Effective October 10, 2026

Read the relationship

Floatflare places sourced release schedules beside market prices. The question is simple: how does price behave as more supply becomes eligible? The measurements keep that question separate from a claim about who sold or why a price moved.

Two lines, two units

The blue line is the daily adjusted closing price in US dollars. The latest incomplete daily bar may appear in the chart and headline but is excluded from the models. The headline quote uses the provider's unadjusted value. Stocks use exchange-local trading dates; tokens use UTC daily dates. Daily histories are requested explicitly and the collector rejects a different interval, instrument or currency.

The orange series is defined separately for every asset. It may be a published maximum, one holder cohort, continuously vesting allocation, reported circulating supply, or net change in escrow. It is not a universal measure of float. CoreWeave's current series is explicitly a 0–100% contract-expiry index, not a share count. “Sources” states each series' scope and exclusions.

Both series share dates but use independent vertical axes with labeled units. Apparent intersections have no numerical meaning. Log price changes the price axis only; it does not change the statistical model. Future extends the scheduled supply view by one year and never projects a price. The full ledger includes more distant events and unresolved conditions.

Month-level estimates and isolated reported balances appear as dots. We do not invent intermediate cliff dates. Conditional releases without a verified date remain outside the curve. Vesting is modeled continuously only where the cited terms support that treatment.

What R² measures

Every regression includes an intercept and uses the selected chart period. The baseline is ordinary least squares of log adjusted price on log cumulative eligible supply, using positive supply only. R² = 1 − residual sum of squares / total sum of squares. A linear supply-versus-price fit is also shown; it can include the zero-supply interval before a scoped cohort unlocks. When the log fit is undefined, the main reading identifies the linear alternative by name.

Cumulative levels and price can share a trend and produce a high R² without a useful causal or predictive relationship. Daily observations are serially dependent. Sample size is not a count of independent events. The site does not report p-values, statistical significance or a fitted price forecast.

The returns model instead fits the forward five-session log return to the fractional change in eligible supply that day. A second version also includes the contemporaneous five-session SPY return for stocks or Bitcoin return for tokens. That market control is observed over the outcome window; it is not known in advance and does not make this a trading strategy.

Held-out diagnostics train on the first 70% of dated observations, skip five observations to purge overlapping outcome labels, then score the rest against a predictor using the training target mean. This score can be negative and is not conventional test-mean R². A sufficiently large, full-rank training and test sample is required. Revised historical schedules are not point-in-time evidence; all scores are retrospective. Dash values mean unavailable, insufficient or constant data, never a hidden zero.

Before and after a release

A dated cliff maps to the first observed trading session on or after its date. A gap greater than four calendar days for a stock or one day for a token makes that event unavailable. The “before” return covers five sessions ending at the close immediately before the event. “After” covers that preceding close through event day plus five subsequent sessions: six close-to-close intervals.

Volatility is the sample standard deviation of daily log returns in each of those windows, reported as the after/before ratio. A turn means that the before and after simple returns have opposite signs. Market-adjusted event return subtracts the benchmark's simple return over the same interval; it is not a factor-model abnormal return.

The summary and event path use chronologically selected windows separated by more than ten sessions. Overlapping releases are retained in the source ledger but do not count as independent summary windows. Very small event counts are descriptive anecdotes. Continuous streams and imprecisely dated observations do not receive invented event studies.

Coverage and freshness

The launch catalog contains ten equities and twelve tokens, including Ethereum and Jupiter as explicit cases without a comparable current cliff curve. Selection is intentional, not an unbiased sample of all assets. No comparison here is evidence that an issuer or protocol is a scam.

The collector requests public market observations hourly, validates them, recalculates the models and atomically replaces one snapshot. An open page checks the snapshot every five minutes while visible. Current does not mean tick-by-tick. The latest source date and actual retrieval time are available in Sources.

Sui's issuer table and Optimism's public supply sheet refresh automatically. Ripple's published balance is tracked separately from gross re-escrow cycles. Other schedules are reviewed transcriptions of cited documents, not a promise to automatically interpret every future amendment. Source age and review scope are disclosed. If a reader fails, the previous dated data remains explicitly delayed; a missing dataset is not fabricated.

Reproduce or correct a result

Use “Data” for the selected historical window or download the complete public JSON snapshot. It includes definitions, source URLs, reviewed dates, source hashes where captured, prices, dated supply, all model coefficients, windows and results. Each snapshot has an identifier. Models exclude incomplete daily prices; the latest displayed quote may therefore be newer than a model's end date.

For a correction, email wonderful@gmail.com with the asset, date, snapshot identifier and primary evidence. We welcome corrections to dates, interpretations and calculations.