Literal-packaging comparison with complete indexed artifacts
Seven public-metadata/runtime/synthetic cases compare conventional baselines, with every original ID and offset/length entry included in SSP1. All 31 emitted .ssp files reconstruct their original bytes. Gzip level9 uses mtime0; zlib level9 and XZ preset6 compress the complete artifact. Compressed sizes do not measure random-access latency, decompression memory or linker/deployment outcomes. Exact inputs/results, finite validation.
| Case | Original → greedy blob bytes | Original → greedy SSP1 bytes | Gzip bytes | XZ bytes | Exact optimum bytes | Paper-count stage |
|---|---|---|---|---|---|---|
| synthetic_four_overlap_fragments | 24 → 16 | 80 → 72 | 75 → 69 | 116 → 112 | 16 | 16 |
| synthetic_periodic_windows | 102 → 22 | 308 → 228 | 163 → 138 | 180 → 168 | unknown | 17 |
| synthetic_greedy_gap | 16 → 10 | 72 → 66 | 60 → 60 | 100 → 100 | 9 | 7 |
| runtime_python_keywords | 155 → 127 | 577 → 549 | 350 → 322 | 352 → 336 | unknown | outside_reduced_length_cap |
| public_source_discipline_labels | 360 → 353 | 566 → 559 | 368 → 365 | 392 → 388 | unknown | outside_reduced_length_cap |
| public_source_first_forty_eight_family_titles | 2535 → 2507 | 3113 → 3085 | 1685 → 1685 | 1616 → 1648 | unknown | outside_reduced_length_cap |
| synthetic_UTF8_zero_duplicate_views | 12 → 9 | 90 → 87 | 76 → 73 | 116 → 112 | 9 | 9 |
The 48-title sample saves 28 raw bytes and no gzip bytes, while XZ grows by 32. Reordering equal raw-size baselines also changes compression. The periodic synthetic sample is favorable to conventional greedy; it establishes no need for a new source backend or paying customer. The generated greedy-gap case has raw optimum nine versus greedy ten; XZ is 100 bytes for both. A smaller blob does not force a smaller complete compressed artifact.
Source family 128 claims factor two for its constructed count/layer/request/cycle algorithm, not these conventional baselines. The separate count-stage implementation was tested against finite occurrence/optimum oracles; its W validity in general depends on the unverified source proof. Full construction, kernel/program acceptance, actual compiler/linker integration, customer workload, access performance and profitable delivery remain open. Detailed business/technical decision.