# UTF-4088 Large Sampling — Executed Result ## Run - samples: **5,000,000** - seed: `0x40880812` - glyph geometry: 8×12 = 96 binary pixels - sampler: `large_summary` equivalent to the repository large-map topology logic - output: `large_sampling_5m_summary.csv` ## Observed result | Metric | Result | |---|---:| | Samples | 5,000,000 | | 4-connected glyphs | 1,316 | | Connected rate | 0.02632% | | 8–40 black-pixel glyphs | 313,927 | | 8–40 density rate | 6.27854% | | Full interpretability candidate filter | **0** | | Candidate rate | **0** | The candidate filter was: `connected AND 8<=black_pixels<=40 AND 8<=transitions<=90 AND edge_count>0` ## Important finding The geometric filter as currently defined is **too restrictive in combination** for random uniform 96-bit sampling: five million samples produced no glyph satisfying all four conditions. This does not contradict the raw 96-bit capacity result. It means that the current candidate definition is a poor random-space filter for discovering the desired population. The connected rate is consistent with the earlier million-sample experiment: - earlier: 276 / 1,000,000 = 0.02760% - this run: 1,316 / 5,000,000 = 0.02632% The agreement is useful evidence that the connectivity statistic is stable under the deterministic generator. ## Consequence for the 4-billion target A zero-observation candidate rate cannot be extrapolated into a population estimate. Therefore this run does **not** establish that the current four-condition candidate filter yields four billion symbols. Raw capacity remains overwhelmingly sufficient: `2^96 / 4,000,000,000 ≈ 1.98e19` The next experiment should therefore change the sampling strategy rather than simply increase uniform random sample count. In particular, sample **conditioned/constructed connected glyphs** first, then measure the transition and density distributions. That directly explores the graph-valid region instead of spending almost all samples on disconnected bitmaps. ## Status - Raw capacity: **PASS** - Connectivity feasibility: **SUPPORTED** - Current random-space combined candidate filter: **NOT DEMONSTRATED** - Four-billion interpreted namespace: **NOT PROVEN** The result is intentionally preserved without converting zero observations into an unsupported extrapolation.