Skip to content

Promote dev to main: Serialized onboarding, bench scoring fixes, 001 node reference - #2269

Open
Flotapponnier wants to merge 189 commits into
mainfrom
dev
Open

Promote dev to main: Serialized onboarding, bench scoring fixes, 001 node reference#2269
Flotapponnier wants to merge 189 commits into
mainfrom
dev

Conversation

@Flotapponnier

@Flotapponnier Flotapponnier commented Sep 5, 2026

Copy link
Copy Markdown
Collaborator

Staging state on dev since 2026-09-05, reviewed at https://staging-openchainbench.vercel.app.

What this promotes

Serialized (serialized.xyz) as a provider

Bench integrity fixes, all as separate series so history is preserved

Audit doc docs/methodology/serialized-onboarding-audit.md v1.3 (#2264, #2265): every number, every correction, two methodology defects in our own benches, a verified 5.2x BONK mispricing traced to a pool-ranking unit bug on their side.

What merging to main does beyond the site

The three Railway head-lag services (Aggregator east usa, Agg eu west, agg sgp) build from main, root harnesses/aggregator-head-lag. After this merge, trigger a fresh git build on each (serviceInstanceDeployV2); SERIALIZED_API_KEY is already set on all three. Agg eu west has three failed snapshot-upload deploys since 2026-09-05 14:32 (lstat …/harnesses: no such file or directory), a railway up from a directory without the harness path; a git build replaces it.

Known state at promotion time

  • 004: Serialized is producing (p50 53.3% on the spec query) but the leaderboard still shows the pre-outage snapshot because Mobula/Codex/Jupiter quantiles are NaN while the dead hours (webshare proxy quota, 429) remain in the 24h window; quorum fails 1/4 and the worker keeps the previous render. Self-heals once those hours roll out.
  • 008: Serialized 3rd at 76.75% (n=1,188). 005: last at 19. 090: 4th of 5 at 19 (Sim by Dune is dead: api.sim.dune.com no longer resolves).

🤖 Generated with Claude Code

https://claude.ai/code/session_01CpArutAtXuBb1BVNUDXoYA

@Flotapponnier
fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chain…
c79087c 
…s only (#2187)
@Flotapponnier
fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#…
61c8997 
…2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.
@Flotapponnier
fix: include still-open Gains positions in carry projection (#2189)
e36c3a5 
* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection
@Flotapponnier
fix: update Gains action names v5→v6, handle position size increases (#…
fdfd864 
…2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease
@Flotapponnier
fix: Gains carry projection (v6 action names + still-open positions)
fe90c0a 
* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease
@Flotapponnier
fix: widen coin name column for long tickers (xyz:NVDA)
0b2978a 
* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column in top markets to fit xyz:NVDA
@Flotapponnier
feat(rpc): add Union bench #253 (Nodes.Guru, Stake And Relax, High St…
db108c1 
…akes) (#2194)
@Flotapponnier
fix(union-rpc): remove em dashes from YAML (#2196)
6877507
@Flotapponnier
feat(rpc): Shentu #254 + MANTRA Chain #255 (#2199)
2a45697 
* feat(rpc): add Shentu bench #254 and MANTRA Chain bench #255

* fix: remove accidental dev-portal submodule
@Flotapponnier
feat: Band Protocol #256 + cheqd #257 benches
91faaeb
@Flotapponnier
fix: bench-blob revalidate 300s to fix perp ISR conflict (#2206)
59a872a 
* merge: dev → main (Gains carry fix + vault fee split) (#2191)

* fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chains only (#2187)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection (#2189)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains action names v5→v6, handle position size increases (#2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* feat(rpc): Union bench #253 (#2195)

* fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chains only (#2187)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection (#2189)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains action names v5→v6, handle position size increases (#2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: Gains carry projection (v6 action names + still-open positions)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column for long tickers (xyz:NVDA)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column in top markets to fit xyz:NVDA

* feat(rpc): add Union bench #253 (Nodes.Guru, Stake And Relax, High Stakes) (#2194)

* fix(union-rpc): remove em dashes from YAML

* feat(union): add provider SVG logos and brand colors

* feat: Shentu #254 + MANTRA Chain #255 benches

* fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chains only (#2187)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection (#2189)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains action names v5→v6, handle position size increases (#2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: Gains carry projection (v6 action names + still-open positions)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column for long tickers (xyz:NVDA)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column in top markets to fit xyz:NVDA

