Sticky States, Brief Transitions: A Markov Map of the Composite Score

Revision note (June 2026). This article was first published in May 2026 using a verdict series generated by the data pipeline. An internal audit then found that the pipeline was computing the composite with a superseded weight set, different from the one the dashboard publishes. The pipeline has been unified with the dashboard weights and every number below regenerated. Most findings survived with shifted values. One did not: the previous version's headline claim, that profit-taking episodes recover in a V-shaped jump straight to ACCUMULATE 70% of the time, was an artifact of the old weighting and is retracted. The corrected exit pattern is reported in its place, and the finding the original article flagged as better-supported (profit-taking avoids jumping directly to high conviction) survived the correction intact. Leaving this note visible is the point: a result that dies under a corrected pipeline was never a result.

Of the 741 state changes the composite score has recorded since July 2010, 56% are reversed within three days. The score crosses a threshold, sits across it for a day or two, and crosses back. Whatever the verdict appeared to say, it then unsays.

Treat the composite as a Markov chain, a discrete-state process where today's verdict is the input and tomorrow's verdict is the output. The dashboard already publishes five verdict states with hard cutoffs (5.5 for STRONG BUY, 4.5 for ACCUMULATE, 4.1 for HOLD, 3.8 for REDUCE, anything lower for TAKE PROFIT). Counting how often the score moves between those states across 5,807 days produces a transition matrix. The matrix is dull in places, surprising in others, and untrustworthy in a few specific cells that ought to be flagged rather than smoothed over.

What follows is the descriptive version of that exercise: what nearly 16 years of daily verdicts have done, with the caveats kept visible.

The Score Lives in Two States

Start by counting days. Since July 2010, the score has read STRONG BUY on 3,034 of its 5,807 days and ACCUMULATE on a further 1,839. Together that is 84% of the historical record. HOLD has been recorded on 310 days, TAKE PROFIT on 463, and REDUCE on 161.

The lopsided distribution is partly a function of bitcoin's trajectory: the score is built from indicators that flag accumulation opportunities, and a fifteen-year asset in long-term uptrend offers a lot of them. It is also partly an artifact of the verdict thresholds themselves. The HOLD band is the score range 4.1 to 4.5, a corridor 0.4 wide. REDUCE is 3.8 to 4.1, narrower still at 0.3. The composite score's empirical standard deviation is 1.06, which means the score's typical daily wobble is more than twice the width of either intermediate band. The verdict scheme catches buying opportunities, catches profit-taking opportunities, and treats the in-between as an afterthought.

That should affect how the rest of this analysis is read. The transition rows out of REDUCE come from 84 historical episodes covering 161 days, better than the previous version's 18 episodes but still the thinnest row in the matrix. They are reported because the calculation is well-defined, not because the numbers are dependable.

The 1-Day Matrix Is Nearly Identity

The first cut at the data is the simplest: the fraction of days at state s that are followed by a day at state s'. For the dominant states, that next day is almost always s again.

STRONG BUY days roll into STRONG BUY days 93.9% of the time. TAKE PROFIT into TAKE PROFIT, 83.8%. ACCUMULATE into ACCUMULATE, 84.5%. The composite score is a smoothed quantity, and smoothed quantities do not jump around. A reader who wants to know what the dashboard will probably say tomorrow can read it today and be right roughly nine times out of ten.

The two transitional bands, HOLD and REDUCE, behave differently. Their diagonal probabilities are 64% and 48%. These are not stable regimes. They are thin corridors that the score crosses on its way somewhere else. Most of the off-diagonal mass out of HOLD goes back to ACCUMULATE (24.2% next day). REDUCE leans the other way: its largest off-diagonal cell is TAKE PROFIT at 28.0%, on a base of 161 days.

The takeaway from the daily matrix is not the diagonal. It is the rate at which the matrix decays when the horizon is lengthened.

Persistence Half-Life

Extending the horizon to seven, thirty, and ninety days shows how long today's verdict carries information about future verdicts.

The dominant diagonals decay steadily out to ninety days. TAKE PROFIT decays fastest: by ninety days its self-transition probability of 20.5% is about 2.5x the empirical base rate of 8% (8% is simply the share of all days that are TAKE PROFIT days: the guess you would make knowing nothing about today), meaning some signal remains but most has dissipated. ACCUMULATE at ninety days (36.4%) sits close to its 32% base rate. STRONG BUY holds up best (69.2% against a 52% base), partly because it covers half the sample and partly because it is an unbounded score band less susceptible to drift: the score can sit far above the 5.5 cutoff, so day-to-day wobble does not tip it out.

The useful summary is that the score's "memory" fades over roughly one to three months depending on the state. After ninety days, today's verdict tells you only modestly more about the verdict than the long-run base rates do.

