Tomy sold four small companion puzzles for Eternity II: solve one, submit the solution, and the official site revealed one piece's placement on the main board. What each puzzle was, the broken online checker, the eBay grey market, why puzzles 5 and 6 never came, and the clue puzzles' second life as complex-theory test cases and donor piece sets.
Alongside the 256-piece prize puzzle, Tomy sold four small companion
puzzles, Clue Puzzle 1 through Clue Puzzle 4. Each was an
edge-matching puzzle in its own right, and each one, once solved and
submitted on the official website, revealed the placement of one piece on
the main board. This page is about those four products: what they
contained, how the hint mechanism worked, how they were sold and then
quietly discontinued, and what the community did with them afterwards.
It is not the page about the five clue placements of the main puzzle.
Those placements (the mandatory centre piece 139 at I8 plus the four
optional ones the clue puzzles revealed) are tabulated with provenance in
Known facts & numbers. The two topics meet in
the middle: the four optional rows of that table are exactly what the four
clue puzzles were for.
The mechanism: solve, submit, receive one placement#
The rules published in May 2007 announced four clue puzzles, each carrying
one hint: two on sale with the main puzzle in 2007, two more to follow in
2008 (msg 182). The
Australian rules text spelled the mechanism out: "For each Clue Puzzle you
solve, submit your solution to obtain a hint that gives the correct position
of one piece of the prize puzzle"
(msg 4187). In practice the
reward was the piece number, its square on the 16×16 board and its exact
rotation (msg 7747).
Three details of the design mattered:
The clue puzzles have many solutions. Unlike the main puzzle they
contain duplicated pieces and near-duplicates; Clue Puzzle 1's 36 pieces
carry only 110 distinct edge patterns out of a possible 288
(msg 2875). Any valid
arrangement was accepted.
Everyone got the same hint. Submitting different solutions to the same
clue puzzle returned the same clue piece
(msg 2860); the hints were
fixed properties of the main puzzle, not per-customer secrets.
The hints were optional. Under the contest rules only the starter
piece was mandatory; prize entries were free to ignore the four clue-puzzle
placements (msg 11046), and
the open record line (467→470) does ignore them; the strict five-clue line
that honours all four is tracked separately (best known: 464, see
Records & solvers).
A 6×6 square, launched with the main puzzle in July 2007. It is genuinely
easy: solved by hand in about a hundred moves, and a plain backtracker needs
on the order of 10²–10⁴ nodes
(msg 2873;
Anthony Cole's solver: 108 nodes,
msg 8775). The flip side of
easy is degenerate: apal1969's partial exhaustive search extrapolated to
roughly 3.6 × 10¹⁰ non-degenerate solutions
(msg 8052). Its CRC-16
checksums are the only clue-puzzle checksums ever published on the list
(msg 6769; see below).
A 12×6 rectangle, also from July 2007, and by consensus the hardest of the
four. "How to solve Clue2?" was a thread within months
(msg 3381). A computer does
it in milliseconds (msg 3386),
but by hand it stumped people who had breezed through Clue 1
(msg 3773), and apal1969
reported roughly a day of hand-solving as late as 2010
(msg 8137). It is also the
only clue puzzle whose solutions were never exhaustively counted: an
8-hour run had found ≈ 1.4 × 10⁶ non-degenerate solutions "and continuing"
(msg 8142), and Brendan Owen
judged the space too large for a complete search
(msg 8143). For a solver it
is still trivial: 1,637,300 nodes, about one second
(msg 8775).
A second 6×6, advertised on the UK site (with Clue 4) in January 2008
(msg 3858); early buyers
had solved it and claimed the hint by mid-January
(msg 3982). Its claim to fame
came later: it is the only clue puzzle whose solution count is known
exactly, at 2,195,647,488, from apal1969's ~1-day exhaustive run
(msg 8168). That count made
it the sharpest of the clue-puzzle test cases for
complex theory (next section). A
backtracker finds a first solution in 78 nodes
(msg 8775).
