A label such as HELD @10 looks simple. A structure break happened, ten confirmed bars passed, and price closed on one side of the old level. The danger begins when that compact observation is stretched into a claim it never made: that price held continuously, that a trade won, or that the same result is likely next time.
A fixed-bar endpoint is useful precisely because its meaning can be stated narrowly. It answers one reproducible question at one disclosed time. It does not describe everything that happened between the break and that endpoint, and it does not define a trading result.
This distinction applies well beyond one indicator. Whenever a chart tool reports what happened after N bars, read three things separately: the event that started the clock, the path price took during the window, and the single observation made at the end.
Three different objects: event, path, endpoint
The first object is the event definition. Structure Participation Matrix does not begin its count whenever price merely touches or wicks through a level. A swing must first become eligible after its full confirmation delay. A break is then recorded only when a confirmed close moves beyond the armed swing level plus the selected ATR buffer.
That rule establishes bar zero. Change the swing strength, buffer, symbol, timeframe, or available history and you may change which bars qualify as events. Before comparing any later percentages, make sure the starting event means the same thing in both samples.
The second object is the path. Price can continue directly, return through the level, oscillate around it, and move back again before the endpoint arrives. Two breaks can finish on the same side of a level at bar ten while taking very different routes to get there. The endpoint label does not preserve that route.
The third object is the endpoint observation. After exactly N confirmed bars, the script compares that bar’s close with the broken level once. For an upward break, a close above the level is HELD; otherwise it is FAILED. The comparison is reversed for a downward break. The result is then fixed.
These three objects should not be collapsed into one story. The event tells you what started the measurement, the path tells you what happened in between, and the endpoint tells you where one later close finished relative to the level.
What HELD @N actually says
HELD @10 means: at the close of the tenth confirmed bar after the recorded break, price was still on the broken side of the level. It does not mean that every one of those ten closes stayed there. It says nothing about the highest high, lowest low, maximum excursion, drawdown, volatility, or the order in which crossings occurred.
FAILED @10 is equally narrow. It means the endpoint close was not on the broken side. It is not a stop-out, a losing trade, or proof that the original structural observation had no descriptive value.
Neither label contains an entry price, position direction, stop, target, position size, fees, or exit rule. Without those definitions there is no trade to score. Calling HELD a win or FAILED a loss silently adds a strategy that the measurement never specified.
The endpoint can still support disciplined chart study. Because N and the comparison are fixed in advance, the tool does not search the future for a convenient high, low, or later bar that makes the event look favourable. That removes one common form of after-the-fact selection. It does not turn the endpoint into performance evidence.
A fixed horizon does not remove path dependence
Suppose two upward breaks are both marked HELD @10. The first closes above the level on every bar. The second falls below the level for several bars and recovers only before the tenth close. The endpoint records are identical, but the paths are not.
That matters whenever the question involves risk, execution, or how a move developed. A trader using an unstated stop could experience different results across those paths. A chart reader studying acceptance around the level would also want to see the intervening candles. The endpoint alone cannot answer either question.
The honest use is to treat HELD or FAILED as an index into further inspection. Freeze the same chart context, then look at the bars between the event and endpoint. If the path matters to the question, measure it explicitly instead of assuming the endpoint contains it.
Pending and UNSCORED records change the denominator
Not every visible break belongs in a resolved score bucket immediately. A new event remains pending until N confirmed bars exist. Counting it as HELD or FAILED early would use information the method has not observed yet, so it stays outside the resolved total.
Some events remain UNSCORED. Structure Participation Matrix freezes four disclosed readings at the break close: PATH, CLOSE, REL VOL, and BALANCE. The composite uses weights of 30%, 25%, 25%, and 20%. If required chart volume is missing, or the structural leg exceeds the bounded measurement history, the break remains visible but does not receive a score bucket. It is excluded from the bucket statistics rather than being converted into a zero.
This exclusion is a form of selection that should be visible, not hidden. The displayed bucket rate describes resolved, scored events. It does not describe every break that appeared on the chart. Check the pending state and UNSCORED count before interpreting the denominator.
The volume limitation matters too. REL VOL uses volume reported by the chart feed, which may be tick volume. BALANCE weights that volume by close location as an OHLCV proxy. Neither field is bid/ask delta, footprint data, or evidence of who participated.
Sample-aware is better, but still historical
A minimum-sample guard prevents a bucket from printing a percentage while it contains too few resolved examples. Until that threshold is reached, the matrix says collecting. This is a valuable restraint: it keeps the first few outcomes from being dressed up as a stable rate.
Passing the threshold does not change the type of claim. The percentage remains a description of resolved events in the history currently loaded on that chart. It is not a forward probability, an accuracy score, or a win rate.
The sample can change when the symbol, timeframe, inputs, or loaded-history boundary changes. More resolved events can also update the current panel while an older screenshot preserves an earlier count. Both may be correct for their own data windows.
For a broader treatment of changing denominators and chart-local frequencies, read How to Read Pullback Base Rates Without Turning Them Into Signals. The recurring rule is simple: keep the event definition, sample, and observation window attached to every rate.
A six-step comparison checklist
- Freeze the setup. Record the symbol, timeframe, relevant inputs, outcome window N, and loaded-history boundary before comparing examples.
- Locate the confirmed event. Identify the break close that actually started the count. Do not substitute a wick, the pivot bar, or a visually convenient later candle.
- Separate the four readings from the outcome. PATH, CLOSE, REL VOL, BALANCE, and the composite describe the leg at the break. HELD or FAILED is added later.
- Inspect the intervening path. Count N confirmed bars and note crossings or adverse movement instead of inferring continuous holding from the endpoint label.
- Include unfavourable and unavailable cases. Compare HELD, FAILED, pending, and UNSCORED records. Do not build a conclusion from the cleanest examples alone.
- Report the denominator honestly. State the resolved count, minimum-sample status, and exclusions. Describe the result as chart-history evidence, not a forecast or trading-performance statistic.
The full Structure Participation Matrix guide explains the confirmed-break record and disclosed components. Its open-source TradingView publication lets you inspect the rules directly.
FAQ
Does HELD @10 mean price stayed above the level for ten bars?
No. For an upward break, it means only that the tenth confirmed bar closed above the broken level. Price may have crossed below it during the interval.
Is FAILED @N a losing trade?
No. FAILED is an endpoint comparison. The indicator does not define an entry, stop, target, size, fees, or exit rule, so the label is not a trade result.
Why exclude UNSCORED events?
The score requires sufficient measured-leg history and valid chart-reported volume. When those inputs are unavailable, showing the break while excluding it from score-bucket statistics is more honest than inventing a value.
Does a larger sample make the rate predictive?
No. A larger resolved sample can make the historical description less dependent on a handful of events, but it does not establish future probability or profitability.
What should I inspect after reading the endpoint?
Review the intervening candles, the event’s frozen component mix, the scored or UNSCORED state, and the sample definition. Measure any path property you care about separately.