Half of All Transitions Are Noise

Now the uncomfortable finding. Of the 741 state changes recorded across the full history, 417 are reversed within three days. The score moves from A to B, sits in B for a day or two, and returns to A. The composite did not change regimes. It oscillated across a threshold.

That is exactly what the wide standard deviation of the score (1.06) and the narrow widths of the HOLD and REDUCE bands (0.4 and 0.3 wide) would predict. A score that wobbles by half a point on a quiet trading day can flip across the HOLD/ACCUMULATE boundary at 4.5 and back, generating two formal "transitions" that mean nothing. Fifty-six percent of the score's recorded transitions are this kind of churn.

One option is to smooth the verdict with a hysteresis rule: require the score to move past a buffer beyond each cutoff before the verdict flips. That fix introduces a free parameter (how much hysteresis), and any choice is arbitrary. The descriptive analysis here keeps the noise visible rather than papering over it: the daily transition matrix should be read with the understanding that roughly half its off-diagonal mass is threshold artifact.

To see the genuine structure, collapse runs of consecutive identical verdicts into episodes, and ask what each episode transitions to when it ends.

Profit-Taking Exits Up the Ladder, Not in a Leap

The full history contains 742 episodes, runs of consecutive days at the same verdict. Episode-to-next-episode transitions remove the daily diagonal entirely and surface the score's underlying topology. Stripped of all the tomorrow-same-as-today repetition, what is left is the actual moves: not how long the score stays put, but where it goes when it finally leaves.

One row is more interesting than the rest, and it is the row where this article has to retract its previous headline. When a TAKE PROFIT episode ends, it transitions to REDUCE 54.7% of the time, to ACCUMULATE 25.3%, to HOLD 17.3%, and to STRONG BUY just 2.7%. The most common path out of profit-taking is one step up the ladder, not a leap. The earlier version of this article, built on the superseded weight series, reported a 70% jump straight to ACCUMULATE and read V-shaped bottoms into it. Under the corrected series, that pattern does not exist: the score exits its lowest band the same way it entered, through the adjacent states.

Checked against the base rates, the corrected row is even more lopsided than it looks. Across all 742 episode-ending transitions, REDUCE is the destination 11.3% of the time regardless of origin; the 54.7% conditional from TAKE PROFIT is a 4.8x lift (it happens nearly five times as often as chance alone would produce). The avoidance finding from the original article survives and sharpens: direct exits to STRONG BUY happen at 2.7% against a 25.0% base rate, a lift of 0.11x (one ninth as often as chance). Profit-taking almost never recovers directly to high conviction. That conclusion held through a change of weights that destroyed the V-shape claim, which is the kind of robustness worth noting.

The mirror image is the STRONG BUY exit row, which is now absolute: all 184 recorded STRONG BUY episode endings transition to ACCUMULATE. The score has never fallen out of its top band by more than one step in a single day. Tops grind; they do not gap. The asymmetry between how the score enters conviction (one step at a time) and how rarely it returns there from the bottom is structural in the data, not an interpretation.

The HOLD and REDUCE rows of this matrix should still be read with care. HOLD has 113 historical episodes and REDUCE 84, a far better sample than the previous version's 56 and 18, but several individual cells still rest on single-digit counts. The suggestive pattern (REDUCE exits downward to TAKE PROFIT 54% of the time, HOLD exits upward to ACCUMULATE 66%) is consistent with the ladder picture; the precise percentages are not settled.

Dwell Times Are Short on the Median, Long in the Tail

The other quantity worth reporting is how long the score sits in each state per episode.

Every state has a median dwell of one to three days. The means are much higher because the distributions have heavy tails: a single STRONG BUY episode in early bitcoin's history ran 293 consecutive days, dragging the mean to 16.4. TAKE PROFIT's longest episode ran 74 days. The takeaway is that most regime episodes are brief (a flickering across thresholds, in line with the boundary-churn finding), punctuated by occasional long, stable runs. Both behaviors are present in the same data.

A reader looking at today's verdict and asking "how long will this last?" is asking the wrong question. The median says one to three more days. The mean says a week, or three for STRONG BUY. The distribution says it could be any of those, or it could be three months. The signal is not predictive at the level a calendar-based question implies.

The Matrix Holds Up Across Macro Regimes

A first-order Markov chain assumes the transition matrix is stationary across time (the same rules of movement in 2011 as in 2025). Bitcoin's history covers a wide range of macro conditions, from a 4.6% M2 contraction in 2022 to a 26.8% expansion at the COVID peak (M2 is the broad US money supply, roughly cash plus bank deposits). Stratifying the matrix by macro regime tests whether the unconditional version is hiding state-dependent dynamics. Split the sixteen years into easy-money days and tight-money days and rebuild the matrix in each pile; if the one-pile version has been averaging away two different behaviors, the two piles will look different. The expectation, set out earlier in the article, was that it would.

