Non-Repainting Indicators: How to Verify the Claim Yourself

This is the verification tutorial we wish had existed when we started: four mechanisms by which an indicator rewrites its own history, four tests that expose them, and the less obvious version of the same problem — statistics panels computed with hindsight. It requires no programming, one afternoon, bar replay, and a willingness to distrust your vendor. Including us.

What “repainting” actually means — and what it doesn’t

An indicator repaints when what it shows for a past bar differs from what it showed for that bar while it was the current bar. That is the whole definition, and two consequences follow that most buyers get backwards.

A delayed signal is not a repainting signal. An indicator that marks a swing high five bars after the high formed is honest, provided the marker appears five bars late on the live chart too, and stays there. Delay is the price of confirmation.

An unchanging signal can still be a repainting signal if it was placed using information that did not exist at that bar. A marker drawn at the exact top of a swing, on the bar of the swing, is the classic case: it looks perfect on the reloaded chart because the script waited to see which bar turned out to be the top, then plotted backwards. The marker never “moves”; it was simply never available when it claims to have been.

So the question is never “does the chart change?” It is “could I have seen this, here, at the time?” Everything below asks it mechanically.

The four ways an indicator rewrites history

1. Values computed on the unconfirmed bar

The current bar on a live chart updates on every tick. Any condition evaluated on it — a crossover, an RSI threshold, a candle pattern — can be true at 10:31:15 and false at 10:31:58. If the script fires on that intra-bar state and the bar closes on the other side, the signal existed live and not on the reloaded chart, or the reverse.

In Pine Script the guard is barstate.isconfirmed: a condition evaluated only on the confirmed close behaves identically live and historically. A script that runs its logic on every tick shows you signals in real time that a fresh load quietly omits.

2. Higher-timeframe data requested with lookahead

This is what people mean by “the indicator uses future data.” request.security() lets a script read a higher-timeframe series — the hourly close on a 5-minute chart, say. Called with lookahead=barmerge.lookahead_on, historical bars receive the hourly close at the start of the hour, fifty-five minutes before it existed. Anything built on it backtests like a miracle and trades like a coin flip.

Lookahead off is the default in recent Pine versions, but a subtler variant remains: even with lookahead off, historical bars show the confirmed higher-timeframe value while the live bar shows the developing one, so the line looks steadier on a reload than it ever did live. The documented fix is to reference the previous confirmed higher-timeframe bar explicitly, and a vendor who has thought about this will say so in the description.

3. Pivots, swings and structure with a right-side lookback

ta.pivothigh(left, right) confirms a swing high only after right further bars have closed. The repainting happens in how the result is displayed: a script that plots the pivot with a negative offset draws it back on the bar where the high occurred, so the reloaded chart shows a perfectly timed marker that, live, appeared right bars later. Every “market structure” and “order block” tool is built on some form of this, which is why the category is so exposed.

The honest version either plots the marker where it was confirmed, or plots it back-shifted and states plainly that the marker is confirmed N bars after the bar it points to. Both are non-repainting in substance. Only silence about the delay is a problem.

4. Objects that are moved, deleted or replaced

Lines, labels and boxes are mutable. A script can extend a line until it breaks, then delete it; redraw a box as price develops; or keep only the “last valid” signal and remove the failed ones. Each makes the chart cleaner and the history dishonest: the failed signals were on your screen when it mattered and gone when you evaluate the tool. A reload test cannot catch this on its own, because deleted objects are gone on both charts. It needs a record.

Four tests, one afternoon

Test 1 — Bar replay, with the indicator loaded

Open bar replay on your own instrument and timeframe, not the vendor’s demo chart. Step forward and note on which bar each marker, level and label first appears and where it is drawn. Then exit replay, reload, and compare the same stretch.

You are looking for any object that now sits earlier than it first appeared in replay, has moved, or has disappeared. Replay renders bars as confirmed, so it will not reproduce intra-bar flicker — but it catches back-shifted pivots and rewritten objects reliably, and those are most real-world cases.

Test 2 — Alerts as a timestamped log

Bar replay cannot see deletion. Alerts can. Create an alert on the indicator’s signal condition and let it run live for a week; every alert is a timestamped record of what the indicator claimed at the time. Then line the alert log up against the reloaded chart. A signal in the log with no marker on the chart is a deleted object. A marker with no alert is a signal that was not available live. Either one is disqualifying. The same log catches mechanism 1: an alert set to “once per bar close” versus “once per bar” fires at different moments for a script that evaluates on unconfirmed bars.

Test 3 — Read the code, if you can

If the script is open source, you do not need to understand Pine to search it. Five strings tell most of the story:

If the script is invite-only or protected, you cannot run this test — which is itself information; the burden of proof shifts to the other three tests and the vendor’s documentation.