A second 12×6, launched alongside Clue 3 in January 2008, with 5 border and
4 interior colours (msg 8137).
Easier than Clue 2 for a machine (first solution in 4,951 nodes,
msg 8775); like Clue 2, its
solutions were never fully counted, so McGavin's complex-theory estimates
for the 12×6 puzzles stood in for exact numbers
(msg 10236). Its lasting
mark on the archive is the checker fiasco below.
Clue Puzzle 4· 72 pieces · 12×6
Cole's benchmark message also names a fifth, unofficial sibling: the
16-piece "web teaser" demo on the official site, solved in 34 nodes
(msg 8775).
The clue puzzles' launch was not smooth. When the Clue 3/4 submission forms
opened in January 2008, members probing them found the validation broken:
petzi.petzi typed his Clue 2 solution into the Clue 4 form (with a
typo piece "449" that does not exist in a 72-piece puzzle), and the site
accepted it and handed over the clue-4 hint
(msg 4109,
4119,
4141). The same trick failed
for another member
(msg 4117), so the checker was
not just lenient but inconsistent. The community's best guess at a partial
explanation: the clue puzzles' heavy piece redundancy means wildly many
valid arrangements, so wrong-looking submissions can collide with real
solutions (msg 4118). But
"449" rules out any reading in which the form was validating real pieces.
Nobody ever reported Tomy fixing or acknowledging it.
The hints themselves leaked immediately:
eBay. By January 2008, sellers offered the hint information from Clue
Puzzles 1 & 2 as a product
(msg 3971); one had
reportedly sold it over two dozen times
(msg 3976), with a
brief debate about whether reselling information you bought is even wrong
(msg 3972).
Derivation. In July 2008 danmayoh derived the Clue 3 & 4 hints from a
"possibility matrix" another member had posted. The matrix had been built
using the hints, and the construction could be inverted despite deliberate
typos. Moderator Brendan Owen deleted every message containing the matrix
"to be safe" (msg 5650,
5651); danmayoh confirmed
the trick could not conjure hints that were not already baked in
(msg 5732).
A short shelf life: sold out, and no puzzles 5 and 6#
Distribution was patchy from the start: Clues 1 & 2 were at first
unobtainable in the US
(msg 2855 thread), and
Clues 3 & 4 appeared UK-first, drawing complaints from Sweden, Australia,
Spain and the US
(msg 3858,
4502 and
7517 threads). Then the
supply dried up entirely:
January 2009. After the first scrutiny found no winner, Steve Grant
emailed the Tomy Careline asking when clue puzzles 5 & 6 would come. The
answer: Tomy has no plans to release any further clue puzzles
(msg 6381). Most of the
group was relieved; more hints would have cheapened the challenge.
October 2009. A Swedish buyer was told Clues 3 & 4 sold out in spring
2009 and the supplier could not reorder
(msg 7185); by November no
stock existed anywhere, as if production had stopped
(msg 7232).
May 2010. The puzzles had been off the UK market for around 18 months
and were never released at all in some regions; second-hand was the
only route (msg 7752), and
eBay listings already included the solved hint positions themselves
(msg 7350).
So the clue puzzles were purchasable for roughly two years, in some
countries never, and the information they sold became unbuyable while the
$2M contest still had a year and a half to run.
Mostly no, and the community knew it early. Brendan Owen never bought them
("just a marketing ploy"), calculating that the hints help but "there are
not nearly enough hints" to make E2 tractable
(msg 2842); his hints-theory
simulations put the requirement at roughly 20+ well-placed hints
(msg 1107), later refined to
~12–13 spread-out ones
(msg 5652). Spread-out
beats bunched, a conclusion still being sharpened in 2026
(msg 11736). The
quantitative picture, in three numbers:
One clue cuts the local search space by ~767×: Verhaard counted 3,366
ways to fill the corner rectangle up to a clue piece versus 2,582,369
without it (msg 3543).
