diff --git a/packages/opencode/src/altimate/native/altimate-core.ts b/packages/opencode/src/altimate/native/altimate-core.ts
index a41296ee71..137e28d848 100644
--- a/packages/opencode/src/altimate/native/altimate-core.ts
+++ b/packages/opencode/src/altimate/native/altimate-core.ts
@@ -320,7 +320,13 @@ export function registerAll(): void {
   register("altimate_core.equivalence", async (params) => {
     try {
       const schema = schemaOrEmpty(params.schema_path, params.schema_context)
-      const raw = await core.checkEquivalence(params.sql1, params.sql2, schema)
+      // Pass the optional dialect hint so dialect-specific compiled warehouse SQL
+      // (e.g. Snowflake semi-structured `col:field`) parses and the pair is
+      // decidable instead of abstaining on a syntax error. Supported since
+      // altimate-core@0.5.1. Use `|| undefined` (not `??`) so the default empty
+      // string from ReviewConfig.dialect coerces to "no hint": the engine throws
+      // on an unknown dialect "", and "" must mean auto-detect, not a real dialect.
+      const raw = await core.checkEquivalence(params.sql1, params.sql2, schema, params.dialect || undefined)
       const data = toData(raw)
       return ok(true, data)
     } catch (e) {
@@ -332,7 +338,7 @@ export function registerAll(): void {
   register("altimate_core.migration", async (params) => {
     try {
       // Build schema from old_ddl, analyze new_ddl against it
-      const schema = core.Schema.fromDdl(params.old_ddl, params.dialect || undefined)
+      const schema = core.Schema.fromDdl(params.old_ddl, params.dialect ?? undefined)
       const raw = core.analyzeMigration(params.new_ddl, schema)
       const data = toData(raw)
       return ok(true, data)
@@ -426,7 +432,7 @@ export function registerAll(): void {
   register("altimate_core.column_lineage", async (params) => {
     try {
       const schema = resolveSchema(params.schema_path, params.schema_context)
-      const raw = core.columnLineage(params.sql, params.dialect || undefined, schema ?? undefined)
+      const raw = core.columnLineage(params.sql, params.dialect ?? undefined, schema ?? undefined)
       return ok(true, toData(raw))
     } catch (e) {
       return fail(e)
@@ -447,7 +453,7 @@ export function registerAll(): void {
   // 22. altimate_core.format
   register("altimate_core.format", async (params) => {
     try {
-      const raw = core.formatSql(params.sql, params.dialect || undefined)
+      const raw = core.formatSql(params.sql, params.dialect ?? undefined)
       const data = toData(raw)
       return ok(true, data)
     } catch (e) {
@@ -458,7 +464,7 @@ export function registerAll(): void {
   // 23. altimate_core.metadata
   register("altimate_core.metadata", async (params) => {
     try {
-      const raw = core.extractMetadata(params.sql, params.dialect || undefined)
+      const raw = core.extractMetadata(params.sql, params.dialect ?? undefined)
       return ok(true, toData(raw))
     } catch (e) {
       return fail(e)
@@ -468,7 +474,7 @@ export function registerAll(): void {
   // 24. altimate_core.compare
   register("altimate_core.compare", async (params) => {
     try {
-      const raw = core.compareQueries(params.left_sql, params.right_sql, params.dialect || undefined)
+      const raw = core.compareQueries(params.left_sql, params.right_sql, params.dialect ?? undefined)
       return ok(true, toData(raw))
     } catch (e) {
       return fail(e)
@@ -522,7 +528,7 @@ export function registerAll(): void {
   // 29. altimate_core.import_ddl — returns Schema, must serialize
   register("altimate_core.import_ddl", async (params) => {
     try {
-      const schema = core.importDdl(params.ddl, params.dialect || undefined)
+      const schema = core.importDdl(params.ddl, params.dialect ?? undefined)
       const jsonObj = schema.toJson()
       return ok(true, { success: true, schema: toData(jsonObj) })
     } catch (e) {
diff --git a/packages/opencode/src/altimate/native/sql/register.ts b/packages/opencode/src/altimate/native/sql/register.ts
index 8a337bdcf9..60b565289f 100644
--- a/packages/opencode/src/altimate/native/sql/register.ts
+++ b/packages/opencode/src/altimate/native/sql/register.ts
@@ -359,7 +359,9 @@ export function registerAllSql(): void {
       const sqlA = params.original ?? params.sql_a
       const sqlB = params.modified ?? params.sql_b
 
-      const compareRaw = schema ? await core.checkEquivalence(sqlA, sqlB, schema) : null
+      // `|| undefined`: coerce a default empty-string dialect to "no hint" — the
+      // engine throws on an unknown dialect "".
