Repository navigation
Speculative Lattices
Speculative. A proposal, not adopted.
IsLattice and FindStat's Lattices statistics on the Poset carrier: phase 4 of Posets, after #1354 (the carrier and its first statistics). Lattices(n), the collection, is not here; it restricts Posets(n) and waits for it.
"Measured" means I ran it in Python (stdlib) against the crawl (~/Playground/@enumeratio/findstat-crawl) or against posets I generated. "Assumed" means reasoned, not run. Nothing here has been compiled or interpreted through compute-engine; no Epsil was run.
- The crawl has 34
Latticesstatistics. I recommend 23 (18 that read only the order, joins and meets, 5 that enumerate subsets as a bitmask), leave out 11 (linear algebra, congruences and tolerances, sublattice families, isomorphism, algebra with no code). - Every recommended statistic agrees with every crawl row the crawl holds for it (up to n = 668 for the four with large lattices) under an independent reading, and with every row up to a stated n cap under a second, Epsil-shaped reading that uses only lower covers, the flat strict order and flat join and meet tables. One statistic's title disagrees with its rows:
St000550(below). -
IsLatticeis: no elements, or exactly one minimal element and every pair has a join. Measured equal to "every pair has a join and a meet" on all 96,428 naturally labeled posets through n = 7. A lattice's bottom is element 1 and its top element n, which every definition uses. - The snapshot as #1354 sets it (Lattices through 6 elements) is not enough as wired: the sweep's standard cap is per carrier (5), so the standard run compares only the 10 lattices with at most 5 elements, and
St001719is constant (1) on them andSt000550andSt000551have identical values. Through 6 (25 lattices) fixes both; through 7 (78 lattices) is the margin. See "The snapshot".
Definition. IsLattice(P), for a Poset, is true when P has no elements, or when P has exactly one minimal element and every pair of elements has a join. A pair has a join when it has a least upper bound. In a natural labeling the only candidate is the least-labeled common upper bound m, and the pair has a join exactly when m is below every common upper bound.
Why joins and a bottom, not both sides. In a finite poset with a bottom, every pair has a join exactly when every pair has a meet: the meet of a and b is the join of their common lower bounds, a non-empty set because it holds the bottom. So the one-sided test is complete once the bottom is checked, and costs half the work of the two-sided one. Joins alone are not enough: a poset with a top and every pair joined need not have meets (the V turned upside down). Measured over naturally labeled posets with n = 1..7 elements (1, 2, 7, 40, 357, 4,824, 96,428; counts confirmed):
| n | posets | joins and meets | one minimal and joins | joins only |
|---|---|---|---|---|
| 4 | 40 | 2 | 2 | 7 |
| 5 | 357 | 7 | 7 | 39 |
| 6 | 4,824 | 39 | 39 | 320 |
| 7 | 96,428 | 320 | 320 | 3,637 |
The two correct definitions agree on every poset; "joins only" overcounts. Counted up to isomorphism (least relabeling code), the lattices are 1, 1, 2, 5, 15, 53 for n = 2..7, the known counts, and all 25,183 crawl rows (all 34 statistics) are lattices: none fails the test.
The empty poset. IsLattice(Poset([])) is true: nothing is unjoined. The crawl lists the empty lattice (([],0)) for St001754 and St001990, so FindStat counts it. The guard answers it before the lattice test, as withPoset does.
Cost. One pass for the minimal element (each entry past the first has a lower cover), then for each of n² pairs the common upper bounds (n reads of below) and a test that the least one is below the rest (n more): about 2n³ reads, 1,024 at n = 8. It needs below, so it sits inside withRelation. Measured on the Epsil-shaped reading: agrees with the bitmask definition on every naturally labeled poset through n = 6 (4,824 at n = 6, 0 disagreements).
Where it sits.
