[BigDataDZ]

Fast-path metric inference could select the configured included-tax default for an explicit excluded-tax question (for example, 未税销售额). This can return the wrong metric without consulting the normal workflow.

Resolve explicit tax qualifiers before prompt defaults and generic aliases, using the configured business lexicon. Longest-prefix matching avoids overlapping qualifier substrings, and requests with conflicting bases or no compatible field defer to the normal workflow. Isolate the metric-matching logic so qualification is shared by alias and field matching; no baseline increases or new hardcoded business terms.

Validation: reproduced the wrong-field selection before the fix;18 fast-path/semantic-policy tests pass after it. New checks cover explicit bases, unavailable or conflicting bases, generic aliases, and custom regex-bearing lexicon prefixes. Independent review found no actionable issues and ran additional exact-source assertions. Based directly on main87c2966. Existing main-suite failures remain separately tracked; CI is not claimed fully green.

中文：明确税口径优先于默认值，防止快车道把未税问题编译成含税指标。

Full combined verification of#80–#82 on main87c2966:811 passed,5 failed,15 skipped; unchanged main794 passed,5 failed,15 skipped. The failing test names match exactly;17 new regression tests pass, with no new skips. Diagnostic runs use node --test --test-force-exit because existing failed tests retain resources. Architecture guard36 errors/2 warnings (baseline shape37 errors); fixture eval14/14 with2 L3 judge results pending. Baseline file unchanged. Combined branch codex/oct27-functional-fixes-integration at319fbbb; no upstream merge performed. CI is not claimed fully green.