- Target app:
azura-monorepo/apps/web - Focus routes:
/scope,/markets/spot*,/markets/pair-scanner, and broad/spot/:chainId/:tokencoverage - Trace scope: web -> tRPC -> API -> Talos-v2 and adjacent backends
- Included data:
16labeled staging runs +40-sample scope click probe
Totals across included runs:
- Requests:
83,302 - tRPC entries:
11,679 - OPTIONS preflights:
5,685(48.7%) - Non-OPTIONS tRPC calls:
5,994
Run families:
- Baseline swarm:
staging-swarm-spot-scope - Mid-size batches:
batch-* - Long-wave batches:
long-wave* - Full-manifest passes (54 URLs):
mega-full-cold-1,mega-full-warm-1,mega-full-cold-2
/scope behavior is highly variable and can fan out heavily:
- Worst baseline sample:
584requests,336tRPC entries,41late API starts (>2s) - Across scope runs, last non-OPTIONS tRPC completion:
- avg
4.7s - p95
8.65s - max
12.9s
- avg
Interpretation:
- The tail is caused by asynchronous request cascades, not first paint.
From full-manifest spot route class:
- Cold average:
259.5requests/route,4.52MBtransfer - Warm average:
256.8requests/route,0.136MBtransfer - Load-window difference is small (
220mscold avg vs215mswarm avg)
Interpretation:
- Cache reduces bytes significantly, but request orchestration and backend behavior still dominate perceived latency.
- Batch family preflight ratio:
47.6% - Long-wave family preflight ratio:
48.8% - Full-manifest family preflight ratio:
48.6%
Interpretation:
- Transport overhead is structural in current architecture.
v2.orders.quote aggregate:
- Calls:
353 - Failures:
159 - Failure rate:
45.0% - Status split:
200=194,500=159
Interpretation:
- Quote-path reliability has to be fixed before frontend-only tuning can deliver stable user wins.
Top procedures by call volume:
v1.tokens.getSecurity:1,206v2.twitter.getProfile:752v2.tokens.getPastTwitterUsernames:716v1.trades.initialTrades:522
Interpretation:
- Repeated security/social lookups are a major request amplifier.
Scope path:
- Web scope/memetics handlers trigger
v2.markets.newPairScanner.initialDataTalos - API fans out via Talos prelaunch/state reads
- Sparkline hydration adds additional API pressure (
getMarketCapSparkline)
Spot path:
- Spot page triggers metadata/stats/holders/traders/security/quote families
- Stats path reaches data-engine and Talos optimal-pool/state-service
- Holders/traders paths call Talos state-service endpoints + wallet label enrichment
- Quote path enters
apps/api/src/routers/v2/orders/quote.tsand Zenithar client module
Converged high-confidence priorities:
- Quote-path backend stabilization first.
- Scope and spot fan-out reduction (especially security/social duplication).
- Interaction-gating for secondary social requests.
- Transport reduction (same-origin/proxy + batching).
Key guardrails from skeptical pass:
- Avoid placebo fixes that reduce freshness for live trading views without SLO proof.
- Treat current data as high-confidence hotspot evidence, but use stricter paired design for causal claims.
- Keep backend and frontend changes separated for clear attribution.
- Quote reliability hardening (
timeout + retry policy + circuit behavior + typed degraded response). - Scope security/social dedupe and interaction gating.
- Add timeout budgets for Talos/downstream API wrappers lacking explicit abort control.
- Cut preflight overhead with same-origin routing and batched tRPC transport.
- Collapse high-frequency Talos/state fan-out where sequential upstream calls are avoidable.
- Add request-scoped dedupe on holders/traders wallet label enrichment.
- Fixed URL manifest and stratified token panel for repeated runs.
- Paired cold/warm block protocol with confidence intervals in report output.
- Persist cache/CDN metadata and count transport failures explicitly.
Expanded probe (2026-04-15T19-21-58):
- Samples:
40 - Success:
40/40 - Click-to-spot: avg
108ms, p95117ms
Interpretation:
- The navigation transition itself is fast; delay is primarily post-navigation data work.
- Data aggregate JSON:
apps/web/scripts/profiling/reports/staging-performance-gist-2026-04-15.json
- Human summary:
apps/web/scripts/profiling/reports/staging-performance-gist-2026-04-15.md
- Scope probe raw:
apps/web/scripts/profiling/results/results/2026-04-15T19-21-58/scope-click-probe.json