* feat(rpc): add Union bench #253 (Nodes.Guru, Stake And Relax, High Stakes) (#2194)

* fix(union-rpc): remove em dashes from YAML (#2196)

* feat(rpc): Shentu #254 + MANTRA Chain #255 (#2199)

* feat(rpc): add Shentu bench #254 and MANTRA Chain bench #255

* fix: remove accidental dev-portal submodule

* fix: bench-blob revalidate 300s to fix perp ISR conflict
@Flotapponnier
fix(fee-compare): clearer Gains labels — Position size + vault fee (#…
128e575 
…2207)

* fix(fee-compare): rename Volume→Position size + clearer vault fee label

* fix(fee-compare): show total position value + inline vault fee explanation
@Flotapponnier
fix(fee-compare): total position value + inline vault fee explanation (…
02a01df 
…#2208)

* fix(fee-compare): rename Volume→Position size + clearer vault fee label

* fix(fee-compare): show total position value + inline vault fee explanation
@Flotapponnier
fix(fee-compare): remove vault fee — carry fees settled per-action, n…
46533b3 
…ot OI surcharge (#2209)
@Flotapponnier
feat(rpc): ICON bench #258 (#2210)
1de4e80 
* merge: dev → main (Gains carry fix + vault fee split) (#2191)

* fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chains only (#2187)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection (#2189)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains action names v5→v6, handle position size increases (#2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* feat(rpc): Union bench #253 (#2195)

* fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chains only (#2187)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection (#2189)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains action names v5→v6, handle position size increases (#2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: Gains carry projection (v6 action names + still-open positions)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column for long tickers (xyz:NVDA)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column in top markets to fit xyz:NVDA

* feat(rpc): add Union bench #253 (Nodes.Guru, Stake And Relax, High Stakes) (#2194)

* fix(union-rpc): remove em dashes from YAML

* feat(union): add provider SVG logos and brand colors

* feat: Shentu #254 + MANTRA Chain #255 benches

* fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chains only (#2187)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection (#2189)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains action names v5→v6, handle position size increases (#2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: Gains carry projection (v6 action names + still-open positions)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column for long tickers (xyz:NVDA)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column in top markets to fit xyz:NVDA

* feat(rpc): add Union bench #253 (Nodes.Guru, Stake And Relax, High Stakes) (#2194)

* fix(union-rpc): remove em dashes from YAML (#2196)

* feat(rpc): Shentu #254 + MANTRA Chain #255 (#2199)

* feat(rpc): add Shentu bench #254 and MANTRA Chain bench #255

* fix: remove accidental dev-portal submodule

* feat: Band Protocol #256 + cheqd #257 benches

* fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chains only (#2187)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection (#2189)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains action names v5→v6, handle position size increases (#2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: Gains carry projection (v6 action names + still-open positions)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column for long tickers (xyz:NVDA)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column in top markets to fit xyz:NVDA

* feat(rpc): add Union bench #253 (Nodes.Guru, Stake And Relax, High Stakes) (#2194)

* fix(union-rpc): remove em dashes from YAML (#2196)

* feat(rpc): Shentu #254 + MANTRA Chain #255 (#2199)

* feat(rpc): add Shentu bench #254 and MANTRA Chain bench #255

* fix: remove accidental dev-portal submodule

* feat: Band Protocol #256 + cheqd #257 benches

* fix: bench-blob revalidate 300s to fix perp ISR conflict (#2206)

* merge: dev → main (Gains carry fix + vault fee split) (#2191)

* fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chains only (#2187)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection (#2189)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains action names v5→v6, handle position size increases (#2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* feat(rpc): Union bench #253 (#2195)

* fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chains only (#2187)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection (#2189)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains action names v5→v6, handle position size increases (#2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: Gains carry projection (v6 action names + still-open positions)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column for long tickers (xyz:NVDA)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column in top markets to fit xyz:NVDA

* feat(rpc): add Union bench #253 (Nodes.Guru, Stake And Relax, High Stakes) (#2194)

* fix(union-rpc): remove em dashes from YAML

* feat(union): add provider SVG logos and brand colors

* feat: Shentu #254 + MANTRA Chain #255 benches

* fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chains only (#2187)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection (#2189)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains action names v5→v6, handle position size increases (#2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: Gains carry projection (v6 action names + still-open positions)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column for long tickers (xyz:NVDA)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column in top markets to fit xyz:NVDA

