The Ablation Had Four Arms. None Matched What Shipped.
Originally published on hexisteme notes . I had an ablation script with four defined arms before I wrote a single line of the fix I meant to ship. By the time I'd actually shipped something, I still couldn't point at any one of the four and say "that's it, that's what's running now" — not because the script was thin (it was thorough, versioned, reproducible to the count) but because one boolean…
The ablation script for my coding harness initially included four arms: C0, C1, C2, and C3. Arm C0 replicated the existing audit script, C1 faithfully reproduced the live hook's actual logic, and C3 turned on both fixes ② and ③. Arm C2 was a middle ground between ② and ③ but was never discussed in the early drafts of this post as I never read its count.
All four arms were gated by a single flag, expansion. When expanded, both fixes were activated, while an unexpanded flag resulted in neither fix being applied. The fix I intended to ship was not among the four arms. To verify the situation, I ran the corpus through all four arms and obtained the following numbers: C0 counted 26 fires, C3 counted 92, C1 counted 47, and C2 counted 47.
Initially, I assumed the audit script was undercounting reality by a factor of 3.5x. However, when I reproduced the live hook's actual logic, I found that FIRE was at 47, challenge_hit was at 13,563, and the audit script itself was missing a fallback for user turns that weren't structured as a list. I then split the expansion flag into two independent toggles, creating a fifth arm, C4, which represented the intended fix: C1 plus ②. C4 achieved a FIRE count of 45, two below C1's 47.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
