Half I measured retail pricing on the app. Half II opened one unsigned transaction for router attribution. This piece is a seven-day probe of Jupiter’s managed /order path: selected-router incidence stays dispersed, however the giant leave-one-out gaps sit in a number of pair×dimension×state corners-not throughout the router layer as a complete.
One Interface, Many Execution Sources
Solana liquidity sits throughout AMMs, books, and market makers. Most merchants by no means see that stack. They enter by a small variety of swap interfaces.
Jupiter was the biggest DefiLlama-attributed Solana aggregator sequence in 15/15 months from Might 2025 by July 2026. These adapter sequence will not be a mutually unique partition of execution move, so that they enter right here solely as persistent interface-level attribution-not as market share.
Jupiter can be not a single router. Its managed Meta-Aggregator /order path (Extremely throughout this assortment lineage) runs a range layer containing Metis, JupiterZ, DFlow, and OKX, and returns which label was chosen. Interface adoption and selected-router incidence are completely different objects.
It is a fastened pair×dimension×time probe grid, not a draw from Jupiter’s manufacturing order move. Every examined intent counts as soon as no matter notional. The router percentages under are incidence inside this experiment-not manufacturing move share, notional share, or landed execution share.
An app can even purchase execution outdoors that managed race. Elements I-II retain one unsigned JTX→DFlow Aggregator v4 transaction-an alternate wiring, not proof that anybody supply dominates Jupiter’s inner choice.
Jupiter Runs the Managed Execution Layer
Panel B0-extended measured selected-router incidence throughout 21 UTC home windows (2026–08–03→08–09). Chosen-router HHI stayed between 0.28 and 0.35, with total HHI 0.298 (efficient variety of routers about 3.35). No single router label took half or extra of the examined intents.
The managed /order path decides eligibility ( excludeRouters), compares affords, and returns quantities plus a transaction when construct succeeds. This probe stops on the order stage, so the saved responses don’t help router-level buildability or touchdown comparisons.
JupiterZ is Jupiter-operated. Metis is handled individually: Jupiter formally separated Metis v7 from the Jupiter umbrella in late 2025 whereas nonetheless itemizing Metis as a 2026 Meta-Aggregator engine. Jupiter-operated chosen incidence is subsequently a bound- 31.7%-55.3% -depending on how Metis is assessed.
Who Will get Chosen Is Not Who Issues Most
The probe: six USDC pairs × 4 sizes × three UTC home windows/day for seven days, leave-one-out for Metis, JupiterZ, DFlow, and OKX. 21/21 home windows accomplished; 941 steady matched units of 1008 tried.
Signal conference. Depart-one-out bps:
LOOᵣ = 10,000 × (qᵃˡˡ / q⁻ʳ − 1),
the place qᵃˡˡ is the midpoint of the bracketing all-router quotes (out_all_ref) and q⁻ʳ is the exclude-r quote (out_exclude_r). Constructive means excluding r worsened measured output.
Depart-one-out estimates are conditional on brackets whose all-router endpoint drift stayed under 5 bps; high-drift brackets have been dropped. On the steady units, median |Δₐₐ| is about 0.26 bps (P95 about 2.75). Amongst brackets with each endpoints observable, pre-gate P(|Δₐₐ| > 5 bp) is about 4.8%. Steady-set P(|Δₐₐ| > 5 bp) is censored by the gate — not a tail benchmark.
JupiterZ choice additionally rises in these high-drift brackets after accounting for pair composition (pair-standardized anticipated share about 49%, noticed about 65%). The stable-bracket outcomes subsequently miss some states the place JupiterZ is chosen extra typically.
Throughout profitable all-router responses (A1 and A2 pooled; n = 2014), OKX was chosen in 36.8% of examined intents, JupiterZ 31.7%, Metis 23.6%, DFlow 7.8%.
Combination leave-one-out medians sit close to that decision: Metis and DFlow about 0 bps, OKX about 0.1, JupiterZ about 0.3. Onerous protection failure was virtually absent-another routing supply often returned a sound quote. A sound fallback isn’t at all times an economically shut substitute. Most steady brackets have a alternative path and combination median results close to probe decision. The massive gaps sit in particular pair×dimension×fallback×state cells.
Exclusion scope. Docs hold Metis eligible when excludeRouters=jupiterz, and individually state that JupiterZ V2 MM liquidity is built-in into Metis as hops. Saved Metis routePlan labels by no means expose a JupiterZ hop-under both the all-router or exclude-JupiterZ arm-so the schema doesn’t present whether or not embedded stock can be disabled. No bias course is assigned. The exclude-JupiterZ arm identifies execution-path dependence, not a clear standalone-router valuation.
Dependence Is Concentrated by Pair, Dimension, and Path
Massive exclude-JupiterZ results will not be a basic property of the probe. They happen in two pairs: AAPLX and BONK. Within the different 4 pairs, no steady set exceeds 5 bps.
On steady units, exclude-JupiterZ LOO exceeds 5 bps in 60.4% of AAPLX units (n = 144; median 21.5 bps) and 34.2% of BONK units (n = 146). BONK’s median is small (~ 0.7 bps), however roughly one-third of its steady units nonetheless clear 5 bps. For SOL, USDT, JITOSOL, and CBBTC the >5 bp charge is 0%.
The AAPLX cell is the utmost of 24 pair×router medians; it was recognized put up hoc. The 21.5 bp AAPLX median sits far above its 0.05 bp native drift median-roughly six instances the pair-specific P95 full-bracket drift ( 3.39 bps). Depart-one-day-out medians keep in 16.6–30.6 bps.
The AAPLX hole is strongly size-dependent. Medians: about 0 bps at $100 (n = 39); 19 bps at $1k (n = 41); 185 bps at $10k (n = 36); about 16,800 bps at $100k (n = 28). The $100k cell can be the place the extreme-value audit concentrates.
Onerous protection and alternative high quality are completely different objects. Amongst steady units the place JupiterZ received A1, Metis and OKX fallbacks sometimes protect the all-router reference (median remaining output ~ 1.00). The acute AAPLX tail concentrates within the uncommon JupiterZ→DFlow fallback (n = 18), the place the median fallback returned about one-fifth of the all-router reference (median remaining-output ratio ~ 0.20). The leave-one-out outcome measures the standard of the accessible alternative path-not a standalone price ticket on JupiterZ.
The sharper calendar distinction is weekday versus weekend, not U.S. hours alone: weekday U.S.-hours overlap median 24.4 bps (n = 30); weekday off-hours 54.2 bps (n = 68); weekend 0.01 bps (n = 46), although roughly 28% of weekend observations nonetheless exceed 5 bps. Onchain xStocks are documented as 24/7 tradable; no depth or market-closure mechanism is claimed right here.
Closing
On this fastened /order probe, Jupiter controls the managed entry and comparability layer. Chosen-router frequency is a poor proxy for financial dependence: most single-source exclusions have combination medians close to within-bracket decision, and a sound fallback quote often exists.
The massive gaps sit in corners of the check grid. Beneath excludeRouters=jupiterz, solely AAPLX and BONK produce >5 bp occasions. On AAPLX, the impact rises sharply with examined order dimension and differs between weekday and weekend observations. The acute tail additionally is determined by which fallback path stays accessible.
As a result of that exclusion could not isolate a clear standalone router, these outcomes establish execution-path dependence slightly than a standalone valuation of JupiterZ. They’re order-stage choice and quote-sensitivity outcomes from a set probe-not production-volume shares or landed fills.
Appendix
Pattern. Manufacturing Meta-Aggregator /order, b0x_20260803 (2026-08-03→08-09). Matched set = A1 + 4 randomised LOO + A2. 1008 tried → 941 steady (|A1-A2| < 5.0 bps; span ≤ 20 s). Pre-gate high-drift charge ~4.8%.Decision. Steady |Δₐₐ|: median ~0.26 bps, P95 ~2.75. Pre-gate (n=1006): median ~0.29, P95 ~4.39. Steady P(>5)/P(>10) are gate-censored.Denominators. Incidence swimming pools A1+A2 (n=2014). Substitution counts use the A1 winner on every steady set (n=941).Scope. Metis affiliation unresolved (Jupiter-operated certain 31.7%-55.3%). JupiterZ exclusion semantics unresolved. No landed fills; no production-volume learn. Diagnostics: empirical package deal.Half I: Completely different Apps, Similar Router · Half II: Can a Solana Transaction Show the Router? · Jupiter /order
This put up was initially printed on my private weblog: https://egpivo.github.io/2026/08/23/who-owns-execution-on-solana.html.
Who Owns Execution on Solana? was initially printed in The Capital on Medium, the place individuals are persevering with the dialog by highlighting and responding to this story.
