refactor: stabilise bench-tx.json keys and consolidate cost-table tests

.gitattributes

@@ -1 +1,3 @@
-CHANGELOG.md merge=union
+CHANGELOG.md merge=union
+bin/bench-transaction/bench-tx.json linguist-generated=true
+Cargo.lock linguist-generated=true

CHANGELOG.md

@@ -10,6 +10,7 @@
 
 ### Changes
 
+- Keyed `bench-tx.json`'s per-note cycle counts by note label instead of note commitment, and added `total_cycles` to each scenario ([#3726](https://github.com/0xMiden/protocol/pull/3726)).
 - [BREAKING] Refactored `AccountVaultDelta` to track generic assets. `FungibleAssetDelta`, `NonFungibleAssetDelta` and `NonFungibleDeltaAction` were removed ([3485](https://github.com/0xMiden/protocol/pull/3485)).
 - [BREAKING] Moved the internal shared helpers of `miden::protocol::input_note`, `miden::protocol::active_note`, and the note memory-write helpers into private `input_note_internal` and `note_internal` modules ([#3501](https://github.com/0xMiden/protocol/pull/3501)).
 - [BREAKING] Changed asset callbacks into validation-only interfaces that return no asset value; the transaction kernel retains and uses the original value, preventing callbacks from modifying it. The kernel commitment changes ([#3505](https://github.com/0xMiden/protocol/issues/3505), [#3513](https://github.com/0xMiden/protocol/pull/3513)).

bin/bench-transaction/Cargo.toml

@@ -64,4 +64,3 @@ tokio      = { optional = true, workspace = true }
 
 [dev-dependencies]
 criterion = { features = ["async_tokio", "html_reports"], workspace = true }
-rstest    = { workspace = true }

bin/bench-transaction/README.md

@@ -40,7 +40,7 @@ Regenerate the tables (and `bench-tx.json`) with:
 make update-note-costs
 ```
 
-Freshness is enforced in CI: the `checked_in_cost_matches_benched_cycles` snapshot tests in `src/note_costs.rs` re-execute every priced scenario during the regular test run and fail when a measured cost drifts more than 5% from its checked-in constant. Drift within the tolerance (from unrelated changes landing on the base branch) is absorbed without regeneration - fee-wise this is safe, since the fee is logarithmic in cycles and the pricing safety margin dwarfs it. A PR that meaningfully changes cycle counts must run `make update-note-costs` and commit the updated tables - which doubles as review signal, since cost regressions show up as table diffs.
+Freshness is enforced in CI: the `checked_in_note_costs_match_executed_scenarios` snapshot test in `src/note_costs.rs` re-executes every priced scenario during the regular test run and fails when a measured cost drifts more than 5% from its checked-in constant, or when a note's declared `created_notes` do not match what its scenarios actually create. It walks `PricedNote::all` - so all priced notes are covered. Drift within the tolerance (from unrelated changes landing on the base branch) is absorbed without regeneration - fee-wise this is safe, since the fee is logarithmic in cycles and the pricing safety margin dwarfs it. A PR that meaningfully changes cycle counts must run `make update-note-costs` and commit the updated tables - which doubles as review signal, since cost regressions show up as table diffs.
 
 ### Benchmark Groups
 
@@ -49,8 +49,12 @@ Each of the above transactions is measured in two groups:
 
   For each transaction, data is collected on the number of cycles required to complete:
   - Prologue
+  - `total_cycles`: the sum of the prologue, notes-processing, tx-script and epilogue intervals,
+    and the figure the cost tables above are derived from
   - All notes processing
-  - Each note execution
+  - Each note execution, in consumption order and labelled by note kind (`P2ID`, or `P2ID#0` /
+    `P2ID#1` when a scenario consumes several notes of one kind). Notes whose script is neither a
+    standard nor an agglayer one are labelled `UNKNOWN`.
   - Transaction script processing
   - Epilogue:
     - Total number of cycles