Skip to content

two of bench 313's columns said the opposite of what they measure - #3576

Merged
Flotapponnier merged 1 commit into
devfrom
fix/sellability-column-directions
Nov 7, 2026
Merged

Flotapponnier merged 1 commit into
devfrom
fix/sellability-column-directions

Conversation

@Flotapponnier

Copy link
Copy Markdown
Collaborator

Florent read the board and asked whether KyberSwap was the best. It is last: the headline is lower-is-better and KyberSwap routed into all ten failed tokens. Two columns were pushing him that way.

What was wrong

Quotes overall was marked lower-is-better, which tells a reader that quoting less is a virtue. It is not what this page measures and it is not true anywhere else on the site: coverage is better when there is more of it, and that column exists to explain a score, not to compete with it. Routed with a reservation carried the same inversion, saying that warning you less wins.

The rule applied

The schema has no neutral (higher_is_better: z.boolean().default(false)), so a column that counts a behaviour has to answer whether more of that behaviour helps the reader choosing a router.

panel counts direction
asked evidence behind the row more is better
coverage what it quotes in general more is better (was: less)
screened it refused and named the token more is better
reserved it routed and warned you more is better (was: less)
no_liquidity it had nothing to sell you left as it was

The headline is untouched and still lower-is-better, because it counts the harm.

For the record

KyberSwap at 100% routed into 10 of 10 failed tokens and is last. Its 98.3% coverage is the reason, not a second score: a router that quotes nearly everything routes into nearly every trap without selecting anything. LI.FI leads at 3 of 10 from a 78.7% baseline, which is the one row where the gap between the two columns is real selection.

Cache keys

bench-unfiltered-v190, all-benchmarks-v183 (lockstep), bench-filters-v108 (both tier tabs sort and colour those panels by the corrected direction).

Gates

pnpm validate 249 specs valid · pnpm exec tsc --noEmit clean · pnpm test 1077 pass 0 fail.

Deploy

Spec only: worker rebuild then materialize reset.

🤖 Generated with Claude Code

Quotes overall was marked lower-is-better, which tells a reader that
quoting less is a virtue. It is not what this page measures and it is not
true anywhere else on the site: coverage is better when there is more of
it, and the reason that column exists is to explain a score rather than to
compete with it. Routed with a reservation carried the same inversion,
saying that warning you less wins.

The schema has no neutral: higher_is_better is a boolean and omitting it
means lower-is-better, so a column that counts a behaviour has to answer
whether MORE of that behaviour helps the reader choosing a router.

  asked         more evidence behind a row          more is better
  coverage      what it quotes in general           more is better
  screened      it refused and named the token      more is better
  reserved      it routed and warned you            more is better
  no_liquidity  it had nothing to sell you          left as it was

The headline is untouched and still lower-is-better, because it counts the
harm: KyberSwap at 100% routed into all ten failed tokens and is last, not
first. Its 98.3% coverage is why, not a second score.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier
Flotapponnier merged commit a1f1c2d into dev Nov 7, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant