R2a function-library audit: verdict and scope-lock record (Epic 19 close-out)

[fupelaqu]

Close-out record for Epic 19 — R2a Function-Library Audit (the memo's §9 recommendation 5, lead-approved). The epic ran with per-story issues only where stories touched tracked code (19.1 → #248 / PR #249, 19.4 → #254); this issue is the epic-level record. Planning artifacts live in _bmad-output/planning-artifacts/ (local-only by design — _bmad* is gitignored).

What the audit produced (5/5 stories complete)

• 19.1 — dialect census: 259 syntax forms over 95 function tokens / 149 accepted spellings, every row anchored to source and regenerable by sbt "sql/testOnly *DialectCensusSpec".
• 19.2 — jOOQ OSS probe: day-one OSS feasibility gate (PASS) + 259 rows × 3 mandatory arms. Headline: arm A 117 parses+renders / 83 parse error / 42 SILENT / 16 loud / 1 crash; the one free OSS lever (parseUnknownFunctions=IGNORE) moves 51 rows but mangles INTERVAL args; parse dialect is not a lever.
• 19.3 — classification + estimate: 113 thin / 13 custom / 133 complex; total 749.15 d = 149.8 w = 24.97× the 6-week budget (band [53.3, 150.3] w, both ends in the rule's branch 3); MINIMUM-set-only option still 18.0 w.
• 19.4 — BI emitted-SQL capture: 99 statements (Superset + Tableau, both Tableau dialects); elasticsql accepted 5 / rejected 94, all 94 absent_feature (89 = the identifier-quoting family, Identifier quoting coverage: backticks rejected in every position; double-quoted table names unparseable in FROM #252). Product issues filed from it: Parser.apply can throw ValidationError instead of returning Left on parenthesized WHERE clauses #250 Support FROM-less SELECT of literal expressions — SELECT 1 is the connection/health idiom of Tableau and Superset and is rejected #251 Identifier quoting coverage: backticks rejected in every position; double-quoted table names unparseable in FROM #252 GROUP BY without an aggregate in the SELECT list returns per-document rows instead of one row per group (silent wrong answer) #253, jdbc#32–35, arrow#166–169.
• 19.5 — verdict memo: epic-19-audit-verdict.md, decision rule applied → branch 3 (lead_decision).

The decision (2026-09-01, lead-signed)

Branch 3 resolved by RE-SCOPE, not slip. Parser-ownership ADR decided: option C with option B as the R2a increment — no jOOQ anywhere in R2a. elasticsql's own parser owns all-ES relational closure (subqueries, CTEs, set operators), executed on the shipped JoinPlanner → DuckDB pipeline; jOOQ enters in R2b (Epics 26/27, narrowed per ADR Addendum A), where the option-C fallback door opens at marginal cost.

Consequences:

• The 149.8 w / 24.97× figure is void as an R2a number with no successor multiple (the R2a fallback scope list is empty); it survives only as an upper-bound planning input to R2b.
• Epics 20–23 re-scoped (r2a-rescope-epics-20-23-draft.md): Epic 20 = BI venue enablement (jdbc#32–35, arrow#166–169, connection-guide fixes) · Epic 21 = core SQL front door (Parser.apply can throw ValidationError instead of returning Left on parenthesized WHERE clauses #250 Support FROM-less SELECT of literal expressions — SELECT 1 is the connection/health idiom of Tableau and Superset and is rejected #251 Identifier quoting coverage: backticks rejected in every position; double-quoted table names unparseable in FROM #252 GROUP BY without an aggregate in the SELECT list returns per-document rows instead of one row per group (silent wrong answer) #253) · Epic 22 = native relational closure (phases 22a–22d; JOIN queries without a join-capable extension silently return wrong results #157 loud-error fix first; terms_lookup pushdown kept) · Epic 23 = BI regression matrix seeded with the captured corpus.
• Release thesis unchanged: "Drop in our driver. Tableau just works on Elasticsearch." The audit's diagnosis: the blockers are host bootstrap + identifier quoting + missing grammar — not dialect translation.

Handshake record (content-addressed sign-off)

• Memo as signed: sha256 f6a71ee02de28c0f2816015ce455106c240b50c20f4d2f8c12497ce911584cf5 (byte copy epic-19/verdict/out/memo-as-signed.md)
• Published (gate: CLOSED, frontmatter-only diff, verified): sha256 7d6714fe7451de14601fa514be3a3d36ee93e0aef16dda64c0a339f27758d5b7
• Signer: Stéphane Manciot (signoff.json; gate_outcome: cleared, adr_status_at_signing: C, denominator_confirmed: no — the unit recall was never answered and is moot under the re-scope)
• Roadmap edits: applied by update_roadmaps.py (9 edits; the two byte-frozen passages — the Epic 22 estimate-history clause and the pre-committed decision-rule table — verified unchanged)

Related

Story issues #248 (PR #249) and #254 can be closed as complete now that the epic gate is closed. Product issues #250 #251 #252 #253 (and the jdbc/arrow siblings) stay open — they are Epic 20/21 work items.