safe travels: base64 with its padding left at home, as an eleventh line

huntress ctf 2026, day 11 · warmups · all three modes

tools used: decoding search engine and timed-attempt pipeline.

the prompt: "our usual base went on a url-friendly vacation." the single and the time trial were one bare line each; the mastery was a zip of 53 blobs, each with eleven numbered lines, ten in day 10's encodings and one new. the answer in every set is the decoded message's slug.

the key insight: "url-friendly base" names base64url, the alphabet that swaps + and / for - and _ and usually drops the = padding. not one blob in any mode contained any of those four characters, so the alphabet never mattered. the only real change was the missing padding, and a decoder that insists on it rejects every line.

challenge shape

one message field per set, a pattern answer whose syntax the question states:

single       one blob, one bare base64url line, no padding
time trial   one blob, one bare base64url line, no padding
mastery      a zip of 53 blob-NN.txt files, one message field each
             every blob: 11 numbered lines, line k in that line's encoding

lines 1 to 10 of a mastery blob are the same encodings as day 10, in the same order: base64, rot13, hex, binary, decimal, octal, percent-encoding, html entities, json \u escapes and base32. line 11 is the new one. as on every warmups day, the eleven decoded lines join into one message, and the answer is read from it.

artifact anatomy

a line looks like a number, a tab and the encoded text. an unpadded base64url line is shorter than its padded form by zero, one or two characters, because base64 encodes three bytes as four characters and pads the last group to four:

decoded length mod 3   padded tail   unpadded tail
0                      none          none
1                      "=="          dropped
2                      "="           dropped

most of the mastery's line-11s needed two = and a few needed one; the time trial's line needed none, so it would have decoded even as plain base64. the single needed one =, so strict base64 rejected it.

11	LXdvcmQtNDI      # illustrative: "-word-42" in base64, its "=" left off

line 11 carries the end of every set's slug, so a line-11 reader is required for every answer.

extraction method

the decoder treats every line as a separate decoding problem: it tries every registered transform on the line, scores each reading as english, and joins the best readings of all lines by letter-pair fluency across the line boundaries (the decoding engine describes the beam). nothing in it is line-number specific: the line order of earlier days is a prior, not a rule.

base64url was already a registered transform. it restores the padding before decoding, so the length of the input never matters:

def b64url(text: str) -> str:
    s = _compact(text)
    return as_text(base64.urlsafe_b64decode(s + "=" * (-len(s) % 4)))

_compact removes whitespace and line breaks, and as_text rejects bytes that do not decode as text, so a wrong reading drops out instead of scoring as noise. plain base64.b64decode raises Incorrect padding on the same input, which is why a transform list with only strict base64 finds nothing on line 11.

validation and answer formatting

a line's reading counts only when it decodes to printable text that scores as english. the joined message must end in the slug the question asks for: three lowercase words, hyphens, and a two-digit number. the answer is that slug as printed, with nothing rewritten.

automation across modes

the day 10 solver answered all 55 sets unchanged: the single and the time trial as one line each, and the mastery's 53 blobs line by line. no transform and no line order was added for the day.

lessons

solve times

public scoreboard · day 11 finish: 1st

  • 0:02.2single#2 of 216 solvers
  • 0:03.7time trialfirst blood, +0.31 s
  • 0:05.6masteryfirst blood, +1.9 s
  • #1team day finishof 164 teams that cleared it, in 0:10.1
  • #3single 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 1.5s 1 0
time trial 1.5s 1 0
mastery 1.9s 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 10.0s 0.6s 9.4s
mastery 8.0s 0.8s 7.2s

where we got stuck

flags: single, time trial, and mastery [redacted].

{generated by a clanker}