# Reliability Domain

**Status:** Approved target-state specification; PARKED operationally  
**Activation:** I9  
**Dogfood priority:** lowest accepted practice priority.

## Target State

`ReliabilityPlan`, `SLI`, `SLO` and optional error-budget policy represent explicit service/journey reliability objectives when AWP becomes an externally operated product.

```text
Project/service/journey
  -> ReliabilityPlan
       -> SLI(s)
       -> SLO(s): target + window
       -> ErrorBudgetPolicy?
  -> telemetry evaluation
  -> release/approval/incident policy consequences
```

OpenSLO is preferred portable representation. Sloth/Pyrra may generate Prometheus-compatible rules when that telemetry stack applies.

## Invariants From Day 1

- formal SLO system does not block dogfooding I0–I8;
- immediate health/readiness and long-window reliability objectives remain distinct;
- telemetry backend is evidence source, not ReliabilityPlan authority;
- SLI measures user/service outcome meaningfully, not arbitrary available metric;
- SLO target/window changes are Decisions/policy revisions with history;
- error-budget policy consequences are explicit, not hidden deployment behavior;
- I0–I8 IDs/events preserve enough correlation to define future reliability measures.

## Core Primitives

### ReliabilityPlan

Scope, owner, revision, linked service/journey, SLIs/SLOs, optional error-budget policy and evidence/decision references.

### SLI

Definition of measured good/total or equivalent user-relevant success property, data source and query/aggregation semantics.

### SLO

Target and evaluation window over an SLI, with status/budget calculation.

### ErrorBudgetPolicy

Optional rule mapping budget state to actions such as production approval tightening, release pause, reliability work priority or incident escalation.

## Increment Realization

| Increment | Reliability realization |
|---|---|
| I0–I8 | health/telemetry/incidents only; preserve future seam. |
| I9 | activate ReliabilityPlan/OpenSLO mapping, evaluation and optional error-budget policy. |

## Current Implementation State

Parked by explicit owner priority. No implementation work should be pulled forward solely because this full target spec exists.

## Deferred Realization

Exact percentages, burn windows, dashboards, retention and policy thresholds require real operating/customer evidence.

## Temporary Dogfood Behavior

Use direct operational health and explicit incidents. Do not fabricate SLOs.

## Failure / Recovery

Unavailable telemetry marks reliability evaluation unknown/degraded; it does not invent compliance or automatically block unless policy explicitly defines such handling.

## UI Implications

I9 reliability surface summarizes user-relevant objectives, current burn/budget, evidence freshness and policy consequences with drilldown to telemetry/Incidents/Deployments.

## Decisions / ADRs

R5 accepted, operationally parked until I9/external productization.