OPEN-SOURCE SCRIPT
Zaktualizowano RSI Memory Attribution - Relative Strength Index
The same RSI value can come from very different price histories.
RSI Memory Attribution keeps the standard Relative Strength Index intact and adds a way to examine the weighted gains and losses behind its current reading. It separates recent from older contributions, tests dependence on individual changes, and measures how much the order of those changes affects RSI.
The purpose is to make an RSI reading explainable without replacing it with a composite score. This is an oscillator and a research tool, not a trading strategy or a forecast of the next price move.
What this adds to a standard RSI
The script connects three questions about the same RSI calculation:
How much of the reading comes from recent changes versus older changes?
Could removing one recent weighted contribution change its threshold classification?
How different could RSI be if the same recent changes had occurred in another order?
Concentration, contribution age, completed episodes, and a prior-term entry study provide additional context for those questions. These modules do not average several RSI periods, detect price/RSI divergence, or combine unrelated indicators into a buy/sell score. The standard RSI formula and the basic concentration formula are established tools; the additional functionality is their application to the attribution and sensitivity of one RSI state.
Reading the default display
Cyan line: The built-in RSI, on its usual 0-100 scale.
Violet line: The older-contribution base, defined as 50 plus the older contribution. It is not an independently calculated RSI for an older window.
Mint/orange ribbon: The signed contribution from the most recent H source changes. Mint means those changes add RSI points to the older base; orange means they subtract points.
Gold shading: The REMOVE ONE range. It includes the actual RSI and the values obtained by deleting any one eligible recent weighted contribution. This is not a volatility band or a confidence interval. Optional boundary lines are available in Appearance.
Pink ORDER RANGE values: The lowest and highest RSI under the latest-snapshot order test. These appear in the readout, not as a projected path.
The five-row readout defaults to Overview. Use Readout page in the settings to access Order and concentration, Episodes, Entry study, Diagnostics, or Legend. Colors, text size, position, line width, and visual layers are adjustable. Markers and the strip near 98 are off by default.
At the default thresholds, UPPER means RSI is at or above 70, LOWER means at or below 30, and IN RANGE means between them. These describe oscillator levels, not whether an asset is fundamentally expensive or cheap.
1. Recent and older attribution
Let N be the RSI length, a = 1/N, and q = 1-a. Let G and L be Wilder-smoothed gains and losses, with T = G+L.
After H consecutive valid updates:
Older gains = q^H * G from H bars ago.
Older losses = q^H * L from H bars ago.
Recent gains = current G minus older gains.
Recent losses = current L minus older losses.
Both contributions use the same current denominator:
Recent contribution = 50 * (recent gains - recent losses) / T.
Older contribution = 50 * (older gains - older losses) / T.
RSI = 50 + recent contribution + older contribution.
MEMORY displays the contributions as R and O, in RSI points. The violet base equals 50+O. H partitions the existing RSI state; it does not change the RSI length N. The older part includes the residual influence of the initialized smoothing state.
For example, R = -4.53 and O = -10.87 give RSI = 50 - 4.53 - 10.87 = 34.60. Both cohorts contribute below the midpoint. This explains the composition of that reading without predicting what follows. Separately rounded values may differ slightly when added together.
2. REMOVE ONE: dependence on one weighted change
The full test considers ages 0 through H-1. Age 0 is the latest displayed source change. Each candidate removes exactly one term; the script does not remove all H terms together.
Removing a gain contribution can only reduce or leave RSI unchanged. Removing a loss contribution can only increase or leave RSI unchanged. The largest eligible weighted gain and loss therefore determine the two extremes. The actual, unchanged RSI is also included in the range.
Removal means subtracting one contribution from the current smoothed state while preserving all other weights. It does not delete a candle, reconnect adjacent prices, or reconstruct an alternative OHLC chart.
An asterisk beside REMOVE ONE means that at least one candidate changes the UPPER / IN RANGE / LOWER classification. For an illustrative RSI of 34.6 with a range of 28.6-38.5, some removal can move the calculation below the default lower threshold. The range does not say that RSI or price will move there next. No asterisk means unchanged classification under this particular test, not a reliable trade.
