Batch-add 12 Backlog Models + 18 Backlog Rules#1067
Open
isPANN wants to merge 54 commits into
Open
Conversation
- New skill .claude/skills/auto-pipeline: orchestrator that drives one Backlog issue from quality gate to Final review via fresh-context subagents (check-issue, fix-issue, run-pipeline, review-pipeline). Substantive issue-body problems are routed to codex xhigh; fundamental flaws with no public reference park the issue on OnHold. - check-issue: add Rule Check 5 (Completeness, fail label "Incomplete"). Mandatory literature research + codebase corner-case enumeration + hand-tracing on >= 2 non-canonical instances for every [Rule] issue. - review-structural: add Step 4b (Round-trip Execution, mandatory for Rule reviews). Reviewer must run cargo test by name, paste the "test result: ok" line, and confirm the test exercises the four phases of a real round-trip. - review-quality: promote "closed-loop without round-trip verification" from a Minor flag to Critical, with explicit red flags (is_some-only, target-side-only asserts, unique-optimum instances). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Robotics inverse-kinematics problem: given link lengths l_j, target
g in R^2, per-link sampled orientations Phi_j, and consecutive-pair
admissibility sets A_j, pick indices a_j in {0..m_j-1} with
(a_{j-1}, a_j) in A_j minimizing the squared end-effector distance
||sum_j l_j (cos phi_{j,a_j}, sin phi_{j,a_j}) - g||^2.
- src/models/misc/minimum_discrete_planar_inverse_kinematics.rs:
per-link dims (non-binary), Min<f64> objective, A_j feasibility
returns Min(None), declare_variants! default entry, ProblemSchemaEntry
+ ProblemSizeFieldEntry, canonical example_db spec via inventory.
- src/unit_tests/models/misc/...: creation, evaluate (feasible/
infeasible), brute-force solver, serialization roundtrip.
- problemreductions-cli/: new (f64,f64) and Vec<Vec<(usize,usize)>>
schema parsers; --link-lengths/--target-point/--orientation-samples/
--allowed-pairs flags via the schema-driven create path.
- docs/paper: problem-def block + display-name + worked example;
references.bib entries for Salloum2025 and DaiIzattTedrake2019.
Reference: Salloum et al., "Quantum annealing for inverse kinematics
in robotics", Scientific Reports 2025, doi:10.1038/s41598-025-34346-z.
Closes #994
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The brute-force search space is prod_{j=1}^n m_j, not 2^n — per-link
sample counts m_j are arbitrary. Add a `total_configurations()` getter
that returns the product, and rewrite the declare_variants! complexity
as `num_links * total_configurations` (n vertices in evaluate cost
times the iteration space).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Trivial single-line rewrite to match rustfmt. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-k-plex problem: given graph G=(V,E), vertex weights w, and integer
k>=1, find max-weight subset S subseteq V such that the induced
subgraph G[S] has maximum degree at most k-1 (i.e. every selected
vertex has at most k-1 selected neighbours). Generalizes
MaximumIndependentSet (the k=1 case) and is the complement-graph view
of maximum k-plex from the clique-relaxation literature.
- src/models/graph/maximum_co_k_plex.rs: MaximumCoKPlex<G,W,K>
parameterized by graph type, weight type, and K-multiplier. Only the
KN (runtime-k) variant registered initially per the issue's
"initially KN, K1/K2/... later" plan. Max<W::Sum> objective,
induced-degree feasibility, declare_variants! default + i32 variant,
canonical example via inventory (5-cycle weights (5,1,4,1,3) k=2,
optimum {0,2,4} value 12).
- src/unit_tests/models/graph/maximum_co_k_plex.rs: creation,
evaluate-feasible (issue optimum + smaller feasible), evaluate-
infeasible (degree-2 violation), brute-force solver, serialization.
- problemreductions-cli/src/commands/create/: schema-driven CLI maps
schema field bound_k to existing --k flag with semantic validation.
- docs/paper: problem-def block with C_5 worked example and k=1 ->
MaximumIndependentSet equivalence note; references.bib gains
Hernandez2016MolecularSimilarity and HosseinianButenko2022KDependent.
References: arXiv:1601.06693 (Hernandez et al., 2016) for the
molecular-similarity framing; doi:10.1016/j.dam.2021.10.015
(Hosseinian & Butenko, 2022) for the maximum k-dependent set view.
Closes #1015
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
MCES: given two directed edge-labelled graphs G1, G2, find a partial injective map f: U1 ⊆ V1 → V2 maximizing the number of preserved labelled arcs (u, λ, v) ∈ E1 with f(u), f(v) defined and (f(u), λ, f(v)) ∈ E2. Edge labels must match exactly; set semantics (no multiplicities); disconnected common subgraphs allowed; no secondary tie-break. - src/models/graph/maximum_common_edge_subgraph.rs: local LabelledArc + LabelledDigraph structs (does not extend the existing Graph trait hierarchy in this PR). dims = vec![|V2|+1; |V1|] with the +1 slot encoding ⊥. Max<i64> objective with injectivity feasibility on the matched slots. ProblemSchemaEntry + ProblemSizeFieldEntry for num_vertices_1/_2 and num_arcs_1/_2, declare_variants! default with complexity (num_vertices_2+1)^num_vertices_1. Canonical example via inventory from the issue's 5-vs-4-vertex instance with optimum value 5. - src/unit_tests/models/graph/maximum_common_edge_subgraph.rs: 12 tests covering creation, evaluate-feasible (optimum 5), evaluate-injectivity-violated, evaluate-fewer-preserved, brute-force solver, serialization. - problemreductions-cli/: new --graph-1 / --graph-2 flags with a LabelledDigraph parser; alias MCES. - docs/paper: problem-def block, display-name, MCES worked example. - docs/paper/references.bib: corrected per Crossref against the check-issue warning — Bahiense2012 first names (Laura/Gordana/Breno), Soule2021 author list (Soule/Reinharz/Sarrazin-Gendron/Denise/ Waldispuhl) and venue, Bokhari1981 volume (C-30). References: doi:10.1109/TC.1981.1675756 (Bokhari 1981), doi:10.1016/j.dam.2012.01.026 (Bahiense et al. 2012, polyhedral investigation), doi:10.1371/journal.pcbi.1008990 (Soule et al. 2021, RNA networks application). The direct `MaximumCommonEdgeSubgraph -> ILP` rule (#1019) is out of scope for this PR and will follow separately. Closes #1018 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Exact-cardinality edge-weighted clique: given a simple undirected
graph G=(V,E), edge weights w: E→R, and an integer k with 0≤k≤|V|,
find a vertex subset S with |S|=k forming a clique that maximizes
the sum of weights of edges induced by S. Edge weights may be
negative; k=0 and k=1 are admitted with objective value 0.
