The matrix summary prints a negative suite count: 248 accounted of 246 ran, so "-2 did not"

[OffgridwithJD]

The matrix summary prints a negative count of suites. From today's CI, both majors:

PG17   suites that ran: 243 of 252 (skipped: 9, incomplete: 0)
       of those, 248 accounted for their checks and -5 did not

PG18   suites that ran: 246 of 252 (skipped: 6, incomplete: 0)
       of those, 248 accounted for their checks and -2 did not

A count of suites cannot be −5. The line is run_all_versions.sh:1475:

echo "  of those, $_acc_any accounted for their checks and $((suites_ran - _acc_any)) did not"

The cause, and it is a category error rather than arithmetic

_acc_any counts every suite whose log shows an accounting line. suites_ran counts suites that ran. A skipped suite still prints one: native_repack does check_skip "REPACK", pgc_summary prints the accounting line, and the suite is recorded SKIP. So a skipped suite is counted as accounted and not as ran, and accounted can exceed ran.

The two numbers reconcile exactly that way:

PG17   243 ran + 9 skipped = 252      248 - 243 = 5 skipped suites that also accounted
PG18   246 ran + 6 skipped = 252      248 - 246 = 2

Why it is worth fixing rather than tolerating

The block it sits in is the one #928 built so there would be one answer to "how many suites accounted". It now gives an impossible one. Nothing is gated on it — the population reconciliation three lines up is sound and independent (registered=252 | accounted=248, not dispatched=4, known debt=0, unaccounted=0 | sum=252) — so this is a reporting defect, not a false green.

But a negative count is the kind of number that makes a reader discount the whole block, including the parts that are right. And the fix has to pick a meaning rather than patch the subtraction: either

• count the same population on both sides — accounted-among-ran, so the difference is the suites that ran and said nothing, which is what the sentence claims to be about; or
• say what it now means — "248 of 252 registered suites accounted, including 2 that were skipped" — which keeps the information and drops the subtraction.

I would take the first: the sentence is of those, and those is the suites that ran. The second reads like a number defending itself.

Limits

• Found while reading a CI log for an unrelated failure of my own, not by a guard. Nothing asserts this line's arithmetic, which is why it has been printing since b0109af7.
• It is not test: the accounting breakdown says which LINE it means, because twelve suites were called unseedable and ten of them record #1003's doing, which I checked because I changed a neighbouring line in that block today: git log -S puts this line's last change at b0109af7 (The matrix prints two different "accounted" totals in one report (237 and 235) #928), and test: the accounting breakdown says which LINE it means, because twelve suites were called unseedable and ten of them record #1003 touched the line below it.
• Both majors, one run. I have not checked whether the magnitude tracks the skip count on other majors, though the arithmetic above says it must.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Uf6UoeBRZYLQZa4KxNiw8a