The test displays n/a when no nonzero candidate exists or a candidate leaves an unstable residual denominator. An unavailable test is not classified as resilient.
3. ORDER RANGE: same changes, different sequence
The order test holds the older smoothed state fixed and keeps every signed change in the recent H-change collection. It changes only which recent weight is assigned to each change.
Assigning the largest signed changes to the largest recency weights maximizes the gain component while minimizing the loss component. Reversing this arrangement gives the opposite extreme. Two sorted arrangements produce the bounds without enumerating every permutation.
The sum of the changes stays the same, so the start and end source values remain unchanged. Intermediate rearranged paths need not be economically feasible. This is an arithmetic sensitivity test, not a model of likely future paths.
ORDER EFFECT compares actual RSI with a baseline that divides the recent weight equally among the H changes. That baseline is not the average RSI across random permutations, because RSI is a nonlinear ratio.
POS is the actual reading's location from 0 to 100 inside the order range. It is not a percentile or probability. A zero-width range has no defined position.
Order and concentration values describe only the latest displayed snapshot. Moving the cursor to an older bar does not turn these readout values into historical measurements. They are not copied into past bars or used to assign historical entry-study groups.
4. Concentration and contribution age
For each recent term, m is its absolute source change multiplied by its RSI recency weight.
EQ = (sum of m)^2 / sum of m^2.
EQ describes how many equal-sized contributions would have the same concentration. A value near 1 means recent weighted movement is concentrated in very few terms. A value near H means the weighted contributions are more evenly distributed. It is not a count of statistically independent observations or a confidence score.
AGE is the movement-weighted mean age within the recent cohort, with the newest change at age 0. TOP is the largest recent weighted term as a percentage of recent weighted movement, not total RSI memory. Separate gain and loss ages are available; a side without movement has no defined age.
These values help distinguish a reading dominated by a few changes from one supported by a broader recent sequence. They do not rank the quality of a potential trade.
5. Completed extreme episodes
An episode is a consecutive sequence of confirmed readings on the same extreme side. Duration is measured in bars. A is the accumulated excess beyond the relevant threshold, in RSI-point bars.
Upper and lower completed episodes are stored separately. The Episodes page shows duration medians and the current episode's age and area ranks against retained, completed episodes from the same side. Ties receive half weight in the ranks.
An episode already active when valid sampling begins is marked with a plus sign. Its start was not observed, so it is excluded from completed pools. Missing valid observations or sample boundaries interrupt an episode rather than count it as complete. An active episode is not a completed observation.
The default retains 60 completions per side and requires 20 observations for a summary. Both are adjustable. These completed-only comparisons are not estimates of remaining duration, survival probabilities, or reversal odds.
6. Entry study: prior terms, not the triggering change
Here, entry means entering an RSI threshold region, not opening a trade.
This study deliberately uses a different candidate set from REMOVE ONE. At a new confirmed threshold entry, it holds age 0 fixed and considers deleting one prior term aged 1 through H-1. It therefore requires H of at least 2.
PRIOR-RESILIENT: None of the tested prior-term deletions changes the entry classification.
PRIOR-SENSITIVE: At least one tested prior-term deletion changes it.
These labels concern the RSI calculation, not future price resilience. They are assigned at the entry close without using later outcomes.
The triggering change must be excluded from this comparison. Removing age 0 restores the previous RSI ratio. At a new threshold crossing, including that change would therefore make every evaluable entry sensitive. The prior-only test avoids that structural problem, but either group can still have few or no observations on a particular chart.
K closes after an accepted entry, the script records:
E / END: RSI is on the same extreme side at the endpoint.
R / RUN: RSI remained on that side at every intervening close, including the endpoint.
An episode can end and later restart, satisfying END without satisfying RUN. These percentages describe observed RSI persistence, not profitable trades or a strategy win rate.
Upper and lower groups are reported separately. POOLED BASELINE combines the same selected, classified, matured observations for the chosen side. It is not an independent control group or all market entries.