+      const compareRaw = schema ? await core.checkEquivalence(sqlA, sqlB, schema, params.dialect || undefined) : null
       const compare = compareRaw ? JSON.parse(JSON.stringify(compareRaw)) : null
 
       // Simple line-based diff
diff --git a/packages/opencode/src/altimate/native/types.ts b/packages/opencode/src/altimate/native/types.ts
index f3e72e562e..ad90c7631e 100644
--- a/packages/opencode/src/altimate/native/types.ts
+++ b/packages/opencode/src/altimate/native/types.ts
@@ -724,6 +724,8 @@ export interface SqlDiffParams {
   original: string
   modified: string
   context_lines?: number
+  /** Optional parsing-dialect hint forwarded to the equivalence engine. */
+  dialect?: string
 }
 
 export interface SqlDiffResult {
diff --git a/packages/opencode/src/altimate/review/runner.ts b/packages/opencode/src/altimate/review/runner.ts
index f26b6add65..e4b5e8f0fd 100644
--- a/packages/opencode/src/altimate/review/runner.ts
+++ b/packages/opencode/src/altimate/review/runner.ts
@@ -481,14 +481,18 @@ export function createDispatcherRunner(opts: DispatcherRunnerOptions): ReviewRun
         })
         const data = (res.data ?? {}) as Record<string, any>
         const validationErrors = asArray(data.validation_errors)
-        // Undecidable when: call failed, no schema to resolve columns, or the
-        // engine returned validation errors. Never guess equivalent=true.
+        // Undecidable when: call failed, no schema to resolve columns, the engine
+        // returned validation errors, or the engine itself flagged the comparison
+        // as not decidable (altimate-core@0.5.1 `decidable` field — authoritative,
+        // catches semantic abstentions that carry no validation error). Never
+        // guess equivalent=true.
         const decided =
           res.success === true &&
           typeof data.equivalent === "boolean" &&
           !res.error &&
           !data.error &&
           validationErrors.length === 0 &&
+          data.decidable !== false &&
           !!schema
         if (!decided) return { decided: false }
         // The engine flags column-reorder etc. as `minor` but still sets
diff --git a/packages/opencode/test/altimate/altimate-core-native.test.ts b/packages/opencode/test/altimate/altimate-core-native.test.ts
index 243425ec8c..78bbb507ac 100644
--- a/packages/opencode/test/altimate/altimate-core-native.test.ts
+++ b/packages/opencode/test/altimate/altimate-core-native.test.ts
@@ -816,3 +816,101 @@ describe("Real-world equivalence regressions", () => {
     expect(await structRules("select distinct a from t group by a", "select distinct a from t group by a")).toEqual([])
   })
 })
+
+// ---------------------------------------------------------------------------
+// altimate-core@0.5.1: dialect forwarding + authoritative `decidable` field
+// ---------------------------------------------------------------------------
+//
+// 0.5.1 added an optional `dialect` arg to checkEquivalence and a `decidable`
+// flag on the result. The native equivalence handler now forwards `params.dialect`
+// to the engine so dialect-specific compiled warehouse SQL parses instead of
+// abstaining on a syntax error. These tests exercise the REAL engine through the
+// Dispatcher (no mocks) to prove the arg is wired end-to-end.
+describe("core 0.5.1 dialect forwarding + decidable", () => {
+  beforeAll(async () => {
+    registerAll()
+    // sql.diff lives in the sql/* handler set, registered separately.
+    const { registerAllSql } = await import("../../src/altimate/native/sql/register")
+    registerAllSql()
+  })
+
+  const variantSchema = {
+    tables: { t: { columns: [
+      { name: "x", type: "INT" }, { name: "payload", type: "VARIANT" },
+    ] } },
+  }
+  // Snowflake semi-structured access `payload:f` is a hard PARSE error in the
+  // default dialect (the ':' is rejected) but parses under the snowflake dialect.
+  // The validation-error TEXT therefore differs by dialect — a syntax error only
+  // when the hint is dropped. Pre-fix (dialect not forwarded) BOTH would be syntax
+  // errors; this asserts they diverge, proving the arg reaches the parser.
