What the evidence supports.
A passing sample is evidence about that sample. Here are the denominators, assumptions, and boundaries behind the results.
Supported inputs
| Bundled Solidity compilers | 0.8.30, 0.8.28, 0.8.26, 0.8.24, 0.8.20, 0.8.17, 0.8.10, 0.7.6, 0.6.12, 0.5.17, 0.4.26, 0.4.19 |
|---|---|
| Chain investigation & pinned simulation | Ethereum mainnet and BNB Chain; depends on provider historical-state support. |
| Explorer dashboard | Ethereum only; direct RPC reads, no full-history indexer. |
| Proxy evidence | EIP-1967 implementation/beacon slots and EIP-1167 standard minimal proxy runtime, read at the selected block. Implementation navigation is explicit. |
| Source verification | Sourcify, with Etherscan fallback when configured. Provider runtime is compared with pinned account code when available; mismatches disable automatic source/ABI use. |
| Execution model | Cancun message execution. No signed transaction validation, intrinsic fees, full finalization, later fork rules or BNB consensus reproduction. |
Synthetic execution corpus
126 fixtures · 491 sampled traces · 2 compiler exclusions
| Metric | Matched / expected | Reported result |
|---|---|---|
| EffectTuples | 928 / 928 | 100% |
| EffectOperands | 1772 / 1772 | 100% |
| FunctionEffectOwnership | 928 / 928 | 100% |
| Selectors | 371 / 371 | 100% |
| SelectorPrecision | 371 / 371 | 100% |
| ExecutedInstructions | 78570 / 78570 | 100% |
| ExecutedEdges | 11013 / 11013 | 100% |
| ExecutedEffects | 928 / 928 | 100% |
| ReadableFunctions | 371 / 371 | 100% |
Synthetic corpus and sampled executions; coverage metrics are not semantic equivalence or general accuracy. Structured function regions are a structural proxy, not human readability or equivalent Solidity.
Download original report JSON ↓OpenZeppelin sample corpus
12 fixtures · 60 sampled traces · 2 compiler exclusions
| Metric | Matched / expected | Reported result |
|---|---|---|
| EffectTuples | 192 / 192 | 100% |
| EffectOperands | 388 / 388 | 100% |
| FunctionEffectOwnership | 192 / 192 | 100% |
| Selectors | 128 / 128 | 100% |
| SelectorPrecision | 128 / 128 | 100% |
| ExecutedInstructions | 15250 / 15250 | 100% |
| ExecutedEdges | 2293 / 2293 | 100% |
| ExecutedEffects | 192 / 192 | 100% |
| ReadableFunctions | 128 / 128 | 100% |
Synthetic corpus and sampled executions; coverage metrics are not semantic equivalence or general accuracy. Structured function regions are a structural proxy, not human readability or equivalent Solidity.
Download original report JSON ↓Methodology & interpretation
Compiler-generated contracts and sampled EVM executions provide selectors, instruction/edge coverage, effects, operands and function ownership expectations. The corpus includes synthetic contracts and a small OpenZeppelin sample. Each metric compares a defined expected set with recovered observations; its denominator is shown above.
Execution-dependent oracle bindings are supplied. These are not independent executions of recovered Solidity, and structured-function counts are not a human readability score. Zero misses in this finite corpus do not establish five-nines reliability on production contracts.
These are bundled reports from the repository, packaged at 2026-11-18T04:52:49.360Z. Building this website does not rerun the benchmark. Regenerate fixtures with npm run fixtures:execution, then run the Go quality suite and inspect its output before updating the published reports.
Known gaps and failure modes
- Benchmark percentages describe the included corpus and sampled executions, not arbitrary deployed contracts.
- Oracle bindings are supplied in the execution benchmark; recovered pseudocode is not independently executable Solidity.
- Decompiler uses bounded Cancun legacy-EVM semantics. Original names, comments and source types cannot generally be recovered.
- Pinned simulation uses Cancun message semantics, not full chain-specific consensus or later hardfork behavior.
- Standard proxy detection does not establish all possible routing behavior.
- Address history, token portfolios, full transaction indexing, team accounts and live collaboration are not implemented.
- Decompiler analysis budgets can stop exploration; partial results and unresolved paths must be inspected.
- Source analysis does not expand every modifier, helper, assembly block or storage alias into a complete effect model.
- Unsupported compilers, missing dependencies, unavailable historical state and provider errors are reported explicitly.
- Imported reports are untrusted snapshots. Re-run compilation or replay a scenario before treating them as independently reproduced evidence.