opposite day: atbash, as a thirteenth line
huntress ctf 2026, day 13 · warmups · all three modes
tools used: decoding engine and timed-attempt pipeline.
the prompt: "a met z. they really hit it off." the single and the time trial were one line of text in atbash; the mastery was a zip of 63 blobs, each with thirteen numbered lines in thirteen encodings, the thirteenth line in atbash.
the key insight: day 12 had broken the run of encodings with a whole-text transform, a reversal. that suggested the next lines could be other keyless transforms rather than another base alphabet. atbash had been a registered reading since day 4, when we added it as a guess, and it sat in the per-line order along with the case inversion added on day 12 (a keyless rail fence was added the same day for whole messages only). the prompt names the mapping outright: a stands for z.
challenge shape
single 1 field one atbash line
time trial 1 field one atbash line
mastery 63 fields 63 blobs of 13 numbered lines, line k in day k's form
the field is message (setK.message in mastery): "the exact answer slug its final non-empty decoded line instructs you to submit", a word-word-word-NN slug.
artifact anatomy
atbash maps the alphabet onto itself reversed, a to z, b to y, and so on, separately for each case. it has no key and is its own inverse. anything that is not a letter passes through, so in a slug the hyphens and the number survive while every word turns into a pronounceable non-word:
svool dliow -> hello world
each mastery blob is the message split into thirteen chunks:
01<TAB>chunk in base64 (day 1)
02<TAB>chunk in rot13 (day 2)
...
12<TAB>chunk reversed (day 12)
13<TAB>chunk in atbash (day 13)
the thirteenth chunk holds the end of the message. by day 13 the chunks are short enough that the slug spans the last two or three lines: its end, digits included, sits in the atbash line, and the words before it in the reversed line 12 or the base64url line 11.
extraction method
atbash is arithmetic on the code point: a and z sum to 219 in ascii, A and Z to 155, so one translation table does it:
ATBASH = {
**{c: 219 - c for c in range(97, 123)},
**{c: 155 - c for c in range(65, 91)},
}
def atbash(text: str) -> str:
return text.translate(ATBASH)
for a single message the decoder peels: it tries every transform it knows, keeps readings that are printable, and stops at the first reading that holds a slug of dictionary words. for a numbered blob every line gets every one-step reading, and a beam joins the readings in line order, scoring the whole message by letter pairs across the line boundaries, before looking for the slug.
validation and answer formatting
an atbash line is printable ascii before and after decoding, so printability decides nothing. the dictionary does: the three words of the decoded slug are dictionary words and the atbash text's words are not, and the decoded sentence scores as english while the encoded one does not. the number passes through atbash untouched, which is a second check: the slug's digits are the same in both readings.
failed approaches and pivots
none. the prepared whole-text transforms covered the day, and no reading competed with the right one.
automation across modes
the same decoder answered the single, the time trial and all 63 mastery sets. mastery decodes each blob's thirteen lines independently, so a new line kind costs one table entry, not a new pipeline.
lessons
- after a run of encodings breaks, prepare the transforms that need no key: reversal, atbash, case inversion, word order.
- a translation that keeps text printable is told apart by the dictionary, not by its alphabet.
- a reading that leaves digits unchanged gives a free cross-check on the slug's number.
solve times
public scoreboard · day 13 finish: top 10
- 0:02.7single#2 of 209 solvers
- 0:04.4time trialfirst blood, +0.12 s
- 0:06.8masteryfirst blood, +2.3 s
- #4team day finishof 165 teams that cleared it, in 1:55.2
- #1single orderof the day's 5, by our flag time
from the public scoreboard: 09:00 et release to our flag; the table below counts from our attempt's start.
from when we started working on the mode to its flag.
| mode | work to flag | attempts | wrong submits |
|---|---|---|---|
| single | 1.3s | 1 | 0 |
| time trial | 1.7s | 1 | 0 |
| mastery | 2.4s | 1 | 0 |
the clock
server timer of the winning attempt; used is attempt start to flag, left is what remained on the clock.
| mode | time available | time used | time left |
|---|---|---|---|
| time trial | 30.0s | 0.5s | 29.5s |
| mastery | 80.0s | 1.5s | 78.5s |
where we got stuck
- nowhere: the decoder read all three modes on the first attempt.
flags: single, time trial, and mastery [redacted].
{generated by a clanker}