Distinct from the existing MaximumClique (vertex-weighted, no
exact-k) and KClique (decision problem with threshold |S|>=k).
- src/models/graph/maximum_edge_weighted_k_clique.rs:
MaximumEdgeWeightedKClique<W: WeightElement> with SimpleGraph fixed;
edge_weights vector aligned to graph.edges() order, runtime k field.
dims = vec![2; |V|]. Max<W::Sum> objective; infeasible when |S|≠k
or S is not a clique. declare_variants! default (SimpleGraph,i32)
plus (SimpleGraph,f64). Canonical example via inventory from the
issue's 4-vertex instance with negative weight (clique {0,1,2}
value 8 beats {0,1,3} value 6).
- src/unit_tests/models/graph/maximum_edge_weighted_k_clique.rs:
12 tests covering creation, evaluate-feasible (both optima),
evaluate-infeasible-wrong-size, evaluate-infeasible-not-clique,
brute-force solver, edge cases k=0 and k=1 (value 0), f64 variant,
serialization roundtrip, panic guards.
- docs/paper: problem-def block with worked example highlighting that
the optimum includes a negative edge; display-name entry; cites
Gouveia & Martins 2015 and Hunting/Faigle/Kern 2001.
- docs/paper/references.bib: Crossref-verified Gouveia2015MEWC
(author corrected to Pedro Martins, not Paulo as the issue body
said) and HuntingFaigleKern2001EWC.
References: doi:10.1007/s13675-014-0028-1 (Gouveia & Martins 2015,
sparse-graph compact formulations); doi:10.1016/S0377-2217(99)00449-X
(Hunting, Faigle & Kern 2001, Lagrangian relaxation).
The direct `MaximumEdgeWeightedKClique -> ILP` rule (#1021) is out
of scope for this PR and will follow separately.
Closes #1020
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Given a simple undirected graph G=(V,E), find a minimum-cardinality
edge set F ⊆ E such that every connected component of G - F is either
an isolated vertex or a highly connected graph on ≥3 vertices
(edge connectivity λ(H) > |V(H)|/2, strict). Components of size 2
are explicitly invalid. Weaker than clique-deletion: every K_k for
k≥3 is highly connected, but not every highly connected graph is a
clique.
- src/models/graph/highly_connected_deletion.rs: variables are EDGES
(x_e=1 means delete edge e). Min<i64> objective counts deletions;
infeasibility on any non-singleton component that is not highly
connected (and any 2-vertex component). Private edge_connectivity
helper computes λ via repeated max-flow with unit edge capacities
(fine for small components in tests). ProblemSchemaEntry,
ProblemSizeFieldEntry (num_vertices/num_edges), declare_variants!
default with complexity 2^num_edges. Canonical example via
inventory: K3 with leaf vertex 3 attached to 2 (4 vertices,
4 edges) — optimum deletes only (2,3), value 1.
- src/unit_tests/models/graph/highly_connected_deletion.rs: 17 tests
covering creation, evaluate-optimum, evaluate-zero-deletions-
infeasible, evaluate-delete-all-feasible, evaluate-infeasible
2-vertex-component and infeasible path-component, wrong-length
config guard, brute-force on canonical + a "double triangle"
discriminator instance (two K3's joined at a bridge — optimum 1)
to address the check-issue warning about example discriminatory
power, serialization, variant, plus edge_connectivity helper
tests (single vertex=0, single edge=1, P3=1, K3=2, K4=3).
- docs/paper: problem-def block with the K3-with-leaf worked example,
display-name entry; Crossref-verified BibTeX entries for Hüffner
et al. 2014 (TCBB) and Hartuv & Shamir 2000 (IPL), with proper
umlaut encoding H{"u}ffner per repo convention.
References: doi:10.1109/TCBB.2013.177 (Hüffner et al. 2014, partitioning
biological networks); doi:10.1016/S0020-0190(00)00142-3 (Hartuv &
Shamir 2000, HCS clustering algorithm).
The direct `HighlyConnectedDeletion -> ILP` rule (#1023) is out of
scope for this PR and will follow separately.
Closes #1022
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Classical directed-multigraph satisfaction problem: given D=(V,A) with
parallel arcs and loops allowed, decide whether a directed trail
exists that uses every arc exactly once. No start or end vertex is
fixed by the input. Empty-arc instance is accepted with the empty
trail; isolated vertices are ignored.
Polynomial-time solvable (O(num_vertices + num_arcs)) by the standard
Eulerian criterion plus Hierholzer construction, so this widens the
catalog beyond NP-hard problems.
- src/models/graph/eulerian_path.rs: EulerianPath { graph:
DirectedGraph }. dims = vec![m; m] where m = num_arcs (variable t
picks which arc occurrence is the t-th trail step); the brute-force
search space is m^m but the registry complexity reflects the
linear-time best-known algorithm. Or-typed feasibility: configuration
must be a permutation of {0..m-1} and consecutive arcs must chain
(end of arc t equals start of arc t+1). declare_variants! default
+ ProblemSchemaEntry + ProblemSizeFieldEntry. Canonical example via
inventory from the issue's 3-vertex 4-arc instance with parallel
arcs (yes-instance, witness config [0,2,3,1]).
- src/unit_tests/models/graph/eulerian_path.rs: 11 tests covering
creation, evaluate-valid-witness, evaluate-not-permutation,
evaluate-bad-trail, evaluate-out-of-range, evaluate-wrong-length,
brute-force yes (canonical) + brute-force no (the issue's 2-vertex
4-arc imbalanced counterexample), empty-arcs edge case (Or(true)
with the empty witness), serialization roundtrip, variant + name.
- problemreductions-cli/src/commands/create/schema_support.rs: wire
--graph (DirectedGraph) for EulerianPath via the existing parser.
- docs/paper: problem-def block with both the yes-instance and the
no-instance from the issue; display-name entry; references.bib gains
Crossref-verified BangJensenGutin2009Digraphs (J{\o}rgen Bang-Jensen,
o-slash) and Ebert1988ComputingEulerianTrails (J{"u}rgen Ebert,
u-umlaut) — corrected from the issue body's mojibake.
References: doi:10.1007/978-1-84800-998-1 (Bang-Jensen & Gutin 2009,
digraphs); doi:10.1016/0020-0190(88)90170-6 (Ebert 1988, computing
Eulerian trails).