By default, the earliest classifiable entry is accepted and overlapping entries through its +K bar are skipped across both sides. This choice does not use the eventual outcome. Non-overlap can be disabled for research; it does not establish statistical independence.
SAMPLE STATUS reports the chosen side:
P: Accepted entry whose outcome horizon is still pending.
D: Accepted entry whose matured window includes invalid, missing, or out-of-date observations.
X: Entry with no classifiable prior test, an unstable residual, or a numerical threshold-boundary ambiguity.
O: Otherwise classifiable entry skipped by the non-overlap rule.
These exclusions are not counted as observed failures. Counts remain visible when percentages are withheld below the minimum sample. Event-study counts accumulate over eligible loaded history; the 60-completion setting limits episode pools, not these counts.
Group differences can reflect entry overshoot, prior path, instrument, timeframe, and chosen parameters. Selecting settings after viewing favorable results introduces selection bias. The study does not establish causation, statistical significance, out-of-sample performance, or a trading advantage.
7. A practical research workflow
Begin with standard candlesticks and the default Close source, RSI length 14, H = 7, and thresholds 70/30. H is an attribution horizon, not a second oscillator period.
Read RSI and MEMORY first. Check whether recent and older contributions agree or oppose. Then compare REMOVE ONE with ORDER RANGE: the first changes the membership of the weighted calculation; the second retains the same changes and varies their order.
Use Order and concentration for contribution age and concentration, Episodes for completed-run comparisons, and Entry study for prior-term groups. Check sample counts before interpreting percentages. n/a can be a correct result when data or valid candidates are insufficient.
Material contribution, default 2 RSI points, controls cohort-state classification rather than the RSI formula. The optional width marker uses a separate default threshold of 3 RSI points. The entry-study horizon defaults to 5 closes. These defaults are research settings, not optimized trading parameters.
8. Optional marks and alerts
A gold circle marks a confirmed transition into full REMOVE ONE classification sensitivity. A pink square marks the removal width crossing its configured threshold. These are different events. With Both selected, the circle takes priority on the same bar; the separate alert conditions still exist.
The optional strip at 98 shows cohort relationships: mint for both materially positive, orange for both materially negative, gold for material opposition, violet for one material cohort, and gray for small or mixed contributions. It is categorical, not another RSI value.
Seven close-confirmed alert conditions cover upper and lower threshold entries and exits, material cohort opposition, the appearance of full-test classification sensitivity, and a removal-width threshold crossing. Create the desired alert separately using TradingView's alert dialog. Native threshold alerts are independent of research date filters. These alerts are not order instructions or entry-study outcome predictions.
9. Live behavior, history, and diagnostics
By default, the open bar displays the previous closed RSI and research snapshot. Historical readings stay on their original bars. Turning this option off permits intrabar changes. Episode records, entry-study classifications and outcomes, and alert conditions still commit only at bar close.
A live built-in RSI can therefore differ from this script's default held reading on the current open bar. Compare confirmed bars or use matching live settings when checking values.
Research requires a standard chart, valid consecutive updates, and positive weighted movement. It waits N times the settling setting plus H changes: 147 at the defaults. Those changes can already exist in loaded history. Native RSI is not suppressed while research is settling. The settling convention does not prove convergence to an infinite-history value.
The script uses available chart history without a fixed 6,000-bar execution cap. Changing the dataset start, source, settings, or supplied data can change initialization and statistics. Dates filter close-time observations; they do not load missing history or reset RSI. Ages, durations, and K count bars rather than elapsed clock time.
External indicator sources can repaint upstream. Bar-close confirmation does not prevent source revisions or data-feed corrections, so no absolute non-repainting claim is made.
Diagnostics displays the RSI identity error, direct recent-sum error, full/prior removal comparisons, and whether sorted order bounds contain the actual RSI. Numeric errors show nine decimals and use a tolerance of 0.000001 in the row's stated units. The direct scans are latest-snapshot checks, not certification of every historical bar. n/a is not a pass, and a zero-width order range can be valid without a defined POS.
10. Open-source scope
The source is published under Mozilla Public License 2.0 for inspection and further research. All implemented features are included without an author-side access fee or external service requirement.