Test 4 — The higher-timeframe reload

For any indicator that references a higher timeframe, screenshot the live chart mid-hour, mid-session. Reload and screenshot again. If the reloaded higher-timeframe line is smoother, or “knew” the hourly close earlier than the live version did, the indicator is two different products: the one you trade with and the one you evaluate.

The subtler problem: statistics that repaint

Everything above is about drawings. The version of the same fault that costs more money is in the numbers, and our own research has produced it — twice — before catching it. A statistics panel repaints when the probability it displays was computed with information that was not available when the setup appeared. The drawing is fine; the measurement peeked. Two examples from our published studies:

The VWAP “almost always reverts” number. In our study of session-anchored VWAP on XAU/USD — 830 days, 6,945 ±2σ band touches — the day’s first ±2σ touch reverted to VWAP before the end of the day 97.6% of the time. That number is true and nearly useless, because about 98% of first touches land in the Asian morning, right after the bands stabilize, with the whole day left to revert. Across all 6,945 touches the before-EOD reversion rate is 72%, the median wait is 115 minutes, only 23% revert within 60 minutes, and in the New York session the before-EOD rate falls to 30%. A panel showing the 97.6% is not lying. It is measuring with hindsight — “before end of day” flatters any early touch — and the honest number depends on the session. The full breakdown is in the session-anchored VWAP guide.

The Fair Value Gap “respect rate.” Our pre-registered FVG study initially found that price “respected” a gap at first touch 85.6% of the time. The audit found the artifact: the touching bar’s own approach path was being counted as the bounce — the statistical equivalent of plotting a pivot on the bar of the pivot. Corrected, with the touching bar excluded, respect at first touch is 57.3% for FVGs against 56.8% for placebo zones of the same geometry. The 85.6% never made it into a product, and both numbers are published side by side on our research page.

When a vendor shows you a fill rate or a reversal probability, the question is not whether the number is real. It is: at what moment was the event defined, and what did the definition know? In our sweep study — 2,434 first sweeps of eight level families across 807 trading days — a level only becomes sweepable after its formation period ends, and only the first sweep per period counts. Those rules are the statistical version of barstate.isconfirmed: they stop the study from “sweeping” a London high while London is still being built. The baseline came out near 50% — a coin flip — exactly the result a hindsight-friendly definition would have hidden; see Do Liquidity Sweeps Really Reverse?.

What to ask a vendor before paying

These questions take five minutes and tell you whether the afternoon is worth spending:

1. “On which bar is each signal finalized?” The only acceptable answer names a bar — “on the close of the confirming bar,” “N bars after the pivot.”
2. “Which higher-timeframe values does it request, and how?” A vendor who cannot say whether the live bar sees the developing or the confirmed value has not checked.
3. “Does it ever delete or move an object?” If yes: what, when, and is the deleted signal recoverable from an alert log?
4. “For any displayed probability: what is the sample size, and when was the event defined?” A percentage without an n is an opinion with a decimal point; one with an n but no definition may be a repainting statistic.
5. “Can I run bar replay on it before I pay?” A free tier, a trial or an open-source lite version is the only honest way to let you run Test 1. As we argued in Free vs Paid TradingView Indicators, the non-repainting check is the single most important purchase criterion in this market, and a vendor who makes it hard to perform is telling you something.

How we handle it — and how to check us

Every indicator in the Delta Terminal suite — 18 today, more added every month, developed hand-in-hand with a full-time Gold day trader — documents where each value is finalized. Levels are confirmed after their formation period closes. Sweep Outcome Stats measures outcomes on your own loaded history with the study’s definitions and prints the confidence interval and the n next to the rate, on the chart, because a rate without its uncertainty can mislead without ever technically repainting.

None of that should be taken on our word. The SLR Dashboard Lite is free, and our three free scripts — Session Killzones ICT - Clean, Adaptive Moving Average - Kalman and Gold-DXY Correlation - Light — are published on our TradingView profile. Run Tests 1 through 4 on the free tier before you spend anything on the paid one.

The short version

“Non-repainting” is a claim, not a property, and it is verified by you, not by the vendor. Four mechanisms rewrite history — unconfirmed-bar logic, lookahead on higher-timeframe data, back-shifted structure, mutable objects — and bar replay, an alert log, a code search and a higher-timeframe reload expose them. The same discipline applies to any probability printed on your chart — ask when the event was defined and what it was allowed to know, because the most expensive repainting in this market is not in the drawings.

Delta Terminal ResearchStudies run on 3.5 years of XAU/USD spot data, developed hand-in-hand with a full-time Gold day trader. We publish the negative results too.