The cleanest split is M2 year-over-year growth. Across the joint history of M2 and the daily signals, the YoY series ranges from -4.6% to +26.8% with a median of 5.7%. Splitting daily transitions by whether the from-day sits above or below the median produces two matrices of 2,919 and 2,887 transitions: roughly balanced halves of the history.

The headline finding is that the matrices are nearly identical. Most cells differ by one to four percentage points, and the STRONG BUY diagonal is identical to one decimal in both regimes (93.9%). HOLD → ACCUMULATE is 26.1% loose versus 22.2% tight (+3.9pp). TAKE PROFIT → TAKE PROFIT is 87.2% loose versus 80.1% tight (+7.1pp). These are small differences, not zero, but not the regime-dependent drift the earlier caveat in this article anticipated.

The largest gaps sit in the REDUCE row, and they are the kind that demand a footnote rather than a celebration. REDUCE → TAKE PROFIT reads 35.1% in loose conditions versus 21.4% in tight (+13.6pp), and REDUCE → REDUCE 41.6% versus 53.6% (-12.0pp). Both rest on thin samples (27 to 45 transitions per cell), where the standard errors are wide enough that the intervals overlap. The gaps are not findings; they are the dominant statistical noise in the stratification.

The negative result is the more important one. The unconditional matrix is doing more work than the original caveat assumed. Either the score is genuinely indifferent to macro conditioning at the verdict-state level of granularity, or M2 YoY is the wrong stratifier. Both are possible. The first is the simpler explanation, and in the absence of a stratifier that produces visibly different matrices, the analysis has to default to it.

What This Analysis Cannot Tell You

Three caveats matter more than the numbers above.

The first is sample size. Sixteen years is three full bitcoin cycles. The transition matrix for the dominant states has thousands of observations behind it. The REDUCE row, even after the June 2026 regeneration tripled its day count, rests on 84 episodes and 161 days, and several of its multi-horizon cells are computed from single-digit counts. Cells with fewer than ten transitions are shown in lighter grey in the matrices above for that reason.

The second is look-ahead in the underlying composite. The indicator weights and verdict thresholds were chosen with knowledge of the full price history, so a user dashboarding the score in 2015 would not have seen the same series this analysis is based on. The transition matrix is descriptively accurate for the post-hoc series; it is not a snapshot of what a real-time user would have observed. (A related provenance problem, that the pipeline computed this series with a different weight set than the dashboard's, was found in a June 2026 audit and fixed: the series now uses the dashboard's exact weights, and this article's numbers were regenerated from the unified series. The revision note at the top records what changed.)

The third is the Markov assumption itself. A first-order Markov chain says today's state contains everything needed to predict tomorrow's. Bitcoin's actual dynamics depend on cycle position, macro liquidity, and halving proximity, none of which appear in the verdict alone. The M2 stratification above tests the macro-liquidity slice of this concern and finds the matrix mostly intact, but cycle position (only three complete cycles in the data) and halving proximity (four events) cannot be stratified with statistical power. The Markov assumption survives one test in this article. It is not yet a defended assumption.

What It Can Tell You

Four things, with appropriate confidence levels.

First, the score is sticky on short horizons. If today's verdict is STRONG BUY, tomorrow's verdict will be STRONG BUY about 94 times out of 100. Daily check-ins on the dashboard reveal almost nothing that yesterday's check-in did not already reveal.

Second, 56% of the score's recorded transitions are not transitions in any meaningful sense: they are boundary noise from oscillation across hard cutoffs, whose own pre-2022 optimization look-ahead is disclosed with the backtest. Users acting on every flip of the verdict will be reacting to threshold churn as often as to actual regime change.

Third, when genuine regime changes happen, they move one step at a time, and the strongest regularity is what the score avoids. Profit-taking episodes exit to the adjacent REDUCE band at 4.8x the base rate and jump directly to STRONG BUY at one ninth of it. The exit-through-adjacent-states pattern replaced this article's original V-shape claim when the series was regenerated; the avoidance-of-conviction pattern survived the regeneration, which is the better reason to trust it. It remains the finding most exposed to the look-ahead caveat, since the indicators that flag profit-taking overlap with those that flag accumulation.

Fourth, the matrix is mostly stable across macro liquidity regimes. Splitting the history by M2 YoY changes most cells by one to four percentage points, and the STRONG BUY diagonal not at all. That is a weaker test of stationarity than splitting by cycle position would be, but cycle position cannot be split with three full cycles.

The transition matrix is a description of where the score has been. It is not where the score is going. Anyone who reads it as a forecast is misusing it. Anyone who reads it as a map of the score's own behavior, with the gaps and unreliable cells visible, will find it modestly useful and unsurprising in most places.

The interesting cells are the ones where the data is thin enough to be wrong.