Fix #180 #181: unjudgeable checkers for problems 11, 22 and 210

[xiawubing]

Summary

Three algorithmic checkers score every submission 0 (or cap it) for reasons unrelated to the
submission. Fixes #180 (problem 11, and the identical bug in problem 22 found in review) and
#181 (problem 210); details and reproduction are in the issues. Problem 145 (#182) is tracked
separately because its fix is a different kind of change.

• 11/chk.cc: enlarge the 30-byte message buffer (the formatted verdict is 41 characters, so
  every correct answer aborted in the fortified sprintf).
• 22/checker.cpp: the same buffer and the same sprintf, same one-line fix. A search of
  every checker and interactor under algorithmic/problems finds no other instance.
• 210/chk.cc: loosen the per-base g/c/d/v read bounds; the shipped 4.in-9.in
  carry v up to 250 and the checker FAILed on its own input for 6/10 cases. The observed
  ranges are added to the statement's "Constraints (from released datasets)".

Type of Change

• Bug fix

Testing

Compiled with the judge's flags (g++ -std=gnu++17 -O2 -pipe -I algorithmic/judge/include).

Regression corpus: the exact per-case outputs the judge recorded for 32 archived agent runs
(16 per problem, two harness arms × 8), re-judged with the original and the patched checker.
Feeding the recorded outputs to the original checker reproduces the official per-case ratio
on every checker-judged case (11: 150/150, 210: 160/160), so the comparison is apples to apples.

• 11: 52 cases that aborted with buffer overflow detected now score (all 1.000 — no archived
  path exceeded the length bound); the 45 -1 cases and the 53 wrong-answer cases are
  unchanged. Run rewards move from a uniform 0.300 to 0.3–1.0 (10 of 15 runs change). The ten
  shipped .ans files also score 1.0 / ok, a non-palindrome scores 0, a path 30× over the
  bound scores 0.
• 22: the shipped .ans files are empty (the checker does not read them), so the corpus is
  the 40 judge-recorded outputs of 4 archived agent runs from another campaign, all officially
  *** buffer overflow detected ***: the original checker aborts on 40/40, the fixed one
  returns a verdict on 40/40 (all 1.0; their K is N or ≈ 3N), clean under ASan/UBSan. Unrelated
  to the crash, the score formula on line 120 of the same file has an operator-precedence issue
  (/ 2 * N) that removes the partial-credit gradient for 3N < K < 5N; it does not affect these
  outputs and, since it changes scoring, is left out of this PR.
• 210: the 96 cases that FAILed on input (cases 4–9 × 16 runs) are now judged; the 64 other
  cases are unchanged. Cases 7 and 9 stay 0 for every run because those programs printed
  nothing there. Run rewards move from 0.000–0.187 to 0.000–0.351.

All three patched checkers run clean under AddressSanitizer + UBSan over the whole corpus plus the
shipped .ans files. For 11 the scoring formula was exercised on a synthetic 1×3 grid: length
below / at / between 1× and 2× / above 2× the bound gives 1.0 / 1.0 / 0.870 / 0.0, -1 on a
solvable instance gives 0, and an empty output on a 1×1 grid gives 1.0 as the statement
requires.

No testdata or .ans changes. The validate-algorithmic CI job will report "Missing reference
solution" as it did for #174; these problems ship no reference.cpp.

Checklist

• Code follows the project structure and conventions
• Self-review completed
• Documentation updated (210 statement constraints)

🤖 Generated with Claude Code

https://claude.ai/code/session_01AUL12r3HsmHBUTDGZxXFQr

[whuang369]

LGTM! Thanks