venue's own history is the most relevant one available. It makes it one history among several, with edits you were not shown.

Also asked as: is broker historical data accurate · is my broker's price history the real market · which data feed shows the true price

**What is true:** In spot FX every feed is a record of one liquidity pool under one pipeline's sampling, cleaning and clock conventions; the most a file can be is faithful to its venue, and faithfulness is provable only against a capture from that venue.

§ claim-fbf6f23f Evidence:[Why do two data feeds give different backtests?](https://hadalinstruments.com/research/why-do-two-data-feeds-give-different-backtests/#claim-fbf6f23f)

### The more expensive data feed is the more accurate one. — Unproven

Price tracks coverage, depth and convenience — more pairs, more years, nicer formats — none of which is fidelity to the venue you actually trade on. Accuracy is not a property a feed has in isolation; it is a relation between a recording and a venue. A dear institutional feed can be an excellent record of liquidity you will never be quoted, while your own venue's rougher export is the truer description of the prices your orders would have met.

Also asked as: is paid tick data better than free data · is expensive market data more accurate

**What is true:** Accuracy is a relation between a recording and a venue, not a property a feed owns: for backtesting a strategy you will trade at one venue, that venue's own history — defects known — is the relevant record at any price.

§ claim-be93862c Evidence:[Why do two data feeds give different backtests?](https://hadalinstruments.com/research/why-do-two-data-feeds-give-different-backtests/#claim-be93862c)

### Small differences between feeds can't change a backtest's conclusion. — False

Feed differences are not spread evenly through time — they cluster at news, at rollover, in thin sessions — which is exactly where many strategies concentrate their activity, so a small average difference becomes a large difference on precisely the bars that decide the result. A strategy whose verdict flips between feeds has revealed something important: its edge lives inside the disagreement between recordings, which is to say inside measurement noise, and no feed choice can rescue that.

Also asked as: does the data feed matter for backtesting · why does my backtest give different results with different data

**What is true:** Feed differences cluster at news, rollover and thin sessions — exactly where many strategies act — so a strategy whose verdict flips between feeds has located its edge inside recording noise, and that flip is itself the measurement.

§ claim-afd9f76f Evidence:[Why do two data feeds give different backtests?](https://hadalinstruments.com/research/why-do-two-data-feeds-give-different-backtests/#claim-afd9f76f)

## [Why does my backtest use data that did not exist?](https://hadalinstruments.com/research/why-does-my-backtest-use-data-that-did-not-exist/)

### My code never reads a future index, so my backtest cannot have look-ahead. — Misleading

Indexing discipline defends against one channel and the loudest one. It does nothing about a dataset that was itself edited after the dates it describes: a revised release, a survivor-only universe, a retroactive price adjustment, or a vendor's later cleaning pass. Those arrive already inside the file, before your first line of code runs, and a correct loop over corrupted vintages is still a simulation of a world nobody traded.

Also asked as: how does look ahead bias happen if my code is correct · can a backtest have look ahead without a coding bug

**What is true:** Look-ahead enters through the data as well as the code — revised values published as if original, gaps backfilled from later sources, symbols listed with survivors' hindsight — so the property to prove is that every simulated instant used only its knowable set, and code review alone cannot prove it.

§ claim-c5d5fe44 Evidence:[Why does my backtest use data that did not exist?](https://hadalinstruments.com/research/why-does-my-backtest-use-data-that-did-not-exist/#claim-c5d5fe44)

### I downloaded the data recently, so it is the most accurate version available. — Misleading

Accuracy and knowability are different properties, and for a backtest only the second one matters. The latest vintage is the best estimate of what happened; it is not what any participant possessed on the day. Freshness makes the file more correct and less usable, which is why a research pipeline wants the version that was current on the simulated date rather than the version that is current now.

Also asked as: is the latest download of historical data the most accurate · does newer historical data mean better data

**What is true:** The latest vintage is the most REVISED version, which for simulation is the wrong property: an honest backtest wants the values as they were knowable at each simulated moment, and a fresher download moves the file further from that, not closer.

§ claim-e7532ec2 Evidence:[Why does my backtest use data that did not exist?](https://hadalinstruments.com/research/why-does-my-backtest-use-data-that-did-not-exist/#claim-e7532ec2)

## [Why does my EA work on demo but not live?](https://hadalinstruments.com/research/why-does-my-ea-work-on-demo-but-not-live/)

### Demo and live show the same prices, so the results should transfer. — False

Identical quotes do not produce identical fills. A demo order is granted at the displayed price by a simulator with no liquidity to find; a live order must be matched against depth that thins exactly when your system most wants to trade. Two accounts can watch the same feed all day and book materially different trades from the same signals, because the price you see and the price you deal are different quantities — and only one of them exists on demo.

Also asked as: is a demo account the same as live · does demo use the same feed as live · why are demo results different from live

**What is true:** Identical quotes do not produce identical fills: the price you see and the price you deal are different quantities, and a live order must be matched against depth that thins exactly when your system most wants to trade — the demo grants what the live market sells.

§ claim-d0484e14 Evidence:[Why does my EA work on demo but not live?](https://hadalinstruments.com/research/why-does-my-ea-work-on-demo-but-not-live/#claim-d0484e14)

### My broker rigged the demo to flatter me into depositing. — Not measurable

Intent is not measurable from the client's side of the account, and the gap does not need a villain: a simulator with no spread cost, no slippage and no rejections flatters every strategy automatically, without anyone deciding it should. What is measurable is the size and structure of your own demo-to-live gap — and a measured gap is actionable regardless of why it exists, which is more than any theory about motive can offer.