* feat(rpc): add Union bench #253 (Nodes.Guru, Stake And Relax, High Stakes) (#2194)

* fix(union-rpc): remove em dashes from YAML (#2196)

* feat(rpc): Shentu #254 + MANTRA Chain #255 (#2199)

* feat(rpc): add Shentu bench #254 and MANTRA Chain bench #255

* fix: remove accidental dev-portal submodule

* fix: bench-blob revalidate 300s to fix perp ISR conflict

* fix(fee-compare): total position value + inline vault fee explanation

* fix(fee-compare): remove vault fee — carry fees settled per-action, not OI surcharge (#2209)

* feat(rpc): ICON bench #258 — new icon probe kind, 3 keyless providers
@Flotapponnier
merge main into dev, keep ICON bench additions
410fb2d
@Flotapponnier
fix(rpc): rewrite bench YAMLs 256-258 with proper SpecSchema format
fb90b3b
@Flotapponnier
merge main, keep proper SpecSchema icon-rpc.yml
35bd243
@Flotapponnier
fix: show signed HL funding projection, rename Volume to Total positi…
a2311fd 
…on value, clarify vault fee label (#2212)
@Flotapponnier
fix: Gains position reconstruction for HL funding projection (#2213)
6aa8d84 
* fix: include long-running Gains positions in HL funding projection

* fix: use earliest POSSIZEINCREASE as anchor for positions opened before fetch window
@Flotapponnier
fix: 1yr Gains lookback for accurate HL funding projection (#2214)
b48c980 
* fix: include long-running Gains positions in HL funding projection

* fix: use earliest POSSIZEINCREASE as anchor for positions opened before fetch window

* fix: fetch 1yr Gains history for position reconstruction to find true open timestamps
@Flotapponnier
feat: apples-to-apples fee comparison (HL-comparable coins only)
990da13 
* fix: include long-running Gains positions in HL funding projection

* fix: use earliest POSSIZEINCREASE as anchor for positions opened before fetch window

* fix: fetch 1yr Gains history for position reconstruction to find true open timestamps

* feat: grey out Gains-exclusive pairs in comparison, exclude from HL fee calc

* fix: hide equivFee for Gains-exclusive pairs (coin not on HL)
@Flotapponnier
fix: HL funding projection accuracy + equivFee + sort + coin filter
16e8d86 
* fix: signed HL funding projection, filter exclusive coins, increase coin limit to 20

* fix: sort Gains trades oldest-first for correct size tracking; fix equivFee uses HL rate not Gains rate
@Flotapponnier
fix(citation): allow negative p50 for deviation benches (#2197)
ab9ea00 
* merge: dev → main (Gains carry fix + vault fee split) (#2191)

* fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chains only (#2187)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection (#2189)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains action names v5→v6, handle position size increases (#2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* feat(rpc): Union bench #253 (#2195)

* fix(harnesses): cap LZ/CCIP/Axelar cardinality to bench-tracked chains only (#2187)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit (#2188)

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection (#2189)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains action names v5→v6, handle position size increases (#2190)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: Gains carry projection (v6 action names + still-open positions)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column for long tickers (xyz:NVDA)

* fix(cache): strip heavy fields from aggregate blob to pass 2MB limit

Hub pages only need card data. Drop seriesByRegion24h, metricPanels,
editorial text (seoIntro/faq/findings/methodology/disclaimer/abstract/
perChainExplainer) from the blob aggregate slim. Cuts serialized size
from ~4.3 MB to under 2 MB so unstable_cache actually persists the
entry instead of silently dropping it on every render.

