Monthly RSI divergence

Higher price, weaker momentum — measured on monthly closes

On 2026-08-25 the Fraktalicious cycle model scored monthly RSI divergence at −0.261 — cautionary — and gave it an effective weight of 0.070. That is a contribution of −0.0183 to the day's raw composite of −0.0060, the 2nd-largest of the 15 signals carrying weight today.

What monthly RSI divergence is

A bearish divergence is price making a higher high while the relative strength index makes a lower high: the market reached a new price without the momentum that produced the last one. A bullish divergence is the mirror at lows. The idea is old and the trouble with it is that on a fast timeframe it fires constantly.

This one is measured on monthly closes, with a 14-period RSI. A peak only counts once three bars have closed either side of it and the new high clears the old one by a margin, so a “divergence” here is a structure that took the better part of a year to form and cannot be revised away by next week's candle.

How this model uses it

This carries the largest base weight in the model, and in a bull market or a parabola it is multiplied by 2.2 — reaching an effective 0.154, more weight than any other signal ever holds. The model's strongest single opinion is a momentum divergence measured in months.

Repeat divergences compound twice over. The score scales with the size of the price/RSI gap, and the weight is separately multiplied: about 1.3× on a second consecutive divergence, 1.7× on a third, capped at double. A market printing three lower-momentum highs in a row is treated as a different situation from one printing its first.

Absence is scored as well. When no divergence is active but the level family is reading expensive, and neither the weekly divergence map nor the halving clock contradicts it, the signal returns a constructive +0.35 — an explicit “the valuation looks stretched but the structure has not broken yet” vote, rather than a silent zero.

Where today sits in its own history

Every figure in this table is measured over the model's full graded replay — 5,051 days from 2012-10-27 to 2026-08-25 — and re-measured on every daily refresh.

What this does not tell you

• monthly RSI divergence is one of the 15 signals that carried weight on 2026-08-25. On most days it is not the signal that decides anything, and the composite regularly disagrees with it.
• Two look-ahead bugs have been found in this signal cluster, and both had inflated the published record. The first paired every price candle with the RSI value from 14 bars later — on a monthly map, fourteen months of the future. The second read the current month's bar, which is stamped at the month's first day but valued at its last close, handing every day in a month that month's closing price. Both are fixed, and the fix for the second made the in-sample numbers fall, which is what removing a fake edge looks like. Anyone quoting this project's older figures is quoting the leak.
• Monthly bars are scarce. The entire price history holds roughly 160 of them, and confirmed peaks are a small fraction of that, so the number of genuinely independent observations behind this signal is in the low dozens at best.
• Its weight steps rather than blends. The model ramps weight changes over seven days for the level family only, so a regime flip can move this one from its base weight to more than double it overnight — the largest single weight step anywhere in the table, and a known rough edge rather than a designed behaviour.
• The number that drives the model's allocation is the composite, not this signal. Today's published composite is −0.0060. The current call, and every call before it, is on the call ledger — hash-chained the day it was made and never edited.