All four cut the expected solution count from ≈ 14,702 to ≈ 4 × 10⁻⁸
(complex theory;
msg 11193); the clues buy
uniqueness, by design.
But the expected nodes-per-solution barely moves: McGavin measured
essentially the same backtracker cost with and without the extra
constraint (msg 8924).
Fewer needles, proportionally smaller haystack.
Hence the standing verdict, delivered by Jef Bucas when he finally published
the placements on his viewer: "so far, most progress has been done without
the clues" (msg 10082). The
best board that respects all five placements scores 464 (Benjamin Riotte,
2026); the open record, which respects only the mandatory starter, is 470
(Records & solvers).
Like the main set, the clue puzzles' piece designs fell under the
inventor's copyright claim, so they were never published. The community's
answer, as for the prize puzzle, was checksums instead of data
(msg 983,
msg 1063: Owen's launch-week
CRC-16 scheme). Requests to extend it to the clue puzzles started in 2007
(msg 3417), but only in 2009
did Thomas (E2_Manual's author) publish CRC-16 checksums, for Clue Puzzle 1
alone, asking others to confirm them and cover Clues 2–4
(msg 6768,
6769). No confirmation and
no checksums for Clues 2, 3 or 4 ever appeared on the list: anyone
typing in a second-hand clue set today still has no public way to validate
it. There was also a placement-level checksum: an e2hints.txt CRC to check
your hint data without posting it
(msg 1147).
Once the puzzles left retail, the placements they sold circulated on the
list as unofficial data: a piece/position list in 2009
(msg 6778), re-pointed
to whenever someone asked
(msg 8808,
8809), restated with rotations
in 2012 (msg 9067). The group
owner refused to bless publication
("That would be a violation of Mr. Moncton's copyright!",
msg 9210), but the data
outlived the objection: in January 2021 Jef Bucas added a Clues page to the
community viewer
(msg 10082), piece 181 was
confirmed first-hand by a member who had solved his clue puzzle and received
that hint from Tomy
(msg 10085), and the
rotations were cross-checked once more in 2021
(msg 10542). The resulting
placements are the four optional rows in
Known facts & numbers.
The clue puzzles' second career was as calibration targets for
Brendan Owen's complex theory. In November
2010 apal1969 ran exhaustive counts (Clue 3: exactly 2,195,647,488
solutions) and compared them with Peter McGavin's estimates from the theory;
the initial 10² discrepancy was traced to McGavin forgetting to halve the
edge-type counts, and after the fix the estimates landed within roughly an
order of magnitude, exactly the accuracy Owen claimed for small boards
(msgs 8142–8172; Owen's
accuracy note 8171, the bugfix
8172). McGavin then filed
backtracker estimates for all the clue puzzles in the group's database
(msg 8901). These were the
first of the exhaustive-vs-estimated
benchmarks that eventually made the community
trust the theory's E2-scale numbers. By 2011 the puzzles
themselves were solver warm-ups: all four fall in under a second
(msg 8773,
8775).
Finally, the physical pieces found a use the designers never intended. In
2014 Peter McGavin showed complete 16×16 "solutions" assembled from an E2
set plus clue-puzzle pieces
(msg 9309,
9310), and in 2023 he built a
full 480/480 board from the pieces of one E2 set, one Clue-1 set and one
Clue-2 set (msg 11169),
in Vlastislav W.'s max-distinct-tiles challenge
(msg 11167). To be perfectly
plain: these boards are not Eternity II solutions. They use pieces that
are not in the official 256-piece set, and they are why headline "480"
claims must always be checked against the piece list
(Records & solvers).
The one thing everyone assumed was gone turns out to be recoverable, and not just
for the first two. The community withheld the clue puzzles' piece designs on
copyright grounds, and no one posted the layouts to the list. But two independent
solver projects bundled them as worked examples anyway.
