Glossary

Bitcoin cycle glossary

Plain-English definitions of the terms used on this site. Search, or link straight to a term.

Price averages and crosses

SMA (simple moving average): The plain average of the last N closes. A 50-week SMA averages the last 50 weekly closes; the 200-week SMA is a very slow trend line.
EMA (exponential moving average): A moving average that gives recent prices more weight, so it turns sooner than an SMA of the same length. V3d is a 30-day EMA of its three inputs.
Death cross: A shorter moving average crossing below a longer one, such as the daily 50-day below the 200-day. On our data the daily version is a late bear-market marker, not a bottom signal.
Golden cross: A shorter moving average crossing above a longer one. The daily 50/200 golden cross last fired on 2026-09-08.
MACD: Moving Average Convergence Divergence: a momentum gauge built from two EMAs and a signal line. Its daily bearish cross fires about 12 times a year.
RSI (relative strength index): A 0–100 momentum gauge over 14 periods (here, 14 monthly closes). Bitcoin’s monthly RSI has not gone below 40 since 2012.
Moving-average variant: One specific choice of average type, length and timeframe. Testing many variants and keeping the best is a multiple-comparisons trap.

Valuation and on-chain

Realized price: The average price at which every coin last moved on-chain: a rough cost basis for the whole market. Price traded below it before each of the last 3 bottoms.
MVRV: Market value divided by realized value. Above 1, the average coin is in profit.
MVRV-Z: MVRV rescaled by its own history into a Z-score. It went below zero before each of the last 3 bottoms.
Puell Multiple: Daily miner revenue divided by its one-year average: a gauge of miner stress.
aSOPR: Adjusted Spent Output Profit Ratio: whether coins being moved are, on average, sold at a profit or a loss. Its history here is reconstructed.
Power-law percentile: Where price sits within a long-run power-law trend fitted to Bitcoin’s history, as a percentile. One of V3d’s three inputs; that input is HLB’s corridor version, a different calculation from the Power Law Value card’s [[g:residual-percentile|residual percentile]].
Balanced price: A model-based reference price that Cowen places near $38k.
Drawdown: The percentage fall from a prior peak. Cycle drawdowns were 85%, 83% and 77%.
Capitulation: A wave of forced, panicked liquidation, usually a volume spike plus a reset in on-chain gauges.

Cycle structure

Cycle: One swing from a major top to the following major low. The reference tops are 2013-12-04, 2017-12-16, 2021-11-08 and 2025-10-06.
Halving: The roughly four-yearly cut in half of Bitcoin’s new-coin issuance. The current cycle counts from the 2024 halving.
Halving clock: Days since the last halving, compared with where prior bottoms sat in time. One of V3d’s three inputs.
Midterm year: A year with US midterm elections, such as 2018, 2022 and 2026. Cowen’s timing work treats its fourth quarter as the usual window for cycle lows.
Reference bottom: A cycle low used as a benchmark: 2011-10-20, 2015-01-14, 2018-12-15 and 2022-11-21. The 2026-06-30 low is provisional.
Provisional low: The lowest price so far, not yet confirmed by later price action.
Near a bottom: Within ±120 days of a reference bottom; about 18–20% of all days.

Testing and statistics

Event study: Line up every date a rule fired, then measure what happened next.
Baseline: The result from every day in the sample, signal or not. The median 365-day return over all days was +83.8%, with 70.9% of days positive.
Base rate: How often something happens by chance. A signal must beat it to mean anything.
Hit rate: The share of events followed by a positive return over a set horizon.
Resolved event: An event old enough for its forward return to be measured.
In-sample: Measured on the same history used to choose the rule. Every claim-check verdict is in-sample.
Out-of-sample: Measured on data the rule never saw. It is the stronger test.
Multiple comparisons: Testing many variants and reporting the best; one will look good by luck.
Rank correlation (ρ): Spearman’s rho: whether higher values of one series go with higher (or lower) values of another, from −1 to +1. Near 0 means no ranking skill.
Uncertainty range (confidence interval): The span an estimate could plausibly fall in. A range that includes zero means no effect is a live possibility.
Placebo: A deliberately meaningless comparison, used to check that a result beats chance.
Overfitting: Fitting past data so closely that a model captures noise. A probability model whose weights flip between folds is a sign.
Calibration: Whether a stated probability matches how often things actually happen.
Fold (leave-one-cycle-out): Set one whole cycle aside, fit on the rest, test on the cycle set aside, and repeat.

This site’s gauges

HCP: The site’s headline cycle score (0–100, from 17 inputs). It reproduces exactly but has shown no forward skill, so it is labelled RESEARCH — NOT DECISION-READY.
V3d: The cycle-position gauge: a 30-day EMA of the power-law percentile, price vs the 200-week SMA, and the halving clock.
Drop-one-input range: The spread of readings from removing one input at a time; it shows how much the level depends on the inputs.
Fragility: How far a number jumps on its worst days, measured at the 99th percentile of daily moves. V3d: 0.30 points against a bar of 3.
Point-in-time (PIT): Using only data that was available on each date, with no later revisions.
Reconstructed: A history rebuilt after the fact rather than recorded live. Values may be right, but they cannot be certified as what was knowable then.
Vintage: The version of a data series as it stood on a given date.
Design contamination: Choosing inputs because they scored well on the same history you then test on. Every historical V3d result carries it.
Pre-registration: Writing down the rules, candidates and pass bars before seeing any results, so they cannot be moved afterwards.
Soak: A scheduled daily snapshot, stored once and never revised, that builds a clean forward record. The HCP v2 soak’s first robustness grade is due 2027-01-13; the V3d soak started in mid-October 2026 and is not graded yet.
Peer recompute: An independent re-implementation, written only from the written spec, that checks the main numbers.
INSUFFICIENT / NO EDGE / EDGE: Claim-check verdicts: too few resolved events; no meaningful difference from the baseline; or a difference beyond the bands. All are in-sample.

Power law value

Point-in-time fit (PIT fit): A trend line refit every day using only the closes up to that day, so no past reading uses data from its own future. The Power Law Value card uses an expanding least-squares fit of log price on log days since the genesis block, on Bitstamp daily closes since August 2011.
Residual percentile: Today’s distance from today’s fitted line, ranked among the distances of every earlier day from that same line (today’s fit): 0 is furthest below, 100 furthest above. It decides the zone, Z1 to Z6.
Value zone (Z2): A residual percentile from 5 to 25: price well below the fitted line. A place on the scale, not an instruction. In-sample, days in Z2 had a median 365-day change of +121% against +61.6% for all days (11 episodes); the 95% interval of that gap includes zero, so after correction it is not statistically distinguishable.
Fair value (power-law line): The fitted line’s price for a given day, and the only number this site calls fair value. A statistical centre, not a target and not a forecast. It is the average distance (zero); P50, the median distance, sits below it because the bubble years pull the average up.
Value floor: How the power law has read best so far: low percentiles came before strong years, while each cycle top reached less stretch than the one before, so a high reading is not a reliable top signal.
Reference curve (Santostasi): The thin dashed power-law curve on the Price Map, drawn with fixed published constants chosen with hindsight. Shown for reference only; no reading is taken from it. The Power Law Value card uses HLB’s own point-in-time fit.