percent chance: url percent-encoding and a line decoder that had to get faster

huntress ctf 2026, day 7 · warmups · all three modes

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

the prompt was one line: "my address bar is having a moment." the blob was a sentence written entirely as url percent escapes, %XX per byte, ending in an instruction line that names the answer slug.

the key insight: nothing new had to be decoded. the transform registry already had a url-unquote step, so the day-6 solver read the single and the time trial unchanged. the mastery bundle was where the day was decided, and there the cost was not decoding but ranking: 33 blobs of numbered lines, each line in its own encoding, scored in a way that grew exponentially with the line count.

challenge shape

one answer field per blob: the slug that the final non-empty decoded line tells you to submit. the single and the time trial delivered one blob; the mastery delivered 33 (blob-01.txt ... blob-33.txt) under one clock.

artifact anatomy

the single blob is plain percent-encoding of ascii text:

%68%65%6c%6c%6f%20 ...      # illustrative: "hello " one escape per byte

the mastery blobs reuse the numbered-line layout from day 2: NN<tab>chunk lines in numeric order, each chunk encoded independently. every one of the 33 blobs had seven lines, and line k carried day k's encoding: base64, rot13, hex, binary groups, decimal, octal, then the new percent-encoding on line 7, which holds the end of the slug:

01	TmljZSB3b3JrIGRlY29k
02	vat guvf jnezhc
03	2120546f20736f6c76652c20737562...
...
07	%2d%NN%NN ...                # illustrative: the slug's tail, one escape per byte

extraction method

the solver peels the blob with the registry's transforms until a layer reads as english and its last non-empty line carries a slug. for percent-encoding that is one step:

def url(text: str) -> str:
    out = urllib.parse.unquote(text)
    if out == text:
        raise ValueError("noop")     # not percent-encoded: let other readings try
    return out

numbered-line blobs go through the per-line decoder: sort the lines by number, collect every printable single-transform reading of each line, rank the combinations by english score, and return the first combination whose joined text contains a slug.

validation and answer formatting

the slug is accepted only from the final non-empty line of a layer that also scores as english, which rejects a slug-shaped token in an intermediate layer. the answer is the slug exactly as written.

failed approaches and pivots

none on the answer. the mastery solve was correct but slow: it spent 3.8 of its 9.0 seconds ranking line combinations.

automation across modes

the per-line decoder keeps the four best readings per line and sorts the cartesian product by total english score. the score of each reading was being recomputed for every combination it appeared in: with n lines and four readings each, 4^n combinations times n lines of regex scoring, about 3.8 million regex calls for the mastery bundle. on day 7 each reading is scored once and the combination total is a sum of cached scores:

options = [_line_candidates(chunk, keep) for _, chunk in lines]
scores = {t: english(t) for opts in options for _, t in opts}

def total(combo):
    return -sum(scores[t] for _, t in combo)

the ordering is identical, so every earlier fixture still decodes the same way. the mastery replay went from 3.81s to 0.27s.

lessons

solve times

public scoreboard · day 7 finish: top 3

  • 0:05.2single#4 of 245 solvers
  • 0:08.1time trial#5 of 226 solvers
  • 0:14.7mastery#3 of 221 solvers
  • #3team day finishof 179 teams that cleared it, in 4:44.3
  • #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 2s 1 0
time trial 3s 1 0
mastery 6s 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 7.0s 1.7s 5.3s
mastery 9.0s 5.4s 3.6s

where we got stuck

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

{generated by a clanker}