Anomaly Taxonomy
The Triadic Observer Layer does not accuse, adjudicate, or infer intent.
It classifies patterns of incoherence so humans and institutions can reason about them without collapsing into narrative or blame.
This taxonomy exists to separate signal from speculation.
Purpose of Classification#
Anomalies are not failures.
They are informational discontinuities that deserve context.
The observer classifies anomalies to:
- Preserve legibility under uncertainty.
- Distinguish mistakes from structural stress.
- Prevent premature conclusions.
- Enable proportional response.
Classification is descriptive, not moral.
Classification Principles#
- No category implies wrongdoing.
- Categories may overlap.
- Unresolved anomalies are valid states.
- Reclassification is allowed as new observations arrive.
- Silence is preferable to forced certainty.
Primary Anomaly Classes#
Clerical or Mechanical#
Errors consistent with human or system handling.
Examples:
- Transcription mistakes.
- Formatting mismatches.
- Delayed uploads.
- Duplicate emissions.
Characteristics:
- Localized.
- Low magnitude.
- Often self‑correcting.
Procedural Deviation#
Departures from expected phase order or process flow.
Examples:
- Phase transitions occurring out of sequence.
- Missing intermediate artifacts.
- Unexpected aggregation paths.
Characteristics:
- Structural, not numerical.
- Often policy‑ or process‑related.
- Requires explanation, not accusation.
Temporal Incoherence#
Timing patterns that break expected continuity.
Examples:
- Sudden jumps after long stalls.
- Out‑of‑order timestamps.
- Retroactive corrections without lineage.
Characteristics:
- Time‑axis dominant.
- Often caused by batching or synchronization issues.
- Meaningful even when values are correct.
Statistical Outlier#
Values that diverge significantly from contextual expectations.
Examples:
- Magnitude spikes.
- Distribution anomalies.
- Rate changes inconsistent with prior phases.
Characteristics:
- Context‑dependent.
- Requires comparative scope.
- Never sufficient alone to imply error or intent.
Source Divergence#
Conflicting observations from different sources.
Examples:
- Parallel systems reporting different values.
- External observers disagreeing with internal reports.
- Audit findings diverging from operational data.
Characteristics:
- Expected in distributed systems.
- Often resolved through lineage inspection.
- Valuable for trust restoration.
Unresolved Inconsistency#
Anomalies that persist without sufficient explanation.
Examples:
- Missing artifacts.
- Conflicting phases without reconciliation.
- Corrections without provenance.
Characteristics:
- Legitimate state.
- Requires patience and transparency.
- Not a conclusion.
Composite Anomalies#
Multiple anomaly classes may apply simultaneously.
Example:
- A late correction (temporal incoherence) from a secondary system (source divergence) producing a magnitude jump (statistical outlier).
The observer preserves all classifications without collapsing them.
What the Observer Never Does#
The observer never:
- Infers intent.
- Labels fraud.
- Assigns blame.
- Resolves disputes.
- Suppresses anomalies for coherence.
Resolution belongs outside the observer.
Why This Taxonomy Matters#
Most trust failures occur when:
- Mistakes are treated as malice.
- Uncertainty is treated as incompetence.
- Silence is treated as concealment.
This taxonomy allows systems to say:
“Something is happening here — and we are not pretending otherwise.”
That posture preserves legitimacy.
Anomalies are not threats.
They are invitations to look more closely.
The Triadic Observer Layer exists to make that looking possible without panic.
This taxonomy keeps the observer firmly in a diagnostic, non‑accusatory posture, which is essential if it’s going to be tolerated in sensitive domains like elections while still generalizing cleanly elsewhere.