Armel Le Bail's SHORTER solver, a Fortran program published at cristal.org in
August 2007, ships three: a 36-piece 6×6, a 72-piece 12×6, and the full 256-piece
board. The 256-piece file is the real official Eternity II set, which is how the
provenance checks out: Le Bail was reading genuine Tomy data in the eternity2.net
format. The other two are Clue Puzzle 1 and Clue Puzzle 2, motifs and all.
Clues 3 and 4 turned up in a second, much more recent source: Janos Wortmann's
Eternity2Puzzles.jl, a Julia
package that carries all four clue puzzles as plain piece files (clue1.txt …
clue4.txt). Its clue1/clue2 are the same two puzzles Le Bail bundled: we
canonicalised each into a rotation-invariant, colour-relabelled multiset of pieces
and they match Le Bail's Clue 1 and Clue 2 exactly, piece for piece. A source that
reproduces the two puzzles we can already cross-check is a trustworthy source for
the two we cannot, so its Clue 3 and Clue 4 stand on the same footing.
The files eternity2.net published itself, recovered from the Wayback Machine,
corroborate the sizes and board positions of Clues 1 and 2 but not the motifs.
They lay each out as a block on a 16×16 grid, the 6×6 at rows 5–10 and columns
5–10, the 6×12 at rows 5–10 and columns 2–13, and collapse every interior edge to
one placeholder colour. (The site never hosted files for Clues 3 or 4, which
launched after its last capture.) So the official files fix where Clues 1 and 2
sat; the solver files are the public record of what the pieces are, for all four.
Each one was solved from scratch to confirm they are real, coherent puzzles. The
solve encodes "place every piece so all edges match and grey sits on the rim" as
a SAT instance and hands it to a solver; a separate pass then checks the answer
independently (each piece used once, every internal edge matched, grey only on
the outer ring). All four close perfectly: Clues 1 and 3 on all 60 of their
internal edges, Clues 2 and 4 on all 126. Each solved board is shown in its own
section above (Clue 1,
Clue 2,
Clue 3,
Clue 4), drawn with the real Eternity II motifs.
Each puzzle offers two downloads: the pieces as solver-ready data (the same
instance JSON published in the dataset, edges in
up-right-down-left order, colour 0 the rim) and as a set of per-piece SVGs the same
way the main puzzle's pieces are offered on the puzzle page.
A caveat on the motifs. The pieces are certainly right: they are the recovered
edge data, they cross-check between the two sources for Clues 1 and 2, and each
solves into a complete edge-matched board, which a wrong piece set would not. The
motifs shown are the real Eternity II glyphs, assigned to each clue's colour
numbers in order; they are almost certainly not Tomy's exact printed motifs for
the clue puzzles, which no public source records. So read the boards and downloads
as the correct pieces drawn in Eternity II's own art, not a facsimile of the
physical tiles.
The full recovery, the piece data for all four, and the reproducible solve are in
the clue-puzzle-pieces topic.
This says nothing about the copyright the layouts were withheld over; it simply
records that they were already public in two solvers' example files, and shows
them solved.
The clue puzzles' piece definitions were never posted to the list
(copyright), and the only community checksums cover Clue Puzzle 1 and remain
unconfirmed (msg 6769). They
survive anyway in two solver projects' example files, Clues 1 and 2 in Le
Bail's SHORTER and all four in Wortmann's Eternity2Puzzles.jl, cross-checked
and solved above.
Which puzzle revealed which placement. The archive never cleanly maps
Clue Puzzle n to a specific square among C3, C14, N3, N14. Only piece
181 has a first-hand puzzle-to-hint attestation
(msg 10085), and even
that message does not say which numbered puzzle produced it.
Exact solution counts for Clues 2 and 4. Only estimates and partial
counts exist (msg 8142,
10236).
Production and sales figures. How many clue puzzles were made or sold,
and why production stopped in early 2009, was never stated by Tomy; the
sold-out story is reconstructed from retailer answers
(msg 7185).
The official hint records. Whether Tomy's submission database and the
canonical hint answers survive anywhere is unknown; the placements rest on
community cross-checks, not an official publication.