The Lean candidate proofs workflow checks new proof modules separately from the accepted root imports. A candidate is not a completed collection result. Acceptance also requires exact statement review, registration in Conjectures, required endpoint checks in ProofAudit, and the ordinary kernel-recheck gate.
The base, product, deterministic hyperplane, and complete four-* scopes have since been registered in the accepted package. The separate jobs remain available to check changes to those components. The other lanes retain their candidate status unless the accepted-scope report records a later registration.
Each job builds its project dependencies in order, audits all imported project theorems, including private declarations, for the three allowed standard axioms, rejects project axiom declarations, and independently replays those modules with leanchecker. The only allowed axioms are propext, Classical.choice, and Quot.sound. No native-computation axiom is allowed.
The candidate lanes are:
| Lane | Scope |
|---|---|
base |
Graffiti.pc318 and its sharpness example; consecutive families, higher-speed examples, exceptional eight-speed families, prefix-product bounds, modular search helpers |
product |
Classification and values within the three stated eight-speed product families |
hyperplane |
Exact shifted maximum over the containing real hyperplane, attainment, and the bound for orthogonal integer velocities |
hyperplane-haar |
Distribution of the predicted hyperplane maximum under Haar-uniform shifts; uniform finite-character approximation on compact subsets of the torus |
modular-closed |
Closed-threshold binary search soundness, with the lower-dimensional runner bound and finite certificate explicit |
nearest-plane |
Euclidean rounding to an actual integer combination, with the squared Gram–Schmidt error bound |
primitive-basis |
Completion of a primitive integer vector to an integer basis |
small-kz |
Exact subset-product arithmetic for the smaller-speed norm cutoffs, with geometric inputs explicit |
four-17 through four-61 |
Ten prime residue covers and their exhaustive lifting trees |
four-assembly |
Primitive four-speed classification up to 5832, the independent all-size cutoff, and the resulting unconditional four-speed gap theorem |
kz-9 through kz-13 |
Exact subset certificates for the five product problems |
kz-assembly |
Product minimization under the stated positive prefix-sum and adjacent-ratio constraints |
The KZ arithmetic theorem does not supply the separate lattice-geometric existence and distance bounds. The higher-speed example and family theorems do not establish that all possible configurations are listed. The hyperplane lane does not prove the finite-exception, probability, distribution, or counting assertions in that result.
The workflow selects lanes by the changed files and each lane's full project dependency closure. Changes to the checker, workflow, toolchain, or package configuration run every lane. This avoids restarting unrelated finite tables when another proof worker adds a module. Manual dispatch also runs every lane.
Run one lane locally with:
python3 scripts/verify_lean_candidate.py base
python3 scripts/verify_lean_candidate.py four-17
python3 scripts/verify_lean_candidate.py kz-9Use --plan to inspect the module dependency order without running Lean. Assembly jobs reuse only artifacts uploaded after all three checks succeeded in the same workflow run. The receipt checks the commit, every project source hash, the pinned package configuration, and every reused artifact hash. Lake then checks its normal dependency traces; the assembly job also repeats the axiom audit and independent kernel replay.
The ordinary accepted-proof workflow remains unchanged. Li is still deferred until the other collection results are complete.
Completed line-size analytic components
The complete line-size asymptotic and analytic generating-function closures are registered from their exact passing receipts. The earlier finite geometry, tuple, formal-series, profile, and Euler components were already registered. All line-size tasks remain available as regression gates; they no longer represent an outstanding mathematical obligation for this note. Exact result and receipts.
The Motzkin evaluator, rational-intervals, and relations tasks have passed all three strict gates and are registered with their exact receipts. These tasks remain available for regression checks. The final relation-basis pass completes the stated Motzkin result. Every remaining task requires strict source checking, a complete theorem axiom audit, and independent replay of its exact closure.
The Motzkin character-window closure has now passed all three strict gates and is registered in the accepted package. Its exact five-module/53-theorem scope and receipt are recorded in the accepted status. The task remains available as a regression gate.
The prime-257 minimum-pair-20 split support certificate is registered from its 33-module passing receipt. It proves the original normalized-support exclusion for that one probe. The full six-speed numerical coverage remains incomplete. The complete length-598 obstruction, including Hadamard598Fibers, is now registered. The literal finite-border certificate and universal abelian-cover cycle theorem are also registered; their component tasks remain regression gates. See the precise scope and exact receipts in the accepted status.
The complete conditional fixed-border subgroup and pattern-reduction closures are now registered from their exact e25e0dcd receipts. The Legendre reduction and invariant-quotient closures are registered for their proved order-dividing-27 and invariant-quotient scopes. These tasks remain available as regression gates. The length-1197 bound of nine, full finite quotient/no-lift packet, and two-row construction remain candidates.
Actual Legendre-pair matrix and quotient transport
The construction and quotient-data Legendre tasks are accepted components after their exact strict-source, axiom-audit, and independent-kernel receipts passed. Their scopes are the actual two-row Hadamard matrix and transport from centered order-27-invariant rows to the actual finite quotient data. The full quotient enumeration/no-lift packet and the final multiplier bound of nine remain candidates. Verification record.