What is claimed is measured.
A numerical claim is only valid if it can be replayed. This page brings together the method, tests, benchmarks and what remains, to date, unmeasured.
Replay a run identically.
A run is defined by four ingredients: a seed (64-bit integer), a configuration, an engine version and an initial state. With equal seed and version, SYNE replays the same trajectory bit-for-bit; the run identifier (run-<seed>-<12hex>) carries the seed, which ECHOS does not derive itself.
SYNE benchmarks print an FNV-1a checksum of the state (population, ticks, then oneid;x;y;energy line per entity, sorted by ID). Two passes with the same seed produce the same checksum: this is the operational proof of determinism.
| Ingredient | Role |
|---|---|
| Seed | Initialises the xoshiro256** PRNG; makes two runs comparable. |
| Configuration | World and entity parameters. |
| Engine version | Ensures identical rules between runs. |
| Initial state | Common starting point for comparisons. |
Measured on 19 October 2026.
passing tests
0 failures across local suites.
ECHOS coverage
Line coverage of the observatory (Python package).
SYNE coverage
Aggregated SYNE assemblies (Core 91,17 %, Console 63,46 %).
| Indicator | Value |
|---|---|
| Analysed reference | feature/livex-realignement-divergences-v0.2 · a229c6a |
| Passing tests per component | SYNE 695 · ECHOS 383 · Launcher 227 · syne-mock 51 |
| Continuous integration | 13/13 jobs green, 19 Oct 2026 |
| Windows | Windows support validated locally since 8 Oct 2026 |
These figures describe a development repository on a given date. They do not constitute functional validation or a promise of stability.
Figures, with their conditions.
SYNE V0.1 measurements were repeated on 8 October 2026, in Release, on the reference machine: Intel Core i7-8750H (6 cores / 12 threads), 14 GiB RAM, Ubuntu x86-64, .NET SDK 10.0.401. 500×500 world, 50-cell spatial partitioning, default configuration, 300 ticks, seeds 12345 / 999 / 7, single-threaded.
| Population | Measured throughput (t/s) | Average tick | V2 target | Verdict |
|---|---|---|---|---|
| 50 | 1,354 – 2,257 | 0.44 – 0.73 ms | ≥ 30 | Above |
| 500 | 42.8 – 53.9 | 18.5 – 23.3 ms | ≥ 20 | Above |
| 1,000 | 5.9 – 6.5 | 153 – 169 ms | ≥ 10 | Gap ≈ 1.7× |
Known gap at 1,000 entities. The scalability target is not met at this scale (≈ 6 t/s vs ≥ 10). Optimising perception and batch-processing events is the identified task; it does not block V0.1, whose scenarios run at 100 agents or fewer.
The cost of observability (ECHOS stack connected to SYNE) is measured separately by a dedicated benchmark script: same entities, same ticks, same seed, so any throughput difference is attributable to observation. Sources: docs/syne/PERFORMANCE.md, docs/echos/BENCHMARKS.md.
What we do not claim.
Calibration of constants
Need and threshold constants are based on inherited values, not calibrated. No reference campaign has yet validated these settings.
Sensitivity of the emergence index
The weights of the composite score are published but untested; a sensitivity analysis remains to be done.
Statistical protocol
The number of runs needed for a difference between configurations to be more than an observation is not determined.
Single reference machine
Performance, determinism and acceptance rely on a single machine and a single runtime. A single measurement is not a law.
A figure is not a conclusion.
Tick throughput measures speed, not relevance. Test coverage measures the extent of checks, not the correctness of the model. An emergence index measures dispersion, not proof. Every result here carries its version, date and source; a figure without a source is not published.
| Claim | Status | Source |
|---|---|---|
| 1,356 passing tests, 0 failures as of 19/10/2026 | Verified | LIVEX-STATUS §1, §5 |
| ECHOS coverage 91.32% / SYNE 81.20% as of 19/10/2026 | Verified | LIVEX-STATUS §1 |
| 13/13 CI jobs green on 19/10/2026 | Verified | LIVEX-STATUS §1 |
| Gap ≈ 1.7× below target at 1,000 entities | Verified | PERFORMANCE §9.2 |