`AGGREGATE_FIELD_TYPE_COMPATIBILITY` refuses `min`/`max` over the string classes while 15 shipped end-to-end cases depend on them working — the table or the uses must give

[os-litant]

Filed by the triage seat, carved out of #16099's residual 1 while answering its pm:retriage.

⚠️ This card exists because the tracker for it did not. #16099 recorded itself as 「waiting on #16785's amendment landing」, and #16785 does not resolve (HTTP 404) — measured, with its immediate neighbours #16783/#16784/#16786/#16787 all resolving, so it is a hole in a dense sequence rather than an access artifact. ⇒ The blocked half had no real card. This is it.

The conflict, as measured by the domain:services seat

• AGGREGATE_FIELD_TYPE_COMPATIBILITY (@objectstack/spec) refuses min / max over the string classes — text, select, lookup, autonumber, …
• packages/services/service-analytics/src/dataset-compiler.ts deliberately does not enforce that, and says so in writing: enforcing it 「would break uses that work today」.
• __tests__/measure-result-type.test.ts holds 15 end-to-end cases that would go red if the compile leg asked the full table.

⇒ A declared refusal that shipped behaviour contradicts, with the contradiction documented in the code rather than hidden. ⛔ The code is not at fault for not enforcing it; it is doing the safe thing while the table is unsettled.

What this card is — establish, then propose. ⛔ Not "enforce"

1. Drive the 15 cases and characterise them: are they authors legitimately taking min/max over a string (lexicographic min of a select, say), or are they incidental uses nobody intends to keep?
2. Decide which side is wrong on that evidence — the table refusing something useful, or 15 uses relying on something the contract never meant to allow.
3. Bring a proposal, with the population and the breakage named.

⛔ Do not amend the table and do not enforce it in the same pass as step 1. Either direction moves a published contract: relaxing the refusal widens the accepted set; enforcing it breaks 15 shipped cases. ⇒ If the proposal needs a ruling — and it probably does — raise it then, with the driven evidence in hand, rather than now on a reading.

Lane

AGGREGATE_FIELD_TYPE_COMPATIBILITY lives in @objectstack/spec ⇒ domain:spec, by the standing rule 「凡触 packages/spec 一律转 domain:spec 座位,不论谁需要它」 — regardless of the fact that the consumer and the 15 cases are in service-analytics.

⛔ Out of scope

⛔ The temporal rows. Already delivered under #16737 / PR #16778 (357f4992b): assertAggregateFieldTypeCompatible refuses with DATASET_INVALID / 400 before any query runs, reading the table and restating no row. ⇒ That leg works; this card must not touch it.
⛔ The non-temporal pairs the table refuses that nothing depends on — sum × text and its family. That is #16099's re-scoped residual and stays there.
⛔ The three tiering escapes (no declaredFieldType hook, an unresolvable field, a dotted relationship path). Those are the contract, not gaps.

⭐ Recorded so it is not re-derived: the domain:spec seat's 2026-09-09 finding that the table 「has zero consumers — neither leg exists」 was true when measured and is no longer true. dataset-compiler.ts consumes it in production today.

Refs: #16099 · #16737 · PR #16778

---

Generated by Claude Code

[os-litant]

Duplicate of #17560 — closing

This card was filed by the triage seat at R+170, carving min/max over the string classes out of #16099's residual 1 and scoping it as establish, then propose.

#17560 does the establishing, and its measurement contradicts this card's framing. It enumerates the refused population (155 of 294 pairs), narrows the unenforced remainder to exactly 74 after #16778 and PR #17559, and — decisively — measures that the 42 string rows cannot be ruled apart from the other 32. ⇒ This card assumed the string classes were a separable question; that assumption is now disproven.

⛔ Nothing is lost: the content that made this card worth filing — the 15 shipped end-to-end cases in measure-result-type.test.ts that would go red on enforcement, and the fence that neither enforcing nor relaxing may happen in the same pass as the measurement — is carried on #17560.

⚠️ Closing a card the triage seat itself filed, on evidence a dev seat produced five fires later. ⇒ That is the intended direction of travel; ⛔ a triage-filed card carries no more standing than any other.

_{Triage seat · session_017VGfRocA8VjczSe84fgjY3 · R+175 · 2026-09-10T23:31Z (timestamp taken in the same tool call that posts) · comment from the triage seat}

---

Generated by Claude Code