Prospective campaign and negative runtime outcome
Preserved research record. This public derivative removes local-only path links; scientific wording and original status claims remain historical. Internal AI reviews are not external peer review. Original SHA-256: c5c45df63195777ff21f0f5541f8d08b67093018e79dd5c3e89e4f9553dba466.
Equations in these preserved Markdown records retain their TeX notation where present; this is not a newly typeset proof edition.
Prospective live-vtree integration campaign — final record
The new, separately frozen synthetic campaign completed on Linux. Mechanism
GO passed; algorithmic GO failed. Live PySDD moves preserved every matched
fresh-search decision and made the search stage cheaper, but they did not make
the selected complete exact factorization method faster than the selected
baseline on the held-out split. The Phase 5B negative result remains unchanged.
Provenance and execution
The registered CAMPAIGN_PROTOCOL.md and 36-record
candidate_manifest.json were frozen before any validation row. The freeze
receipt SHA-256 is
e7b1250c3a512fa7b2d1bbf76cf46a4ff8b0e2654c7231035fb29411f252c587.
The local Linux Docker run used python:3.11-slim image digest
sha256:e41613d42d4891e4930f79523f93f81bbc7632584ec65e36ab055f41a800b41e,
Python 3.11.16, pinned packages including PySDD 1.0.6, 4 GiB container
memory, 2 CPUs, a 2 GiB worker RLIMIT_AS and a 20-second watchdog. The
freeze hard-cap negative control and capped PySDD exactness smoke passed.
Earlier Phase 5B and integration directories were mounted read-only.
Validation completed all 252 planned rows. The registered selection rule
locked B2 as baseline and L1 as challenger; the validation-results SHA-256
is 35ff39d90574722392b31995ad5fdc77bc145e8cdcb5a2e345bc4d0fb56ba168.
Held-out then completed all 504 planned rows, with SHA-256
f667606603bd2f6383e73ccd61c31c8ebe77f0d5a8130bdfc6706c207b23b21e.
All 756 rows had status ok, matched their frozen job identities and
resource metadata, and passed the runner's independent semantic and carrier
partition checks. Cumulative worker process-wall time was 270.31 seconds,
well below the registered 90-minute ceiling; no row was censored. The
maximum reported worker peak RSS was 31.8 MB.
Registered decision
| Gate |
Registered measure |
Result |
| Complete-time primary |
L1 / B2 held-out paired geometric ratio |
1.03798 |
| Complete-time uncertainty |
One-sided 95% bootstrap upper ratio |
1.14154 |
| Fresh/live M1 parity |
F1/L1 matched pairs and mismatches |
72/72; 0 |
| Fresh/live M2 parity |
F2/L2 matched pairs and mismatches |
72/72; 0 |
| M1 search mechanism |
L1 / F1 geometric ratio; upper 95% |
0.18050; 0.20167 |
| M2 search mechanism |
L2 / F2 geometric ratio; upper 95% |
0.18087; 0.20094 |
The complete-time ratio fails the required <=0.90 improvement and its upper
bound fails the <1 condition. Both matched search mechanisms satisfy their
registered criteria with complete pairing and no decision mismatches. The
saved campaign_analysis.json accordingly records mechanism_go: true
and algorithmic_go: false; there were no method failures.
Exposure diagnostic and scope
As a post-campaign diagnostic on saved rows, B3's direct partition missed
the exact semantic partition on 12 of 24 held-out instances. L1 missed on
8 of 24. L1 reached the exact direct partition in four B3-miss cases, all
four sizes of the equivalence-ladder-bridge family. This exposure result was
not used to select methods or change the registered objective. It establishes
some genuine factor recovery in the new synthetic inputs, while the primary
complete-time result remains negative. These small, constructed CNFs do not
support a public-corpus or large-scale performance claim.
The source hashes, complete rows, frozen selection and analysis were checked
again after the run. The disposable container exited and was removed. No
Runpod resource was used.