-
IsLatticeis a predicate head onposetin theposets/area (Is…). The statistics machinery'sDefinitionhas an optionalresulttype for answers that are not numbers; whether it carries a boolean, and how an area declares a predicate that is not a statistic, is not checked (open question 1). - A guard helper beside
withPoset:withLattice(body, otherwise = 0)iswithPosetofIf(IsLattice(_x), body, <declines>). The empty poset answersotherwisefirst. How a definition declines (leaves the call unevaluated, notNaNand not 0) was probed first and is open (question 1): a definition's own else branch cannot do it. - Two further helpers bind flat n² tables once, as
withRelationbindsbelow:withJoins(joins, entry(a − 1)·n + bis the least element above both: the leastzwitha ≤ zandb ≤ z) andwithMeets(meets: the greatestzbelow both). No sort is needed: the least label among upper bounds is the join in a natural labeling, andMinoverIf(cond, z, n + 1)picks it without a filter. Cost n³ each. They are valid only after the guard.
The crosscheck. findstat/src/objects.ts already has "Lattices": "Poset" in COLLECTION_CARRIER (#1354), so the reader reads a lattice's spelling as a Poset and the statistic runs on it. Every crawl object is a lattice (measured), so no statistic is evaluated off its domain and skipped stays []; there is no row for a non-lattice to disagree with. A lattice statistic's findstat-data.ts row is { head, on: "Poset", findstat: [id], values }. One consequence of keying by carrier: see "The snapshot".
Rows are the crawl's; "complete" means every lattice of that size is present (1, 1, 1, 1, 2, 5, 15, 53, 222 for n = 0..8, measured as the union of spellings over all 34 statistics, equal to the known counts). n = 9 has 905 distinct spellings of 1,078 lattices, so it is partial everywhere.
| Id | Title (short) | Sage | Rows | Complete through | Bucket |
|---|---|---|---|---|---|
| 550 | modular elements | yes | 78 | 7 | a |
| 551 | left modular elements | yes | 656 | 7 (+ some to 499) | a |
| 1613 | log2 of the center size | yes | 1,200 | 8 | a |
| 1615 | join-prime elements | yes | 1,200 | 8 | a |
| 1616 | neutral elements | yes | 1,200 | 8 | a |
| 1617 | dimension of the space of valuations | yes | 1,200 | 8 | d (linear algebra) |
| 1618 | size of the Frattini sublattice | yes | 1,200 | 8 | c (family of sublattices) |
| 1619 | non-isomorphic sublattices | yes | 1,200 | 8 | e (isomorphism) |
| 1620 | sublattices | yes | 1,200 | 8 | c |
| 1621 | atoms | yes | 78 | 7 | a |
| 1622 | join-irreducibles | yes | 554 | 7 (+ some to 64) | a |
| 1623 | doubly irreducibles | yes | 78 | 7 | a |
| 1624 | breadth | yes | 78 | 7 | c (bitmask over join subsets) |
| 1625 | Möbius invariant | yes | 78 | 7 | b |
| 1626 | maximal proper sublattices | yes | 78 | 7 | c (family of sublattices) |
| 1630 | global dimension of the incidence algebra | no | 76 | 3..7 | d |
| 1651 | Frankl number | yes | 586 | 2..7 (+ some to 64) | a |
| 1666 | non-isomorphic subposets that are lattices | yes | 1,200 | 8 | e |
| 1677 | non-degenerate subsets with meet 0 | yes | 1,200 | 8 | c |
| 1679 | subsets with meet 0 | yes | 1,200 | 8 | c |
| 1681 | minimal subsets with meet 0 | yes | 1,200 | 8 | c |
| 1719 | shortest chains of small intervals | yes | 1,200 | 8 | a (DP fold) |
| 1720 | minimal length of a chain of small intervals | yes | 2,050 | 8 (+ some to 668) | a |
| 1754 | tolerances | no | 26 | n = 0..6 | c (bits over pairs) |
| 1820 | image of the pop stack operator | yes | 1,200 | 8 | a |
| 1833 | linear intervals | yes | 1,200 | 8 | a |
| 1845 | join-irreducibles minus rank | yes | 1,200 | 8 | a |
| 1846 | elements without a complement | yes | 1,200 | 8 | a |
| 1875 | simple modules, projective dimension at most 1 | no | 76 | 3..7 | d |
| 1876 | 2-regular simple modules | no | 60 | none (3..4) | d |
| 1877 | injective modules, projective dimension 2 | no | 76 | 3..7 | d |
| 1878 | projective dimension of the minimum's simple | no | 76 | 3..7 | d |
| 1881 | factors as a Cartesian product | yes | 1,200 | 8 | a (center) |
| 1990 | zero divisors | yes | 79 | 0..7 | a |
Bucket counts: (a) 17, (b) 1, (c) 8, (d) 6, (e) 2 = 34. The design page counted 18 meet/join-local, 1 Möbius, 8 sublattice enumeration, 5 algebraic and 2 isomorphism. Differences: breadth (St001624) needs a bitmask, so it is (c) here; St001881 moves from sublattice enumeration to (a), because its rows equal max(1, St001613) on all 1,200 (measured) and a direct product's factors are the central elements; St001617 (linear algebra) joins the algebraic six and St001754 (subsets of pairs) the enumerations.
