feat: Add setup transactions example

Cargo.toml

@@ -14,8 +14,8 @@ rustls = { version = "0.23", default-features = false, features = ["ring"] }
 serde_json = "1"
 futures = "0.3"
 serde = { version = "1", features = ["derive"] }
-simulator-api = { git = "https://github.com/nitro-svm/nitro-stream", rev = "d157f1b9" }
-simulator-client = { git = "https://github.com/nitro-svm/nitro-stream", rev = "d157f1b9" }
+simulator-api = "0.18.0"
+simulator-client = "0.18.0"
 solana-account = "4.1.0"
 solana-account-decoder = "4.1.0"
 solana-address = "2.6.0"

src/bin/counterfactual_flow/README.md

@@ -89,17 +89,17 @@ per-hop `ammKey`s in any baseline run's `routePlan`.
 ## What we measured
 
 The 10,000-slot range above, against **BisonFi** (`8FnX3xo2yYw3EUE6w3nQA4GfXGS9wpK6oj3veJpbFzLo`),
-a SOL/USDC prop AMM. The pair carries **14,513** re-quotable swap legs in that window — every
-SOL↔USDC swap a supported router placed, not just the ones this venue won.
+a SOL/USDC prop AMM. The pair carries **13,615** re-quotable swap legs in that window — every
+SOL↔USDC swap an admitted router placed, not just the ones this venue won.
 
-**Fresh, the venue won 1,516 of them. Ten slots slower, it won 2.**
+**Fresh, the venue won 1,341 of them. Ten slots slower, it won none.**
 
 | | legs won |
 |---|---|
-| the venue as it really ran | 1,516 |
-| the same venue, 4s slower | 2 |
+| the venue as it really ran | 1,341 |
+| the same venue, 4s slower | 0 |
 
-Ten slots of staleness costs this venue **99.9%** of the flow it captures. That is one run per
+Ten slots of staleness costs this venue **all** of the flow it captures. That is one run per
 side; the router is not deterministic, so see [below](#reading-the-output-and-what-not-to-claim)
 for how many runs a smaller difference would need.
 
@@ -110,26 +110,31 @@ books:
 
 | | legs |
 |---|---|
-| SOL/USDC legs re-quoted | 14,513 |
-| **…that routed through the venue on chain** | **2,595** |
-| re-quoted fresh → router picks the venue | 1,516 |
-| re-quoted 10 slots stale → router picks the venue | 2 |
+| SOL/USDC legs re-quoted | 13,615 |
+| **…that routed through the venue on chain** | **2,595** † |
+| re-quoted fresh → router picks the venue | 1,341 |
+| re-quoted 10 slots stale → router picks the venue | 0 |
+
+† The on-chain figure is carried over from an earlier measurement. The run JSONL records only the
+route Metis chose, never the one the original took, so this row cannot be re-derived from the
+output of the runs above.
 
 Two things to read here, and the first one bounds what any of this can say.
 
 **This only sees aggregator flow.** Swaps are detected by decoding router instructions — Jupiter,
-OKX, Titan and DFlow. A taker hitting your pool directly, through an unsupported aggregator, or via
-private flow never enters the sample at all, so the totals here are a slice of your book, not your
-book.
+OKX, Titan and DFlow — and of those, only the venues the run admits are re-quoted, which is Jupiter
+alone unless `--reroute-venues` names more. A taker hitting your pool directly, through an
+unsupported aggregator, or via private flow never enters the sample at all, so the totals here are a
+slice of your book, not your book.
 
-**The baseline is close to history, but it is not a replay of it.** The venue held 2,595 legs on
-chain and the fresh re-quote gives it 1,516 — 58%. It is not the same 58%, either: it keeps 660 of
-the legs it really won and takes 856 it did not, because the re-quote runs against a different
-router than the one that originally placed each swap, with its own market coverage and quote timing.
-Read the baseline as *"what this router would do"*, not *"what happened"*.
+**The baseline is close to history, but it is not a replay of it.** The fresh re-quote gives the
+venue 1,341 legs against the 2,595 it held on chain — and they are not the same legs, because the
+re-quote runs against a different router than the one that originally placed each swap, with its own
+market coverage and quote timing. Read the baseline as *"what this router would do"*, not *"what
+happened"*.
 
 **Which is why the comparison is baseline-vs-modified, never modified-vs-L1.** Both arms run through
-the same router with the same disagreements, so those cancel; what remains is your change. 1,516 → 2
+the same router with the same disagreements, so those cancel; what remains is your change. 1,341 → 0
 is not a haircut, it is elimination.
 
 ## What it means for a maker
@@ -175,7 +180,7 @@ Rows 1–5 are exact: every way of splitting ten runs into two disjoint arms of
 are estimated by resampling, since two disjoint arms of six would need twelve runs.
 
 So a single run is not a precise measurement of *how much* flow moved, even though the example above
-separates so far — 1,516 against 2 — that one run per side already carries the direction. If you ever
+separates so far — 1,341 against 0 — that one run per side already carries the direction. If you ever
 see a minor difference between the baseline and the override, run both sides several times to take
 run-to-run variance out of the answer.