Verify exact production traces and record rollout
This commit is contained in:
@@ -1,215 +1,97 @@
|
||||
{
|
||||
"uid": "duocthu-grounding",
|
||||
"title": "Dược thư — Grounding & Retrieval",
|
||||
"tags": ["duocthu", "rag"],
|
||||
"uid": "duocthu-observability",
|
||||
"title": "Dược Thư — Request path observability",
|
||||
"tags": ["duocthu", "rag", "opentelemetry"],
|
||||
"timezone": "browser",
|
||||
"schemaVersion": 39,
|
||||
"version": 1,
|
||||
"version": 2,
|
||||
"refresh": "10s",
|
||||
"time": { "from": "now-1h", "to": "now" },
|
||||
"panels": [
|
||||
{
|
||||
"id": 1,
|
||||
"type": "stat",
|
||||
"title": "Số lần LLM bịa số và bị chặn",
|
||||
"description": "Generations discarded because they stated a figure that does not appear character-for-character in the cited source. This is the measured form of the claim that the answer layer cannot invent a dose. Non-zero is not a failure — it is the guardrail doing its job.",
|
||||
"gridPos": { "h": 6, "w": 6, "x": 0, "y": 0 },
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum(duocthu_generation_rejected_total{reason=\"ungrounded_number\"})",
|
||||
"legendFormat": "blocked"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "short",
|
||||
"decimals": 0,
|
||||
"color": { "mode": "thresholds" },
|
||||
"thresholds": {
|
||||
"mode": "absolute",
|
||||
"steps": [
|
||||
{ "color": "green", "value": null },
|
||||
{ "color": "orange", "value": 1 }
|
||||
]
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"graphMode": "area",
|
||||
"textMode": "value",
|
||||
"colorMode": "value",
|
||||
"reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false }
|
||||
}
|
||||
"title": "Request rate",
|
||||
"gridPos": { "h": 5, "w": 6, "x": 0, "y": 0 },
|
||||
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
||||
"targets": [{ "refId": "A", "expr": "sum(rate(duocthu_requests_total{route=\"/v1/rag/query\"}[5m]))", "legendFormat": "requests/s" }],
|
||||
"fieldConfig": { "defaults": { "unit": "reqps", "decimals": 2 }, "overrides": [] },
|
||||
"options": { "graphMode": "area", "colorMode": "value", "reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false } }
|
||||
},
|
||||
{
|
||||
"id": 2,
|
||||
"type": "stat",
|
||||
"title": "Tỷ lệ câu trả lời có kiểm chứng",
|
||||
"description": "Share of served answers that were LLM-generated and passed grounding verification. The remainder are served as verbatim source text — safe, just less readable.",
|
||||
"gridPos": { "h": 6, "w": 6, "x": 6, "y": 0 },
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum(duocthu_generation_served_total) / clamp_min(sum(duocthu_generation_served_total) + sum(duocthu_answer_extractive_total), 1)",
|
||||
"legendFormat": "verified"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "percentunit",
|
||||
"decimals": 1,
|
||||
"min": 0,
|
||||
"max": 1,
|
||||
"color": { "mode": "thresholds" },
|
||||
"thresholds": {
|
||||
"mode": "absolute",
|
||||
"steps": [
|
||||
{ "color": "red", "value": null },
|
||||
{ "color": "orange", "value": 0.5 },
|
||||
{ "color": "green", "value": 0.8 }
|
||||
]
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"graphMode": "area",
|
||||
"textMode": "value",
|
||||
"colorMode": "value",
|
||||
"reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false }
|
||||
}
|
||||
"title": "Request p95",
|
||||
"description": "The latency histogram carries Tempo trace exemplars. Click an exemplar in the latency panel below to inspect the complete request path.",
|
||||
"gridPos": { "h": 5, "w": 6, "x": 6, "y": 0 },
|
||||
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
||||
"targets": [{ "refId": "A", "expr": "histogram_quantile(0.95, sum by (le) (rate(duocthu_request_duration_seconds_bucket{route=\"/v1/rag/query\"}[5m])))", "legendFormat": "p95" }],
|
||||
"fieldConfig": { "defaults": { "unit": "s", "decimals": 2 }, "overrides": [] },
|
||||
"options": { "graphMode": "area", "colorMode": "value", "reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false } }
|
||||
},
|
||||
{
|
||||
"id": 3,
|
||||
"type": "stat",
|
||||
"title": "Tỷ lệ từ chối trả lời",
|
||||
