Observability engineering · 2026 · Solo — research, systems design, build, verification

Tail Scope

Tune error, latency, route, fallback, and capacity policy, then see exactly which completed traces survive.

The problem

Tail sampling can use information from a completed distributed trace, but ordered criteria and resource pressure make configuration difficult to reason about from text alone. Tail Scope turns that bounded decision into a local comparison without claiming OpenTelemetry Collector parity or production capacity estimates.

Architecture

Key decisions

01

Evaluate the whole trace in one pure engine

Span aggregation, ordered matching, stable hashing, and budget eviction return one evidence record. The UI does not re-derive policy decisions, which keeps behavior deterministic and testable.

02

Preserve high-signal decisions under pressure

When capacity drops, error and latency matches outrank route and probabilistic fallback matches. Every eviction still records the rule it matched before the budget gate.

03

Bound the model instead of overstating it

The release uses six English-only fixtures and four rule kinds. It explicitly excludes OTLP ingestion, Collector configuration parity, production memory estimates, export, and persistence.

04

Make publication evidence transactional

The private product repository is separate from this portfolio site. Production behavior, rendered English, responsive screenshots, and the Vercel status attached to the pushed Git SHA passed before this case study was added.

Metrics

6
automated tests passing
6
completed traces compared
4
ordered policy kinds
0
production browser console errors