The direct `EulerianPath -> ILP` rule (#1025) is out of scope for
this PR and will follow separately.
Closes #1024
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Biology-paper prize-collecting Steiner forest: given a network G with
nonnegative vertex prizes p(v), nonnegative edge costs c(e), and
tradeoff coefficients beta, omega, find a forest subgraph
F = (V_F, E_F) minimizing
beta * sum_{v notin V_F} p(v) + sum_{e in E_F} c(e) + omega * kappa(F)
where kappa(F) is the number of tree components (singleton selected
vertices count). Generalizes prize-collecting Steiner tree from one
connected tree to a forest; the artificial-root trick is deliberately
kept out of the base model and will live in the companion reduction
rule.
- src/models/graph/prize_collecting_steiner_forest.rs:
PrizeCollectingSteinerForest<G, W> with dims = vec![2; n+m] (vertex
bits then edge bits), Min<W::Sum> objective. Feasibility checks
edges-incident-to-selected-vertices and forest acyclicity; infeasible
→ Min(None). Canonical example via inventory from the issue's
3-vertex path with optimum [1,1,1, 1,0] value 5 (cost 1 + omega*2
components). declare_variants! default (SimpleGraph,i32) plus
(SimpleGraph,f64). Complexity 2^(num_vertices+num_edges).
- src/unit_tests/models/graph/prize_collecting_steiner_forest.rs:
13 tests — creation, evaluate-optimum, evaluate-full-path (value 9),
evaluate-three-singletons (value 6), evaluate-empty-forest
(value 12), evaluate-edge-without-endpoint-infeasible,
evaluate-cycle-infeasible (triangle selected entirely),
brute-force solver, serialization, f64 variant, panic guards.
- problemreductions-cli/: new --vertex-prizes / --edge-costs / --beta /
--omega flags via schema-driven create; mapping and fixture updates.
- docs/paper: problem-def block with worked example breakdown
(omitted-prize / edge-cost / component terms summing to 5);
display-name; Crossref-verified BibTeX for both Tuncbag et al.
papers (JCB 2013 and RECOMB 2012).
References: doi:10.1089/cmb.2012.0092 (Tuncbag et al. 2013, JCB);
doi:10.1007/978-3-642-29627-7_31 (Tuncbag et al. 2012, RECOMB).
The direct `PrizeCollectingSteinerForest -> SteinerTree` rule (#1027)
is out of scope for this PR and will follow separately.
Closes #1026
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Lexicographic objective on a directed multigraph with source s, sink t,
arc capacities u_e, and arc costs c_e: first maximize the s-t flow
value |f|, then among maximum-value flows minimize total arc cost
sum_e c_e f_e. Captures the CellRouter (Lummertz da Rocha et al. 2018)
model directly.
Architectural carve-out: the repo's Problem trait is discrete, so the
implementation restricts to INTEGRAL flows with dims=[c_e+1; m]. This
mirrors the existing MinimumEdgeCostFlow precedent and stays sound
for any rational instance by scaling. Documented in the module doc.
Lex encoding: Min<i64> with combined scalar
score = M * (max_possible_flow - flow_value) + cost
where M = sum_e c_e * u_e + 1 dominates any feasible cost — so lower
scores always prefer higher flow value first, breaking ties by lower
cost. Infeasible (capacity / conservation violations) → Min(None).
- src/models/graph/minimum_cost_maximum_flow.rs:
MinimumCostMaximumFlow { graph: DirectedGraph, source, sink,
capacities: Vec<i64>, costs: Vec<i64> }. Inherent helpers
flow_value(config) and total_cost(config) for tests. Canonical
example via inventory: V={0,1,2,3}, arcs [(0,1),(0,2),(1,2),
(1,3),(2,3)], capacities [2,1,1,1,2], costs [1,0,0,1,2] — optimum
config [2,1,1,1,2] with value 3 and cost 7. ProblemSchemaEntry +
ProblemSizeFieldEntry (num_vertices, num_arcs). declare_variants!
default with complexity (num_vertices+num_arcs)^6 (a conservative
polynomial placeholder justified by the LP formulation).
- src/unit_tests/models/graph/minimum_cost_maximum_flow.rs: 9 tests
covering creation, evaluate-optimum, evaluate-suboptimal-feasible,
evaluate-capacity-exceeded (infeasible), evaluate-conservation-
violated (infeasible), brute-force solver returning value 3 cost 7,
serialization, and the lex-tiebreaker test on a 4-vertex bottleneck
instance where two distinct max-value flows (value 1) exist with
costs 1 and 5 — brute-force must pick the cheaper one. The
tiebreaker test directly addresses the check-issue warning that
the original example admits a unique max-flow.
- problemreductions-cli/: new --source / --sink (usize) flags wired
via schema-driven create; --graph (DirectedGraph), --capacities,
--costs reused from the MECF wiring.
- docs/paper: problem-def block explaining the lex objective and the
integral-flow restriction, worked example with value/cost
breakdown; display-name; Crossref-verified BibTeX for
Lummertz da Rocha et al. 2018 (doi:10.1038/s41467-018-03214-y);
MIT 6.854 min-cost-flow notes as a @misc entry with URL.
References: doi:10.1038/s41467-018-03214-y (CellRouter, Nature Comms
2018); MIT 6.854 scribe notes for the standard min-cost-flow
equivalence.
The direct `MinimumCostMaximumFlow -> MinimumCostCirculation` rule
(#1031) is out of scope for this PR and will follow separately.
Closes #1029
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Classical minimum-cost circulation on a directed multigraph with
finite arc capacities u_e ≥ 0 and signed arc costs a_e ∈ R: find
g: E→R_{≥0} satisfying capacity bounds (0 ≤ g_e ≤ u_e) and flow
conservation at every vertex, minimizing sum_e a_e g_e.
This is the exact companion target for `MinimumCostMaximumFlow`
(reduction #1031) — the standard MCMF → MCCirc reduction uses a
sufficiently negative return arc from sink to source, which is why
signed costs must be supported in the base model.
Architectural carve-out (same as MinimumCostMaximumFlow #1029 and
MinimumEdgeCostFlow): the discrete Problem trait restricts to
INTEGRAL circulation with dims=[c_e+1; m]; sound for any rational
instance by scaling. Documented in the module doc.
- src/models/graph/minimum_cost_circulation.rs:
MinimumCostCirculation { graph: DirectedGraph, capacities: Vec<i64>,
costs: Vec<i64> } — no source/sink, conservation at every vertex.
