Repository navigation
two venues crowned on numbers their own benches disown - #3082
Merged
Merged
Conversation
From the SEO/GEO audit (.saudit/sitemap, verdict FIX THEN SHIP). Both are claims four surfaces repeat and the bench behind them contradicts. A 7.5% gap published as a tie. /api/stat/aggregator-head-lag served "Mobula and Serialized both post the lowest head lag, tied at 0.3 s" with `leader` naming Mobula alone at 0.304262 and `leaders` listing Serialized at 0.327105 beside it. 34,488 samples each. leaders() treats two providers as tied when their FORMATTED values match, and that rule is right: a page must not claim a difference its own number does not show. The defect is upstream. The `sec` branch of fmtUnit printed one decimal below 10 s, and its comment called 0.1 s "fine grain enough for finality comparison" - true at that scale, false on a bench measured in milliseconds, where it merges 304 ms and 327 ms. Below a second the formatter now keeps two decimals, 0.30 s against 0.33 s, so the tie disappears at the source rather than being special-cased per surface. The contradiction had reached the meta description, the TL;DR, /api/stat's headline and /llms.txt, while the same page's at-a-glance box said "Leader Mobula", /data-api showed 295 ms against 328 ms and /products/mobula showed 293 ms at rank #1. GMX crowned on slippage by a bench that excludes GMX. /perps/eth served "GMX fills a $100,000 ETH market order with the least slippage at 0.00 bps" three sentences above its own FAQ answer saying "oracle-priced venues (GMX v2, Gains) have no book to walk". The bench agrees with the FAQ: /api/stat/perp-execution-quality?chain=ETH names Hyperliquid at 0.4746 across 9 ranked venues, and the spec's methodology says the excluded venues' "tiers are flat and would read as zero or constant slippage". The page does not read the bench's ranked set. It computes the figure itself, as perp_fees_all_in_bps_tier minus the taker fee, and an oracle-priced venue publishes a flat tier, so the subtraction gives zero and zero wins a lowest-is-best comparison. The existing guard could not catch it: it drops a venue whose book failed to absorb $100k on more than a tenth of ticks, and a venue with no book never fails that test. The two venues are listed, because a flat tier is indistinguishable from a genuinely deep book until you know why it is flat. The list must stay in step with the exclusion in the bench spec, and the comment says so.
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
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.
From the SEO/GEO audit (
.saudit/sitemap, verdict FIX THEN SHIP). Both areclaims four surfaces repeat and the bench behind them contradicts.
A 7.5 % gap published as a tie
/api/stat/aggregator-head-lagserved, live:34,488 samples each.
leaders()treats two providers as tied when theirformatted values match, and that rule is right: a page must not claim a
difference its own number does not show. The defect is upstream. The
secbranch of
fmtUnitprinted one decimal below 10 s, and its comment called0.1 s "fine grain enough for finality comparison" — true at that scale, false
on a bench measured in milliseconds, where it merges 304 ms and 327 ms.
Below a second the formatter now keeps two decimals (0.30 s against 0.33 s), so
the tie disappears at the source rather than being special-cased per surface.
The contradiction had reached the meta description, the TL;DR,
/api/stat'sheadline and
/llms.txt, while the same page's at-a-glance box said"Leader Mobula",
/data-apishowed 295 ms against 328 ms, and/products/mobulashowed 293 ms at rank #1.GMX crowned on slippage by a bench that excludes GMX
/perps/ethserved "GMX fills a $100,000 ETH market order with the leastslippage at 0.00 bps" three sentences above its own FAQ answer saying
"oracle-priced venues (GMX v2, Gains) have no book to walk".
The bench agrees with the FAQ:
/api/stat/perp-execution-quality?chain=ETHnames Hyperliquid at 0.4746across 9 ranked venues, and the spec's methodology says the excluded venues'
"tiers are flat and would read as zero or constant slippage".
The page does not read the bench's ranked set. It computes the figure itself,
as
perp_fees_all_in_bps_tierminus the taker fee, and an oracle-priced venuepublishes a flat tier, so the subtraction gives zero — and zero wins a
lowest-is-best comparison. The existing guard could not catch it: it drops a
venue whose book failed to absorb $100k on more than a tenth of ticks, and a
venue with no book never fails that test.
The two venues are listed rather than detected, because a flat tier is
indistinguishable from a genuinely deep book until you know why it is flat. The
list must stay in step with the exclusion in
perp-execution-quality.yml, andthe comment says so.
Gates: 252 specs valid, tsc clean, 626 tests pass (623 before; the three new
ones pin the formatter at the two real head-lag values).