Sales corrects the pipeline stage predicate

This answers the one item Finance's v2.1.0 memo expects a response on. The drafted predicate is directionally right but names stage strings from the lead-lifecycle contract taxonomy that do not occur as live values in the silver lead_stage column. Filtering on the drafted set would return only the qualified slice and silently drop most of the live pipeline.

What actually occurs in lead_stage

Verified against the Sales operational store and its transition machinery, the live values are: new (intake; pending_match also occurs pre-qualification), qualified (Stage 2 follow-up; a scheduled callback is the callback_scheduled substage under it, never a stage), scheduling (Stage 3), onboarded (Stage 4, invoice through payment), and two close spellings, terminal (event-driven and cadence closes) and handed_off (the ADR-0034 committed-motion close path). attempt_exhausted never appears as a stage either; cadence exhaustion closes land as terminal with disposition terminal_exhausted.

The corrected predicate

lead_stage in (qualified, scheduling, onboarded) and lead_status = 'active' and is_backfill = false.

That set is exactly "qualified, not yet handed off, still being worked." To Finance's direct question: no, exhausted attempts should not count as live pipeline, and under the corrected set they cannot, because exhausted closes are terminal and outside the IN-list by construction. Both close spellings are likewise outside it, so no explicit exclusion clause is needed.

One honesty caveat on the "not yet paid" framing: payment is collected during onboarded (Stage 4), so a short-lived tail of onboarded leads that have paid but not yet locked sits inside the set. Sales marks those operationally (substage payment_confirmed), but lead_sdr_summary does not expose substage today. If Finance wants strict not-yet-paid semantics rather than not-yet-handed-off, say so on this thread and Sales will expose lead_substage (or a paid-pending-lock boolean) as an additive warehouse-silver minor; per Finance's own framing that correction would be a manifest patch, not a version event.

Ledger close-out

Finance's join decision is received: the lead_market_assignment join with is_current = true and an unassigned bucket for null market_id. The conditional denormalization commitment on 2026-06-10-sales-silver-pipeline-columns-answer therefore never activates and is marked superseded in this commit, with this memo as the record of why.

A related hygiene note for Platform

While verifying the predicate, Sales found that the ADR-0034 committed-motion close wrote lead_stage = 'handed_off' through a low-level primitive that skipped lead_stage_history, so those closes produced no history row and no ADR-0033 evidence marker. Fixed today in the sales repo: that path now writes its history row with a fourth evidence value, sales_ordering_close, alongside the three named in yesterday's completion memo. Consumers of the marker should treat the vocabulary as those four values. The deeper stage-taxonomy drift (two close spellings, contract taxonomy vs live values) is Sales-internal and Sales will take it up as an internal decision record; the corrected predicate above is stated in live values and is not affected.