fix(simulator): OpcodeStats Add/Sub must use union of keys

[meyer9]

Problem

OpcodeStats.Add and OpcodeStats.Sub (in runner/payload/simulator/simulatorstats/types.go) only iterate the other argument, silently dropping keys present in the receiver o but absent from other:

func (o OpcodeStats) Add(other OpcodeStats) OpcodeStats {
    result := make(OpcodeStats)
    for opcode, count := range other {          // iterates 'other' only
        result[opcode] = o[opcode] + count
    }
    return result
}

Calling OpcodeStats{"A": 1, "B": 2}.Add(OpcodeStats{}) returns {} — the receiver is gone.

Downstream impact in sendTxs

The simulator worker accumulates per-tx config like:

expected := t.payloadParams.Mul(float64(t.numCalls+1) * t.scaleFactor)  // has all base opcodes + precompiles
blockCounts := expected.Sub(actual).Round()                             // first tx: actual is empty
...
t.actualNumConfig = t.actualNumConfig.Add(blockCounts)

For the first tx, actual is empty. The buggy Sub iterates actual (empty) → produces blockCounts with empty Opcodes/Precompiles maps. Then actual.Add(blockCounts) likewise iterates blockCounts (now also empty for these fields) → still empty.

Result: every tx in every block has empty Precompiles in its SimulatorConfig call, so the on-chain Run() skips all precompile work that the YAML config specified.

Observable consequence

Workloads with precompiles in their config — base-mainnet-simulation is the only one in mainnet-config.yml — under-fill the gas budget because per-tx gas is far below what calcNumCalls predicted from the gas-estimation call (which DID include precompiles). Concrete data from production run test-1779411190779930:

Run	Payload	GasLimit	gas/per_block	Fill %
-0	base-mainnet-simulation	150 M	49 M	33%
-1	base-mainnet-simulation	200 M	49 M	24%
-2	base-mainnet-simulation	250 M	51 M	20%

Note the flat ~49–51 M ceiling regardless of GasLimit — that's the per-tx cost of the non-precompile portion (storage + accounts + opcodes) × 100 txs, which is what the worker actually executes. Other workloads (storage-create-full-block, etc.) don't declare precompiles in their config, so they're unaffected and fill ~98%.

Fix

Both Add and Sub now iterate the union of keys:

 func (o OpcodeStats) Add(other OpcodeStats) OpcodeStats {
-    result := make(OpcodeStats)
+    result := make(OpcodeStats, len(o)+len(other))
+    for opcode, count := range o {
+        result[opcode] = count
+    }
     for opcode, count := range other {
-        result[opcode] = o[opcode] + count
+        result[opcode] += count
     }
     return result
 }

Sub matches: keys only in other produce negative entries (correct arithmetic semantics).

Tests added

runner/payload/simulator/simulatorstats/types_test.go:

• TestOpcodeStatsAdd_UnionOfKeys / TestOpcodeStatsSub_UnionOfKeys — direct union-of-keys check
• TestOpcodeStatsAdd_EmptyOther / TestOpcodeStatsAdd_EmptyReceiver / TestOpcodeStatsSub_EmptyOther — empty-side edge cases
• TestStatsSubAdd_FirstTxBlockCountsIncludePrecompiles — regression test for the production usage pattern in sendTxs

All pass. go test ./... clean across the repo.

Expected outcome after this PR + #192 land together

base-mainnet-simulation at 150M/200M/250M should fill to ~98% (gas budget becomes the cap, not the missing precompiles). Other workloads unchanged. The 1.05x→2.0x pre-init buffer in #192 covers the slightly higher consumption that comes from actually executing precompile work per tx.

Draft because end-to-end validation needs one CI run.

[cb-heimdall]

🟡 Heimdall Review Status

Requirement	Status	More Info
Reviews	🟡 0/1	Denominator calculation Show calculation 1 if user is bot 0 1 if user is external 0 2 if repo is sensitive 0 From .codeflow.yml 1 Additional review requirements Show calculation Max 0 0 From CODEOWNERS 0 Global minimum 0 Max 1 1 1 if commit is unverified 0 Sum 1

[meyer9]

Superseded by #194 which combines this OpcodeStats fix with the pre-init buffer bump and observed-gas recalibration into one coordinated change.