back to table · edit · history · where entries came from · files · download
4642 bytes, as of the version from 2026-10-24 20:50 (current). Recorded here, not run.
Build handoff: T466, non-phi4 Feynman periods
Status: successful build. T466 contains 511 entries. Generator verification reports 511/511 matched, 0 differing, 0 missing and 0 extra. Independent checks passed on every stored value. The website audit returns clean: true with no findings. This handoff is for the controller's separate critique and repair stages. No publication, offer, review, Git commit, source edit or deployment was performed.
Scope and conventions
- Family: numberdb-data#199, saved as family-199.txt. Title and parameters loops,index follow that family.
- Source: Panzer–Schnetz arXiv:1603.04289v2, PeriodsNonPhi4; SHA-256 f735c62370e0e1561e0cdffc28d86226d6d0fb0be138729abd44f74191d1966a. The unversioned download has the same digest.
- 629 census labels. All 511 evaluated labels are included: 1 at five loops, 3 at six loops, 50 at seven loops, 457 at eight loops. The 118 unknown eight-loop labels remain omitted and documented.
- One named quantity: the projectively normalized Feynman period of the indexed primitive log-divergent non-phi4 graph. Every label remains, even when values agree.
- The source's weighted completions are preserved in its attached raw file. Braces mean weight +1, brackets weight -1. Deleting the largest vertex gives the graph for the defining integral. Every decoding was tested against weighted degree four, connectivity, edge count, loop count, a degree above four and all proper induced vertex subsets.
- 100 significant digits; heuristic (agreement-checked), not certified integration. PARI uses the source's descending MZV indices directly. Sage Multizeta would reverse them. Alternating MPLs at -1 use polylogmult(s,[-1,1,...,1]); no nonreal root-of-unity approximation is needed.
Evidence
- screen.json and cheap-checks.json: already_here, already_asked, three numerical API lookups, and existing period, physics, transcendental tags. No duplicate of this family was found. Claim creation returned T466, unpublished and not ready for review.
- numerical-results.json retains every computed 100-digit number, source discrepancy and precision agreement. Worst source discrepancy: 0.4978582329 ULP at (7,31). Minimum relative agreement between 150- and 200-digit working fields, each with 32 guard bits: 155.3503594 digits at (8,43).
- checks-before.log: all 511 source values, all 629 graphs, 129 Arb single-zeta evaluations, all four wheel periods (5,1), (6,1), (7,21), (8,162), exact expression equalities and nonzero MZV/MPL controls passed.
- 71 labels equal 25 distinct values in companion T464. cross-equality-proofs.json records both exact expressions, compared as polynomials over rational coefficients. The generator records these as equals, not repeats. Another 274 rows point internally to earlier labels; 191 distinct expressions are retained.
- verify-summary.txt and stored-check-results.json: verification and independent checks of the actual database values; stored-document.json preserves the readback. audit.json records the clean site audit.
- dry-run.log: clean; 511 entries, longest reported representation 164 characters, about 113.4 KB. The site Programs snippet was executed and compared with the independently parsed source expression.
Review attention
- T464 is currently a draft. HREF{T464} is a supported stable identifier (confirmed in the audit resolver), not a guessed slug. The tables have cross-equalities and explicit links, so their eventual publication ordering must avoid a public page pointing to a private draft. Nothing in T464 was edited.
- The checks evaluate the source's period formulas; they do not independently integrate all graphs. The wheel theorem and graph properties supply independent evidence about normalization and graph identity; the 129 Arb evaluations independently check the arithmetic library.
- The family said 100 digits and MZVs, but its non-phi4 data also needs alternating MPLs. These are supported explicitly and controlled by Li_{1,1}(-1)=log(2)^2/2 and Li_3(-1)=-3*zeta(3)/4.
- Table prose carries no request provenance. The controller should manage the family queue after its critique/repair workflow; this run does not offer a draft.
Reproduction
Download the attached generate.py, parse_census.py, checks.py, check.py, PeriodsNonPhi4, cross-equalities.json, cross-equality-proofs.json, and cross-source-values.json into one directory. Run `sage -python checks.py`; run `sage -python generate.py` with an authorized API key to verify the draft. `checks.py --stored` checks the database values. The build environment uses agents/sage.sh with those explicit inputs and credentials piped through stdin.