Data and Methodology
Transni translation reliability report: 30-day operational baseline
Quick Answer
This page publishes Transni's 30-day aggregate operational reliability baseline for text, voice, and image translation flows. The report emphasizes request success rate, p95 latency, and first-token responsiveness where available. Metrics are operational indicators, not direct claims about language correctness across all domains or every travel context.
Metrics Snapshot (30 Days)
Text request success rate
99.2%
Aggregate request success over 30 days after exclusion rules.
Voice request success rate
97.8%
Aggregate request success over 30 days after exclusion rules.
Image request success rate
96.4%
Aggregate request success over 30 days after exclusion rules.
Text request p95 latency
1.24s
p95 derived from route-level duration histogram in Prometheus.
Voice request p95 latency
2.82s
p95 derived from route-level duration histogram in Prometheus.
Image request p95 latency
3.41s
p95 derived from route-level duration histogram in Prometheus.
Text stream median first token
0.42s
Median first token for supported streaming text routes.
Voice stream median first token
0.69s
Median first token for supported streaming voice routes.
Image stream median first token
0.88s
Median first token for supported streaming image routes.
Scenario Breakdown
Text translation baseline (30 days)
Success rate: 99.2%. P95 latency: 1.24s. Median first token: 0.42s.
Voice translation baseline (30 days)
Success rate: 97.8%. P95 latency: 2.82s. Median first token: 0.69s.
Image translation baseline (30 days)
Success rate: 96.4%. P95 latency: 3.41s. Median first token: 0.88s.
How Transni Handles It
Method: 30-day rolling window
Metrics cover requests from 2026-02-01 through 2026-03-01 in production traffic.
Method: aggregate by translation mode
Request outcomes and latency were grouped by text, voice, and image routes using Prometheus counters and histograms.
Method: apply exclusion rules
Excluded malformed client payloads, duplicate retries within 2 seconds, and known synthetic monitoring probes.
Limitations
- Operational success does not equal semantic correctness for every language pair.
- Metrics are aggregated and do not represent each market, provider, or acoustic condition equally.
- First-token values are available for supported streaming paths and may be absent in non-streaming fallback paths.
Method and Exclusions
Data sources
- transni_api_requests_total
- transni_api_request_duration_seconds_bucket
- transni_api_request_duration_seconds_count
- transni_stream_errors_total
Exclusion rules
- Malformed client payloads rejected before translation pipeline execution.
- Duplicate retries received within 2 seconds of an identical request hash.
- Synthetic monitoring and internal health probes.
FAQ
- Are these numbers a translation quality benchmark?
- No. They are operational reliability metrics, not a full linguistic quality benchmark.
- How often is this page updated?
- This report is designed for weekly refresh with a rolling 30-day window.
- Do these metrics include every failed request type?
- No. The report excludes malformed payloads, rapid duplicates, and synthetic monitoring traffic.
- Can I compare these values directly to other tools?
- Only with caution, because request shapes, routing, and exclusion rules differ across systems.