Methods / sources / limits

What the numbers mean

The site draws exported observations. Every result needs a run, a data source, a sample label and its costs. Missing evidence stays missing.

Data and provenance

A source chip identifies the exported run, feed and sample. Its date range belongs to that figure. A signed upload establishes who published the bytes; it does not prove that a strategy works.

Synthetic fixtures exercise the pages and charts. They are illustrations, never trading evidence. A zero commit hash in a fixture is a placeholder. When no real bundle is published, the site does not substitute fixture results for performance.

The project build notes describe real MNQ bars at five-minute resolution from August 2026 onward, with a QQQ price proxy for earlier history. The source recorded on each run takes precedence over this project description. A real-price backtest remains hypothetical.

Fixture illustrations

The styleguide uses deterministic seeded bars and exported synthetic outcomes to test rendering. Its commission and slippage values are test settings, not production defaults. No strategy is evaluated by these fixtures.

The QQQ proxy

A scaled equity ETF series cannot reproduce a futures market's tick structure, spread, session coverage or executions. Agreement in a short overlap does not establish equivalence outside it.

The build notes report an overlap check of 43 days and 26 trades on the same side. No validating run ID, dates or underlying trade table accompanies that statement here. Treat it as an unverified project note, not a published result.

Run id not supplied · project build notes §7.8 · reported proxy check, unverified · Methodology

Keep samples separate

In sample: Historical observations available during development or parameter selection. Results can reflect those choices.
Out of sample: Historical observations reserved from that selection stage. Once inspected and used to revise a model, they are no longer untouched evidence for the revision.
Paper: Virtual trades recorded as time moves forward. No funded execution is implied.
Live: Actual funded execution, only when a published source explicitly establishes it. A field named live_r in a parity file is not enough.

The Backtests explorer keeps the exported in/out labels and shows their boundary. Combining them for an explicitly labeled full-run view does not turn development data into independent evidence. Paper and backtest totals are never combined.

Fills and costs

Entries, stops, targets, exits and net trade values come from c3.lab. The browser aggregates exported outcomes; it does not decide fills or run trading rules. Prices and costs must be read with the contract and run that produced them.

Figure captions give round-trip commission in USD per contract and entry/stop slippage in ticks when exported. These assumptions can differ by run. Target fills, ambiguous bar order, gaps and unfilled orders depend on the exporter; bar OHLC alone cannot prove an executable fill. Missing fill assumptions are a limit, not zero cost.

Net USD and net R are already after the exported costs. The site does not subtract them a second time. Legacy paper snapshots report commission-adjusted virtual results but may omit slippage details; those omissions are labeled.

Walk-forward evaluation

A walk-forward study selects a configuration using an earlier window, fixes it, and evaluates it on the next window. Repeating that sequence can show how a method behaves across time. The run must supply its actual window boundaries and selection procedure to support that claim.

The site displays the exported walk_forward kind and sample labels. It does not perform selection or independently certify the procedure. If window details are absent, the label alone cannot establish that leakage was avoided.

Locked holdout

A holdout is a reserved period that must remain unavailable while a candidate is developed. Lock the rules before reading it. Repeated inspection makes selection easier and weakens independence, even when each individual test appears reasonable.

The build notes specify a maximum of six holdout looks per candidate. That is a proposed research protocol, not a verified property of the published results. No verified look ledger is published here. A sample label cannot certify compliance.

Random-walk check

A randomized baseline asks whether an apparent pattern could arise without the claimed structure. Its generator, seed, sample count, costs and comparison statistic belong with the exported study. A favorable comparison does not establish a market edge or predict a future result. No baseline result is claimed when those artifacts are absent.

The paper track record

The Track record page uses only the signed paper snapshot. Older snapshots can provide totals and monthly values without daily equity. In that case the curve stays absent. A missing run ID or market-data source is shown as unavailable, never borrowed from a backtest.

