mμ│LLM

Chat Dashboard Performance Accuracy Games

PERFORMANCE

Real-time pipeline performance from scoring_log.jsonl Refreshes every 5s

Loading performance data...

LATENCY PERCENTILES

P50 (median)

—

Half of queries faster than this

P95

—

95% of queries faster than this

P99

—

Worst 1% tail latency

Average

—

Mean latency across all queries

THROUGHPUT & HEALTH

Requests/min (5m window)

—

Queries/sec

—

Tokens/sec

—

Error rate

—

Cache hit rate

—

Cost/hour

—

Uptime

—

Total requests

—

SLOWEST QUERIES

Latency	Tier	Model	Category	Actions
Loading...

TIER DISTRIBUTION

MODEL USAGE

CATEGORY DISTRIBUTION

REAL BROWSER PERFORMANCE (Playwright — actual user experience)

Chat Full Load

—

Navigation to load event

Chat FCP

—

First Contentful Paint

Chat Interactive

—

Input accepts typing

Dashboard Load

—

Full page load

Unblock Load

—

Full page load

Query Response

—

Type → click → render (avg)

Run npx playwright test tests/perf-benchmark.spec.js to update these numbers. Last run stored in localStorage.

BENCHMARK SNAPSHOTS Archived milestone comparisons

BEFORE (baseline)

AFTER (optimized)

QUERY TRACE

Time	Query	Tier	Model	Latency	Cost
No queries traced yet

Time	Dir	Method	Params	Status
No MCP calls traced yet

muLLM Performance Monitor · Data from scoring_log.jsonl · Raw JSON