"description": "Share of requests the system declined. A medical reference tool is expected to abstain — symptom questions, invented drug names and out-of-scope asks all land here by design.",
|
||||
"gridPos": { "h": 6, "w": 6, "x": 12, "y": 0 },
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum(duocthu_abstention_total) / clamp_min(sum(duocthu_abstention_total) + sum(duocthu_generation_served_total) + sum(duocthu_answer_extractive_total), 1)",
|
||||
"legendFormat": "abstained"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "percentunit",
|
||||
"decimals": 1,
|
||||
"min": 0,
|
||||
"max": 1,
|
||||
"color": { "mode": "fixed", "fixedColor": "blue" }
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"graphMode": "area",
|
||||
"textMode": "value",
|
||||
"colorMode": "value",
|
||||
"reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false }
|
||||
}
|
||||
"title": "Answerable ratio",
|
||||
"gridPos": { "h": 5, "w": 6, "x": 12, "y": 0 },
|
||||
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
||||
"targets": [{ "refId": "A", "expr": "sum(rate(duocthu_decision_total{decision=\"answerable\"}[5m])) / clamp_min(sum(rate(duocthu_decision_total[5m])), 0.000001)", "legendFormat": "answerable" }],
|
||||
"fieldConfig": { "defaults": { "unit": "percentunit", "min": 0, "max": 1, "decimals": 1 }, "overrides": [] },
|
||||
"options": { "graphMode": "area", "colorMode": "value", "reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false } }
|
||||
},
|
||||
{
|
||||
"id": 4,
|
||||
"type": "stat",
|
||||
"title": "Câu đi đúng đường section route",
|
||||
"description": "Retrievals resolved by section filter — the route measured at 16/16 on human-written questions. The remainder fall back to similarity, measured at hit@1 0.544.",
|
||||
"gridPos": { "h": 6, "w": 6, "x": 18, "y": 0 },
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum(duocthu_retrieval_route_total{route=\"section\"}) / clamp_min(sum(duocthu_retrieval_route_total), 1)",
|
||||
"legendFormat": "section"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "percentunit",
|
||||
"decimals": 1,
|
||||
"min": 0,
|
||||
"max": 1,
|
||||
"color": { "mode": "thresholds" },
|
||||
"thresholds": {
|
||||
"mode": "absolute",
|
||||
"steps": [
|
||||
{ "color": "red", "value": null },
|
||||
{ "color": "orange", "value": 0.6 },
|
||||
{ "color": "green", "value": 0.85 }
|
||||
]
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"graphMode": "area",
|
||||
"textMode": "value",
|
||||
"colorMode": "value",
|
||||
"reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false }
|
||||
}
|
||||
"title": "Trace-write failures",
|
||||
"gridPos": { "h": 5, "w": 6, "x": 18, "y": 0 },
|
||||
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
||||
"targets": [{ "refId": "A", "expr": "sum(increase(duocthu_trace_write_failed_total[1h]))", "legendFormat": "fail-open writes" }],
|
||||
"fieldConfig": { "defaults": { "unit": "short", "decimals": 0 }, "overrides": [] },
|
||||
"options": { "graphMode": "area", "colorMode": "value", "reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false } }
|
||||
},
|
||||
{
|
||||
"id": 5,
|
||||
"type": "timeseries",
|
||||
"title": "Vì sao một bản sinh bị loại",
|
||||
"description": "Every reason a generation was discarded before reaching a clinician. `ungrounded_number` is a fabrication caught; `provider_unavailable` is an outage; `evidence_insufficient` is the model correctly declining.",
|
||||
"gridPos": { "h": 9, "w": 12, "x": 0, "y": 6 },
|
||||
"title": "End-to-end latency — click exemplar for Tempo trace",
|
||||
"gridPos": { "h": 9, "w": 12, "x": 0, "y": 5 },
|
||||
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum by (reason) (rate(duocthu_generation_rejected_total[5m]))",
|
||||
"legendFormat": "{{reason}}"
|
||||
}
|
||||
{ "refId": "A", "expr": "histogram_quantile(0.50, sum by (le) (rate(duocthu_request_duration_seconds_bucket{route=\"/v1/rag/query\"}[5m])))", "legendFormat": "p50" },
|
||||
{ "refId": "B", "expr": "histogram_quantile(0.95, sum by (le) (rate(duocthu_request_duration_seconds_bucket{route=\"/v1/rag/query\"}[5m])))", "legendFormat": "p95" },
