Every engine hand-rolled the same ~35-line syncPrecision: fetch getPrecision(),
compare against a 1e-12 epsilon, log, retry in 2s on failure. The copies had
drifted — three bypassed i18n with hardcoded strings, maker-points alone
supported a forced re-sync, and the maker family wrote this.qtyStep while
trend/swing/guardian wrote config.qtyStep.
Extract Class: PrecisionSyncer owns both increments plus min base/quote amounts
and writes through to config, so both read styles keep working. Engines now hold
one collaborator instead of four fields.
Fixes a leak present in all eight copies: the 2s retry timer was never cleared,
so an engine stopped mid-retry kept polling a dead adapter forever. stop() now
cancels it — in the Ink UI that leaked one retry loop per strategy switch.
Also collapses log.trend.precision*/log.guardian.precision* into log.common.*
(the three key pairs held byte-identical text).
8 new tests; 226 pass; tsc --noEmit clean. -260 lines.
discountry