The 14-day Relative Strength Index (RSI) is a momentum oscillator bound between 0 and 100 that quantifies velocity and directional price persistence. While conventional technical analysis evaluates static RSI levels (overbought above 70, oversold below 30), this engine tracks the inter-session rate of change alongside the ending baseline level.
ΔRSI = RSItoday − RSIyesterday
RSI State = f(RSItoday, ΔRSI)the session's read depends on both the ending level and the one-day change
Two readings describe the event, and neither is sufficient on its own:
Current RSI level: where the oscillator landed at today's close.
Single-session move: how many points RSI traveled, and in which direction.
Why one metric alone is insufficient
Evaluating either metric in isolation leads to false equivalences across different volatility and trend regimes:
Both moves are the same size, +15 points. One lifts a stock off its oversold low; the other drives it into overbought. A tool that read only the change would file them together.
Ignoring the level: a +15 point jump from 25 to 40 is a reversal off oversold extremes, whereas the same +15 point jump from 58 to 73 is momentum expansion into overbought territory.
Ignoring the move: a stock that has drifted at RSI 52 for two weeks is a completely different setup from one that surged +3.6 points to reach RSI 52 this afternoon.
Both readings come off the same line: where RSI finished, and how far the last session moved it.
Worked example: NVDA
At the close on 21 July 2026 (charted above), NVDA's 14-day RSI stood at 52.2 after a single-session move of +3.6 points. The prior session's level follows directly:
RSIyesterday = 52.2 − 3.6 = 48.6
NVDA moved from an equilibrium baseline of 48.6 to close at 52.2, a modest expansion that leaves it right around the midpoint of the oscillator.
To establish forward-looking statistical expectations, the engine identifies historical sessions for that specific ticker where both the directional move and the baseline level aligned with today's readings.
The dual-filter isolation mechanism
Strict directional separation: sessions with positive RSI moves match exclusively against positive historical moves. Bullish and bearish momentum shifts are never mixed.
Self-referential volatility scaling: point moves are calibrated to the stock's own historical volatility profile. A 5-point RSI move is routine for a high-beta growth name but a statistical anomaly for a steady, low-volatility one.
Regime level filter: keeps historical sessions whose ending RSI sat near today's level. This stops an oversold reading from being explained by a session that ended deep in overbought.
Impact of the level filter (NVDA)
At the close on 21 July 2026, NVDA registered a +3.6 point RSI increase to close at 52.2. Matching on the move alone is not enough:
Matching on the move alone returns a pool scattered across the whole oscillator. Adding the level filter keeps only the sessions that ended near today's 52.2, a far smaller and closely comparable set.
Matching strategy
Sample size
RSI levels of the matched sessions
Move only (ΔRSI ≈ +3.6)
600+ sessions
scattered from ~28 to ~88
Move + level
54 sessions
50.5 to 53.8, around today's 52.2
Core insight
Screening for move size alone returns over 600 sessions occurring all over the oscillator scale. Adding the baseline level filter isolates the 54 sessions that reflect NVDA's current momentum and baseline structure.
Ticker-specific normalization
RSI momentum cannot be generalized across different equities or different price regimes. A 4-point RSI expansion generates unique historical probabilities for each individual stock, based on its own price history and its own distribution of moves.
Chapter 3
Price Projections
The engine projects across three fixed horizons: the next trading day, one week (5 trading days) and one month (21 trading days). Every card shows the exact future date it refers to, so there is no ambiguity about which session is being measured.
Three measures on each card
Green / red odds: the split between comparable days that closed higher and those that closed lower over that horizon.
Avg green return: the average gain of only the comparable days that closed higher.
Avg red return: the average loss of only the comparable days that closed lower.
NVDA's read after that 3.6 point jump:
Horizon
Target date
Odds
Avg green
Avg red
Next day
22 July 2026
42% / 58%
+2.28%
−2.43%
One week
28 July 2026
58% / 42%
+4.96%
−7.63%
One month
20 August 2026
55% / 45%
+13.97%
−12.04%
Measured on NVDA at the close on 21 July 2026, from 54 comparable days.
Read across the row rather than down the column. The one-week line leans green at 58%, and yet its average losing week is the bigger of the two figures beside it. More winners, larger losers. The odds and the averages answer different questions, and either one on its own would have pointed the wrong way here.
Always read the averages next to the odds
The two average rows are one-sided: avg green averages only the up outcomes, avg red only the down outcomes. Never read them by themselves.
Note also how far apart the three horizons can sit. This reading leans red tomorrow and green at both of the longer horizons, which is normal, and it is a reminder that each card is its own measurement rather than three steps along one path.
Chapter 4
The Three Charts
One Chart View toggle, three ways of looking at the same sample of comparable days: the Price Cone, the Bar Graph and the Chart Log.
1. Price Cone
The default. It projects forward 45 sessions from today's price, drawing the percentile bands of what the comparable days actually did, converted into prices.
Median line (50th percentile): the central outcome of the sample, not a target.
Cone width (10th to 90th): how much the comparable days disagreed. A narrow cone means consistent outcomes; a wide one means high variance.
2. Bar Graph
This asks a different question: how often did price travel a set distance, and how quickly? You choose the move size, and the chart answers in both directions at once. Green bars above the line show the share of comparable days that rose that far within a given number of days; red bars below show the share that fell that far.
By Touch vs By Close
Setting
What counts
Use it for
By Touch
Any intraday high or low that reached the level, even if price came back before the close
Path risk: stop-loss triggers or intraday margin
By Close
Only a daily close beyond the level
Settlement risk: holding-period targets or trend sustainability
Touch numbers are always equal to or higher than close numbers. A wide gap between the two flags heavy intraday movement that did not hold into the close.
3. Chart Log and the event register
The Chart Log shows the full price chart with every comparable day marked, so you can see the wider context each time the same setup appeared. Below it, the event register lists every comparable day with its date, the RSI before and after, the point change, and what price did the next day, one week later and one month later.
Any column sorts, which makes it easy to find the largest gains and worst drawdowns in the sample. Those extremes often matter more than the average when you are sizing a position.
Sorting by Point Change is the quickest way to see whether today's move sits at the gentle end of the sample or the violent end, which the odds alone will not tell you.
The front page runs the same analysis across every ticker at once. Every stock moves its RSI a little every day, so unlike the detection tools in this family the list is long, and sorting is how you use it.
The RSI Move column
One cell, both readings, stacked. The point change on top, colored green when RSI rose and red when it fell, with the level it reached printed underneath. It is the same pair of numbers Chapter 1 described, and the reason the level is there is that the point change alone would not tell you what kind of day it was.
Sorting the column puts the sharpest RSI moves of the day at the top. Sorting it the other way brings up the largest drops.
Price Projections
The same price projections from Chapter 3 appear here: avg green, avg red, and the odds, shown for next day, next week and next month.
Each figure is calculated from that ticker's own comparable days. The numbers are therefore not comparable side to side in the way they look. Two rows both showing 54% are each describing their own stock.