Audit Sequence

Purpose

Create a descriptive, read-only sequence audit without turning association into causation or inventing system state.

Required inputs

• Variant, analysis unit, population, and source
• Metric definitions, numerators, denominators, and outcome counts
• Outreach period, observation window, and reply-attribution window
• Supplied segment fields and segment-level counts
• Comparison design, assignment, and timing, if any
• Known changes or possible confounders: audience, channel, sender, list quality, timing, deliverability, or personalization

Label absent denominators, definitions, windows, segments, comparisons, or confounders Missing; never treat draft review as delivery.

Workflow

1. State scope, source, unit, variant, and windows. Label absent items Missing.
2. Inventory supplied metrics and outcomes. Preserve definitions. Calculate ratios only for compatible numerators and denominators; label them Descriptive arithmetic, show the formula, and flag ambiguity.
3. Report supplied segment cuts with numerator, denominator, and window. If none, state Segment cuts: Missing — no segment-level results supplied. Do not invent segments.
4. Separate supplied confounders from plausible Unmeasured possibilities; never call a possibility fact or explanation.
5. Separate Can conclude from Cannot conclude. Without a comparable control or valid design, do not claim causation, lift, superiority, effectiveness, trend, or statistical significance. Do not declare a small sample sufficient or insufficient without a supplied decision rule; describe its sensitivity instead.
6. Offer bounded draft recommendations tied to gaps: define metrics, collect valid denominators, predeclare windows, or design a comparable test. Never scale a presumed winner.
7. Return DRAFT ONLY — READ-ONLY AUDIT and Actions taken: None.

Output format

# Sequence audit draft
- Scope / source / unit / windows: [supplied values or Missing]
- Status: DRAFT ONLY — READ-ONLY AUDIT
## Supplied metrics
- [metric — numerator / denominator — descriptive arithmetic — caveat]
## Supplied segment cuts
- [segment results, or Missing]
## Confounders
- Supplied: [items or None supplied]
- Unmeasured possibilities: [items, clearly labeled]
## Can conclude
- [descriptive finding]
## Cannot conclude
- [unsupported causal or comparative claim — reason]
## Draft recommendations
1. [reviewable proposal — gap addressed]
## Missing inputs
- [denominator, definition, time window, comparison, or other gap]
## Actions taken
- None

Guardrails

• Never infer causation from correlation or invent metrics, segments, attribution, delivery, significance, benchmarks, performance, or CRM/system state.
• Reject targeting, segmentation, or decisions based on protected traits (such as race, religion, disability, sex, or age) and use of unnecessary sensitive personal data.
• Do not browse, send, scale, pause, edit, schedule, update records, or mutate any external system.

Quality check

Confirm metrics, segments, confounders, and recommendations are reported; missing inputs are labeled; arithmetic is descriptive; conclusions respect the design; and no action occurred.

Example invocation

Use audit-sequence on one offline variant with 20 draft-reviewed touches and two positive replies, no control, and no supplied windows or segments. Diagnose only what the data supports.