bridge quote benches (002, 264): headline was a random Prometheus cell; aggregated queries, EU-West pinned, meme route and exec quotes excluded, mean fixed, Mobula timing/auth, NI 3 s wait disclosed - #2532
Merged
Conversation
…ies now aggregate the recorded cells, region pinned to EU-West, meme route and execution-path quotes excluded, mean fixed, Mobula timed like the others and keyed by header, Near Intents' fixed 3 s wait disclosed
ocb-bench-audit on bridge-quote-latency-solana (and confirmed on the
production bridge-quote-latency page):
- Blocker: the recording rules keep (bridge, region, chain, amount_usd)
and the spec queries selected on bridge only, so
histogram_quantile(...{bridge="mobula"}) returned 39 series and
prom.scalar took result[0], which Prometheus orders by Go map
iteration. The live production page read Mobula p50 31 / p90 57 /
p99 4172 ms and Near Intents p99 108 below p90 4700: a different cell
on every materialization. Every p50/p90/p99 query is now
histogram_quantile(q, sum by (le) (...)), mean is
sum(rate_sum)/sum(rate_count) (the previous clamp_min(count_rate, 1)
floored a per-second rate to 1 and printed ms per second, 300x too
small), success is avg(...), n is round(sum(increase)). Verified on
the VPS Prometheus: one series, Mobula 43.7 / 345.8 / 982.6 ms, mean
118.3.
- Region pinned to eu-west on every headline query: the us-east probe is
live (the copy said paused) and was part of the random draw; sgp is
down. Copy corrected on both specs.
- Rules (infrastructure/monitoring/prometheus/recording_rules/
ocb_bridge_quote.yml, source of truth for the VPS file that was only
on the host, and ocb_bridge_quote_solana.yml) exclude the R4 TRUMP to
BRETT meme route the same loop quotes (it pooled a cross-asset swap
into the Base USDC cells: Mobula 34.6 ms USDC-only vs 139.3 pooled)
and the $3 / $30 execution-path quotes.
- Harness: execution-path quote latency goes to its own histogram
(bridge_exec_quote_latency_ms); Mobula's stopwatch stops after the
body is read like the five other providers; the Mobula key travels in
the Authorization header instead of the query string (a transport
error wrapped the full URL into the log and Slack); the quote
histogram gains 500 ms buckets between 1 and 5 s.
- Near Intents: its dry quote waits a fixed 3 s solver window
(quoteWaitingTimeMs: 3000), so its row measures that wait; disclosed
in the methodology and the finding, the "bimodal, sub-100 ms" claim
removed. Robinhood coverage bullet (four providers quote it now) and
the Across / Near Intents coverage (Base and Arbitrum only, our maps)
corrected.
Deployed: rules reloaded (promtool 6 + 5), harness rebuilt, Mobula
quotes verified with header auth. validate 236, go vet/build/test.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Found by the 264 audit and confirmed on production bench 002: the recording rules keep chain/region/amount labels, the spec queries filtered on bridge only, and
prom.scalartookresult[0](Go map order) → a different cell on every render (prod showed Mobula p99 4172 vs p90 57; Near Intents p99 < p90).histogram_quantile(q, sum by (le) (...)); meansum/sum(theclamp_min(rate,1)made it 300x too small); successavg; nround(sum(increase)). One deterministic series (Mobula 43.7 / 345.8 / 982.6 ms).Deployed on the VPS. This one should go to main (002 is live in prod with the random draw).
🤖 Generated with Claude Code
https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw