Trading confluence — dependencies, vetoes and decision time

Confluence is agreement among features considered in an analysis. It can organize a decision, but more checked boxes do not automatically provide more independent information. Record each feature’s source, role and time of availability before counting confirmations.

Different names can repeat one fact

Many tools reuse the same price series. RSI uses price changes; the MACD main line is a difference between two averages. Names do not establish statistical independence. A technical-analysis overview supplies broader context, rather than a probability assigned by a count of signals.

Assign EMA12 = 1.0854 and EMA26 = 1.0850 for the same EUR/USD series, settings and instant. Main MACD = 0.0004. Here “EMA12 above EMA26” and “MACD above zero” are algebraically equivalent. The difference is 4 pips under pip 0.0001, not a forecast of a four-pip move. The histogram compares MACD with its signal line instead.

RSI and MACD need not be identical or always agree. Shared inputs do not prove perfect correlation, yet do not justify independence either. Do not multiply assumed probabilities or add a fixed certainty percentage for each additional label.

Give conditions roles and identify vetoes

Our example organizes four roles: context defines where a candidate is considered; location specifies a previously recorded zone; timing specifies an available event; execution checks specified constraints. Distinct roles do not guarantee independence or an edge.

A support/resistance reference and a retracement reference can describe the same location. Their coincidence does not count institutional orders. MetaTrader’s Forex Volumes description concerns price changes in a period; a spike is not by itself evidence of large-capital participation.

Separate mandatory conditions from optional descriptions. In a plan requiring a spread limit, favorable price conditions cannot outvote failure of that limit. Missing required data means unknown, not permission to proceed.

A higher hit rate can accompany a worse mean result

Consider ten fictional already-costed outcomes in a fixed R unit: +0.1, +0.1, +0.1, +2, +2, +2, −1, −1, −1, −1. A hypothetical added filter retains the three small wins and one loss. It is not a measured market filter or a new backtest.

Before the filter: 0.60 × 1.05 − 0.40 × 1 = +0.23 R. After it: 0.75 × 0.10 − 0.25 × 1 = −0.175 R. Hit rate increases by 15 percentage points while the mean turns negative because the win sizes also change.

For unchanged data and an unchanged feature list, raising a count threshold can reduce or preserve qualification count; it does not guarantee higher precision or better results. No universal two/four/six-signal trade frequency, day-job suitability or hundred-trade certificate follows.

Preserve availability and rule versions

A condition using a completed candle becomes available after that candle ends. Current-bar and higher-timeframe values can change; a later historical shape cannot justify an earlier decision. Specify whether a previous signal expires instead of collecting favorable screenshots from separate moments.

The diagram assigns context [09:00,09:05), location [09:04,09:07) and trigger [09:06,09:08). Each occurs, but their common interval is empty. Extending context to [09:00,09:07) in the quiz creates [09:06,09:07), excluding 09:07. These are fictional validity policies, not recommended market waiting times.

The list need not be written one day before; it must exist before the evaluated event and be reproducible. Create a new version for future assessments while retaining prior decisions and the conditions actually known then. Extra justifications discovered after an outcome do not rewrite the original rule.

Three positive conditions do not cancel a failed spread limit

Use a separate fictional Bid EUR/USD series, UTC and pip 0.0001. Before 09:00 record zone 1.0848–1.0852, width 4 pips. Assigned anchors 1.0800 and 1.0900 have midpoint 1.0850. That midpoint describes the location; it is not an extra mandatory vote or proof of zone effectiveness.

At 09:00, completed H1 [08:00,09:00) has assigned Close 1.0853 and EMA20 1.0850. Later M15 [09:00,09:15) has O 1.0850, H 1.0858, L 1.0849, C 1.0856. The prior closed M15 High is 1.0854. Assess only after those completed data arrive at 09:15, assuming the next supplied quote is available then.

Spread is (1.0857 − 1.0855)/0.0001 = 2 pips. Reject the candidate at that decision under this policy; do not search for a fifth favorable indicator. The 1.5-pip limit is an example rather than a universal recommendation, and spread alone does not cover all costs or execution uncertainty.

In a later separate calculation, keeping Bid 1.0855 while assigning Ask 1.08565 gives spread 1.5 pips, satisfying ≤. If other model conditions remain valid, all four pass. That establishes qualification, not a fill or profit. A plan must separately define signal expiry and any later reassessment.

Record the complete decision, including rejections

Retain timestamps, source inputs, each required condition’s status, unknown fields, rejection reasons and rule version. Keep observations without trades as well as accepted candidates. Test whether an added feature changes outcomes rather than only duplicating an existing condition.

Use a setup checklist to make roles and vetoes explicit. More labels do not justify increasing exposure; a complete count still lacks win size, loss size, costs and dependence.

Compare complete rules on consistent data

Specify inputs, formation and expiry times, required conditions, entry/exit and execution assumptions before comparing alternatives. Evaluate all outcomes under the same cost and R conventions, including missed opportunities and rejected cases. Reserve data not used to select the filter.

A trade-count or calendar quota does not create a valid estimate or oblige continued trading. Investigate data faults and operational limits when they appear. Distinguish a useful description, reproducible qualification and evidence about net outcomes.