Reach of a bitmask over elements at n ≤ 8 (the standard run reaches 5..7, the deep run 7):
-
1624breadth,1677,1679,1681,1620: yes. One mask per subset (256 at n = 8, 512 at n = 9), a meet or join fold of n steps for the first four, a closure test over pairs for1620. Modeled at 100% agreement in Python. -
1618Frattini and1626maximal proper sublattices: modeled at 100% through n = 9 (1,200 and 78 rows) but they need the family of sublattice masks and a maximality test of every one against all others, a mask-subset test with no bit operations. Reachable at n ≤ 7 and clumsy; left out. -
1754tolerances: the unknowns are subsets of pairs, 2^15 at n = 6 and 2^28 at n = 8. Modeled at 100% for n ≤ 6 (26 rows), the only rows there are; not reachable at n = 8 and no row beyond 6. Left out. -
1619,1666: need to decide isomorphism of the sublattices or subposets. Left out untilIsIsomorphicPosetexists. -
1617: nullity of a rational matrix (Birkhoff: the number of projective-prime-interval classes of the modular quotient). Modeled at 100% by exact Gaussian elimination over Q with v(bottom) = 0 (n ≤ 9). No row reduction in the core. Left out. -
1630,1875..1878: homological algebra of the incidence algebra; no Sage code. Left out.
All values are on the lattices: 1-chain, 2-chain, 3-chain, B2 ([[],[1],[1],[2,3]], the diamond), N5 (the pentagon [[],[1],[1],[3],[2,4]]), M3 ([[],[1],[1],[1],[2,3,4]]). Agreement is the Python reading against all crawl rows, under the independent reading (readings.py) and the Epsil-shaped one (fast.py, with its n cap). "Needs" lists what the Epsil definition reads beyond _x.
Head names are checked with git ls-files | grep '/reference/<Name>/index.md' on origin/main and against the reference dirs #1354 adds; "free" means neither has it.
| Id | Head | Reading that matches the rows | Needs | 1-ch, 2-ch, 3-ch, B2, N5, M3 | Rows agree (readings / Epsil-shaped) |
|---|---|---|---|---|---|
| 550 |
ModularElementCount (free) |
elements a with a ∧ (y ∨ z) = (a ∧ y) ∨ z for every y and every z ≤ a. Not the title's "(x, y) and (y, x) are modular pairs" |
joins, meets, below
|
1, 2, 3, 4, 4, 5 | 78/78 (all, n ≤ 7) / 78/78 |
| 551 |
LeftModularElementCount (free) |
a with (y ∨ a) ∧ z = y ∨ (a ∧ z) for every y ≤ z
|
joins, meets, below
|
1, 2, 3, 4, 4, 5 | 656/656 (n ≤ 499) / 532/532 (n ≤ 20) |
| 1613 |
LatticeCenterRank (free; CenterSize is taken) |
log2 of the center, the center being the neutral elements that have a complement; as the central elements that cover no other non-bottom central element |
joins, meets; neutral test below |
0, 1, 1, 2, 1, 1 | 1,200/1,200 (n ≤ 9) / 1,200/1,200 |
| 1615 |
JoinPrimeCount (free) |
non-bottom a with a ≤ p ∨ q ⇒ a ≤ p or a ≤ q for all p, q; the top counts |
joins, below
|
0, 1, 2, 2, 2, 0 | 1,200/1,200 / 1,200/1,200 |
| 1616 |
NeutralElementCount (free) |