Min<i64> objective; capacity-or-conservation violations → Min(None).
ProblemSchemaEntry + ProblemSizeFieldEntry (num_vertices, num_arcs).
declare_variants! default with conservative polynomial placeholder
(num_vertices+num_arcs)^6. Canonical example via inventory — a
3-vertex two-cycle instance discriminating between four feasible
alternatives (zero, cycle-A-only -2, cycle-B-only -3, both at
capacity -5) so round-trip tests have real discriminatory power,
addressing the check-issue warning about the issue's 2-vertex
example being too small.
- src/unit_tests/models/graph/minimum_cost_circulation.rs: 11 tests
covering creation, evaluate-optimum (config [2,2,1,1] → Min(-5)),
evaluate-zero, evaluate-cycle-A-only (-2), evaluate-cycle-B-only
(-3), evaluate-half-cycle-A (-4), evaluate-infeasible (capacity
exceeded, conservation violated), brute-force solver, serialization,
negative-cost-only-cycle smoke.
- problemreductions-cli/: --graph (DirectedGraph), --capacities,
--costs reused from MECF/MCMF wiring; new schema mapping for MCCirc.
- docs/paper: problem-def block with the two-cycle worked example
spelled out (per-unit costs, capacity bottlenecks), display-name
entry; reuses the existing mit6854MinCostFlow @misc bib entry
added with MCMF — no new references.bib changes.
References: MIT 6.854 (S2021) min-cost flow algorithms notes (shared
with #1029).
The direct `MinimumCostMaximumFlow -> MinimumCostCirculation` rule
(#1031) is out of scope for this PR and will follow separately.
Closes #1030
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Consensus-string problem under Hamming distance: given an alphabet
Σ = {0,...,q-1} and n equal-length strings s_1,...,s_n ∈ Σ^m, find a
center string c ∈ Σ^m minimizing max_i d_H(c, s_i). NP-hard (Frances
& Litman 1997, Lanctot et al. 1999), with extensive FPT and PTAS
literature (Gramm & Niedermeier 2003, Ma & Sun 2009, Li/Ma/Wang 2002).
Distinct from ClosestSubstring — every input string has the same
length as the center, so there is no window-selection decision.
- src/models/misc/closest_string.rs: ClosestString { alphabet_size,
strings: Vec<Vec<usize>> }. Validating constructor panics on
length mismatch or out-of-alphabet symbol. dims = vec![q; m].
Min<i64> objective (always feasible — every config in the cube is
a syntactically valid center). Inherent getters alphabet_size,
num_strings, string_length, total_length. ProblemSchemaEntry +
ProblemSizeFieldEntry with all four size fields. declare_variants!
default with complexity alphabet_size^string_length. Canonical
example via inventory from the issue's 4-string binary length-3
instance (optimal center [0,0,0], radius 2).
- src/unit_tests/models/misc/closest_string.rs: 11 tests covering
creation, evaluate at three different centers (c=000 → 2, c=100
→ 3, c=111 → 3), brute-force solver returning radius 2 over 8
candidates, three panic guards (empty input, length mismatch,
out-of-alphabet symbol), a q=3 length-2 ternary smoke test
(radius 2 over 9 candidates), and serialization.
- problemreductions-cli/: schema-driven create wires --alphabet-size
(usize) and --strings (Vec<Vec<usize>>) — reuses the existing
Vec<Vec<usize>> parser added with #994.
- docs/paper: problem-def block with all four Hamming distances
spelled out for c=000; display-name entry; Crossref-verified
Li/Ma/Wang 2002 (JACM) bib entry.
Reference: doi:10.1145/506147.506150 (Li, Ma & Wang 2002, JACM).
The direct `ClosestString -> ILP` rule (#1034) is out of scope for
this PR and will follow separately.
Closes #1032
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Window-selection generalization of ClosestString (#1032): given an alphabet Σ, n strings (NOT necessarily equal length), and a substring length ℓ, find a center c ∈ Σ^ℓ and start positions p_i selecting length-ℓ windows s_i[p_i .. p_i+ℓ) minimizing max_i d_H(c, s_i[p_i .. p_i+ℓ)). Motif-discovery model — NP-hard, no PTAS in general (Li, Ma & Wang 2002 JACM; Marx 2008). ClosestString is the special case where every input string has length exactly ℓ (single window per string). - src/models/misc/closest_substring.rs: ClosestSubstring { alphabet_size, strings: Vec<Vec<usize>>, substring_length }. Validating constructor panics on empty input, substring_length > min |s_i|, or out-of-alphabet symbol. dims concatenates ℓ center slots (domain {0..q-1}) with n window-start slots (domain {0..W_i-1} where W_i = |s_i| - ℓ + 1). Min<i64> objective, always feasible since every config in the cube is syntactically valid. Inherent getters alphabet_size, num_strings, substring_length, total_length, total_num_windows, num_window_choice_product (with saturating multiplication). ProblemSchemaEntry + ProblemSizeFieldEntry. declare_variants! default with complexity alphabet_size^substring_length * num_window_choice_product. Canonical example via inventory from the issue's 3 binary strings with ℓ=3 — optimum center [0,1,0] with window picks (0,1,0), radius 1 over 216 candidate configs. - src/unit_tests/models/misc/closest_substring.rs: 11 tests covering creation, evaluate at optimum (radius 1), evaluate at center [0,0,0] with all-zero windows (radius 2), evaluate at center [1,1,1] (radius >=1), brute-force solver, ClosestString reduction validation (substring_length = string_length → matches the #1032 canonical's radius 2), three panic guards (empty input, length mismatch, out-of-alphabet symbol), and serialization roundtrip. - problemreductions-cli/: schema-driven create wires --alphabet-size + --strings (reused from #1032) plus the new --substring-length (usize) flag. - docs/paper: problem-def block with the worked example listing all three window picks and per-window Hamming distances; display-name entry. Reuses the existing Li/Ma/Wang 2002 JACM BibTeX entry added with #1032 — no references.bib changes. Reference: doi:10.1145/506147.506150 (Li, Ma & Wang 2002, JACM) shared with #1032. The direct `ClosestSubstring -> ILP` rule (#1035) is out of scope for this PR and will follow separately. Closes #1033 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Classical protein-structure contact-map alignment: given two ordered
contact graphs G_1=(V_1,E_1) and G_2=(V_2,E_2), find an order-preserving
partial injective map f: V_1 → V_2 ∪ {unmatched} maximizing the number
of contacts {i,k} ∈ E_1 such that both i, k are matched and
{f(i), f(k)} ∈ E_2. Aliases: CMO, MaxCMO.