The script does not request external market series, import libraries, backdate pivot signals, place orders, or project the order range into future bars. Its historical engine uses recurrence calculations and rolling built-ins; H-term scans and sorting are limited to the latest snapshot.
This publication provides descriptive indicator research. It does not claim a new underlying RSI formula, universal originality, predictive accuracy, guaranteed moderation status, or profitable execution. Its additional value is making the composition and sensitivity of an otherwise familiar RSI reading visible and inspectable.
RSI Memory Attribution keeps the standard Relative Strength Index intact and adds a way to examine the weighted gains and losses behind its current reading. It separates recent from older contributions, tests dependence on individual changes, and measures how much the order of those changes affects RSI.
The purpose is to make an RSI reading explainable without replacing it with a composite score. This is an oscillator and a research tool, not a trading strategy or a forecast of the next price move.
What this adds to a standard RSI
The script connects three questions about the same RSI calculation:
How much of the reading comes from recent changes versus older changes?
Could removing one recent weighted contribution change its threshold classification?
How different could RSI be if the same recent changes had occurred in another order?
Concentration, contribution age, completed episodes, and a prior-term entry study provide additional context for those questions. These modules do not average several RSI periods, detect price/RSI divergence, or combine unrelated indicators into a buy/sell score. The standard RSI formula and the basic concentration formula are established tools; the additional functionality is their application to the attribution and sensitivity of one RSI state.
Reading the default display
Cyan line: The built-in RSI, on its usual 0-100 scale.
Violet line: The older-contribution base, defined as 50 plus the older contribution. It is not an independently calculated RSI for an older window.
Mint/orange ribbon: The signed contribution from the most recent H source changes. Mint means those changes add RSI points to the older base; orange means they subtract points.
Gold shading: The REMOVE ONE range. It includes the actual RSI and the values obtained by deleting any one eligible recent weighted contribution. This is not a volatility band or a confidence interval. Optional boundary lines are available in Appearance.
Pink ORDER RANGE values: The lowest and highest RSI under the latest-snapshot order test. These appear in the readout, not as a projected path.
The five-row readout defaults to Overview. Use Readout page in the settings to access Order and concentration, Episodes, Entry study, Diagnostics, or Legend. Colors, text size, position, line width, and visual layers are adjustable. Markers and the strip near 98 are off by default.
At the default thresholds, UPPER means RSI is at or above 70, LOWER means at or below 30, and IN RANGE means between them. These describe oscillator levels, not whether an asset is fundamentally expensive or cheap.
1. Recent and older attribution
Let N be the RSI length, a = 1/N, and q = 1-a. Let G and L be Wilder-smoothed gains and losses, with T = G+L.
After H consecutive valid updates:
Older gains = q^H * G from H bars ago.
Older losses = q^H * L from H bars ago.
Recent gains = current G minus older gains.
Recent losses = current L minus older losses.
Both contributions use the same current denominator:
Recent contribution = 50 * (recent gains - recent losses) / T.
Older contribution = 50 * (older gains - older losses) / T.
RSI = 50 + recent contribution + older contribution.
MEMORY displays the contributions as R and O, in RSI points. The violet base equals 50+O. H partitions the existing RSI state; it does not change the RSI length N. The older part includes the residual influence of the initialized smoothing state.
For example, R = -4.53 and O = -10.87 give RSI = 50 - 4.53 - 10.87 = 34.60. Both cohorts contribute below the midpoint. This explains the composition of that reading without predicting what follows. Separately rounded values may differ slightly when added together.
2. REMOVE ONE: dependence on one weighted change
The full test considers ages 0 through H-1. Age 0 is the latest displayed source change. Each candidate removes exactly one term; the script does not remove all H terms together.
Removing a gain contribution can only reduce or leave RSI unchanged. Removing a loss contribution can only increase or leave RSI unchanged. The largest eligible weighted gain and loss therefore determine the two extremes. The actual, unchanged RSI is also included in the range.