e such that for all p, q and each of the 3 ways to pick r from {e, p, q} (the others u, v): (u ∨ v) ∧ r = (u ∧ r) ∨ (v ∧ r) and the dual |
joins, meets
|
1, 2, 3, 4, 2, 2 | 1,200/1,200 / 1,200/1,200 |
| 1621 |
AtomCount (free) |
entries whose only lower cover is 1 |
_x only |
0, 1, 1, 2, 2, 3 | 78/78 / 78/78 |
| 1622 |
JoinIrreducibleCount (free) |
entries with exactly one lower cover |
_x only |
0, 1, 2, 2, 3, 3 | 554/554 (n ≤ 64) / 553/553 (n ≤ 32) |
| 1623 |
DoublyIrreducibleCount (free) |
exactly one lower cover and exactly one upper cover |
_x, upperCoverCount (as #1354) |
0, 0, 1, 2, 3, 3 | 78/78 / 78/78 |
| 1625 |
MoebiusInvariant (free) |
μ(1, n): the last entry of the Möbius row of element 1 (the PR's moebiusRow, one row not n) |
below, Möbius recurrence |
1, −1, 0, 1, 1, 2 | 78/78 / 78/78 |
| 1651 |
FranklNumber (free) |
max over join-irreducible x of `n − 2· |
[x, n] | `; no value on the 1-element lattice (no join-irreducible, no row) | below |
| 1719 |
ShortestSmallIntervalChainCount (free) |
chains 1 = e₀ < … < eₖ = n with the fewest elements in which each eᵢ₊₁ is a join of upper covers of eᵢ: j > e is such a step iff the upper covers of e below j join to j; count by a fold from the top over (elements, chains) |
joins, below
|
1, 1, 1, 1, 1, 1 | 1,200/1,200 / 1,200/1,200 (n ≤ 9) |
| 1720 |
ShortestSmallIntervalChainSize (free) |
the elements on that chain, both ends counted (1-chain: 1): iterate e ← join of the upper covers of e from 1 to n; the title says "length" |
joins |
1, 2, 3, 2, 2, 2 | 2,050/2,050 (n ≤ 668) / 1,741/1,741 (n ≤ 30) |
| 1820 |
PopStackImageSize (free; PopStackSortingPasses is a permutation statistic) |
distinct values of x ∧ (meet of x's lower covers), the empty meet being x
|
meets |
1, 1, 2, 1, 2, 1 | 1,200/1,200 / 1,200/1,200 |
| 1833 |
LinearIntervalCount (free) |
pairs x ≤ y (including x = y) whose interval is a chain |
below |
1, 3, 6, 8, 12, 11 | 1,200/1,200 / 1,200/1,200 |
| 1845 |
JoinIrreducibleCountMinusRank (free) |
join-irreducibles less the length of the longest chain (PosetRank's reading) |
_x, chain sizes as #1354 |
0, 0, 0, 0, 0, 1 | 1,200/1,200 / 1,200/1,200 |
| 1846 |
UncomplementedElementCount (free) |
x with no y having x ∧ y = 1 and x ∨ y = n
|
joins, meets
|
0, 0, 1, 0, 0, 0 | 1,200/1,200 / 1,200/1,200 |
| 1881 |
DirectFactorCount (free) |
`max(1, log2 | center | ): the center is a Boolean lattice 2^kandLis a product ofkdirectly indecomposable lattices;St001881 = max(1, St001613)` on all 1,200 rows |
as 1613 |
| 1990 |
ZeroDivisorCount (free) |
non-bottom a with some non-bottom b and a ∧ b = bottom; the empty lattice gives 0 (its row exists) |
meets |
0, 0, 0, 2, 3, 3 | 79/79 (n ≤ 7, includes the n = 0 row) / 79/79 |
Subsets are bitmasks folded through Mod and Divide as PosetWidth and AntichainCount do in #1354. Cheap at the sizes the sweep reaches, grows as 2^n.