* fix: include still-open Gains positions in carry projection

* fix: update Gains open action names to v6 API, handle TradePosSizeIncrease

* fix: widen coin name column in top markets to fit xyz:NVDA

* feat(rpc): add Union bench #253 (Nodes.Guru, Stake And Relax, High Stakes) (#2194)

* fix(union-rpc): remove em dashes from YAML

* feat(union): add provider SVG logos and brand colors

* fix(citation): allow negative p50 for deviation benches
@Flotapponnier
fix: fee-compare accuracy — paginated HL funding, multi-slice positio…
5a4dda1 
…ns, comparable notional (#2217)

F1: paginate fetchHlFundingHistory (up to 5 pages × 500) to cover 180d+ windows
F4: multi-slice reconstructGainsPositions — one PositionSlice per size-change period
    so pre-increase periods use smaller notional (more accurate HL equiv funding)
F5: track comparableNotionalUsdc (HL-listed coins only) and use it in walletStats
    so the HL taker-equiv fee isn't inflated by PONS/Gains-exclusive notional
@Flotapponnier
report: fastest Robinhood Chain RPC (Chainstack vs Alchemy vs QuickNode)
240ddae
@Flotapponnier
Merge pull request #2218 from ChainBench/report/robinhood-rpc
d606dce 
report: fastest Robinhood Chain RPC
@Flotapponnier
fee-compare: signed funding, maker/taker-aware equiv, symmetric coin …
def7455 
…filter
@Flotapponnier
report: retitle Best Robinhood Chain RPC, scope claims to 24h median
b598cc3
@Flotapponnier
Merge pull request #2220 from ChainBench/report/robinhood-rpc-v2
5eef8e2 
report: Best Robinhood Chain RPC (median-scoped)
Flotapponnier and others added 30 commits September 15, 2026 00:56
@Flotapponnier @florent-sys @claude
bench page: metric view tabs carry their definition as a tooltip; act…
a877d40 
…ive view's definition as a caption (#2367)

The View pills above the chart (P/F ratio, FDV / fees, P/S, ...) only
had the Prom gauge name as a native title. Each pill now shows the
panel's spec `description` on hover and keyboard focus, and the active
view's description is repeated under the row so the definition is
readable on touch screens.

The headline tab reads a new optional `panel_main_description` spec
field (schema, type, loader); bench 265 sets it to the P/F definition.
Panels without a description render as before.


Claude-Session: https://claude.ai/code/session_01L4rjxemTvgRwrCG6D2iK5D

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
compare pages: two-line trend chart on every shared bench card; hero …
3a99aea 
…drops the duplicate logo row (#2368)

Each compare card now carries the 7d/30d/90d history behind its two
headline numbers, read from /api/series filtered to the pair, fetched
when the card scrolls into view. The perp volume hero already sits
under the provider header cards, so its own logo row went.


Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
compare: per-card trend charts only on perp venue pairs, with a y axis (
807140d 
#2369)

Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
compare: drop the per-card trend charts, the perp volume hero is the …
f172e89 
…chart (#2370)

Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
bench 265 copy: meta description no longer renders "1.559xx"; intro c…
c1231f4 
…laims matched to live floats and multiples (#2371)

- seo_description and two findings appended a literal "x" after
  {{best_p50}} / {{worst_p50}}, which already carry the unit: the meta,
  OG and Twitter descriptions read "GNS trades at 1.559xx". Description
  also rewritten to fit the 158 char cap instead of being cut with an
  ellipsis.
- seo_intro: HYPE/LIT/ASTER float is 23/25/35%, so "a quarter to a
  third of supply"; the expensive end is dYdX and Aster, with Paradex's
  multiple flagged as an adapter artifact (Derive sits mid-table now).
- SynFutures row labelled "F (SynFutures)": the bare ticker made
  compare links read "GNS vs F".


Claude-Session: https://claude.ai/code/session_01L4rjxemTvgRwrCG6D2iK5D

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
compare: trend chart back on every shared bench card, with view tabs …
8e4baa1 
…and a proper axis (#2372)

Each card now carries the history behind its two numbers: view tabs
mirroring the bench page (headline metric plus every metric panel with a
tab), 7d / 30d / 90d range, y axis with nice ticks, area fills, latest
or hovered values in the legend. Fetched on scroll-in from /api/series
filtered to the pair, cached per view and range.


Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
compare charts: accent on the selected view/range, hidden when either…
8360d93 
… side has no data, isolated points drawn (#2373)

Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
perp-volume-history: sources run concurrently; dYdX honours 429 and g…
ee1963d 
…ives up early (#2374)

The dYdX indexer rate-limited the VPS IP mid-backfill and the source
spent an hour walking every market through the retry ladder while the
four venues behind it waited. Sources now run in parallel (one goroutine
each, flush + publish as each lands), 429s honour Retry-After, and the
dYdX walk stops after a run of rate-limited calls at one request a
second.


Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
bench 265: un-gate perp-pf-ratio on dev too, so the worker's sitemap …
43a40bb 
…emits it (shipped to prod in #2375) (#2376)

Claude-Session: https://claude.ai/code/session_01L4rjxemTvgRwrCG6D2iK5D

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
perp-volume-history: Extended skips a host after three consecutive fa…
3629692 
…ilures (the retired Ethereum deployment answers 504) (#2378)

Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
bench 265: Open interest ledger column, so pre-TGE rows show fees and…
e080720 
… OI (Float stays a chart tab; six columns max) (#2379)

Claude-Session: https://claude.ai/code/session_01L4rjxemTvgRwrCG6D2iK5D

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
bench 266 copy: Gains series is the Dune view, refresh runs in parall…
54ec353 
…el, DeFiLlama check in the intro (#2381)

Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
social cards: the leader claim is highlighted in orange on bench, rep…
3a85d8d 
…ort and product cards (#2382)

The OG / Twitter card body was one italic sentence; nothing in it caught
the eye in a feed. The claim now sits on its own line under an orange
marker (#ffb257), with the qualifier in the usual muted italic:

- bench cards (OG, Twitter, per-chain OG): "GNS posts the lowest p/f
  ratio at 1.559x" highlighted, then "(p50, 24h)". The title is no
  longer repeated under the claim.
- report cards: the hero finding's first clause (split at the first
  ", " / ". " / "; " / " - " between 24 and 84 chars) highlighted, the
  remainder within the previous ~140 character budget.
- product cards: "#1 on 11 of 13 benchmarks" (or "N benchmarks
  measured" when no win) highlighted.

headlineSentence() is unchanged in output; it is now built from the new
headlineParts() so the cards and the citation string stay in lockstep.


Claude-Session: https://claude.ai/code/session_01L4rjxemTvgRwrCG6D2iK5D

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
bench 262 fiat-onramp-cost: live, headline cell BTC by card at 500 EUR (
9a5dc0f 
#2385)

USDC by SEPA was the headline but no keyed provider fills it: MoonPay's
test key answers 'Currency not supported in test mode' for USDC and
Transak's quote endpoint has no SEPA. BTC by card is the one cell every
provider in the cohort quotes, so every row is scored on the same thing.
SEPA, USDC and ETH stay on the grid as their own series. Copy, FAQ,
findings and every selector updated; status draft -> live.

Live on the VPS harness: MoonPay 810 bps, Transak 410 bps (24h p50,
198 samples). Ramp, Onramper and Meld rows wait for keys.


Claude-Session: https://claude.ai/code/session_01L4rjxemTvgRwrCG6D2iK5D

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
bench 262: only keyed providers declared (MoonPay, Transak) so the wo…
2a0ddbf 
…rker quorum passes; Ramp, Onramper, Meld return with their keys (#2386)

The worker needs min(ceil(declared/2), 3) ranked rows to publish; with
five declared and two keyed it kept the draft render forever.


Claude-Session: https://claude.ai/code/session_01L4rjxemTvgRwrCG6D2iK5D

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
bench 262: view tabs and ledger columns per grid cell (declared fee, …
7989d22 
…hidden spread, BTC SEPA, BTC 100 EUR, ETH card/SEPA, USDC card) (#2387)

Every cell was already measured and only readable in Prometheus; the
metric_panels expose them as the View pills and six ledger columns.
Declared fee and hidden spread split the headline number the way the
bench's own story tells it.


Claude-Session: https://claude.ai/code/session_01L4rjxemTvgRwrCG6D2iK5D

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
head-lag: Solana headline becomes the lag behind the first feed to re…
87dd01a 
…port the trade; First to report panel (#2388)

Solana has no on-chain timestamp below one second, so measuring each
provider against the timestamp it sends compared conventions: Mobula's
`date` is its ingestion time (a constant 0.10 s from Singapore, 0.01 s
from Paris), Serialized's `at` is blockTime in whole seconds (0.75 s).
Against a common clock the two feeds are 10 to 30 ms apart. No RPC
WebSocket we can hold precedes their geysers: mainnet-beta and Helius
land ~110 ms after both on every trade, Alchemy is level on half the
trades and 3.5 s late or absent on the other half.

race.go runs a per-transaction race on every chain from the deferred
matcher's emissions, our node included: head_lag_first_total,
head_lag_races_total and a rolling 24 h head_lag_first_share_pct. On
raceChains (Solana) head_lag_seconds is the lag behind the first arrival;
late pollers (GeckoTerminal, 15-100 s) are measured against the same
first arrival for ten minutes. Validated from Paris over 11 trades:
Mobula first 91 %, Serialized 27 % (ties count for both), node 0 %;
headline Mobula 0.000 s, Serialized 0.010 s.

