Process vs outcome

Definition

In any probabilistic activity, decision quality and outcome quality are different variables: good decisions lose regularly and bad decisions win regularly, because the outcome adds noise the decision never controlled. Process-vs-outcome grading scores each trade on BOTH axes — was the decision right given what was knowable, AND how did it land — producing a two-by-two whose off-diagonal cells (good decision/bad outcome, bad decision/good outcome) carry all the learning. The documented failure it prevents is OUTCOME BIAS (Kahneman): judging decisions by results, which in noisy domains teaches superstition with conviction.

How it works / structure

• The grading grid (engine-native):
  • Good process / good outcome: the target state — reinforce, but audit for luck’s share.
  • Good process / bad outcome: variance paid its visit — the CRITICAL cell: changing sound process after unlucky outcomes is the documented skill-destroyer; the grade protects the process.
  • Bad process / good outcome: the most dangerous cell — rewarded rule-breaking trains repetition at growing size until the variance bill arrives (bias-overconfidence compounding); the grade flags wins for censure.
  • Bad process / bad outcome: the honest teacher — cheap tuition if the lesson is recorded (disc-journaling-review).
• What “process” means concretely: adherence to the plan’s clauses (disc-trading-plan) — thesis stated with falsifier, size within rules, entries/exits per specification; process grading is mechanical against the record, not vibes about effort.
• The statistical frame: per-trade outcomes are draws from the strategy’s distribution — expectancy (mean) emerges only across samples (philosophy-geometric-compounding path arithmetic); the documented sample-size discipline: no strategy verdicts on fewer than dozens of independent trades, no process amendments on single outcomes.
• The platform translation: agents grade automatically — every decision logs its rule compliance, and outcome attribution separates thesis correctness, execution quality, and regime luck; the human version of this discipline is the machine version’s design spec.

When it applies

Every review cycle (the journal’s grading rubric); strategy evaluation windows (verdicts on process metrics and SAMPLES, not last-month P&L — bias-recency control); post-loss decision hygiene (the good-process/bad-outcome cell is where discipline either holds or dies); post-win audits (the bad-process/good-outcome flags are the cheapest disasters ever prevented).

Risk profile & failure modes

• Outcome bias (the target failure): results-based grading in a noisy domain — documented to corrupt both confidence and process; the grid is the antidote.
• Process worship: a consistently losing strategy executed flawlessly is still losing — process grades protect against NOISE, not against a wrong edge thesis; expectancy review across samples remains the edge’s judge.
• Grading dishonesty: post-hoc thesis softening (“I basically followed the plan”) corrupts the dataset — mechanical grading against the written record is the defense.
• Sample impatience: acting on off-diagonal cells before samples accumulate re-introduces the noise the discipline exists to filter.

Evidence & limits

Outcome bias and the illusion of validity are peer-reviewed (Kahneman and the judgment literature); the grading framework is decision-science standard applied to trading (practitioner form labeled). The discipline’s value is measurable in deviation costs and verdict stability — auditable per account.

Falsifiable-thesis examples

Illustrations only, not signals:

• “Strategies evaluated on 50+ trade samples show more stable quarter-over-quarter verdicts than last-10-trade evaluations (noise-filter check)” — falsified by the verdict-stability comparison.
• “Bad-process/good-outcome trades repeat at larger size when ungraded vs graded (censure-loop check)” — falsified by the tagged-ledger pattern.

Cross-references

• The dataset: disc-journaling-review; the clauses: disc-trading-plan
• The statistics: philosophy-geometric-compounding, risk-kelly-criterion (edge estimation)
• The biases at issue: bias-overconfidence, bias-recency