u0577-observability-regression-sentinel

# Observability Regression Sentinel

## Why This Skill Exists
We need this skill because decisions are only as good as the quality and visibility of data. This specific skill prevents unnoticed quality drift after updates.

## When To Use
Use this skill when the request explicitly needs "Observability Regression Sentinel" outcomes in the Data Quality and Observability domain.

## Step-by-Step Implementation Guide
1. Define the scope and success metrics for `Observability Regression Sentinel`, including at least three measurable KPIs tied to data drift and blind spots.
2. Design and version the input/output contract for freshness, drift, schema health, and telemetry coverage, then add schema validation and failure-mode handling.
3. Implement the core capability using baseline-delta detection, and produce regression watchlists with deterministic scoring.
4. Integrate the skill into swarm orchestration: task routing, approval gates, retry strategy, and rollback controls.
5. Add unit, integration, and simulation tests that explicitly cover data drift and blind spots, then run regression baselines.
6. Deploy behind a feature flag, monitor telemetry/alerts for two release cycles, and iterate thresholds based on observed outcomes.

## Deterministic Workflow Notes
- Core method: baseline-delta detection
- Archetype: detection-guard
- Routing tag: data-quality-and-observability:detection-guard

## Input Contract
- `freshness` (signal, source=upstream, required=true)
- `drift` (signal, source=upstream, required=true)
- `schema health` (signal, source=upstream, required=true)
- `telemetry coverage` (signal, source=upstream, required=true)
- `claims` (signal, source=upstream, required=true)
- `evidence` (signal, source=upstream, required=true)
- `confidence traces` (signal, source=upstream, required=true)

## Output Contract
- `regression_watchlists_report` (structured-report, consumer=orchestrator, guaranteed=true)
- `regression_watchlists_scorecard` (scorecard, consumer=operator, guaranteed=true)

## Validation Gates
1. **schema-contract-check** — All required input signals present and schema-valid (on fail: quarantine)
2. **determinism-check** — Repeated run on same inputs yields stable scoring and artifacts (on fail: escalate)
3. **policy-approval-check** — Approval gates satisfied before publish-level outputs (on fail: retry)

## Failure Handling
- `E_INPUT_SCHEMA`: Missing or malformed required signals → Reject payload, emit validation error, request corrected payload
- `E_NON_DETERMINISM`: Determinism delta exceeds allowed threshold → Freeze output, escalate to human approval router
- `E_DEPENDENCY_TIMEOUT`: Downstream or external dependency timeout → Apply retry policy then rollback to last stable baseline
- Rollback strategy: rollback-to-last-stable-baseline

## Handoff Contract
- Produces: Observability Regression Sentinel normalized artifacts; execution scorecard; risk posture
- Consumes: freshness; drift; schema health; telemetry coverage; claims; evidence; confidence traces
- Downstream routing hint: Route next to data-quality-and-observability:detection-guard consumers with approval-gate context

## Required Deliverables
- Capability contract: input schema, deterministic scoring, output schema, and failure modes.
- Orchestration integration: task routing, approval gates, retries, and rollback controls.
- Validation evidence: unit tests, integration tests, simulation checks, and rollout telemetry.

## Production Trigger Clarity
- Use only when this capability produces production-facing outcomes with measurable acceptance criteria.
- Do not invoke for exploratory brainstorming or unrelated domains; route those requests to the correct capability family.

## Deterministic Tolerances
- Repeated runs on identical inputs must remain within **<=1% output variance** for scoring fields and preserve schema-identical artifact shape.
- Any variance beyond tolerance is a hard failure and must trigger escalation.

## Fail-Closed Validation Gates
1. Schema validity gate (required inputs present and valid).
2. Determinism gate (variance within tolerance).
3. Policy/approval gate (required approvals satisfied).

If any gate fails: **block output publication and fail closed**.

## High-Risk Human Sign-Off
- Any high-risk change, policy-impacting output, or publish-level action requires explicit human sign-off before release.
- Missing sign-off is a blocking condition.

## Explicit Handoff Contract
- **Produces:** normalized artifacts, decision scorecard, risk/confidence metadata.
- **Consumes:** validated upstream inputs for this capability.
- **Next hop:** route only to declared downstream consumers with gate/approval context attached.