Removal means subtracting one contribution from the current smoothed state while preserving all other weights. It does not delete a candle, reconnect adjacent prices, or reconstruct an alternative OHLC chart.
An asterisk beside REMOVE ONE means that at least one candidate changes the UPPER / IN RANGE / LOWER classification. For an illustrative RSI of 34.6 with a range of 28.6-38.5, some removal can move the calculation below the default lower threshold. The range does not say that RSI or price will move there next. No asterisk means unchanged classification under this particular test, not a reliable trade.
The test displays n/a when no nonzero candidate exists or a candidate leaves an unstable residual denominator. An unavailable test is not classified as resilient.
3. ORDER RANGE: same changes, different sequence
The order test holds the older smoothed state fixed and keeps every signed change in the recent H-change collection. It changes only which recent weight is assigned to each change.
Assigning the largest signed changes to the largest recency weights maximizes the gain component while minimizing the loss component. Reversing this arrangement gives the opposite extreme. Two sorted arrangements produce the bounds without enumerating every permutation.
The sum of the changes stays the same, so the start and end source values remain unchanged. Intermediate rearranged paths need not be economically feasible. This is an arithmetic sensitivity test, not a model of likely future paths.
ORDER EFFECT compares actual RSI with a baseline that divides the recent weight equally among the H changes. That baseline is not the average RSI across random permutations, because RSI is a nonlinear ratio.
POS is the actual reading's location from 0 to 100 inside the order range. It is not a percentile or probability. A zero-width range has no defined position.
Order and concentration values describe only the latest displayed snapshot. Moving the cursor to an older bar does not turn these readout values into historical measurements. They are not copied into past bars or used to assign historical entry-study groups.
4. Concentration and contribution age
For each recent term, m is its absolute source change multiplied by its RSI recency weight.
EQ = (sum of m)^2 / sum of m^2.
EQ describes how many equal-sized contributions would have the same concentration. A value near 1 means recent weighted movement is concentrated in very few terms. A value near H means the weighted contributions are more evenly distributed. It is not a count of statistically independent observations or a confidence score.
AGE is the movement-weighted mean age within the recent cohort, with the newest change at age 0. TOP is the largest recent weighted term as a percentage of recent weighted movement, not total RSI memory. Separate gain and loss ages are available; a side without movement has no defined age.
These values help distinguish a reading dominated by a few changes from one supported by a broader recent sequence. They do not rank the quality of a potential trade.
5. Completed extreme episodes
An episode is a consecutive sequence of confirmed readings on the same extreme side. Duration is measured in bars. A is the accumulated excess beyond the relevant threshold, in RSI-point bars.
Upper and lower completed episodes are stored separately. The Episodes page shows duration medians and the current episode's age and area ranks against retained, completed episodes from the same side. Ties receive half weight in the ranks.
An episode already active when valid sampling begins is marked with a plus sign. Its start was not observed, so it is excluded from completed pools. Missing valid observations or sample boundaries interrupt an episode rather than count it as complete. An active episode is not a completed observation.
The default retains 60 completions per side and requires 20 observations for a summary. Both are adjustable. These completed-only comparisons are not estimates of remaining duration, survival probabilities, or reversal odds.
6. Entry study: prior terms, not the triggering change
Here, entry means entering an RSI threshold region, not opening a trade.
This study deliberately uses a different candidate set from REMOVE ONE. At a new confirmed threshold entry, it holds age 0 fixed and considers deleting one prior term aged 1 through H-1. It therefore requires H of at least 2.
PRIOR-RESILIENT: None of the tested prior-term deletions changes the entry classification.
PRIOR-SENSITIVE: At least one tested prior-term deletion changes it.
These labels concern the RSI calculation, not future price resilience. They are assigned at the entry close without using later outcomes.
The triggering change must be excluded from this comparison. Removing age 0 restores the previous RSI ratio. At a new threshold crossing, including that change would therefore make every evaluable entry sensitive. The prior-only test avoids that structural problem, but either group can still have few or no observations on a particular chart.
K closes after an accepted entry, the script records:
E / END: RSI is on the same extreme side at the endpoint.