Spec: methodology and FAQ rewritten for Solana, First to report metric
panel, probe placement note (all three probes ran in Singapore until
2026-09-15).


Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
head-lag race: a race needs two participants (a lone GeckoTerminal po…
d226251 
…ll was winning by default) (#2390)

Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
head-lag: Solana opens on First to report; leader lag floored at 1 ms…
d1646ef 
…; success ratio on the real 30 s scrape (#2392)

- spec chart.default_panel_by_chain: a chain tab can open on a metric
  panel; Solana opens on First to report, Base keeps the head lag view.
  A ?view= in the URL still wins on first paint.
- The site drops rows whose headline is exactly 0 (Singapore: Mobula won
  every race and vanished from the ranked chart); the race lag is now
  floored at 1 ms, below the tie window.
- Presence ratio divided by 5,760 assumed a 15 s scrape; the store is
  scraped every 30 s, so every provider capped at 49.5 %. Now 2,880.


Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
head-lag: Solana offers only First to report (spec chart.hide_headlin…
42d25f5 
…e_by_chain) (#2394)

Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
bench page: chart config read across seeded variants; cache keys bump…
8c5d321 
…ed so stale objects without Benchmark.chart do not keep the Head lag tab on Solana (#2396)

Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
head-lag: First to report is scored between feeds only; panels sort b…
a4906ea 
…y their own direction (#2398)

On Base and BNB the flashblock / node reference preceded every feed on
almost every trade, so the feeds' shares added up to ~5 % and read as
broken. The share is now a race between feeds (ties within 5 ms count
for each); the node still sets the zero point of the lag figures and is
counted apart when it beat every feed. Trades with a single feed are not
scored. The ranked chart sorted a higher-is-better panel ascending
(0 % rows first) because the panel projection kept the bench's
direction.


Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
Serialized: brand blue (was the palette orange next to Mobula), logo …
eb3591f 
…mark centred in its box (#2400)

Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
head-lag: redact keyed reference URLs in the subscription log (#2401)
ab5e852 
Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
perp-volume-history: read through the data cache (revalidate 300), no…
380f4d2 
…t no-store (#2404)

Same change as the prod hotfix #2403: a no-store fetch inside an ISR page
can flip the route dynamic.


Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
share cards: gauge benches drop "by p50" and the p99 tails; benches 2…
59e15fe 
…65/234 count scrapes as samples, not fees in dollars (#2405)

On perp-pf-ratio the ranking card read "Product ranking by p50" with
"p99 1.480x" under every bar, and the footer "124,240,646 samples": the
spec pointed sample_size at perp_protocol_fees_30d_usd, so the card
summed 30-day fees in dollars and called them samples. Both benches now
count scrapes of perp_protocol_health over 24h (about 1,440 per row),
and the methodology says so.

Cards: isGaugeBench() (every ranked row has p50 = p90 = p99) hides the
p99 captions, drops "by p50" from the ranking and leaderboard subtitles,
labels the compare panes with the metric name, and hides the p99 cell.
Latency benches unchanged.

Logos: 20 files under public/logos were WebP bytes with a .png/.jpg
name; Satori rendered them blank (the SynFutures column had an empty
gap). Converted to real PNG/JPEG, same names. Long labels on dense
ranking cards (SynFutures) shrink instead of clipping mid-word.


Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
benches 265/234: scrape cadence is 30 seconds (2,880 samples per row …
f88ad44 
…per 24h), not 60s (#2407)

Claude-Session: https://claude.ai/code/session_01HJgbZCqjR4nvCfcJSzofbw

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
fiat-onramp-cost: Meld adapter verified live (sandbox): POST + JSON b…
5bab24c 
…ody, BASIC base64(key), Meld-Version header; partnerFee nullable (#2409)

The adapter was written from docs and would have failed twice: GET gives
405 (POST only) and base64(key + ":") gives 403 (the dashboard key is
already id:secret). One local cycle with MELD_SANDBOX=true produces 6
quotes with the right labels. Sandbox rates are synthetic (+3,749 bps on
BTC), so Meld stays out of the spec until a production key arrives.