| Id | Head | Reading that matches the rows | Needs | 1-ch, 2-ch, 3-ch, B2, N5, M3 | Rows agree (readings / Epsil-shaped, n cap) |
|---|---|---|---|---|---|
| 1620 |
SublatticeCount (free) |
subsets closed under join and meet, the empty set included (1-chain: 2) |
joins, meets, masks |
2, 4, 8, 13, 23, 20 | 1,200/1,200 (n ≤ 9) / 300/300 (n ≤ 8) |
| 1624 |
LatticeBreadth (Breadth is taken) |
the most elements in a join-irredundant subset (no element can be dropped without changing the join), at least 1; the same value over subsets of join-irreducibles only (78/78) |
joins, masks |
1, 1, 1, 2, 2, 2 | 78/78 / 78/78 (all rows are n ≤ 7) |
| 1677 |
NondegenerateBottomMeetSubsetCount (free) |
non-empty subsets with neither bottom nor top and meet bottom |
meets, masks |
0, 0, 0, 1, 3, 4 | 1,200/1,200 / 300/300 |
| 1679 |
BottomMeetSubsetCount (free) |
non-empty subsets with meet bottom; equals 2^(n−1) + 2·St001677 for n ≥ 2 (all 1,199 rows) |
meets, masks |
1, 2, 4, 10, 22, 24 | 1,200/1,200 / 300/300 |
| 1681 |
MinimalBottomMeetSubsetCount (free) |
subsets with meet bottom, none smaller; "meet bottom" is upward closed, so dropping one element tests every proper subset |
meets, masks |
1, 1, 1, 2, 3, 4 | 1,200/1,200 / 300/300 |
Neutral and central elements: St001616 and St001613 agree with all 1,200 rows under the identity form (the Epsil-shaped reading) and under the literal definition ("the sublattice generated by e, x, y is distributive for all x, y", by closure) in readings.py.
Quirks that the records should say, since rows beat titles:
-
St000550: title and description say "modular pair both ways"; the rows say the one-sided law above. The title's reading agrees with 39 of 78 rows (the smallest miss: the pentagon, title reading 3, row 4). The rows also differ fromSt000551on 28 of 78 objects. -
St001720counts elements, not steps (the 1-chain is 1, the 2-chain 2), asShortestMaximalChainSizein #1354. -
St001620includes the empty sublattice;St001833includes the trivial intervals;St001615counts the top and not the bottom;St001624is at least 1 on the one-element lattice. -
St001651has no row on the one-element lattice, so it is undefined there (decline), not 0.
| Id | Reason |
|---|---|
| 1617 | Nullity of a rational linear system. Modeled 1,200/1,200 by elimination over Q; needs a rank, which the core has no way to write. |
| 1618, 1626 | Family of sublattice masks plus maximality of each against the others; modeled 1,200/1,200 and 78/78. Clumsy at n = 8; wait for a subset-family vocabulary. |
| 1619, 1666 | Isomorphism classes of sublattices or subposets; second slice (IsIsomorphicPoset). Not modeled. |
| 1754 | Subsets of pairs (2^28 at n = 8); modeled 26/26, rows only to n = 6. |
| 1630, 1875, 1876, 1877, 1878 | Incidence algebra homological invariants, no Sage code; St001876's rows list only some lattices per size. |
- Declining off the domain. Probed: a definition cannot return "unevaluated" (question 1). The fallback is not 0; the ways on are listed there.
-
The empty lattice. It is a lattice here (vacuous). Counts then give 0;
MoebiusInvariant,FranklNumber, the two small-interval statistics andLatticeBreadthhave no natural value on it, and FindStat has no row for them there. Recommendation: they decline on the empty poset; onlyZeroDivisorCounthas a row (0). -
Natural labeling carries the load. Bottom is 1, top is n, and a join is the least-labeled upper bound. This is why no sort is needed, and it is only valid for a naturally labeled value (the carrier's
valid, the reader). A definition run on a non-natural list would be wrong; the guard checks the unique minimal element only. -
Two samples for the records. #1354's sample
[[],[],[1],[1,2]]is not a lattice (two minimal elements), so a lattice record's example would show an unevaluated call. A lattice sample is needed (the pentagon[[],[1],[1],[3],[2,4]]: 550 and 551 are 4, the Frankl number 1, 1620 is 23); whetherSAMPLEStakes two entries for one carrier is not checked. -
Cost, unmeasured through compute-engine.