R / RUN: RSI remained on that side at every intervening close, including the endpoint.
An episode can end and later restart, satisfying END without satisfying RUN. These percentages describe observed RSI persistence, not profitable trades or a strategy win rate.
Upper and lower groups are reported separately. POOLED BASELINE combines the same selected, classified, matured observations for the chosen side. It is not an independent control group or all market entries.
By default, the earliest classifiable entry is accepted and overlapping entries through its +K bar are skipped across both sides. This choice does not use the eventual outcome. Non-overlap can be disabled for research; it does not establish statistical independence.
SAMPLE STATUS reports the chosen side:
P: Accepted entry whose outcome horizon is still pending.
D: Accepted entry whose matured window includes invalid, missing, or out-of-date observations.
X: Entry with no classifiable prior test, an unstable residual, or a numerical threshold-boundary ambiguity.
O: Otherwise classifiable entry skipped by the non-overlap rule.
These exclusions are not counted as observed failures. Counts remain visible when percentages are withheld below the minimum sample. Event-study counts accumulate over eligible loaded history; the 60-completion setting limits episode pools, not these counts.
Group differences can reflect entry overshoot, prior path, instrument, timeframe, and chosen parameters. Selecting settings after viewing favorable results introduces selection bias. The study does not establish causation, statistical significance, out-of-sample performance, or a trading advantage.
7. A practical research workflow
Begin with standard candlesticks and the default Close source, RSI length 14, H = 7, and thresholds 70/30. H is an attribution horizon, not a second oscillator period.
Read RSI and MEMORY first. Check whether recent and older contributions agree or oppose. Then compare REMOVE ONE with ORDER RANGE: the first changes the membership of the weighted calculation; the second retains the same changes and varies their order.
Use Order and concentration for contribution age and concentration, Episodes for completed-run comparisons, and Entry study for prior-term groups. Check sample counts before interpreting percentages. n/a can be a correct result when data or valid candidates are insufficient.
Material contribution, default 2 RSI points, controls cohort-state classification rather than the RSI formula. The optional width marker uses a separate default threshold of 3 RSI points. The entry-study horizon defaults to 5 closes. These defaults are research settings, not optimized trading parameters.
8. Optional marks and alerts
A gold circle marks a confirmed transition into full REMOVE ONE classification sensitivity. A pink square marks the removal width crossing its configured threshold. These are different events. With Both selected, the circle takes priority on the same bar; the separate alert conditions still exist.
The optional strip at 98 shows cohort relationships: mint for both materially positive, orange for both materially negative, gold for material opposition, violet for one material cohort, and gray for small or mixed contributions. It is categorical, not another RSI value.
Seven close-confirmed alert conditions cover upper and lower threshold entries and exits, material cohort opposition, the appearance of full-test classification sensitivity, and a removal-width threshold crossing. Create the desired alert separately using TradingView's alert dialog. Native threshold alerts are independent of research date filters. These alerts are not order instructions or entry-study outcome predictions.
9. Live behavior, history, and diagnostics
By default, the open bar displays the previous closed RSI and research snapshot. Historical readings stay on their original bars. Turning this option off permits intrabar changes. Episode records, entry-study classifications and outcomes, and alert conditions still commit only at bar close.
A live built-in RSI can therefore differ from this script's default held reading on the current open bar. Compare confirmed bars or use matching live settings when checking values.
Research requires a standard chart, valid consecutive updates, and positive weighted movement. It waits N times the settling setting plus H changes: 147 at the defaults. Those changes can already exist in loaded history. Native RSI is not suppressed while research is settling. The settling convention does not prove convergence to an infinite-history value.
The script uses available chart history without a fixed 6,000-bar execution cap. Changing the dataset start, source, settings, or supplied data can change initialization and statistics. Dates filter close-time observations; they do not load missing history or reset RSI. Ages, durations, and K count bars rather than elapsed clock time.
External indicator sources can repaint upstream. Bar-close confirmation does not prevent source revisions or data-feed corrections, so no absolute non-repainting claim is made.