Daily equity is cumulative closed-trade net USD per micro contract since the export's start. Dates must ascend. When supplied, daily changes, closing drawdown and the final total must reconcile. An inconsistent series is withheld while signed aggregate totals remain visible. This is a consistency check, not a strategy evaluation.

Paper trades do not establish real fills, capital requirements or funded performance. Virtual commission and slippage assumptions can understate actual costs. Receipt time is when the site received the signed snapshot; it is not a trade timestamp.

Parity is a diagnostic

Each exported pair compares recorded observed R with hypothetical backtest R for a day. The two values appear only in the separate parity section. They never enter the paper performance totals or equity curve.

The export's verdict and causes are shown as saved. A mismatch may reflect feed differences, an uptime gap, bar timing or fill assumptions; the site does not infer a cause. Observed R may itself be virtual. Agreement does not prove funded execution or future agreement.

How to read the engine

The NQ engine's reversal and continuation work is in build. Its process diagram is not a performance claim. MNQ strategies and their evidence are labeled separately. A node has no operational state unless one was exported.

Feed · §3: Closed bars from the configured data feed. The site receives exports and does not fetch a live price stream.
Structure · §4: Saved swings, breaks and reference levels. Only exported observations appear in a replay.
Context · §5: The recorded bias, trend and session reference levels available at that point in time.
Intraday · §6: Exported time windows describe when candidates could arm. The browser does not authorize a trade.
Models · §8–9: A candidate's model and its published explanation. Theory-only models have no performance figures.
Score · §10: The candidate's exported score and, when supplied, its threshold. The dial never calculates a trading score.
Gates · §11: Saved pass/fail checks. The displayed verdicts do not rerun the engine's tests.
Tracker · §12: A virtual trade's saved entry, stop, targets, exit and costs.
Alerts · §13: Notifications for a person. The site does not place orders or connect to an execution account.
Ledger and storage · §14: The exported record of closed trades and decisions, versioned by a bundle.
Site · publish contract: The Worker checks signed uploads, file sizes and hashes before changing the published pointer. Pages read that record and label its limits.

Known limits

The real-bar history described in the build notes is short. Results from a few sessions or trades can change sharply.
A QQQ proxy differs from futures. Overnight coverage, missing bars, time zones and session boundaries can change which setups appear.
Bar data cannot establish intrabar event order or available liquidity. Replays are observations of an export, not an execution simulator.
Strategy selection, repeated testing and later revisions can make historical results look stronger. An out-of-sample label does not erase that history.
Drawdown here uses daily closing results with a zero starting balance. It excludes intraday and open-position losses, margin and financing. Net/drawdown is undefined when closing drawdown is zero.
A small sample warning is a display cue. It is not a test of significance. Best, worst and median sessions describe the export; none is an expectation.
Run IDs and source chips make evidence traceable. They do not certify data accuracy, independence or profitability.

Alert-only. Educational, not financial advice. Futures carry substantial risk of loss. Past and hypothetical results do not guarantee future results. Hypothetical trades do not represent actual execution and can understate the effects of liquidity, costs and financial risk.

Glossary

R: Trade outcome relative to its defined initial risk. Read the exported net R and cost assumptions; the site does not reconstruct risk.
MFE / MAE: Maximum favorable / adverse excursion: the largest recorded move for or against a trade while open. Units and bar resolution belong to the export.
Drawdown: The fall from a previous equity peak. This site's daily closing measure starts from zero and can miss deeper intraday losses.
Profit factor: Gross positive outcomes divided by the absolute gross negative outcomes in the same units. Undefined when there are no negative outcomes.
ORH / ORL: Opening-range high / low. The strategy defines the opening-range window.
PDH / PDL: Previous-day high / low. Session definitions affect the levels.
ONH / ONL: Overnight high / low. The export defines the overnight window.
FVG: Fair value gap: a three-bar imbalance label, commonly a gap between the outer bars' price ranges. The strategy's exact definition belongs to its rules.
MSS / BOS: Market structure shift / break of structure: names for recorded breaks of reference swings. A label alone does not specify a trigger or establish predictive value.