withJoinsandwithMeetsare n³ each; the neutral test is n³ identity checks over n² (e, p, q) triples with 3 permutations;1616and1613are the heaviest of tier 1. Tier 2 is 2^n. The standard run reaches the lattices of 7 elements at most (below); the deep run's crawl has rows to n = 9 and, for four statistics, to hundreds of elements, which the cap keeps out. -
A snapshot of 78 lattices is not a proof. It is every lattice through 7, which includes N5, M3 and the first non-modular, non-distributive, non-complemented cases, but the counts that need 8 elements to vary (
LatticeBreadthof 3 is B3, 8 elements) are checked only in a deep run with the crawl. - n = 9 is partial (905 of 1,078 spellings), for every statistic; the deep cap of 7 never reaches it.
-
The unique-minimal check is not a naturally-labeled check. A poset whose labeling is not natural fails
isPosetin the reader tests, so it never reaches a statistic from the crawl.
#1354 keeps every Lattices row through 6 elements (OBJECT_WISE_THROUGH, Lattices: 6). That is 25 distinct lattices (1, 1, 1, 2, 5, 15 for n = 1..6; 26 with the empty lattice that two statistics list). But OBJECT_WISE.Poset = { standard: 5, deep: 7 } caps by carrier, and these statistics are on Poset, so the standard run compares only lattices with at most 5 elements: 10 lattices. Measured on the recommended statistics:
| Through | Lattices | Statistics constant | Statistics with identical value vectors | Rows added (23 statistics) | Added size, as statistics.json indents |
|---|---|---|---|---|---|
| 5 (the standard run) | 10 |
St001719 (always 1) |
St000550 and St000551
|
230 | about 9 KB |
| 6 (the snapshot) | 25 | none | none | 575 | about 29 KB |
| 7 (the deep cap) | 78 | none | none | 1,794 | about 111 KB |
Weakest even at 6: St001719 (2 values), St001624, St001881, St001845 (2 to 3 values), because they first vary at n = 4 to 6. Through 7, St001719 has 3 values. The file is 949 KB today.
Recommendation: key the cap by FindStat collection, not carrier (the item already carries collection), with Lattices at least 6 in the standard run and 7 in a deep run, and keep the snapshot through 7. If the standard run's cost of 78 lattices times 23 statistics is too much, take 6 and accept two statistics that first vary at 6 (St000550 vs St000551, St001719) being checked on a single lattice class each.
All are decided as recommended (Dean, 2026-11-15), and the probe of question 1 has since come back: the decline it assumed cannot be written, so how a statistic declines is open again.
-
How does a lattice statistic decline, and how is
IsLatticedeclared? Decided: probe before any statistic is written;IsLatticeas an area definition withresult: "boolean". The probe (the builder's first step, 2026-11-15):-
The predicate works. A definition with
result: "boolean"is declared(poset) -> booleanand gaveTrueandFalseon two posets. Not checked: what a boolean head does in the carrier's statistics table, the generated entries, the compile guard and the coverage tests. -
A definition cannot decline.
applyDefinition(combinatorics/src/statistics/declare.ts) always replaces the call with the definition's value. An else branch ofNothingorUndefinedbecomes the answer;Holdof the call stays as aHold; the head's own call recurses without end or fails to type. The compiled route is no obstacle: it falls back to the interpreter when it throws. -
Open (Dean's call). The lane stopped here with no commits. The ways on:
-
(a) A placeholder value (
UndefinedorNothing) as the answer off the domain. No change to the machinery. It is an answer, not a decline, and a sum orMaxover the values of a collection then meets it. -
(b) A sentinel the machinery reads. The definition's else branch is a designated value and the head's
evaluateleaves the call when it sees it. One file, and the lane's recommendation. The body's type becomes a number or the sentinel, which is theIf(cond, z, sentinel)shape this page already doubts will compile to typed code, and the thirty-odd files that callapplyDefinitiondirectly (tests, the finder, the area scripts) see the sentinel as a value. -
(c) A
Latticecarrier. Against answer 7 (one carrier, two domains), and FindStat'sLatticesrows are read asPosetvalues. -
(d) The domain as data. A definition carries a predicate over the subject (
where: IsLattice(_x), with a size condition where a statistic has no value on the smallest lattices). One helper, called by the head and by the carrier's statistics table, evaluates it first and leaves the call when it is false. The body stays a number on its domain and compiles as the poset statistics do; the record can print the domain, and the crosscheck and the finder can read it. It touches more: theDefinitiontype, the two call sites indeclare.ts, the entries generator, and the build's compile cross-check, which must take its subjects from the domain. Not checked: that cross-check's sampling.
