Evaluate an indicator with a reproducible journal

A useful indicator evaluation starts with a question narrow enough to answer. "Does this chart look good?" is not reproducible. "Was each Phantom Combo Buy event visible at the completed bar under these settings?" is a question you can check.

Download the indicator evaluation CSV journal and keep a separate copy for each configuration. The worksheet is a manual observation aid. Phantom Flow does not automatically populate it or provide a built-in backtest through it.

Define the observation before looking ahead#

Record the edition and script version, symbol and exchange, chart type, session, timezone, timeframe, and settings. Choose the event name and a fixed observation interval in advance. If you are investigating gaps, include the requested gap timeframe.

Decide what counts as a new event and how you will handle an unconfirmed candle. Do not count an ongoing green trend as a new Buy signal on every bar. For confluence, use the actual same-bar definition.

Keep every eligible occurrence#

For each event, preserve its timestamp, first observed time, bar-confirmation state, and a screenshot reference. Include events followed by flat movement and movement in the opposite direction. Record missing observations as missing instead of silently dropping them.

Use a notes field to explain anomalies, such as an alert created after the event or a changed data feed. A row with incomplete evidence can still be useful, but it should not be treated as a fully verified observation.

Separate timing from hypothetical outcomes#

First check whether the event was available at the time you recorded. A label drawn on an earlier pivot is not proof of an earlier signal. Core higher-timeframe FVG history has a specific timing caveat described in repainting and confirmation.

If you separately study hypothetical price outcomes, define the measurement time, observation horizon, and cost assumptions before collecting results. Record that these are simulated observations. Spread, fees, slippage, and delayed reaction can change results; the indicator's label does not establish an executable fill.

This is a research-recording method, not a recommendation to use any entry, exit, instrument, or position size.

Keep development and evaluation separate#

Changing settings after viewing outcomes makes that interval part of your development process. Freeze the new configuration and inspect a different, previously unseen interval. Keep both records. Repeatedly choosing only the most attractive result hides how many alternatives you tried.

Report the sample size, missing rows, date range, settings, and observation method alongside any summary. A few screenshots cannot establish a win rate, and a high percentage alone cannot establish profitability.

A completion checklist#

• The exact version and configuration are recorded.
• Every eligible event in the chosen interval is accounted for.
• First availability is distinguished from the drawing's location.
• Changed or missing observations remain visible in the journal.
• Any hypothetical outcome uses rules fixed before inspection.
• The conclusion stays within the sample and behavior actually checked.

A valid conclusion may be that timing remains unverified or that the sample is insufficient. Keeping that limit explicit makes the next check more useful.