NP-hard with substantial literature on exact algorithms and integer
programming (Andonov, Malod-Dognin & Yanev 2011; Xie & Sahinidis 2007).
- src/models/graph/maximum_contact_map_overlap.rs:
MaximumContactMapOverlap { num_vertices_1, contacts_1, num_vertices_2,
contacts_2 }. Validating constructor normalizes each pair to sorted
form (u<v), rejects self-loops, duplicates, and out-of-range
endpoints. dims = vec![num_vertices_2 + 1; num_vertices_1] (value 0
encodes unmatched; value j+1 maps to vertex j of G_2). Max<i64>
objective; non-injective or non-order-preserving matched values →
Max(None). ProblemSchemaEntry + ProblemSizeFieldEntry; inherent
getters num_vertices_1/_2 and num_contacts_1/_2. declare_variants!
default with complexity (num_vertices_2+1)^num_vertices_1.
Canonical example via inventory: G_1 with 4 vertices and contacts
{(0,2),(1,3)}, G_2 with 5 vertices and contacts {(0,2),(0,3),(1,4)}
— optimum [1,2,4,5] preserves both contacts → Max(Some(2)).
- src/unit_tests/models/graph/maximum_contact_map_overlap.rs: 17 tests
covering creation, evaluate at optimum, all-unmatched, single-match,
non-injective Max(None), non-order-preserving Max(None), suboptimal
feasible (config [1,2,3,4] preserves 1 of 2 contacts), brute-force
solver returning Max(2), wrong-length and out-of-range guards,
serialization, alias resolution for CMO/MaxCMO, and three panic
guards (self-loop, duplicate contact, endpoint out of range).
- problemreductions-cli/: schema-driven create wires --num-vertices-1
/ --num-vertices-2 / --contacts-1 / --contacts-2 (Vec<(usize,usize)>
parser) via the existing CreateArgs + flag_map + tests fixture.
- docs/paper: problem-def block with the alignment table and the two
preserved-contact bullets; display-name; Crossref-verified BibTeX
for both Andonov-Malod-Dognin-Yanev 2011 and Xie-Sahinidis 2007
JCB papers (with N{\"o}el encoded per repo umlaut convention).
References: doi:10.1089/cmb.2009.0196 (Andonov, Malod-Dognin & Yanev
2011, JCB); doi:10.1089/cmb.2007.R007 (Xie & Sahinidis 2007, JCB).
The direct `MaximumContactMapOverlap -> ILP` rule (#1044) is out of
scope for this PR and will follow separately.
Closes #1043
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
isPANN
changed the title
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## main #1067 +/- ##
==========================================
+ Coverage 97.93% 97.98% +0.05%
==========================================
Files 966 1026 +60
Lines 100154 105325 +5171
==========================================
+ Hits 98083 103200 +5117
- Misses 2071 2125 +54 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Net -83 lines across the four skill files touched by 694ef0c. auto-pipeline (-79): - Collapse Step 0a + 0b into a single picker block; the only difference was "filter by number" vs "sort and pick top" — now branched by whether the ISSUE env var is set. - Extract the boilerplate that was duplicated across all five subagent prompts (output-only-JSON-block contract, universal don'ts, malformed-JSON retry policy, severity vocabulary) into a single "Subagent Contract" section near the top. Each prompt now states only its scope and JSON shape. Drop the trailing "Reporting Contract" section (merged into the new one). - Trim the Board states table from 8 GraphQL IDs to the 3 columns the orchestrator actually writes (ready, on-hold, plus the Backlog it reads in Step 0). The IDs for In Progress / Review pool / Under review / Final review live in run-pipeline / review-pipeline where they are used. - Drop three rows from Common Mistakes that just echoed the spec (codex retry cap, increment SUBSTANTIVE_RETRIES, re-check after auto-fix); keep only the non-obvious cross-cutting traps. check-issue (-9 net): - Rule Check 5a no longer re-lists the literature fallback chain; one-line reference to Check 3c suffices. - Rule Check 5c verdict table drops the (severity: ...) annotations on Fail rows — severity classification is owned by auto-pipeline's Subagent Contract, not by check-issue itself. - Rule Check 5c drops the "cited reference does not contain the reduction" row that explicitly admitted overlap with Check 3c; add a one-line note that 3c handles that case. review-structural (-19): - Step 4b-4 (pred --via spot-check) was hedged out of its own purpose with a "fall back to 4b-2" escape hatch. Rewrite as a short focused step that uses pred --via when wired and is skipped (with a note) otherwise — no padding. review-quality (~2 net): - Replace the 4-criterion expansion of the round-trip rule (which was copy-pasted from review-structural Step 4b-3) with a single pointer to that source-of-truth section. Per the existing feedback_skill_no_duplication memory. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- MinimumDiscretePlanarInverseKinematics: simplify complexity string from "num_links * total_configurations" to "total_configurations" so it matches issue #994's stated O(prod_{j=1}^n m_j) baseline literally (the extra num_links factor was per-config feasibility-check work, not configs). - MinimumCostCirculation: add test_minimum_cost_circulation_issue_example_1030 that constructs issue #1030's verbatim 2-vertex example (arcs 0->1 cap=2 cost=3 and 1->0 cap=1 cost=-5, optimum -2). The existing richer 3-vertex canonical instance is kept as the primary discriminator.
Adds Sahni's multiplier-flow gadget: each Partition element becomes an item vertex whose multiplier amplifies a binary source choice into either 0 or a_i units entering a relay. A single bottleneck arc of capacity S/2 converts the target's "net inflow at least R" condition into the exact equality needed by Partition. Odd-S inputs reduce to a fixed infeasible 3-vertex instance. - src/rules/partition_integralflowwithmultipliers.rs: reduction impl with odd-S/even-S branches and witness extraction from source arcs - src/unit_tests/rules/partition_integralflowwithmultipliers.rs: 5 tests (closed-loop, structure on even-total YES, even-total NO exercises bottleneck, odd-total fixed NO, witness extraction) - src/rules/mod.rs: register module and example specs - docs/paper/reductions.typ: full theorem with construction, correctness proof, and worked example using the canonical fixture Closes #363. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Plaisted (1976) reduction from Decision Minimum Vertex Cover to
Comparative Containment (Garey & Johnson SP10). Given a unit-weight VC
instance (G = (V, E), K), the universe X = V is encoded as:
- For each vertex v, a reward set R_v = V \\ {v} with weight 1, so the
total R-weight equals n - |Y|.
