Subjects
0 / 64
Included subjects passing evidence gates in the last 24h view.
Uptimer Observatory
Regional Truth Gap
Uptimer Observatory study measuring regional disagreement, latency spread, quiet failures, partial outages, and advisory anomaly lead signals.
Methodology: rtg-v1 | Last Updated: 2026-08-28 10:11:12Z
Evidence Quality: Poor
Headline metrics are suppressed due to insufficient evidence.
- Included subject coverage 0.0 % is below 60 %.
- Fallback-compromised rate 40.0 % exceeds 2 %.
Evidence is still being gathered...
What This Page Shows
This observatory compares how the same public website appears from multiple regions over time. It helps answer: "Do all regions agree, or is something inconsistent?"
- Higher disagreement means regions often report different health for the same subject.
- Higher latency spread means some regions are much slower than others at the same moment.
- Higher missing/fallback rates mean lower evidence quality, so conclusions should be treated cautiously.
- Anomaly lead is advisory only. It is an early warning signal, not proof of an incident.
Top-line Metrics (24h)
Regional Disagreement
N/A
How often regions disagreed on health state in the same bucket. Lower is better.
Quiet Failure Rate
N/A
Share of valid samples that were "HTTP success but degraded". Lower is better.
Anomaly Lead Rate
N/A
How often advisory anomaly signals appeared before visible degradation.
Missing Sample Rate
0.1 %
Share of region buckets with no usable sample. Lower is better.
Fallback-Compromised Rate
40.0 %
Samples excluded from regional truth because execution provenance was compromised. Lower is better.
Partial Outage Windows
0
Windows where mixed regional state persisted beyond a single bucket.
Methodology
Bucket size: 2 minute(s)
Aggregation cadence: 5 minute(s)
Configured regions: 5
Regional truth requires per-bucket quorum (>= max(3, 60% regions)); full quorum is >= 80%. Headline metrics are suppressed when evidence quality gates fail. Regional evidence excludes fallback-compromised samples, and advisory anomaly/predictive signals are non-deterministic.
Top Subjects by Disagreement / Latency Spread
Interpretation: disagreement and quiet failure are rates over recent buckets. Latency spread is the largest same-bucket difference between fastest and slowest valid region.
No public subject summaries available yet.
Recent Divergence Examples
These rows include divergence signals only (not data-quality alerts). They are signals to investigate, not legal/compliance conclusions.
No recent public incident examples available.
Data Quality Alerts
These alerts indicate reduced evidence quality (for example missing regional samples). Treat nearby divergence metrics with additional caution.
| Created | Type | Severity | Subject | Title | Why It Matters |
|---|---|---|---|---|---|
| 2026-08-28 10:10:33Z | data-quality | warning | TensorFlow | Data quality degradation | Evidence quality was reduced by missing samples or fallback-compromised execution provenance. |
Caveats and Limitations
Regional truth calculations exclude fallback-local execution samples from regional evidence and track them separately as compromised quality.
Advisory anomaly and predictive signals are non-deterministic indicators and are not deterministic guarantees of incidents.
Public outputs are intentionally minimized and do not include target credentials, internal tenant metadata, private monitor targets, or admin/compliance/billing records.