How to reason about it
- The documentation corpus and targeted runtime tests check named examples and negative diagnostics.
- The integrated denominator is exactly 144 cases, made up of 38 differential, 61 conformance, 7 warning, 6 resource, and 32 generated cases run inside fixed limits.
- Receipts record exact inventory, runners, observations, replay, mismatch disclosure, provenance, artifact identity, delegation, capabilities, and limits.
- A separate scaled campaign stores 11,088 generated cases across native and WebAssembly routes of the same Rust lineage, with zero natural divergence and zero infrastructure failure. A route is one way of running a rule.
- A focused Rust engine court compares tree and verified-bytecode VM outcomes, exercises all 205 primitive dispatch rows, and records zero forbidden fallback. A court is one run of checks gathered into a single verdict, and a VM is a virtual machine, the program that runs an already compiled list of instructions. This court is separate from the historical three-family receipt.
- Synthetic drifts and mutation controls demonstrate that checkers can detect selected failure classes.
The conformance court has several layers
- Authored examples pin one behavior at a time.
- Generated cases explore a larger valid and deliberately invalid space.
- Mutation controls prove that selected contract damage is rejected.
- Resource cases keep termination inside fixed limits visible as an observation.
- Diagnostic cases compare error codes, messages, spans, bytes, and warning order.
- Differential runs compare the same case across admitted implementation families.
These layers complement one another. A generated count does not replace a
clear authored example, and a mutation control does not prove that every defect
is detectable. Any new pair mismatch begins as unclassified, and the
maintained court fails closed until its observation and primary reason are
recorded.
Measured evidence lines
| Measured area | Inventory | Strongest honest reading |
|---|---|---|
| Documentation conformance | 122 result annotations plus diagnostic/TCO/control cases | named examples match the reference interpreter |
| Integrated backend denominator | 38 differential + 61 conformance + 7 warning + 6 resource + 32 generated inside fixed limits = 144 cases | one exact inventory reused for LIL, LIT admission, and the all-family matrix, with generated seed 550633217 |
| LIL hosted backend | a current contract of 205/205 primitive rows and 18/18 guest-calling rows, while the 144-case released receipt records the prior 84-row boundary | focused courts cover prior convergence, exact integers, higher-order callbacks, output, and 4 success plus 3 negative warning-alias observations for ordering, call-site deduplication, shadowing, warning-before-error, channel separation, and zero forbidden fallback. The historical receipt remains unchanged |
| LIT transliteration | 84/205 primitive capability rows and 6/18 guest-calling rows, with one LIT product route in the 144-case receipt | Rust/LIT agrees on 65 cases and diverges on 79, and no source-independence claim is made |
| LIT four-host seal | Topaz interpreter, generated Rust, generated Python, and Web routes from one exact source | evidence collected inside fixed limits closes its exact versioned set of checks. These are host variants, not four implementations |
| Three-family joint receipt | 144 cases, 432 installed-backend executions, and 432 family-pair comparisons | 59 all-family agreements and 85 cases with at least one divergence. 253 pair agreements and 179 pair divergences remain explicit |
| Scaled same-lineage campaign | 11,088 stored generated cases across native and WebAssembly Rust routes | zero natural divergence and zero infrastructure failure on that inventory. This is not a Rust/LIL/LIT campaign |
| Rust tree/VM court | focused semantic and control programs, all 205 primitive dispatch rows, strict malformed-bytecode negatives, exact source/Core IR/bytecode identities, and zero fallback instrumentation | the explicit VM matches the tree observation on this maintained same-Rust-lineage area. It is not an independent witness, a Vouch receipt, or a whole-profile proof |
A common mistake
A zero mismatch count is scoped to its stored or executed inventory only.
Current boundaries
- The 11,088-case zero-divergence campaign is not the Rust/LIL/LIT joint receipt and does not establish implementation independence or whole-language equivalence.
- The focused LIL courts change current LIL coverage only. They do not recompute or upgrade the historical 144-case receipt.
- The Rust tree/VM comparison report is diagnostic material. It does not join the backend denominator or grant execution authority.
Keep going
The observation matrix holds the pair results, and Current and Deferred Scope separates the language profile from what each backend implements.
Bytecode and Rust VM · Backend Observation Matrix · Current and Deferred Scope