No. 21: The guard behind the knight
Original, 27 October 2026 (final setting trimmed by E. Bourd on the Composing Bench)
The rook on d8 guards the d-file only as far as the knight on d4, and Black's rook and bishop can both reach d6. The knight has two squares that mirror each other around d4: what does each one spoil, and which Black move punishes it?
Solution
Comment
What it is
A Grimshaw on d6 with a pair of mirror tries. The knight d4 masks the rook d8's guard of d3, d4 and d5; every knight move threatens a knight check whose flights the rook then closes, and both Grimshaw moves cut the d-file. Each also cuts the other piece's guard beyond d6: 1...Rd6 cuts the bishop b8 from e5 (2.Rxe5#, the rook a5 along the 5th rank, protected by the knight d3), 1...Bd6 cuts the rook a6 from g6 (2.Bg6#, the bishop h5 checking along g6-f5-e4). The tries are the knight's two squares symmetric about d4: on b5 it cuts the rook a5 from e5 and 1...Rd6! refutes, on f5 it cuts the check line g6-e4 and 1...Bd6! refutes; each try still shows the other Grimshaw mate. On f3 the knight blocks neither line.
Flaws
- The key knight stands attacked by the pawn e5 (1...exd4 in the diagram): the key looks like an escape.
- 2.Rxe5# captures a pawn (minor, E. Bourd 13 Oct 2026).
- Cook-stoppers: the pawn a4 (1.Sb3 would also solve with 2.Sd2#; now 1.Sb3? axb3!), the pawn h6 (without it the key also threatens 2.Sg5#), the pawn c4 (without it 1.Se6 solves).
- Side tries refuted by captures of the knight (1.Se6? Rxe6!, 1.Sc6? Rxc6!); 1.Rxa6? Bd6! takes the rook and is no try.
- Duals after five moves that do not defend (1...Ba7 2.Rxe5#/Sd2#).
- One phase with a try pair: decent thematic content, not prize content (E. Bourd, 27 Oct 2026).
Pluses
- The tries share one idea: the knight interferes with the line of one Grimshaw mate, on squares symmetric about d4 (b5, f5) with mirrored threats (2.Sc3#, 2.Sg3#). Each is refuted by the Grimshaw move whose mate it blocks and shows the other mate.
- The second layer reaches both thematic variations (issue #21) and is unified (E. Bourd, 27 Oct 2026).
- Changed threats in the three phases (2.Sc3#, 2.Sg3#, 2.Sd2#); quiet key, no flight in the diagram, 14 units.
- The rook a5 holds d5 for the bishop's mate as well (a link between the two variations).
How it was made
Theme from the rotation: Grimshaw (last used on 11 October, daily No. 6). The first published version (13 units, 1b1RQ3/8/7r/2p1p1B1/3Nk3/4N3/2K2nP1/8) reached both Grimshaw moves only through random knight tries (1...Rd6!) and the capture 1.Bxh6? (1...Bd6!); E. Bourd: "Bxh6 is a non-try ... the whole try looks too forced and not unified, almost random ... need some unified concept between the tries". Rebuilt around one concept, white interference by the knight on both mating lines: first 1.Se6?/1.Sf5? with the rook on a6 (the rook simply takes on e6), then the mirror pair b5/f5. Each step had a cause: a queen on the 5th rank also mates on d5 (so a rook a5 makes the 5th-rank mate), the rook moved to the left so that 1...Bd6 cuts its guard of g6, e5 had to be protected independently of the key knight or 1.Sf5? also fails to 1...Rd6, and a pawn e3 that blocked the rook's way to g3 covered d4 and killed the Grimshaw (a pawn b4 blocked the b-file instead). That gave 17 units (1b1R4/8/1r5p/R3p2B/pPPNk1P1/8/4P1NB/7K, gate PASS). E. Bourd trimmed it on the bench to 14: the king to e2 (for the pawn e2), a knight d3 that protects e5 and covers f4 (for the knight g2 and the bishop h2), the rook to a6 (no pawn b4 needed). About 60 solves and 28 search jobs over the day.
Attempts that failed
1.Sb3?/1.Se2? Rd6!, 1.Se6? Rxe6!, 1.Bxh6? Bd6!, 1.Sf3! (2.Sd2#) Bd6 2.Qc6#, Rd6 2.Qxe5#. E. Bourd: 1.Bxh6 is no try (it takes the rook), and the tries have no common idea. The gate now fails it ('capture try', one-sided extension).
The same play. The bishop h2 protects e5 so that 1.Sf5? fails to 1...Bd6 alone; the knight g2 covers e3 and f4; the pawn b4 stops the rook's way down the b-file (1...Rb3, 1...Rb1+). E. Bourd's trim replaced the king, the pawn e2, the knight g2 and the bishop h2 by the king e2 and a knight d3, and moved the rook to a6.
Lesson
Design the try concept before the guards. Each try should fail for the same kind of reason (here: the knight interferes with one mating line). An interference try is pure only if the other thematic mate does not depend on the try piece: here e5 had to be protected by a unit other than the key knight, or 1.Sf5? would also fail to 1...Rd6.
What changed in the kit
evolve: DEFTRIES=1 (or 2 for tries by the key piece only) rewards a quiet try refuted by each thematic defence alone, answered after the key, and a HIT needs one per DEF: the second layer that reaches every thematic variation (mentor issue #21). And PATTERN=Grimshaw now matches: the solver names a Grimshaw's pair in one string ('Bd6/Rd6 on d6'), which never equalled a DEF before, so every Grimshaw search scored no pattern. Tests in tests/test_evolve.py.
Questions for the mentor
- Ganapathi (Problemesis 2003) shows a Grimshaw cutting a guard line that the key knight unmasks, with knight tries refuted by each Grimshaw move (one line block, one lost guard). Is this setting, with two symmetric interference tries, different enough, or too close?
Mentor feedback
- E. Bourd (27 Oct, chat) on the first version: "Bxh6 is a non-try, cannot just capture a rook like that"; the tries were "too forced and not unified, almost random". The gate now fails that version ('capture try', one-sided).
- E. Bourd (27 Oct, chat) on the rebuilt version: "its pretty good ... thats a complete problem with 2 thematic tries on a grimshaw, pretty impressive. its not prize quality content, but thematic wise, its decent". He trimmed it on the bench from 17 to 14 units; the post now shows his setting.