-
(a) A placeholder value (
- Lean: (d), then (b). (b) is the smaller change and unblocks the lane as written; (d) keeps the compiled route out of the question and makes the domain something the page and the tools can read, which the next partial statistics (a graph statistic on connected graphs only) will want too.
-
The predicate works. A definition with
-
Is the empty poset a lattice? Decided: yes. FindStat's rows say it is (
([],0)forSt001754,St001990), the definition is vacuous, and the guard order is unchanged. Counts give 0; statistics with no natural value decline. -
Cap the sweep by collection? Decided: yes, Lattices 7 deep, 6 or 7 standard, with the snapshot through 7. Why: with the carrier's cap of 5 the standard run holds
St001719to one value and cannot tellSt000550fromSt000551(measured above). - Slice size. Decided: all 23, tier 2 last so it can drop without touching tier 1. Why: tier 1 is where the verification is strongest and the cost lowest, tier 2 is 2^n but needs nothing new. If compile or standard-run time of tier 2 is too high at n = 7, ship tier 1 and keep tier 2 for the next batch.
-
Names. Decided:
LatticeCenterRank(the rows count the rank of the Boolean center; the title says binary logarithm of its size) andLatticeBreadth(Breadthis a permutation statistic). Why: bare names where free, prefixed only forBreadth, andCenterSizeis taken. Alternatives:CenterBinaryLogarithm;BreadthOfLattice. -
ZeroDivisorCountbare or prefixed? Decided: bare. Free today; a ring statistic with the same name would take the prefix then, asGraphLeafCountdid. -
A separate area file? Decided:
posets/src/statistics.lattices.ts,LATTICE_STATISTICS, joined into the area's list, with its own recorddomain: "Lattice statistics". Why: one carrier, two domains; thePosetcarrier record andLattices(n)stay untouched, andLattices(n), a restriction ofPosets(n), waits for that collection (out of scope here). -
A
Latticeprotocol bridge (GreatestLowerBoundandLeastUpperBoundon aPoset): out of scope; the joins and meets here are tables, not those heads.
Measured (Python, stdlib, one run at a time, each under a minute): the 34 statistics, titles and Sage-code presence; rows per size and the union of spellings per size (1, 1, 1, 1, 2, 5, 15, 53, 222, and 905 of 1,078); that all 25,183 rows relabel to a natural, transitively reduced poset and form a lattice; the readings of 27 statistics against every crawl row (all agree except the title of St000550); the Epsil-shaped readings of 23 against the rows (all agree, with an n cap where noted); the alternative reading for St000550 (78/78); the identities St001679 = 2^(n−1) + 2·St001677, St001881 = max(1, St001613); breadth over join-irreducibles only; IsLattice three ways on 96,428 posets with the counts of lattices up to isomorphism; the Epsil-shaped IsLattice on all posets through 6; per-statistic distinct values on the lattices through 5, 6 and 7; snapshot sizes; the head-name collision checks; and that the sweep caps Poset objects by carrier (read from #1354's crosscheck-sweep.ts).
Probed in the repository (the lattice lane, 2026-11-15): IsLattice can be a predicate with a boolean result; a definition cannot decline (question 1).
Assumed or unverified: that any of this compiles; that Min and Max over If(cond, z, sentinel) map compile to the typed list shape that withRelation did; cost in compute-engine; that the entries generator takes a second sample; the estimate that 78 lattices times 23 statistics fits the standard run; FindStat's own Sage rank() for St001845 (I matched the rows with longest chain length, not Sage's code); the names (LatticeCenterRank in particular); that no ring statistic will want ZeroDivisorCount; the 1-element and empty values for the statistics that have no row there.