|
||||
{ "refId": "C", "expr": "histogram_quantile(0.99, sum by (le) (rate(duocthu_request_duration_seconds_bucket{route=\"/v1/rag/query\"}[5m])))", "legendFormat": "p99" }
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "reqps",
|
||||
"custom": {
|
||||
"drawStyle": "line",
|
||||
"lineWidth": 2,
|
||||
"fillOpacity": 12,
|
||||
"showPoints": "never",
|
||||
"stacking": { "mode": "normal", "group": "A" }
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": { "displayMode": "table", "placement": "right", "calcs": ["sum"] },
|
||||
"tooltip": { "mode": "multi", "sort": "desc" }
|
||||
}
|
||||
"fieldConfig": { "defaults": { "unit": "s", "custom": { "drawStyle": "line", "lineWidth": 2, "fillOpacity": 10, "showPoints": "never" } }, "overrides": [] },
|
||||
"options": { "legend": { "displayMode": "table", "placement": "bottom", "calcs": ["lastNotNull"] }, "tooltip": { "mode": "multi", "sort": "desc" } }
|
||||
},
|
||||
{
|
||||
"id": 6,
|
||||
"type": "timeseries",
|
||||
"title": "Vì sao hệ thống từ chối trả lời",
|
||||
"description": "Abstentions by the reason retrieval gave. `drug_not_resolved` dominating means most refusals are questions that never named a drug in the formulary — symptom questions and invented names.",
|
||||
"gridPos": { "h": 9, "w": 12, "x": 12, "y": 6 },
|
||||
"targets": [
|
||||
{
|
||||
"refId": "A",
|
||||
"expr": "sum by (reason) (rate(duocthu_abstention_total[5m]))",
|
||||
"legendFormat": "{{reason}}"
|
||||
}
|
||||
],
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"unit": "reqps",
|
||||
"custom": {
|
||||
"drawStyle": "line",
|
||||
"lineWidth": 2,
|
||||
"fillOpacity": 12,
|
||||
"showPoints": "never",
|
||||
"stacking": { "mode": "normal", "group": "A" }
|
||||
}
|
||||
},
|
||||
"overrides": []
|
||||
},
|
||||
"options": {
|
||||
"legend": { "displayMode": "table", "placement": "right", "calcs": ["sum"] },
|
||||
"tooltip": { "mode": "multi", "sort": "desc" }
|
||||
}
|
||||
"title": "Stage p95 latency",
|
||||
"gridPos": { "h": 9, "w": 12, "x": 12, "y": 5 },
|
||||
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
||||
"targets": [{ "refId": "A", "expr": "histogram_quantile(0.95, sum by (stage, le) (rate(duocthu_stage_duration_seconds_bucket[5m])))", "legendFormat": "{{stage}}" }],
|
||||
"fieldConfig": { "defaults": { "unit": "s", "custom": { "drawStyle": "line", "lineWidth": 2, "fillOpacity": 8, "showPoints": "never" } }, "overrides": [] },
|
||||
"options": { "legend": { "displayMode": "table", "placement": "right", "calcs": ["lastNotNull"] }, "tooltip": { "mode": "multi", "sort": "desc" } }
|
||||
},
|
||||
{
|
||||
"id": 7,
|
||||
"type": "timeseries",
|
||||
"title": "Decision and reason rate",
|
||||
"gridPos": { "h": 9, "w": 12, "x": 0, "y": 14 },
|
||||
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
||||
"targets": [{ "refId": "A", "expr": "sum by (decision, reason) (rate(duocthu_decision_total[5m]))", "legendFormat": "{{decision}} · {{reason}}" }],
|
||||
"fieldConfig": { "defaults": { "unit": "reqps", "custom": { "drawStyle": "line", "lineWidth": 2, "fillOpacity": 12, "showPoints": "never", "stacking": { "mode": "normal", "group": "A" } } }, "overrides": [] },
|
||||
"options": { "legend": { "displayMode": "table", "placement": "right", "calcs": ["sum"] }, "tooltip": { "mode": "multi", "sort": "desc" } }
|
||||
},
|
||||
{
|
||||
"id": 8,
|
||||
"type": "timeseries",
|
||||
"title": "Provider failures",
|
||||
"gridPos": { "h": 9, "w": 12, "x": 12, "y": 14 },
|
||||
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
||||
"targets": [{ "refId": "A", "expr": "sum by (provider, operation, reason) (rate(duocthu_provider_failure_total[5m]))", "legendFormat": "{{provider}} · {{operation}} · {{reason}}" }],
|
||||
"fieldConfig": { "defaults": { "unit": "reqps", "custom": { "drawStyle": "bars", "lineWidth": 1, "fillOpacity": 35, "showPoints": "never" } }, "overrides": [] },
|
||||
"options": { "legend": { "displayMode": "table", "placement": "right", "calcs": ["sum"] }, "tooltip": { "mode": "multi", "sort": "desc" } }
|
||||
}
|
||||
]
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user