Tape classification · evidence before narrative

6S Liquidity Labels: Define Sweeps, False Breaks and Drifts

By Kyle ParrottUpdated August 21, 2026

Three traders can watch the same 6S move and call it a stop sweep, a failed breakout and a slow drift. Those labels are not interchangeable observations. Each needs a frozen reference level, path, clock and required data fields—or it is only a story attached after the chart is complete.

Freeze the labels Open the replay worksheet

Operational definitions

Make Every Label Computable From Available Data

The definitions below are protocol patterns, not claims that 6S exhibits them at a stable rate. Thresholds, reference levels and horizons must be fixed before replay.

Label	Required path	Required outcome	Do not infer
Sweep candidate	Price trades through a predeclared level by a minimum distance within a maximum time	Classify only the traversal; continuation or rejection is a separate field	Stops, manipulation, trader identity or intent
False-break candidate	Price first satisfies the frozen break rule beyond a predeclared level	Then returns through a frozen reclaim boundary within the registered horizon	That the break “trapped” participants or will keep reversing
Slow-drift candidate	Signed movement accumulates with bounded pullbacks and a declared path-efficiency threshold	Persists for the frozen duration without a disqualifying jump/event	Low risk, informed flow or inevitable continuation
Unclassified	Required fields, level history or clock state are missing/conflicted	No label	Nearest-looking category

“Liquidity sweep” is not an exchange event type.

Trades through a level are observable. The resting orders that canceled, filled or arrived can be studied only with suitable order-level data. The beneficial owners, motives and off-book stops generally remain unknown.

Claim-to-field discipline

Use the Minimum Data That Can Support the Claim

Candles can define price paths at their resolution but cannot reconstruct the book. Quote and order data add fields; they do not reveal intent.

Bars

Path summary

OHLC, volume, exact boundary and completion time. Intrabar ordering remains unresolved unless finer data are used.

Trades

Executed sequence

Exchange time, price, quantity, aggressor logic if defined, corrections and sequence gaps.

Quotes/depth

Displayed state

Bid, ask, quantities, levels and updates. Displayed liquidity can cancel before an order arrives.

Own orders

Experienced execution

Decision, send, acknowledgment, fills, partials and cancels are needed for realized shortfall.

Preserve the dated contract. A continuous series can splice different order books and create artificial level events.
Version the roll. Declare when the study changes expiries and how open intervals are handled.
Audit sequence gaps. Missing updates cannot be silently treated as an unchanged book.
Keep zero and missing distinct. No trade, no quote and no data are three states.

Causality needs a clock

Record the Sequence Before Naming the Pattern

The same high and low can arise from incompatible sequences. Store when the level became known, when it was first touched, what happened beyond it and when the outcome became classifiable.

T−1

Register the level

Name the source, price, creation time, expiry and whether future data can revise it.

First eligible touch

Record book state, event tags and the first trade/quote satisfying the rule.

Traverse or reject

Measure signed distance, elapsed time, traded quantity, quote changes and gaps.

Resolve the horizon

Apply the frozen continuation, reclaim, timeout or unclassified outcome.

absolute net movesum of absolute interval movespath efficiency from 0 to 1

Metric boundary: path efficiency is a descriptive candidate, not a profitable drift signal. Sampling interval, missing-data policy, event exclusions and threshold must be registered before observing outcomes.

Labels can fail

Write Disqualifiers Into the Definition

A robust classifier says when it cannot decide. These conditions prevent a vivid chart from being forced into the nearest narrative.

Reference failure

The level was not knowable

Pivot uses future bars
Level moved after contact
Continuous-chart splice
Unspecified tolerance

Data failure

The path is incomplete

Sequence gap
Crossed/stale book
Unresolved intrabar order
Wrong contract month

State failure

A confound dominates

Scheduled release collision
Exchange pause/holiday
Roll migration
Threshold changed later

Reproducible replay

One Record Per Candidate, Including Rejections

A second reviewer should reproduce the label from the stored fields without seeing the later trade result.

6S tape-classification record

Identity: Dated contract, source, schema, checksum, timezone and CME trade date.
Reference: Level price, source rule, creation timestamp, expiry and revision policy.
Eligibility: Event, holiday, roll, spread, depth and data-integrity flags at first touch.
Path: Signed excursion, duration, trades, volume, quote/depth changes and sequence gaps.
Outcome: Sweep candidate, false-break candidate, slow-drift candidate or unclassified, with rule version.
Later result: Stored separately after classification; never used to relabel the candidate.

Develop

Choose definitions and cleaning policy on an early chronological block.

Lock

Freeze parameters, hierarchy when labels collide and every disqualifier.

Replay

Classify forward using only information available at each timestamp.

Audit

Report counts, overlaps, unclassified states and reviewer disagreement.

From label to decision

A Description Is Not an Entry Signal

Even a reproducible label needs a separately tested strategy, order rule, costs and risk veto. Do not map “false break” directly to “fade” or “drift” directly to “follow.”

ClassifierWhat observable path occurred?Forecast testDoes the frozen label add out-of-sample information?Execution testCan an order arrive and fill within cost limits?Risk vetoDoes size, gap stress and invalidation fit?

Research status as of August 21, 2026

No original 6S frequency, transition probability, expectancy, preferred label, order-book finding or trading result is reported. The valid output is unclassified or no trade whenever the required evidence or a separately validated strategy is absent.

Sources, methods and editorial disclosure — reviewed August 21, 2026

CME DataMine historical-data catalog for distinctions among settlement, trade, Market by Price, Market by Order and PCAP data.
CME Market by Order FAQ for order-level message and priority fields.
CME Group trading-hours and holiday schedules for venue-state controls.
CME Rulebook Chapter 254 for the dated 6S contract identity.
CFTC advisory on commodity trading systems and hypothetical results for hindsight and simulated-execution limits.

Sources and methods were reviewed August 21, 2026. No CME dataset was purchased or analyzed for this page. The article distinguishes observations, operational labels, causal claims, forecasts and trading applications.