- For each edge e = {u, v}, a penalty set S_e = V \\ {u, v} with weight
n + 1 that dominates the maximum possible reward whenever the edge is
uncovered.
- One budget set S_0 = V with weight n - K, which encodes the bound
through the resulting inequality K - |Y| >= (n + 1) * (#uncovered).
Source assertions mirror decisionminimumvertexcover_hamiltoniancircuit.rs:
unit weights are required, a negative bound emits a fixed unsatisfiable
target, and K >= n is handled as a trivial-YES instance (empty universe,
no R/S sets). Adds a ProblemSizeFieldEntry for ComparativeContainment
since it had no declared size_fields, and a paper theorem entry. Eight
unit tests cover structure counts, closed-loop YES/NO, witness
extraction, both trivial-YES branches, the trivial-NO branch, and the
weight-assertion guard.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Implements Salloum et al. 2025 quantum-annealing IK encoding: one-hot binary lifting of sampled orientations, quadratic position-error term, one-hot exactly-one penalty, and pair-feasibility penalty for forbidden adjacent (a,b) pairs. Penalty constants are chosen above the maximum possible position-error savings, so every QUBO minimizer decodes to a feasible source configuration. Also adds num_orientation_samples() getter on the source model. Closes #995. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…nTime reduction (#472) Implements Lawler 1978 precedence-constrained 1||sum w_j C_j encoding: each vertex becomes a unit-length task with weight 1, each edge becomes a zero-length task with weight 1, and the precedence constraints enforce the OLA ordering. The minimum weighted completion time recovers the OLA objective plus an additive shift d_max*n*(n+1)/2. Also relaxes the target model's validator to allow zero-length edge jobs needed by Lawler's construction. Closes #472. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
) Direct partition-matroid embedding: a 3DM matching is precisely the common independent set of three partition matroids, one over each coordinate axis (X, Y, Z). The reduction copies the triple set, builds the three partition matroids by grouping triples that share each coordinate, and uses identity solution extraction. Closes #857. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…on (#848) Identity instance mapping: a clique cover {C_1, ..., C_k} of G is a valid intersection-graph basis with the same cardinality k, since each vertex's labels are exactly the cliques it belongs to and each edge is witnessed by a shared label. The witness extraction labels each edge by a shared intersection slot. Closes #848. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…reduction (#827) Standard G&J transformation: a Numerical 3DM instance with triples (X, Y, Z, bound) maps to an NMTS instance where the X and Y triples form the pair sizes and each Z element becomes a target sum equal to bound + z. The witness is recovered by multiset matching against the source's third coordinate. Source u64 sizes are checked-cast to i64 with a documented overflow guard. Closes #827. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Direct boolean ILP encoding for the Co-k-Plex problem: each vertex gets
a binary variable, the objective maximizes the weighted sum, and a
per-vertex constraint bounds the induced degree by k - 1 using the
linear inequality sum_{u in N(v)} y_u <= (k - 1) + M * (1 - y_v).
Closes #1016.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Boolean ILP encoding finds a one-to-one vertex correspondence between G1 and G2 that maximizes the number of preserved edges. Variables x[i,k] encode whether v_i in G1 maps to v_k in G2, with row/column sum-at-most-1 constraints enforcing a partial injection, and edge-preservation variables y[(i,j),(k,l)] linearized via the standard product-with-binaries pattern. Closes #1019. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Boolean ILP encoding with vertex selectors x_v, edge selectors y_uv via McCormick linearization, an exact-cardinality constraint sum x_v = k, and non-edge clique constraints x_u + x_v <= 1 for every non-edge. The lower-bound McCormick inequality is essential because edge weights may be negative. Closes #1021. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Set-partitioning ILP encoding: each candidate cluster is a feasible highly-connected vertex subset (verified via edge-connectivity > n/2), and a partition constraint forces each vertex to belong to exactly one selected cluster. The objective minimizes the number of deleted edges (those whose endpoints are in different clusters). Also exposes pub(crate) helpers is_feasible_cluster and induced_edge_count on the source model. Closes #1023. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
MTZ-style ILP encoding for EulerianPath: integer variables y_{a,b}
encode compatible arc-pair successions, plus per-arc start/end/position
variables. Constraints enforce in-degree/out-degree balance for the
Eulerian trail, unique start/end arcs, and a position-strict ordering
that prevents subtours. The empty-arc source instance maps to the
empty ILP.
Closes #1025.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Textbook reduction: keep all original arcs unchanged, append a return
arc (t, s) with capacity U = sum_{e in delta^+(s)} u_e and cost
-(1 + sum_e c_e). The deeply negative return-arc cost makes maximizing
flow on (t, s) the priority — i.e. the circulation must saturate
s -> t flow before optimizing residual cost. Solution extraction
discards the return arc.
Closes #1031.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Replace serde_json::json! comparisons with plain Vec<usize> values to match the SolutionPair.source_config/target_config field types. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Position-character ILP encoding for ClosestString: binary variables
x_{j,a} encode the consensus string character at position j (with
one-hot assignment constraints), and a single integer radius variable
R is bounded by per-string Hamming-distance constraints. The objective
minimizes R.
Closes #1034.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Position-character ILP encoding for ClosestSubstring: binary variables
x_{r,a} for the center substring's character at each position, window
choice indicators y_{i,p} (exactly one window per source string), and
an integer radius R bounded by per-window Hamming-distance constraints
plus a tight R <= ell upper-bound constraint critical for ILP solver
performance.