+  const colonSql = "select payload:f::int as v from t"
+  const equivCall = (dialect?: string) =>
+    Dispatcher.call("altimate_core.equivalence", { sql1: colonSql, sql2: colonSql, schema_context: variantSchema, dialect })
+  const hasSyntaxError = (data: any) =>
+    (data.validation_errors ?? []).some((e: string) => /Syntax error|Expected end of statement/i.test(String(e)))
+
+  test("dialect hint is forwarded to the engine (changes parse outcome)", async () => {
+    const noDialect = await equivCall(undefined)
+    const snowflake = await equivCall("snowflake")
+    expect(noDialect.success).toBe(true)
+    expect(snowflake.success).toBe(true)
+    // Without a dialect the colon is a syntax error; snowflake accepts it.
+    expect(hasSyntaxError(noDialect.data)).toBe(true)
+    expect(hasSyntaxError(snowflake.data)).toBe(false)
+  })
+
+  test("decidable=false is surfaced when the engine cannot parse/plan", async () => {
+    const noDialect = await equivCall(undefined)
+    expect((noDialect.data as any).decidable).toBe(false)
+  })
+
+  test("decidable=true is surfaced for a cleanly decided pair", async () => {
+    const r = await Dispatcher.call("altimate_core.equivalence", {
+      sql1: "select count(*) from t",
+      sql2: "select count(1) from t",
+      schema_context: variantSchema,
+    })
+    expect(r.success).toBe(true)
+    const d = r.data as any
+    expect(d.decidable).toBe(true)
+    expect(d.equivalent).toBe(true)
+  })
+
+  test("empty-string dialect is coerced to no-hint (does NOT throw)", async () => {
+    // ReviewConfig.dialect defaults to "" (auto-detect). The engine throws on an
+    // unknown dialect "", so the handler must coerce "" → undefined (`|| undefined`,
+    // not `??`). Regression guard: before the coercion this returned success=false
+    // and silently disabled equivalence checking on the default config.
+    const r = await Dispatcher.call("altimate_core.equivalence", {
+      sql1: "select count(*) from t",
+      sql2: "select count(1) from t",
+      schema_context: variantSchema,
+      dialect: "",
+    })
+    expect(r.success).toBe(true)
+    const d = r.data as any
+    expect(d.decidable).toBe(true)
+    expect(d.equivalent).toBe(true)
+  })
+
+  test("sql.diff accepts and forwards the dialect hint without throwing", async () => {
+    // sql.diff's runtime shape (success/diff/equivalent) predates its declared
+    // SqlDiffResult type, so read through `any`. The assertion that matters: the
+    // dialect param is plumbed into the handler's checkEquivalence call (Edit in
+    // sql/register.ts) and the snowflake path is reachable end-to-end.
+    const call = (dialect?: string) =>
+      Dispatcher.call("sql.diff", {
+        original: colonSql,
+        modified: colonSql,
+        schema_context: variantSchema,
+        dialect,
+      } as any) as Promise<any>
+    const noDialect = await call(undefined)
+    const snowflake = await call("snowflake")
+    expect(noDialect.success).toBe(true)
+    expect(snowflake.success).toBe(true)
+    expect(typeof noDialect.diff).toBe("string")
+    expect(typeof snowflake.diff).toBe("string")
+  })
+})
diff --git a/packages/opencode/test/altimate/review-equivalence-e2e.test.ts b/packages/opencode/test/altimate/review-equivalence-e2e.test.ts
new file mode 100644
index 0000000000..f0031d17cb
--- /dev/null
+++ b/packages/opencode/test/altimate/review-equivalence-e2e.test.ts
@@ -0,0 +1,143 @@
+import { afterAll, beforeAll, describe, expect, test } from "bun:test"
+import { writeFileSync } from "node:fs"
+import path from "node:path"
+import { runReview, DEFAULT_RUBRIC, DEFAULT_REVIEW_CONFIG, type ChangedFile } from "../../src/altimate/review"
+import { createDispatcherRunner } from "../../src/altimate/review/runner"
+import { registerAll } from "../../src/altimate/native/altimate-core"
+import { tmpdir } from "../fixture/fixture"
+
+// ---------------------------------------------------------------------------
+// Full-stack E2E: dbt PR review pipeline → real Dispatcher → real altimate-core
+// 0.5.1 engine. NO mocks. Exercises the complete chain that the dialect wiring
+// (core 0.5.1 `dialect` arg) and `decidable` handling run through:
+//
+//   runReview → semanticChangeLane → runner.equivalence(old, new, dialect)
+//             → Dispatcher → altimate_core.equivalence handler
+//             → core.checkEquivalence(sqlA, sqlB, schema, dialect)
+//
+// A real dbt manifest supplies the schema so the engine can resolve columns and
+// actually DECIDE equivalence rather than abstain.
