Superpermutation of 7 symbols shorter than 5906

Download the artifact, strip whitespace, and normalize the alphabet to digits 1-7 exactly as the mission's format describes (reject if it mixes alphabets). Slide a length-7 window across the string, and for each window that is a permutation of 1..7 add it to a set:
      s = normalize(artifact_text)
      seen = set()
      for i in range(len(s) - 6):
        w = s[i:i+7]
        if len(set(w)) == 7:
          seen.add(w)
      assert len(seen) == 5040
    Compare len(s) to the claimed score and tier. A dishonest or broken submission typically is missing a small number of the 5040 permutations (easy to miss without building the full set, since 'looks long enough' is not the same as 'covers everything'), or was produced by concatenating two shorter strings with a stray separator character that breaks the windows at the seam.

Background. A superpermutation on n symbols is a string containing every one of the
n! permutations of those symbols as a contiguous substring. Minimal known superpermutation
lengths for small n: n=1: 1, n=2: 3, n=3: 9, n=4: 33, n=5: 153. These five are proven
minimal. For n=6, the shortest known is length 872, found by Robin Houston in 2014
(improving on a widely believed but ultimately non-optimal conjectured value of 873); 872
is now believed to be minimal for n=6 based on later exhaustive-style verification efforts,
but state this as a strong belief rather than a fully settled theorem when citing it. For
n=7, the best known length is 5906, due to Greg Egan (2019), building on constructions and
ideas from Houston and from an anonymous poster on the /sci/ board of 4chan whose 2011
argument established a lower bound. That lower bound, generalized as
n! + (n-1)! + (n-2)! + n - 3, evaluates to 5884 for n=7 (5040 + 720 + 120 + 7 - 3 = 5884).
So the true minimum length for n=7 is known to lie in the interval [5884, 5906]; closing
this gap is an open problem, and Egan's website and associated mailing-list / blog
discussions (with Houston, Egan, and others, often referenced under "the
superpermutation problem") document both the construction and the lower-bound argument in
detail.

Artifact format. A single string over the symbols; digits "1" through "7" are canonical.
Whitespace anywhere in the artifact is ignored (stripped before validation), so a string
may be wrapped across multiple lines. Also accepted, and normalized automatically: digits
"0" through "6" (each shifted up by one to become "1".."7"), or letters "a" through "g"
(case-insensitive, mapped to "1".."7" in order). Do not mix alphabets within one artifact;
the whole (whitespace-stripped) string must belong entirely to exactly one of the three
accepted alphabets. Example of a valid (but tiny, n=3-style) fragment showing the idea for
n=3 with symbols 1,2,3: "123121321" contains 123, 231, 312, 213, 132, 321 as consecutive
length-3 windows -- length 9, matching the known minimum for n=3.

Validity: the checker slides a window of length 7 across the (normalized) string; for
each window that happens to be a permutation of 1..7 (i.e. all 7 digits distinct and in
range), it is added to a set. The artifact is valid iff this set reaches all 5040 = 7!
distinct permutations. Score is the string's length after whitespace normalization; lower
is better. Reject artifacts longer than 20000 characters outright. target=5906 (mission
"solved" once you find something strictly shorter than the current best known, i.e. 5905
or less), record=5905 (this is the number to beat to claim a new best-known result;
verify very carefully and consider having independent re-verification before treating any
sub-5906 claim as final, since it would be a genuine, citable improvement on the
literature).

Attack strategies.
1. Literature/data retrieval: Greg Egan's website publishes explicit superpermutations of
   length 5906 for n=7 (and the underlying constructions for smaller n); retrieving the
   exact string and reformatting it (stripping whitespace, translating symbol alphabet if
   needed) is the most direct way to obtain a valid, near-target artifact. Note that 5906
   itself does not beat target (need strictly less than 5906).
2. Understand and reimplement the construction method: Egan's approach builds on
   Houston's method of gluing together permutations via a graph-theoretic / Hamiltonian-
   path-style construction on the permutation graph, together with a case analysis that
   improves packing efficiency over naive concatenation. A careful reimplementation might
   reveal opportunities to trim a handful of characters, especially at the boundary of
   the construction.
3. For smaller n, the recursive constructions (based on inserting each symbol into a
   superpermutation of n-1 symbols following certain patterns) achieve the proven-minimal
   lengths for n<=5 and the best-known length for n=6; studying these recursive patterns
   in detail (and verifying them with your own checker) builds confidence and tooling
   before tackling n=7 modifications.
4. Local trimming search: starting from a known valid 5906-length string, try deleting
   individual characters or short substrings and checking whether the resulting shorter
   string still covers all 5040 permutations; since removing characters can only break
   coverage of specific permutations, this is a computationally cheap way to probe for
   slack in the known-best construction.
5. Read up on the lower-bound argument (5884) to understand exactly why no superpermutation
   of n=7 can be shorter than that; this may suggest where the known 5906 construction is
   "wasting" characters relative to the theoretical floor, guiding a targeted search.

Pitfalls. The checker treats overlapping windows, so a string of length L contains
exactly L-6 windows of length 7 (for n=7); do not confuse "number of permutations
attempted" with "number of distinct permutations covered" -- repeated permutations do not
help. Symbol alphabet confusion is a common transcription error: if a source uses 0-indexed
symbols (0..6) or letters (a..g), make sure every character in the artifact is consistently
in that one alphabet; do not mix, e.g., a string that is mostly "1".."7" digits but has one
stray "0" or letter, since that will make the artifact fail to parse as any single
accepted alphabet. Whitespace inside the string is ignored, but this means accidentally
concatenating two separate strings with a single space between them will be silently
treated as one long string -- check total length carefully after normalization, since an
unintended concatenation could accidentally reach or exceed the 20000 character reject
limit, or could spuriously appear valid despite not being a genuine single construction.
Finally, remember the checker is specific to n=7 (alphabet size and required permutation
count of 5040 are hardcoded); an n=6 or n=4 superpermutation submitted here will simply
fail (it lacks digit "7" or contains too few distinct permutations), which is expected
behavior, not a bug.