The draft-preview evaluator answers `0` for an `avg` over a group with no numeric values — `emptyGroupValueFor` rules that null, and the live path answers null

[os-warren]

Found while working #16203. ⛔ Not that card's defect class (that one is the operand's TYPE) and ⛔ deliberately left unchanged there; the min/max half of this same idiom moved with that card because min/max were its subject.

The divergence, driven

packages/services/service-analytics/src/preview-evaluator.ts, aggregate():

case 'avg': return nums.length ? nums.reduce((a, b) => a + b, 0) / nums.length : 0;

The : 0 is the whole defect. emptyGroupValueFor (packages/spec/src/data/aggregation-policy.ts) is the platform's own ruling on this question and says the opposite in prose that reads like it was written for this line:

Counting no rows is 0 and summing them is 0: those are measured facts, not missing data. Averaging, minimising or maximising no rows is undefined — there is nothing to average — and must stay null rather than be flattened to a zero that reads as a real measurement.

Measured on one dataset, one row set, two AnalyticsService instances differing only in draftRowsResolver (the live half is NativeSQLStrategy's SQL on a real SQLite through sql.js, seeded from the same rows). The meals group's amount is NULL in every row:

LIVE     [{"category":"meals","avg_amount":null,"sum_amount":0, ...}]
PREVIEW  [{"category":"meals","avg_amount":0,   "sum_amount":0, ...}]

sum agrees at 0 — that is the identity, and fillEmptyGroups writes it deliberately. avg does not: the preview reports a measurement nobody made, and 0 is a plausible-looking average that a reader cannot tell from a real one.

Scope note

The same : 0 idiom stood on the min/max arms until #16203, which replaced it with null on those two and cites the same policy function. avg was left untouched there on purpose: it was outside that card's statement, and moving it is a value change on a published wire surface that deserves its own review rather than a rider.

Dedup

Searched (search_issues, repo-scoped; repo-scoped REST answers 403 for this session). The zero-match is a reading, not a silent zeroing — the same query returned #15546, #11065, #11151 and #6815, four adjacent aggregate-divergence cards known to exist. ⛔ Not a duplicate of #15546: that is sum (where 0 IS the ruled identity) across driver-sql and the objectql in-memory tier — a different aggregate on different faces. The relationship is worth reading together, though: both are "which face decides what an empty group means".

Related

#16203 · #15546 · #4708 (the card fillEmptyGroups and emptyGroupValueFor came from).