OverviewFilesCommitsIssuesDocsReleasesPeers

atomspace-layer-discovery commits / 19845da

experiments/ladder/port: stage C parser statistical check PASS
lgsect vs Linas's recorded G0 parses, with a fresh run of Linas's own code as the noise yardstick, on all 5,853 parses: links/parse, ANY share, mean FMI, connectors/section and section-type cosine all within the pre-registered bounds. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
19845dab659c2f16445d37d1dbf9ebb0cf62eb83
by plankatron <fleeky@gmail.com> · 2026-10-18T18:12:42.000Z

Changed files (1) +14 -0

experiments/ladder/port/STAGE_C_CHECK.md +14 -0
……@@ -23,3 +23,17 @@ different parses. Linas's stage C is not reproducible run to run, so no port can
2323**Pass:** for every metric, |LS − G0| ≤ 2 × |LF − G0| + 1% of the G0 value, and
2424cosine(LS, G0) ≥ cosine(LF, G0) − 0.02. LF − G0 is Linas-vs-Linas sampling noise; the port may not add
2525more than that.
26
27## Result (2026-10-18 14:12) — PASS
28
29All 1,951 windows × 3 parses = 5,853 parses in each run. `statsect.py` output:
30
31| metric | G0 | LF (Linas, fresh) | LS (`lgsect`) | \|LF−G0\| | \|LS−G0\| | pass |
32|---|---|---|---|---|---|---|
33| links / parse | 16.4422 | 16.4929 | 16.4661 | 0.0507 | 0.0239 | yes |
34| ANY share | 0.1009 | 0.1002 | 0.1018 | 0.0006 | 0.0009 | yes |
35| mean FMI (non-ANY) | 3.9742 | 3.9751 | 3.9806 | 0.0009 | 0.0064 | yes |
36| connectors / section | 2.1631 | 2.1698 | 2.1662 | 0.0067 | 0.0031 | yes |
37
38Section-type cosine to G0: LF 0.9152, LS 0.9138 (LF vs LS 0.9150) — pass. The port's parses differ from
39Linas's recorded run no more than a second run of Linas's own code does.