Range state · forward distribution · robustness

6B Compression and Expansion: A Reproducible Test

Suppose the last twelve five-minute bars span 18 ticks while the same clock window’s trailing median span is 42 ticks. The observed ratio is 0.43. That calculation can describe compression. It does not say expansion is next, choose a direction, or prove a tradable edge.

Input
Past-only range
Output
Forward distribution
Direction
Separate test
Original result
None
Illustrative state18 / 42 = 0.43
Current window
18 ticks
Matched median
42 ticks

Example onlyNo forecast

Direct calculation

Start With the Range, Not the Story

Fact: realized movement can be summarized from consistent price observations. Observation: a declared recent window can be small relative to a relevant historical distribution. Hypothesis: that state may change the later distribution of movement or execution quality. “Energy building” is a metaphor, not a measurable mechanism.

state ratio = current realized rangematched trailing median range

The denominator matters. Comparing a quiet overnight hour with the all-day distribution can classify the clock rather than the state. Match the weekday, exact timezone-aware window, bar length, event status, and contract-roll condition. Use only observations available before the candidate time when calculating a rolling percentile or median.

Use ticks for execution, returns for comparability

6B’s standard outright tick is 0.0001, worth $6.25 per contract. A range in ticks maps directly to order distance and dollar exposure; a log return is easier to compare across long histories with different price levels. Preserve both and state which one drives the label.

Past-only state

Define Compression Without Peeking at the Break

Compression must be completely knowable at the decision timestamp. A symmetric centered window, a future-confirmed swing, or a box drawn around the bars that preceded a known breakout all leak future information.

Definition familyFrozen ruleUseful propertyLimitation
High-low spanMaximum high minus minimum low over the last n fixed barsSimple and transparentOne extreme can dominate
True-range summaryMedian or mean true range over the last n barsIncludes gaps across bar boundariesSmooths the path and depends on lookback
Realized varianceSum of squared fixed-interval log returnsUses the intrawindow pathHigh-frequency noise and bad ticks matter
Range percentileCurrent measure below a training-sample clock-matched percentileAdapts scale to the historical distributionThreshold and lookback can be overfit
Multi-scale agreementTwo predeclared measures must both be below their thresholdsCan reject a one-measure artifactMore choices create more search degrees of freedom

Store the continuous value even when the live rule uses a binary threshold. If the result appears only below the 14.7th percentile and vanishes at the 10th and 20th, the threshold may be fitting noise.

Three separate outcomes

Define Expansion on Size, Direction, and Cost Separately

A large forward range is not the same as a profitable directional break. It can be one clean move, a breakout and reversal, or two-sided churn with poor fills.

Movement

Did the distribution of range change?

Measure forward realized range, realized variance, tail quantiles, and time until a frozen range threshold is reached. Compare the full distribution, not only the average.

Direction

Was movement one-sided?

Measure signed return, close location, path efficiency, breakout-side follow-through, and reversal frequency. Direction must be declared independently of compression.

Execution

Could an order capture it?

Measure spread, displayed depth, trade-through distance, slippage, missed passive fills, and latency. A wider range can arrive with worse execution.

Economics

Did the net rule beat a baseline?

Specify entry, invalidation, exit, sizing, commission, fees, spread, slippage, and queue assumptions. Gross range is not net expectancy.

Predeclare several forward horizons because a state can affect the next 15 minutes but not the next four hours. Treat every horizon as part of the multiple-testing family.

Event sample

Construct One Observation per Eligible State

Repeated low-range bars inside one quiet episode are dependent. Counting each bar as a fresh signal can inflate the sample and reward long-lasting compression.

Episode-based primary sample

Enter an episode when the frozen compression rule first becomes true. Keep it active until an exit threshold is met. Record one anchor timestamp per episode, impose a declared cooldown, and retain duration as a variable. A secondary bar-level analysis can be reported with dependence-aware uncertainty.

Unit
Compression episode
Anchor
First knowable timestamp
Cooldown
Predeclared
Uncertainty
Block by day or episode
1

Clean data

Audit gaps, duplicates, crossed quotes, outliers, and contract identity.

2

Fit baseline

Use training dates only to estimate clock-matched distributions.

3

Label episodes

Apply the frozen entry, exit, overlap, and cooldown rules.

4

Score forward

Measure movement, direction, cost, and failure on untouched dates.

Rival explanations

Control the Clock, News, and Contract

A low-range state often reflects a known information schedule or a thin market. The later expansion may be caused by a release, reopening, or roll transition rather than the compression itself.

Clock window

Match in an IANA timezone and preserve the daylight-saving regime. Do not pool unlike local hours.

Official events

Tag ONS, Bank of England, BLS, BEA, and Federal Reserve times from the contemporaneous official calendar.

Event distance

Store minutes before and after each release. A pre-news lull is a distinct state, not a generic compression result.

Contract roll

Use dated securities and an ex-ante roll policy. Report nearby and next-contract results around migration.

Market quality

Condition on spread, depth, quote-update rate, and outages. Zero trades can mean missing data.

Volatility regime

Compare within longer-run states so a low intraday percentile in a crisis is not equated with ordinary calm.

Stress the result

A Finding Should Survive Reasonable Definitions

Robustness is not rerunning parameters until one works. Declare the grid before opening the holdout and publish all cells.

Stress testWhat changesFailure signal
SamplingOne-, five-, and fifteen-minute inputs with noise-aware constructionEffect exists only at the noisiest frequency
Threshold neighborsAdjacent percentiles and durationsSign or magnitude flips at small changes
Outcome horizonSeveral frozen forward windowsOnly a post-selected endpoint looks favorable
Event exclusionAll dates, non-event dates, and named event familiesGeneric claim is entirely an event effect
Contract constructionDated contracts, two ex-ante roll rules, no stitch barsContinuous-series artifact drives the result
Cost scenariosObserved and stressed spread/slippage, missed fillsNet result fails under ordinary friction
ChronologyDevelopment, validation, final holdout, rolling re-estimationPerformance decays outside the fit period

Report effect sizes and uncertainty intervals, not just a p-value. If testing many definitions, horizons, sessions, and trade rules, use a multiple-testing procedure or present the study as exploratory. Keep the final holdout sealed until the analysis code and exclusions are frozen.

Evidence hierarchy

What the Evidence Can and Cannot Support

The strongest permissible conclusion depends on which layer has actually been measured.

1Descriptive fact

A named window’s realized range was low relative to a frozen matched distribution.

2Conditional observation

In a disclosed sample, a declared forward outcome differed from its matched baseline with uncertainty reported.

3Validated rule

A fully specified, costed rule remained stable on untouched data and across reasonable variants.

4Live use

Current data quality, liquidity, margin, and risk limits match the conditions under which the rule was validated.

Research status

Layer 1 is a measurement definition. Layers 2 through 4 are not claimed. No original 6B dataset, event count, expansion frequency, directional edge, backtest or live performance result was produced. The 18-tick, 42-tick and 0.43 values are a hypothetical calculation, not a market observation.

Sources, method and editorial disclosure

Sources and time-sensitive facts were reviewed August 13, 2026. This is original, unsponsored editorial analysis.