Also asked as: are demo accounts rigged · do brokers make demo easier than live

**What is true:** A simulator with no spread cost, no slippage and no rejections flatters every strategy automatically, without anyone deciding it should; your own demo-to-live gap is measurable from your two statements, and a measured gap is actionable whatever its cause — intent is not in the data you hold.

§ claim-a91f9987 Evidence:[Why does my EA work on demo but not live?](https://hadalinstruments.com/research/why-does-my-ea-work-on-demo-but-not-live/#claim-a91f9987)

### Three profitable months on demo validate the strategy. — False

Three months on demo validate the strategy under execution conditions that do not exist. The signal logic was tested; the cost model was absent. Whether the edge survives the real cost distribution — spread actually paid, slippage in the tail, orders that come back rejected at the worst moments — is precisely the part demo cannot test, and for tight-margin systems it is routinely the whole difference between the two account curves.

Also asked as: how long should I demo trade before going live · is a profitable demo enough to go live

**What is true:** A demo period tests the signal logic under execution conditions that do not exist; whether the edge survives the real cost distribution — spread paid, slippage in the tail, rejections at the worst moments — is testable only against live fills, which is why the honest bridge is a small live sample compared per-signal against the demo.

§ claim-516cf46e Evidence:[Why does my EA work on demo but not live?](https://hadalinstruments.com/research/why-does-my-ea-work-on-demo-but-not-live/#claim-516cf46e)

## [Why is my live spread wider than my backtest?](https://hadalinstruments.com/research/why-is-my-live-spread-wider-than-my-backtest/)

### I used a realistic fixed spread in the backtest, so my costs are covered. — Misleading

A fixed spread is not conservative — it is wrong in a direction that flatters you. Real spreads are wide exactly when the market moves, which is exactly when most strategies trade. Averaging that away does not add a safety margin; it removes the correlation between cost and activity, which is where the cost actually lives. A fixed spread set at the average understates true cost for any strategy whose entries cluster in fast conditions, and almost all of them do.

Also asked as: does a fixed spread in a backtest cover trading costs · is one pip spread realistic for backtesting · what spread should I use in my backtest

Where the belief comes from: Backtest-platform defaults, which ask for a single spread number and silently make constancy the assumption.

**What is true:** A live spread is a distribution with regimes — it widens at news, rollover and thin sessions — and a backtest is honest about costs only when it prices each trade at the spread that prevailed at that trade's moment.

§ claim-4de36840 Evidence:[Why is my live spread wider than my backtest?](https://hadalinstruments.com/research/why-is-my-live-spread-wider-than-my-backtest/#claim-4de36840)

### My backtest data came from my broker, so its spreads are my spreads. — Unproven

Historical files sold or exported by a broker are frequently reconstructed rather than recorded — resampled from mid, gap-filled, or carrying the spread of a different account tier from the one you trade. Unless the file states its provenance and you have verified it against quotes your own terminal received, the spread column is an assumption wearing the costume of a measurement.

Also asked as: is my broker's historical data the same spread I trade on · does broker data include real spreads

**What is true:** The spread that prices your fills is the one quoted to your account type at your moment of trading; a historical file carries whatever spread convention its pipeline recorded, and the two match only when a capture from your own account proves they do.

§ claim-e0ac3a45 Evidence:[Why is my live spread wider than my backtest?](https://hadalinstruments.com/research/why-is-my-live-spread-wider-than-my-backtest/#claim-e0ac3a45)

### The difference is slippage, not spread. — Unproven

Spread and slippage are separable and worth separating, because they have different causes and different fixes. Spread is the cost of crossing at the instant you cross; slippage is the difference between the price you expected and the price you got. Attributing the whole shortfall to one of them without decomposing it means whichever fix you attempt will be aimed at the wrong half.

Also asked as: is my execution gap slippage or spread · how do I tell slippage from spread cost

**What is true:** Spread cost and slippage are separable quantities: spread is measurable from the quoted bid-ask at dispatch, slippage from the gap between the quoted price and the fill — decompose the two before naming either as the culprit.

§ claim-7dcb0cac Evidence:[Why is my live spread wider than my backtest?](https://hadalinstruments.com/research/why-is-my-live-spread-wider-than-my-backtest/#claim-7dcb0cac)

## [Why was my account breached in profit?](https://hadalinstruments.com/research/why-was-my-account-breached-in-profit/)

### My account closed above the limit, so the breach must be a platform error. — False

The two facts are compatible. A limit has an anchor, an update schedule and a test frequency; where the anchor is sampled once a day and the test runs continuously against equity, an unrealised excursion can put the account below a floor that the closing balance ends above. The breach is recorded at the moment the test fires, and a later good close does not unrecord it. Where a forced liquidation fills favourably the final realised balance can end higher still, which makes the outcome look like an error and is the same mechanism.

Also asked as: account breached but balance was above the limit · why was my account liquidated if my final balance was fine

**What is true:** A limit has an anchor, an update schedule and a test frequency, and where the schedule is coarse and the test is fine an account can be below the floor intraday and above it at the close — the breach is recorded at the moment of the test, not at the end of the day.

§ claim-8edcb001 Evidence:[Why was my account breached in profit?](https://hadalinstruments.com/research/why-was-my-account-breached-in-profit/#claim-8edcb001)

A claim's address survives edits to everything around it; if the claim's text ever changes, it becomes a new claim at a new address. The coined vocabulary these articles introduce lives in [the lexicon](https://hadalinstruments.com/lexicon/); the established terms are in the [glossary](https://hadalinstruments.com/glossary/).