Service review

Platform reliability report

A static, PDF-ready report generated from trusted data.

Executive summary

Availability held at target while p95 latency improved after the June migration. One incident, fully recovered, is reviewed below.

Availability 99.97% +0.04 pts
p95 latency 172 ms −48 ms
Open incidents 1 +1 this week

Trend

Quarter over quarter, throughput rose +12.4% and error rate fell −0.3 pts. Cost moved +$1.2k (up is unwelcome here), while support backlog was unchanged.

Before / after

Before migration

p95 latency 220 ms

After migration

p95 latency 172 ms −48 ms

Findings

Major finding — retry policy still needs review

The migration improved latency, but retry volume rose during the incident window. Keep the migration and schedule a retry-policy review before the next rollout.

Impact: the risk is bounded today, but could consume error budget during a larger rollout.

Remediation: review retry-policy thresholds and archive the outcome with this report.

Caveat: the conclusion assumes the next rollout keeps traffic shape comparable.

The migration improved p95 latency while availability remained above target.

Evidence state: supported

Scope: service traffic reviewed through Jun 2, 2026

Basis: metrics export and reviewed incident note both support the claim.

Limit: retry volume still requires a separate threshold review.

Sources: S1 S2

Metric trend

Observation

Prometheus export · verified

Window: 7 days

−48 ms p95 latency

p95 latency fell 48 ms while availability stayed above target.

Source: S1

Incident log

Log

On-call note · reviewed

Recovery completed within the same hour with no manual rollback.

The on-call note says the alert opened at 09:12 and closed at 09:54 after automated retries drained. No manual rollback was required, but the reviewed note still recommends revisiting retry-policy thresholds before the next production rollout.

Source: S2

Assumption

Assumption

Capacity model · generated

The next rollout keeps traffic shape comparable to this review window.

Treat as a planning assumption until the next rollout profile is fixed.

Evidence

Claim, evidence, and caveat ledger
Claim Evidence Source Trust Relation Caveat
Latency improved p95 fell 48 ms while availability stayed above target. S1 Verified Supports Retry-volume review remains open.
No rollback needed Incident closed in the same hour. S2 Reviewed Supports Single incident window.
Fig 1 — weekly request mix
Weekly request mix Reads account for 68 percent of requests and writes account for 32 percent. Reads Writes 68% 32%
  • Reads
  • Writes
Request mix, last 7 days
Series Share Δ wk
Reads 68% +2
Writes 32% −2

Incident timeline

  1. Latency alert fired for p95 over threshold.
  2. Latency returned to baseline; incident closed.

Review retry-policy threshold before the next production rollout.

Owner: platform lead

Due:

Success criteria: threshold change accepted or the current policy is documented as intentionally retained.

Evidence: C1

Open

Archive this report with the source export and generated PDF.

Owner: report author

Due:

Success criteria: HTML, PDF, source export, and visual review notes are stored together.

Ready

Sources

Generated from verified operational exports No remote runtime assets

  1. Metrics export

    Prometheus · availability and latency window

    Weekly availability, latency, request mix and retry-volume readings.

  2. Incident review note

    On-call review · human reviewed

    Recovery timeline, mitigation notes and follow-up ownership.

Notes

Figures are illustrative. Measured against the modern Chromium print/PDF target.