+// ---------------------------------------------------------------------------
+
+describe("E2E: review pipeline + real engine equivalence (core 0.5.1)", () => {
+  beforeAll(async () => {
+    process.env.ALTIMATE_TELEMETRY_DISABLED = "true"
+    // The DispatcherRunner calls the REAL native handlers; register them in case
+    // another file reset the Dispatcher.
+    registerAll()
+    const { registerAllSql } = await import("../../src/altimate/native/sql/register")
+    registerAllSql()
+  })
+  afterAll(() => {
+    delete process.env.ALTIMATE_TELEMETRY_DISABLED
+  })
+
+  // A real manifest with typed columns so resolveSchema() yields a usable schema.
+  async function reviewerWithManifest() {
+    const tmp = await tmpdir()
+    const manifestPath = path.join(tmp.path, "manifest.json")
+    writeFileSync(
+      manifestPath,
+      JSON.stringify({
+        metadata: { adapter_type: "snowflake" },
+        nodes: {
+          "model.demo.orders": {
+            resource_type: "model",
+            name: "orders",
+            original_file_path: "models/marts/orders.sql",
+            config: { materialized: "table" },
+            depends_on: { nodes: [] },
+            columns: {
+              id: { name: "id", data_type: "integer" },
+              amount: { name: "amount", data_type: "integer" },
+              status: { name: "status", data_type: "varchar" },
+            },
+          },
+        },
+        sources: {},
+      }),
+    )
+    return { tmp, runner: createDispatcherRunner({ manifestPath }) }
+  }
+
+  async function review(oldSql: string, newSql: string, dialect = "snowflake") {
+    const { tmp, runner } = await reviewerWithManifest()
+    try {
+      const files: ChangedFile[] = [{ path: "models/marts/orders.sql", status: "modified", diff: "+changed" }]
+      const env = await runReview({
+        changedFiles: files,
+        config: { ...DEFAULT_REVIEW_CONFIG, reviewers: ["semantic_change"], dialect, ai: false },
+        rubric: DEFAULT_RUBRIC,
+        mode: "comment",
+        runner,
+        getContent: async (_f, side) => (side === "new" ? newSql : oldSql),
+        getCompiled: async (_f, side) => (side === "new" ? newSql : oldSql),
+        generatedAt: "2026-06-10T00:00:00Z",
+      })
+      return env
+    } finally {
+      await tmp[Symbol.asyncDispose]?.()
+    }
+  }
+
+  const eqFindings = (env: Awaited<ReturnType<typeof review>>) =>
+    env.findings.filter((f) => f.evidence?.tool === "altimate_core.equivalence")
+
+  test("real engine DECIDES a non-equivalent rewrite through the full pipeline", async () => {
+    // `amount > 5` vs `amount > 6` is a genuine row-changing predicate change.
+    const env = await review(
+      "select id from orders where amount > 5",
+      "select id from orders where amount > 6",
+    )
+    const findings = eqFindings(env)
+    expect(findings.length).toBeGreaterThan(0)
+    const f = findings[0]
+    expect(f.category).toBe("semantic_change")
+    expect((f.evidence?.result as any)?.equivalent).toBe(false)
+    // The engine names the concrete predicate difference (not a vague abstention).
+    const diffs = (f.evidence?.result as any)?.differences ?? []
+    expect(diffs.length).toBeGreaterThan(0)
+  })
+
+  test("real engine PROVES an equivalent refactor → lane stays silent (no false positive)", async () => {
+    // AND-conjunct reorder is provably equivalent; the reviewer must not nitpick it.
+    const env = await review(
+      "select id from orders where amount > 5 and status = 'x'",
+      "select id from orders where status = 'x' and amount > 5",
+    )
+    expect(eqFindings(env)).toEqual([])
+    expect(env.verdict).toBe("APPROVE")
+  })
+
+  test("identical compiled SQL → equivalence lane is skipped entirely", async () => {
+    const sql = "select id from orders where amount > 5"
+    const env = await review(sql, sql)
+    expect(eqFindings(env)).toEqual([])
+    expect(env.verdict).toBe("APPROVE")
+  })
+
+  test("column projection change (drop a column) is decided NOT equivalent", async () => {
+    const env = await review(
+      "select id, amount from orders",
+      "select id from orders",
+    )
+    const findings = eqFindings(env)
+    expect(findings.length).toBeGreaterThan(0)
+    expect((findings[0].evidence?.result as any)?.equivalent).toBe(false)
+  })
+
+  test("DEFAULT config dialect (empty string) still DECIDES — no engine throw", async () => {
+    // ReviewConfig.dialect defaults to "". The engine throws on an unknown dialect
+    // "", so without coercion the lane would abstain on EVERY change under the
+    // default config. This drives the full pipeline with dialect="" and asserts
+    // the non-equivalent change is still caught (decided), proving the coercion.