Closes #1035.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Replace `6 + 0` and `6 + 6 + 0` with `6` and `6 + 6`. Surfaced during structural review of #363 commit (the lint blocks `cargo clippy --all-targets -- -D warnings`). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The relaxed-validator test asserted Min(14) but the actual weighted completion time for lengths [0,1,3], weights [3,5,1], schedule [0,1,2] is 3*0 + 5*1 + 1*4 = 9. Surfaced during structural review of #472. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…ersection (#857 review) Surfaced during structural review: rule was missing canonical_rule_example_specs() and was not extended into the rule example aggregator in mod.rs. Adds the q=3 5-triple feasible instance from the closed-loop test as the canonical example. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Surfaced during structural review of #1021: - MaximumEdgeWeightedKClique has both i32 and f64 ReduceTo impls but only an i32 canonical example was registered; the example-db coverage test failed. Add an f64 spec mirroring the i32 instance. - MaximumCoKPlex has both i32 and One ReduceTo impls but only an i32 canonical example existed (also flagged by the coverage test). Add a One-weighted spec on the same C5 graph. - Paper cited `@ParkLeePark1996EWClique` and `@GouveiaMartins2015EWClique` but neither bib key existed. Add both entries to references.bib while keeping the original `@GouveiaMartins2015MEWC` entry that the problem-def section still cites. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
isPANN
changed the title
PR #472 (commit f13bbb3) relaxed SequencingToMinimizeWeightedCompletionTime's validator to accept zero-length tasks for the Lawler OLA reduction. Two CLI tests still asserted the removed "task lengths must be positive" rejection, so CI broke on the workspace test job. The new accept-zero-length semantics are positively covered by unit tests in src/unit_tests/models/misc/sequencing_to_minimize_weighted_completion_time.rs. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Three mechanical paper-build fixes: - Replace math-mode `intersect` with Typst's `inter` (set-intersection) in the MinimumCoveringByCliques → MinimumIntersectionGraphBasis proof (5 occurrences on lines 18087-18091). - Drop orphan citation @deGastinesKnippel2024MCES from MaximumCommonEdgeSubgraph → ILP (no matching bib entry; the rule still cites the valid @Bahiense2012MCES McCormick formulation). - Fix typo @lawler1978a → @lawler1978 on the OLA → SequencingToMinimizeWeightedCompletionTime rule. The first CI run was masked by the cli_tests failure (set -e); now that tests pass, the paper compile step exposed these issues. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
PR #1067 hit two CI failures the per-item phases could not see: - Two CLI integration tests asserted a model-validator behaviour that the new OLA -> SequencingToMinimizeWeightedCompletionTime reduction intentionally relaxed; closed-loop and unit tests for the new rule all passed. - `make paper` blew up on `intersect` (Typst expected `inter`), an orphan bib key, and a typo'd key; no phase ran the Typst compile. Adds an orchestrator-owned Step 2.5 that runs `cargo test --workspace --features "ilp-highs example-db"` and `make paper` on PR HEAD between Phase 2 (run-pipeline) and Phase 3 (review-pipeline). Any failure parks the card on OnHold for human triage — codex rescue is not appropriate because the failing artefact lives outside the issue's files. In batch mode (multiple issues stacked on one branch with the PR opened at the end), this gate is the only thing that catches accumulated cross-item breakage before review. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
CI-class failures (stale tests, typo'd bib keys, math-mode typos) are small and mechanical; hand them straight to codex-rescue with a one-line failure summary rather than walking the orchestrator through a long JSON contract, PR-comment template, and explicit OnHold dance. Re-run Step 2.5 once after codex; OnHold only if still failing. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Bumps review-pipeline to Step 4. Diagram, intro paragraph, cross-step references, and Common Mistakes table updated accordingly. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Step 4b mixed three different concerns into review-structural (a read-only structural skill): - 4b-2 ran `cargo test --exact <closed_loop>` for the new rule — redundant with auto-pipeline Step 3's workspace-wide `make check`. - 4b-3 read the test source to score it against four criteria — that's a test-quality judgment, not a structural check; review-quality is the right home for that and it already covers test quality. - 4b-4 ran `pred solve` end-to-end via the new rule — overlaps with the agentic feature tests review-pipeline already runs. - 4b-1 (grep for the closed_loop test) is subsumed by Step 4's existing checklist of required rule artefacts. Restoring review-structural to its main version. Also removing the Phase 4 prompt's mandatory-Step-4b clause from auto-pipeline. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The PR's edit added a Critical-severity rule for `[Rule]` PRs that deferred to review-structural Step 4b-3 — now a dead link since that section was reverted. Main's original bullet already captures the intent: verify the extracted solution is optimal via brute-force on both source and target. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Removes the codex:codex-rescue dependency from both substantive-rewrite (Step 1c-sub) and integration-gate-failure (Step 3) paths. Both now dispatch a general-purpose subagent with the same JSON contract. Wins: - Step 1c-sub prompt drops from ~45 lines of nested codex-exec / <<<BODY>>> / FUNDAMENTAL_FLAW sentinel escaping to a direct subagent prompt with the issue body and check report inlined. - Phase 3 failure path is consistent with the rest of the skill — one subagent, one JSON return, orchestrator owns side effects. No behaviour change to Step 2's stop-on-failure stance (run-pipeline still moves the card to OnHold itself; orchestrator just halts). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The skill is scoped to one issue at a time; batching is an external concern. Step 3's rationale only needs to justify why catching workspace-wide breakage locally beats waiting for CI. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Adds `load-example` + `example: true` + `example-caption` + `extra:` (pred-commands + witness summary) to the paper entries that previously shipped without a rendered example, matching the existing `mvc_mis` pattern. Brings the 13 outstanding rules in PR #1067 in line with the already-rendered #363, #848, and #995. Rules updated: - DecisionMinimumVertexCover -> ComparativeContainment (#385) - OptimalLinearArrangement -> SequencingToMinimizeWeightedCompletionTime (#472) - Numerical3DimensionalMatching -> NumericalMatchingWithTargetSums (#827) - MinimumCoveringByCliques -> MinimumIntersectionGraphBasis (#848 - was using older inline style) - ThreeDimensionalMatching -> ThreeMatroidIntersection (#857, resolves sibling-inconsistency) - MaximumCoKPlex -> ILP (#1016, i32 variant) - MaximumCommonEdgeSubgraph -> ILP (#1019) - MaximumEdgeWeightedKClique -> ILP (#1021, i32 variant) - HighlyConnectedDeletion -> ILP (#1023) - EulerianPath -> ILP (#1025) - MinimumCostMaximumFlow -> MinimumCostCirculation (#1031) - ClosestString -> ILP (#1034) - ClosestSubstring -> ILP (#1035) Verified: `make paper` compiles cleanly. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Classical Yannakakis-Gavril (1980) reduction establishing NP-completeness of Achromatic Number (G&J GT5): for bipartite G, ach(complement(G)) = |V| - mm(G). Adds BipartiteGraph variant of MinimumMaximalMatching as a prerequisite, the complement-graph reduction with the inverse coloring -> maximal-matching extractor, a closed-loop test plus identity tests on several bipartite instances, the canonical P4 example (acknowledged check-issue warning that the P4 canonical example has only one suboptimal maximal matching; kept for tutorial clarity), and the matching paper entry under docs/paper/reductions.typ. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Address three review nits on commit 1826867: 1. Canonical example richness: swap the path P4 canonical example for a "T-tree" on 5 vertices (spider v0-v1-v2-v3 with extra leaf v1-v4), which exposes two strictly suboptimal maximal matchings besides the minimum (mm = 1 vs. two size-2 maximal matchings), comfortably passing the >=2-suboptimals rule-of-thumb. Update the canonical builder, the paper worked example, and the closed-loop test to match. 2. extract_solution: replace the per-call HashMap rebuild with a single pass over source_edges that checks whether each edge's endpoints share a color. For bipartite G all color classes have size <= 2, so this is equivalent and avoids any auxiliary allocation. 3. Unit test imports: drop the catch-all "use super::*" in favour of an explicit "use" list mirroring the MVC->MIS reference test. cargo test rules::minimummaximalmatching_maximumachromaticnumber: 5 passed (closed-loop, complement structure, known coloring, suboptimal recovery, identity); make paper builds cleanly. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Add `q()` helper on ExactCoverBy3Sets and use it in the rule and tests instead of recomputing `universe_size / 3`. - Add a `debug_assert!` in `reduce_to` confirming that root-to-set edges occupy indices 2..2+m, so a future reordering will surface a failure in `extract_solution`. - Simplify the `no_instance` test: drop the duplicate assertion and the `unwrap_or_default` fallback; assert directly that `find_witness` is `None` and that the brute-force aggregate is `Or(false)` for an infeasible Or-valued target. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
isPANN
changed the title
isPANN
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Serial batch-add of 12 Backlog
[Model]issues + 18 Backlog[Rule]issues on a single branch (batch-add-models), driven by the new/auto-pipelineorchestrator. Each commit: full Rust impl + unit tests + schema-driven CLI + Typst entry + Crossref-verified BibTeX.Closes
Merging this PR closes the following 30 issues (12 models + 18 rules):
Closes #363, closes #385, closes #472, closes #827, closes #846, closes #847, closes #848, closes #857, closes #913, closes #994, closes #995, closes #1015, closes #1016, closes #1018, closes #1019, closes #1020, closes #1021, closes #1022, closes #1023, closes #1024, closes #1025, closes #1026, closes #1029, closes #1030, closes #1031, closes #1032, closes #1033, closes #1034, closes #1035, closes #1043.
Models added (12)
MinimumDiscretePlanarInverseKinematicsMaximumCoKPlexMaximumCommonEdgeSubgraphMaximumEdgeWeightedKCliqueHighlyConnectedDeletionEulerianPathPrizeCollectingSteinerForestMinimumCostMaximumFlowM*(max-flow)+costencoding; integral-flow carve-out (mirrors MECF)MinimumCostCirculationClosestStringq^mbrute-forceClosestSubstring[q;ℓ]++[W_i]MaximumContactMapOverlapRules added (18)
Partition → IntegralFlowWithMultipliersMinimumVertexCover → ComparativeContainmentOptimalLinearArrangement → SequencingToMinimizeWeightedCompletionTimeNumerical3DimensionalMatching → NumericalMatchingWithTargetSumsMinimumMaximalMatching → MaximumAchromaticNumberMinimumMaximalMatching → MinimumMatrixDominationMinimumCoveringByCliques → MinimumIntersectionGraphBasisThreeDimensionalMatching → ThreeMatroidIntersectionExactCoverBy3Sets → BoundedDiameterSpanningTreeMinimumDiscretePlanarInverseKinematics → QUBOPMaximumCoKPlex → ILPMaximumCommonEdgeSubgraph → ILPMaximumEdgeWeightedKClique → ILPHighlyConnectedDeletion → ILPis_feasible_cluster/induced_edge_countEulerianPath → ILPMinimumCostMaximumFlow → MinimumCostCirculation(t,s)with capacityU, cost-(1+Σc_e)ClosestString → ILPClosestSubstring → ILPR ≤ ℓupper bound for HiGHS perfStructural review pass (post-impl)
Each new rule commit was re-reviewed via
/review-structural. Most passed clean; surfaced regressions (all fixed in follow-up commits):test_..._zero_length_taskassertedMin(14)(actualMin(9)on schedule[0,1,2])canonical_rule_example_specs(); not extended intorules/mod.rsaggregatorcanonical_rule_examples_cover_exactly_authored_direct_reductions); bib keys@ParkLeePark1996EWClique/@GouveiaMartins2015EWCliquenot inreferences.bib. Same review also caught #1016 missing One-weight example.identity_oplints (6 + 0,6 + 6 + 0) blockedclippy --all-targets -D warningsVec<usize>againstserde_json::Value(compile error under--features example-db)use super::*extract_solutionbrute-forced overu64mask (silent overflow >64 edges); paper Step 4 omitted EDS→IEDS transform caveatno_instancetest duplicated assertion +unwrap_or_defaultpanic;extract_solutionlayout-coupled with no debug_assert;q = universe_size/3recomputedOnHold
Total Backlog rules triaged in this session: ~70. Of those:
DecisionMinimumVertexCover(register_decision_variant!insrc/models/graph/minimum_vertex_cover.rs:185); also wrong for the actuali32-weighted source variant. Framing-level flaws, not rewritable.#1046
MaximumAcyclicAgreementForest(from foundation): substantive errors (wrong worked-example outcome, wrong cited complexity3.18^kvs Whidden-Beiko-Zeh's3^k); codex rescue failed in sandbox. Diagnostic at #1046 (comment).Final-review fixes (commit e346156)
declare_variants!complexity fromnum_links * total_configurationstototal_configurationsso it matches the issue's literalO(prod_{j=1}^n m_j)baseline.test_minimum_cost_circulation_issue_example_1030which constructs the issue's verbatim 2-vertex example (arcs0→1cap=2 cost=3 and1→0cap=1 cost=-5, optimum -2). Existing richer 3-vertex canonical kept as the primary discriminator.Test plan
make fmtcleanmake clippyclean (incl.--all-targets --all-features -D warnings)make testgreen — full workspace passes on the final commit (5201+ lib tests after [Rule] MINIMUM MAXIMAL MATCHING to MinimumMatrixDomination #847 nit fix)make paperbuilds the Typst PDF cleanly (re-verified after each of the 3 new rules + nit fixes; no warnings)pred create --example <Model> | pred solve --solver brute-forcefor each of the 12 new models — 11/12 work via bare name;MaximumCoKPlexrequires explicit/i32disambiguation (same pattern as existingMaximumClique)pred path Source Target --json+ via-reduction solve for each of the 18 new rules — all agree with source-direct brute force / ILP🤖 Generated with Claude Code