Skip to main content

Provenance · field semantics · quarantine

6M Data Traps: A Dataset Acceptance Audit

Two vendors can publish different daily 6M candles without either row being a simple typo. One may switch expiries before the other, one may back-adjust the history, one may use settlement while the other uses a last trade, and their day boundaries may differ. Until those choices are documented and reconciled, neither candle is evidence for a market claim.

Null field
Failure
Derived time
Labeled
Unknown roll
Quarantine
Original audit
None

Begin with definitions

A Candle Disagreement Is a Lineage Question

Do not average the vendors or choose the prettier chart. Reconstruct the row from raw records and metadata. The first goal is to determine whether each series measures the same instrument, interval, price type, and timestamp at all.

01

Identify

Resolve exchange product, dated expiry, venue, feed, schema and file version.

02

Define

Record quote direction, price type, bar clock, inclusion rules and correction handling.

03

Rebuild

Regenerate the candle from eligible raw rows without manual edits.

04

Reconcile

Classify the difference as expected construction, source defect, unknown, or noncomparable.

Acceptance principle

A provider passes only when its actual payload supplies every field and timing requirement for the intended use. Inferred timestamps, polling windows, receipt times, reconstructed events, or confidence labels may be retained as experimental evidence, but they do not replace an authoritative required field.

Evidence ladder

Documentation Describes a Feed; Payloads Prove What Arrived

Authoritative documentation and actual data answer different questions. Certification needs both. A product page can establish contract mechanics, but only inspected records can establish whether a purchased or vendor-delivered file contains usable timestamps, quotes, quantities, corrections, and contract identifiers.

Provider schema and field guide

Versioned documentation defines advertised fields, timestamps and message behavior. Save the exact version relevant to the files.

Actual payload inspection

Sample every relevant regime, contract and boundary. Record nulls, gaps, duplicates, corrections and unexpected values; do not infer advertised fields that are absent.

Independent reconciliation

Rebuild aggregates, compare authoritative reference points where comparable, and document expected basis, timing, and construction differences.

CME DataMine advertises multiple historical dataset types. A catalog description is evidence of availability, not evidence that this page acquired or validated any file. It did not.

Semantics before statistics

Audit Identity, Quote Direction, and Time as Required Fields

Three mistakes can invert or reorder an entire study: using an unidentified contract, confusing MXN/USD futures with a pesos-per-dollar cash series, and assigning timestamps with an assumed offset.

FieldRequired meaningAudit testFailure classification
Product6M Mexican peso/U.S. dollar futuresMap code and venue to authoritative CME documentationPrimary use failed if ambiguous
Contract monthDated delivery month on every recordReconcile symbol parser across year and month transitionsContinuous-only data is secondary or experimental
Quote conventionU.S. dollars per Mexican peso for 6MUnit-check prices and P&L formulas; do not rely on pair-name orderReject inverted or undocumented values
Price typeTrade, bid, ask, settlement, open, high, low, close, or derived midpointReconcile each field to provider definitions and raw messagesUnknown “price” cannot support execution claims
TimestampProvider event time with precision, timezone and clock semanticsTest ordering, duplicates, transitions and source latency fieldsReceipt time cannot substitute for event time
Trade dateExchange session label stored apart from civil dateInspect overnight and holiday boundariesReject joins made on an undocumented date
Correction stateInsert, cancel, correction, sequence-reset or equivalentReplay message rules and reconcile totalsUnprocessed corrections contaminate aggregates

6M futures

USD per MXN

The CME FX Product Guide lists MXN/USD futures in U.S. dollars per Mexican peso. A higher 6M quote represents more U.S. dollars per peso under that convention.

Banxico cash reference

MXN per USD

Banco de México’s SIE exchange-rate table publishes pesos per U.S. dollar for named reference series. A reciprocal can align units, but it remains a different instrument, timestamp, and methodology from a futures trade.

The word “close” is overloaded

Keep Trades, Settlements, and Continuous Futures Separate

A settlement can be an exchange-defined daily reference rather than the last trade. A vendor close can be the final record before its own boundary. A continuous price can be a transformation across expiries. Store these as separate fields and never use one name for all three.

Observed market

Trade or quote

Retain timestamp, dated contract, side-relevant quote, quantity and feed state. This is the input required for an executable simulation.

Official daily reference

Settlement

Record the named exchange settlement and applicable methodology or notice. Do not imply it was the final executable trade.

