Grid, offset-maker, liquidity-maker and maker-points wrote their trade log and
Telegram alerts as Chinese literals, so LANG=en changed the menu but not a
single runtime message. src/ now holds zero user-facing Chinese literals.
Reuses rather than duplicates: the four engines' subscription boilerplate maps
onto the existing log.subscribe.* / log.process.* keys, and the spot-maker
wording shared by offset-maker and liquidity-maker became one log.spotMaker.*
set instead of two.
Exchange gateways were treated differently on purpose: aster's exception and
console.error text is developer diagnostics, not UI, and its eight sibling
gateways already used English — so those were translated to English rather than
added to the table.
Adds tests/i18n-coverage.test.ts to hold the line: it fails on any new CJK
literal in src/, on a key missing either language, and on a duplicate key.
Several existing tests asserted the Chinese literals, which is what let the gap
persist; they now assert the resolved key and hold in either language.
214 new keys (300 -> 514). 275 pass; tsc and oxlint clean.
checkDataStaleAndDefense mixed the question (are these feeds trustworthy?) with
the answer (probe over REST, cancel everything, start polling), inside a 2063-line
class, so the rule could only be exercised through a live adapter.
Separate Query from Modifier: maker-points-defense.ts takes feed ages and health
flags and returns a verdict; the engine keeps every action. Follows the existing
maker-points-logic.ts / grid-logic.ts convention.
Three other enterDefenseMode call sites each spelled out the same 14-field stale
info, 11 fields of which were false/0/null padding. defenseReasonsFor() states
only the known cause.
16 new unit tests cover what previously had none — the age-0 startup case, the
threshold boundary, and the probe-then-defend sequence for a quiet account feed.
250 pass; tsc and oxlint clean. Engine 2063 -> 1939 lines.
discountry