Claude-Session: https://claude.ai/code/session_01L4rjxemTvgRwrCG6D2iK5D

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
products: one page per product; /hyperliquid/<slug> and /perp/<slug> …
d76e10f 
…fold into /products/<slug> as views behind the pill bar (#2410)

A product belongs to several categories (FOMO is a Hyperliquid
frontend, a Solana trading app, a memecoin platform and an app-store
listing), so no category may own its page. Until now /products/<slug>
308'd HL builders to /hyperliquid/<slug> and perp venues to /perp/<slug>,
and those pages showed only that category's numbers.

- /products/<slug> is the canonical page for every product. The two
  redirects out of it are gone; the reverse redirects live in
  next.config (/hyperliquid/:slug -> /products/:slug#hl, /perp/:slug ->
  /products/:slug#perp, both 308). The /hyperliquid and /perps hubs are
  untouched.
- The pill bar (VenueKpiToggle) is the category selector. It now reads
  the URL hash to open a view and rewrites it on click, so deep links
  and the old URLs land on the right tab. Panels stay in the DOM.
- Hyperliquid view: new HlFrontendSection carries everything the
  /hyperliquid/<slug> page had (30d fees hero, first day active, peak,
  HyperTracker dashboard, peer group, degraded modes when Prom or the
  history blob is missing). Peer links point at /products/<slug>#hl.
- Perpetuals view: PerpVenueSection now carries everything the
  /perp/<slug> page had on top of the KPI strip and bench cards: venue
  data, daily volume (bench 266 / cohort ring / venue API) and fees
  charts, all-time totals, vault and collateral table, other venues on
  the same benchmarks, footer links. Rankings and badges are not
  duplicated: the product page already lists every appearance.
- Links: HL cohort leaderboard, HL frontend cards, apps leaderboard and
  perp venues leaderboard point at the product page with the hash.
- Sitemap: HL builders and perp venues are product routes again; the
  /hyperliquid/<slug> entries are gone (they would 308). The worker no
  longer strips those slugs from providerSlugs; the builder unions them
  in anyway so older blobs still produce a complete sitemap.
- Route directories src/app/hyperliquid/[slug] and src/app/perp/[slug]
  (pages and OG images) deleted.

Verified locally against the live Prometheus: /products/fomo#hl opens
on the Hyperliquid view with the full dashboard and peer group, the 8
other benches FOMO appears on are listed underneath; /hyperliquid/fomo,
/perp/gains and /perp/gmx answer 308 to the product page. tsc clean,
269 tests pass.


Claude-Session: https://claude.ai/code/session_01L4rjxemTvgRwrCG6D2iK5D

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@Flotapponnier @florent-sys @claude
products: Trading app, Data API and Bridge views join the pill bar; e…
fea2499 
…very hub now has a per-product view (#2411)

The category selector on /products/<slug> covered perp, prediction
markets, data feeds, Hyperliquid and RPC. The three remaining hubs
(/trading-apps, /data-api, /bridge) had no per-product view, so a
product like FOMO showed its Hyperliquid numbers but not its trading-app
row, and Mobula showed nothing of the data-API pivot.

- Trading app (TradingAppSection): the platform's six KPIs (24h volume,
  swap tx, avg trade, active wallets, fee rate, app rating) with rank in
  the cohort, then the cohort table. Platforms, columns and formatters
  moved to src/lib/trading-apps.ts, shared with the /trading-apps hub
  (loadTradingAppMatrix computes rows, per-column ranks and bests once).
- Data API (DataApiProviderSection): the provider's row of the
  /data-api pivot opened up per group: bench, p50, rank, region and
  chain wins.
- Bridge (BridgeProviderSection): all-in fee and quote latency
  p50/p99/success with cohort rank, fee per corridor, latency per
  origin region.

All three are snapshot-only and hide-if-empty, same rule as the RPC
view. Verified locally: FOMO = [Hyperliquid | Trading app], Mobula =
[Data API | Bridge], Relay = [Data API | Bridge], LI.FI = [Bridge],
Codex = [Data API], Terminal and pump.fun = [Trading app].


Claude-Session: https://claude.ai/code/session_01L4rjxemTvgRwrCG6D2iK5D

Co-authored-by: Flotapponnier <florent@mobula.io>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@Flotapponnier