Analytical transform

Continuous series

Record roll trigger, splice date, adjustment direction, cumulative adjustment, source contracts and rebuild code.

TrapHow it entersDiagnosticRequired treatment
Artificial jumpUnadjusted expiry switchReturn aligns exactly with roll and contract-price gapExclude from signal returns or model the roll explicitly
Rewritten levelBackward adjustmentHistorical price changes when a new roll is appendedVersion data and avoid level rules on transformed prices
Hindsight rollSwitch uses same-day final volume or later informationDecision timestamp occurs before required input was knownReplace with an ex-ante rule or reject for backtesting
Split liquidityNearby-only feed during migrationOutgoing activity falls as next expiry risesPreserve both contracts and analyze migration

Field-specific evidence

Liquidity Claims Require Quotes, Depth, and Missingness

Volume alone cannot establish a tight spread, deep book, passive fill, or low slippage. A top-of-book feed cannot reconstruct full depth or exact queue position. Market by Price and Market by Order provide different granularity. Match each claim to the minimum field set it requires.

ClaimMinimum evidenceStill not established
Completed volumeDated-contract trades with quantity, corrections and complete interval coverageDisplayed or hidden liquidity; participant motive
Bid-ask spreadSynchronized best bid and offer with stale/crossed-book policyFill at either quote after latency
Displayed depthDepth updates with price levels, quantity, sequence and replay rulesCommitted liquidity or undisplayed quantity
Queue positionOrder-level data and a validated book reconstruction appropriate to the venueTrader-specific latency and private order state
SlippageArrival benchmark, actual or replayed order, size, fills, latency and feesFuture slippage under a different state or size

CME’s Market by Order FAQ says Market by Price consolidates quantity at price levels, while Market by Order supplies anonymous order-level details and priority information. A source missing the required granularity fails the corresponding use; a model estimate cannot be relabeled as observed queue data.

Do not repair silently

Quarantine Ambiguous Records and Preserve the Reason

A cleaning rule should be deterministic, versioned, and reversible. Keep the raw record immutable. Every exclusion or transformation needs a reason code, original values, transformed values, code version, and count by contract and date.

Failed for primary use

Required evidence is absent

  • Contract month missing
  • Event time inferred from receipt time
  • Quote side unavailable for fill claim
  • Unreconciled gaps or correction semantics

Secondary context only

Narrow facts remain usable

  • Daily settlement without intraday book
  • Cash reference with different instrument
  • Volume without spread or depth
  • Continuous chart with documented construction

Experimental only

Derived fields stay labeled

  • Estimated event time
  • Reconstructed quote state
  • Probabilistic roll mapping
  • Vendor claim awaiting payload proof
  • Do not convert missing quotes to zero spread. Preserve the interval as unavailable and report missingness.
  • Do not forward-fill through closures or outages. A stale price is not a live market.
  • Do not delete extreme moves by magnitude alone. Reconcile sequence, corrections, neighboring records, official events, and contract state first.
  • Do not overwrite source files. Store cleaned data as a new lineage with reproducible code and checksums.

Ending gate

6M Dataset Acceptance Checklist

Run this gate separately for every intended use. A source may pass daily settlement research and fail intraday execution research. Acceptance does not transfer across questions.

Authority

Current CME rule/specification and provider documentation are saved with review dates.

Payload

Actual records prove every required field, precision, timestamp and correction state.

Identity

Product, venue and dated contract resolve without inferred symbol mapping.

Units

Quote direction, currency units, price type and P&L use consistent definitions.

Time

Event time, UTC conversion, local offset, trade date, bar boundaries and DST are reproducible.

Coverage

Gaps, duplicates, stale records, outages, corrections and attrition are quantified.

Roll

Every continuous observation maps to dated contracts under a declared ex-ante rule.

Liquidity

The feed’s quote, depth and order granularity matches the exact execution claim.

Decision

Outcome is failed for primary use, secondary context only, experimental only, or accepted for the named use.

Research status as of August 13, 2026

No original study is reported. No vendor dataset or CME market-data file was acquired, inspected, reconciled, or certified for this article. The vendor-candle example is illustrative. This page defines a fail-closed audit; it does not certify any named provider or data history. Reconcile verified mechanics against the canonical 6M contract-specification guide.

Sources, methods and editorial disclosure — reviewed August 13, 2026

Sources and methods were reviewed August 13, 2026. This page is unsponsored editorial analysis and presents a dataset-audit protocol, not a provider certification or original market finding.