+    const env = await review(
+      "select id from orders where amount > 5",
+      "select id from orders where amount > 6",
+      "", // <- default ReviewConfig.dialect
+    )
+    const findings = eqFindings(env)
+    expect(findings.length).toBeGreaterThan(0)
+    expect((findings[0].evidence?.result as any)?.equivalent).toBe(false)
+  })
+})
diff --git a/packages/opencode/test/altimate/review-runner.test.ts b/packages/opencode/test/altimate/review-runner.test.ts
index 40e38c8ad8..538b2d7631 100644
--- a/packages/opencode/test/altimate/review-runner.test.ts
+++ b/packages/opencode/test/altimate/review-runner.test.ts
@@ -88,3 +88,77 @@ describe("review manifest loading", () => {
     expect(seenParams.dialect).toBe("duckdb")
   })
 })
+
+describe("runner honors engine `decidable` flag (core 0.5.1)", () => {
+  // Manifest with a typed column so resolveSchema() yields a non-null schema
+  // (the runner abstains when no schema is available, independent of decidable).
+  async function runnerWithManifest() {
+    const tmp = await tmpdir()
+    const manifestPath = path.join(tmp.path, "manifest.json")
+    writeFileSync(
+      manifestPath,
+      JSON.stringify({
+        metadata: { adapter_type: "duckdb" },
+        nodes: {
+          "model.demo.orders": {
+            resource_type: "model",
+            name: "orders",
+            original_file_path: "models/orders.sql",
+            config: { materialized: "table" },
+            depends_on: { nodes: [] },
+            columns: { id: { name: "id", data_type: "integer" } },
+          },
+        },
+        sources: {},
+      }),
+    )
+    return { tmp, runner: createDispatcherRunner({ manifestPath }) }
+  }
+
+  function mockEquivalence(data: Record<string, any>) {
+    dispatcherSpy = spyOn(Dispatcher, "call").mockImplementation((async () => ({
+      success: true,
+      data,
+    })) as any)
+  }
+
+  test("decidable=false with NO validation errors → abstains (decided:false)", async () => {
+    const { tmp, runner } = await runnerWithManifest()
+    // The engine says equivalent=true but flags the comparison as not decidable.
+    // We must NOT clear the change as safe on an authoritative abstention.
+    mockEquivalence({ equivalent: true, confidence: 0.95, differences: [], validation_errors: [], decidable: false })
+    const result = await runner.equivalence("select id from orders", "select id from orders", "duckdb")
+    expect(result).toEqual({ decided: false })
+    await tmp[Symbol.asyncDispose]?.()
+  })
+
+  test("decidable=true → decided verdict is returned", async () => {
+    const { tmp, runner } = await runnerWithManifest()
+    mockEquivalence({ equivalent: true, confidence: 0.95, differences: [], validation_errors: [], decidable: true })
+    const result = await runner.equivalence("select id from orders", "select id from orders", "duckdb")
+    expect(result).toEqual({ decided: true, equivalent: true, differences: [], confidence: "high" })
+    await tmp[Symbol.asyncDispose]?.()
+  })
+
+  test("decidable absent (0.4.0 legacy shape) → still decided (backward compatible)", async () => {
+    const { tmp, runner } = await runnerWithManifest()
+    mockEquivalence({ equivalent: true, confidence: 0.95, differences: [], validation_errors: [] })
+    const result = await runner.equivalence("select id from orders", "select id from orders", "duckdb")
+    expect(result).toEqual({ decided: true, equivalent: true, differences: [], confidence: "high" })
+    await tmp[Symbol.asyncDispose]?.()
+  })
+
+  test("decidable=false overrides even a non-equivalent verdict (no false block)", async () => {
+    const { tmp, runner } = await runnerWithManifest()
+    mockEquivalence({
+      equivalent: false,
+      confidence: 0.5,
+      differences: [{ description: "maybe changed", severity: "semantic" }],
+      validation_errors: [],
+      decidable: false,
+    })
+    const result = await runner.equivalence("select id from orders", "select 1 as id from orders", "duckdb")
+    expect(result).toEqual({ decided: false })
+    await tmp[Symbol.asyncDispose]?.()
+  })
+})