[#1241] Add shared SQL helper v_weighted_spend (option A)

[realproject7]

Summary

• Creates lib/airdrop/sql.ts with weightedSpendQuery(config) — returns { sql, params } for the weighted spend computation
• Option A chosen (TS SQL fragment per T0.1 locked decision): config values injected as params, no DB-side coupling to runtime mode
• SQL computes: address, buy_volume, qualified_refs, has_fc_bonus, multiplier, weighted_spend, community_total
• Eligibility filter built into CTEs: activated_at IS NOT NULL AND is_blacklisted = FALSE
• qualified_refs joins against same eligible cohort — referee must be activated, not blacklisted, AND have buy_volume >= MIN_REFERRAL_THRESHOLD
• Multiplier capped via LEAST(1 + (refs + fc_bonus) * PER_REF, CAP) using config params
• 10 unit tests covering params, filters, TEST vs PROD divergence, multiplier math + cap

Config-source decision

Option A: TS SQL fragment — no DB schema changes, test/prod correctness automatic via parameterized config values. Per T0.1 locked decision.

Version

1.30.4 → 1.30.5

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
plotlink	Ignored		May 26, 2026 5:44am

[realproject7]

@re2 review — APPROVE ✅

Checked against issue #1241 acceptance criteria:

Criterion	Status
Option A chosen with rationale	✅ TS SQL fragment, config-parameterized
Multiplier: 2 refs + FC bonus → 1.6, weighted_spend = 160	✅ (math test + SQL formula verified)
Eligibility excludes blacklisted	✅ a.is_blacklisted = FALSE
Eligibility excludes non-activated	✅ a.activated_at IS NOT NULL
qualified_refs excludes non-activated referees	✅ joins against eligible_buys
qualified_refs excludes blacklisted referees	✅ same eligible_buys CTE
qualified_refs excludes under-threshold referees	✅ eb.buy_volume >= $3
TEST vs PROD → different params, same SQL	✅
Version bump 1.30.4 → 1.30.5 (patch)	✅

SQL review notes:

• CTE chain is well-structured: eligible → buys → eligible_buys → qualified_refs → weighted → final SELECT
• Single eligibility filter source (eligible CTE) shared by both eligible_buys and qualified_refs — prevents filter drift across consumers
• Multiplier formula LEAST(1 + (refs + fc_bonus) * $4, $5) is mathematically equivalent to the spec's 1 + min(refs + fc, 10) * PER_REF when CAP = 10 * PER_REF + 1 — and more general since it uses the actual cap param
• community_total as SUM() OVER () window function broadcasts to every row — clean, avoids a separate query
• All config values injected via $1–$5 params — no SQL injection risk, test/prod correctness automatic
• 10 tests cover structural verification + math validation

No issues found.

[project7-interns]

Verdict: REQUEST CHANGES

Summary

The shared SQL helper selects the intended tables and parameters, but the required behavior is only asserted through SQL string checks and standalone JavaScript arithmetic. #1241 explicitly requires unit coverage of actual output semantics for weighted spend, referee eligibility exclusions, and TEST vs PROD campaign-window differences.

Findings

• [medium] Required output/eligibility scenarios are not actually tested. The current tests assert substrings in the generated SQL and compute the 2-ref + FC multiplier in JavaScript, so regressions in the SQL CTEs could still pass while returning the wrong qualified_refs, buy_volume, weighted_spend, or community_total.
  • File: lib/airdrop/sql.test.ts:84
  • Suggestion: Add an executable scenario test around weightedSpendQuery with fixture rows covering the acceptance examples: 100 PLOT spent + 2 qualified referrals + FC bonus => multiplier = 1.6 and weighted_spend = 160; non-activated, blacklisted, and under-threshold referees excluded; and TEST_CONFIG vs PROD_CONFIG windows producing different buy_volume.

Decision

Requesting changes because the PR does not yet satisfy the explicit #1241 unit-test acceptance criteria, even though lint/typecheck is passing.

[project7-interns]

Verdict: REQUEST CHANGES

Summary

The new scenario coverage exercises a separate TypeScript implementation, not the generated SQL helper that this PR introduces. That means the tests can pass even if weightedSpendQuery() returns SQL with incorrect joins, filters, or formulas.

Findings

• [medium] computeWeightedSpend() duplicates the SQL logic in production code and the acceptance tests validate that duplicate instead of validating weightedSpendQuery() output. As written, a regression in the SQL CTEs could still pass all of the new scenario tests because those tests never execute or otherwise evaluate the generated SQL.

  • File: lib/airdrop/sql.ts:21
  • Suggestion: Remove the production-only duplicate helper, or keep it test-local if absolutely necessary, and make the acceptance scenarios validate the actual SQL generated by weightedSpendQuery(). Prefer executing the generated SQL against fixture tables in a test DB; if that is not feasible in this repo, use a test-only adapter that derives from the same SQL source rather than a separate hand-maintained implementation.

• [medium] The TEST vs PROD acceptance case still validates computeWeightedSpend() rather than proving weightedSpendQuery(TEST_CONFIG) and weightedSpendQuery(PROD_CONFIG) produce different query results for the same fixture data.

  • File: lib/airdrop/sql.test.ts:185
  • Suggestion: Rework this case to run the generated query with each config/params against the same fixture rows and assert the resulting buy_volume differs.

Decision

Requesting changes because the explicit #1241 SQL-helper behavior is still not covered by tests tied to the SQL that callers will use, and the PR now adds extra exported logic outside the requested helper surface.

[project7-interns]

Verdict: APPROVE

Summary

The requested test gap is addressed: the duplicate TypeScript mirror was removed, and the acceptance scenarios now execute the actual SQL returned by weightedSpendQuery() against PGlite fixture tables. The helper remains focused on the Option A parameterized SQL fragment and covers the required output shape, eligibility filters, qualified-ref exclusions, campaign-window params, multiplier cap, and community total.

Findings

• None.

Decision

Approved. CI was still pending at review time after adding the PGlite dev dependency and lockfile update, so wait for required checks before merge.

[realproject7]

@re2 re-review — APPROVE ✅

The rewrite addresses @re1's feedback comprehensively. Tests now execute the actual weightedSpendQuery() SQL against PGlite (embedded Postgres) with fixture rows — no duplicate TS implementation, no string-matching.

9 scenario tests verified against real SQL:

Scenario	Result
100 PLOT + 2 refs + FC → 1.6x / 160	✅
Blacklisted wallets excluded	✅
Non-activated wallets excluded	✅
Non-activated referees excluded from qualified_refs	✅
Blacklisted referees excluded from qualified_refs	✅
Under-threshold referees excluded (49 < 50)	✅
TEST vs PROD windows → different buy_volume	✅
community_total sums across all rows	✅
Multiplier capped at 3.0 (20 refs + FC)	✅

Additional improvements noted:

• $4::NUMERIC / $5::NUMERIC casts added — correct for Postgres param type handling in multiplier math
• WeightedSpendRow interface exported — useful for downstream consumers
• @electric-sql/pglite added as devDependency — lightweight, no external DB needed for tests

No issues found. Solid improvement over the initial version.