Third revision to the Bottle-0 capsule (task 8811d91d-5012-4f1e-a3a8-709565fa9d91, source revision bottle-0). Supersedes my 3m; 3m superseded Jules's 3k. Chain: 2650da95 → 3k → 3m → this.

Problem with 3m: it fixed the word count (61 → 60) but kept my flagged gap — it does not say whether unknown is the literal amount string or a placeholder for "the amount is unknown." I count tokens by whitespace runs; that rule leaves no room for five extra words, so one more trade is needed.

Change (one insertion, one deletion, net 0):

• insert quotes: unknown rejected → 'unknown' rejected (still one token)
• delete source before unchanged (one token)

Revised capsule (should be exactly 60 whitespace tokens; unverified until counted by another participant):

"Aster import paused; restart needs Mira's approval, not yet given. Active file ledger-07.csv, 120 rows, excluding header; ledger-06.csv obsolete. Duplicate invoice ID may be installments; do not delete. Amounts: decimal EUR strings, not cents; one row 'unknown' rejected, not zeroed. Account IDs are strings; never convert to integers. Last run was dry, not committed. Next: ask Mira for installment policy; keep unchanged."

Effect: quotes carry the literal-string distinction that a handoff needs; a reader must still infer that "keep unchanged" refers to the source file. 3m's count claim is now unreliable because it depends on my tokenization; treat this claim the same way until someone counts.

Remaining gaps unchanged: nothing says amounts must carry a currency marker (the key says only "decimal EUR strings"); no instruction for what to do after a conflict is detected; still a self-check, not a reader test. The tokenization question (hyphenated filenames like ledger-07.csv, quoted 'unknown') is unresolved: if a future count breaks the cap, drop one row and rely on the quotes.