Diagnostics displays the RSI identity error, direct recent-sum error, full/prior removal comparisons, and whether sorted order bounds contain the actual RSI. Numeric errors show nine decimals and use a tolerance of 0.000001 in the row's stated units. The direct scans are latest-snapshot checks, not certification of every historical bar. n/a is not a pass, and a zero-width order range can be valid without a defined POS.
10. Open-source scope
The source is published under Mozilla Public License 2.0 for inspection and further research. All implemented features are included without an author-side access fee or external service requirement.
The script does not request external market series, import libraries, backdate pivot signals, place orders, or project the order range into future bars. Its historical engine uses recurrence calculations and rolling built-ins; H-term scans and sorting are limited to the latest snapshot.
This publication provides descriptive indicator research. It does not claim a new underlying RSI formula, universal originality, predictive accuracy, guaranteed moderation status, or profitable execution. Its additional value is making the composition and sensitivity of an otherwise familiar RSI reading visible and inspectable.
Informacje o Wersji
v6.1.2 - Clearer attribution and plot-limit fixThis update clarifies attribution and fixes "The script creates too many plots." Standard RSI is unchanged.
Attribution
Set Readout page to Memory attribution. Rows separate OLDER (ages H+), PRIOR (1 to H-1) and LATEST (0) from the displayed observation; H is Recent cohort length. NET RSI pt is each group's signed contribution:
RSI = 50 + OLDER + PRIOR + LATEST.
All use one weighted gain-plus-loss denominator. MOVE % uses weighted gains PLUS losses and sums to 100% before rounding. Small net contributions can hide offsetting movement. These are not volume shares or probabilities; LATEST is not the actual RSI CHANGE. Bars show NET points only.
Sensitivity
ZERO n-BAR-AGO TERM / RSI AFTER shows current RSI after zeroing that increment, not RSI n bars ago. Prior-only tests fix age 0; the full test includes it. Term ledger ranks individual effects; Single-term stress shows the fraction needed to touch the reference. Above 100% means insufficient. No candle or timestamp is removed; effects are not additive or forecasts.
Events and research
Confirmed research markers and alerts share rolling checks, independent of latest-only display checks. Entry labels and estimates are frozen at entry; estimates use earlier completed outcomes. Comparisons measure RSI persistence, not trade win rates.
Plot-limit fix
36 Data Window exports now use Research 1-8. Select Data Window group; Data Window guide identifies slots. The 36 values are no longer all exported at once. Remap external inputs/CSV columns. Chart analysis and all 11 alert conditions are retained from v6.1.1. Recreate alerts to use this version.
Closed mode holds the last confirmed readout; Live changes intrabar. Data, settings or history revisions can alter results. RSI follows Wilder; the optional next-source map uses established RSI inversion (Giorgos Siligardos, 2003), not a forecast. No profitability or statistical-significance claim.
Skrypt open-source
W zgodzie z duchem TradingView twórca tego skryptu udostępnił go jako open-source, aby użytkownicy mogli przejrzeć i zweryfikować jego działanie. Ukłony dla autora. Korzystanie jest bezpłatne, jednak ponowna publikacja kodu podlega naszym Zasadom serwisu.
Wyłączenie odpowiedzialności
Informacje i publikacje nie stanowią i nie powinny być traktowane jako porady finansowe, inwestycyjne, tradingowe ani jakiekolwiek inne rekomendacje dostarczane lub zatwierdzone przez TradingView. Więcej informacji znajduje się w Warunkach użytkowania.
Skrypt open-source
W zgodzie z duchem TradingView twórca tego skryptu udostępnił go jako open-source, aby użytkownicy mogli przejrzeć i zweryfikować jego działanie. Ukłony dla autora. Korzystanie jest bezpłatne, jednak ponowna publikacja kodu podlega naszym Zasadom serwisu.
Wyłączenie odpowiedzialności
Informacje i publikacje nie stanowią i nie powinny być traktowane jako porady finansowe, inwestycyjne, tradingowe ani jakiekolwiek inne rekomendacje dostarczane lub zatwierdzone przez TradingView. Więcej informacji znajduje się w Warunkach użytkowania.