fix(pipeline): split gate evaluations from revision rounds in cost estimate

Previously, evaluationsCompleted was passed as a single count and the cost
formula charged (critic + director) per unit. This overcounted when the gate
evaluated but didn't revise (score >= 7 first try): charged for a director
call that never happened.

Now estimateCost takes separate gateEvaluations and revisionRounds params.
Cost formula: evaluations * critic_cost + revisions * director_cost.

Author: Talha Jubair Siam

src/cli/cost-estimator.test.ts

@@ -119,17 +119,28 @@ describe("estimateCost", () => {
     expect(fal.videoCost).toBeGreaterThan(gemini.videoCost);
   });
 
-  it("includes revision cost when revisionRounds > 0", () => {
+  it("includes revision cost with separate evaluation and revision counts", () => {
     const score = makeScore([{ visual_type: "ai_image", script_line: "Test" }]);
-    const noRevision = estimateCost(score, "gemini", "elevenlabs", undefined, "anthropic", "bundled", 0);
-    const withRevision = estimateCost(score, "gemini", "elevenlabs", undefined, "anthropic", "bundled", 2);
-    expect(noRevision.revisionCost).toBe(0);
-    expect(noRevision.details.revisionRounds).toBe(0);
-    expect(withRevision.revisionCost).toBeGreaterThan(0);
-    expect(withRevision.details.revisionRounds).toBe(2);
-    expect(withRevision.totalCost).toBeGreaterThan(noRevision.totalCost);
-    // Revision cost should be approximately 2 × (critic + director) calls
-    expect(withRevision.revisionCost).toBeCloseTo(withRevision.totalCost - noRevision.totalCost, 4);
+    // No gate activity
+    const noGate = estimateCost(score, "gemini", "elevenlabs", undefined, "anthropic", "bundled", 0, 0);
+    expect(noGate.revisionCost).toBe(0);
+    expect(noGate.details.gateEvaluations).toBe(0);
+    expect(noGate.details.revisionRounds).toBe(0);
+
+    // Gate only (score >= 7 on first try): 1 eval, 0 revisions
+    const gateOnly = estimateCost(score, "gemini", "elevenlabs", undefined, "anthropic", "bundled", 1, 0);
+    expect(gateOnly.revisionCost).toBeGreaterThan(0);
+    expect(gateOnly.details.gateEvaluations).toBe(1);
+    expect(gateOnly.details.revisionRounds).toBe(0);
+
+    // Full revision: 2 evals + 1 revision
+    const fullRevision = estimateCost(score, "gemini", "elevenlabs", undefined, "anthropic", "bundled", 2, 1);
+    expect(fullRevision.revisionCost).toBeGreaterThan(gateOnly.revisionCost);
+    expect(fullRevision.details.gateEvaluations).toBe(2);
+    expect(fullRevision.details.revisionRounds).toBe(1);
+
+    // Gate-only should cost less than a full revision (no director call)
+    expect(gateOnly.revisionCost).toBeLessThan(fullRevision.revisionCost);
   });
 
   it("uses gemini LLM pricing when llmProvider is gemini", () => {

src/cli/cost-estimator.ts

@@ -18,6 +18,7 @@ export interface CostBreakdown {
   totalCost: number;
   details: {
     llmCalls: number;
+    gateEvaluations: number;
     revisionRounds: number;
     ttsCharacters: number;
     aiImages: number;
@@ -98,6 +99,7 @@ export function estimateCost(
   videoProvider?: VideoProviderKey,
   llmProvider: LLMProviderKey = "anthropic",
   musicProvider: MusicProviderKey = "bundled",
+  gateEvaluations = 0,
   revisionRounds = 0,
 ): CostBreakdown {
   const aiImageScenes = score.scenes.filter((s) => s.visual_type === "ai_image").length;
@@ -118,10 +120,10 @@ export function estimateCost(
     callCost(TOKEN_ESTIMATES.critic) +
     aiImages * callCost(TOKEN_ESTIMATES.imagePrompter);
 
-  // Revision cost: each evaluation = 1 critic call + 1 director-sized revise call.
-  // Slightly undercounts because revise prompts include the original score JSON (~1-3K extra tokens),
-  // but using the director estimate is a reasonable approximation.
-  const revisionCost = revisionRounds * (callCost(TOKEN_ESTIMATES.critic) + callCost(TOKEN_ESTIMATES.creativeDirector));
+  // Quality gate cost: critic calls for evaluation + director calls for revision.
+  // Split because evaluations > revisions (gate always evaluates, only revises if score < 7).
+  // Slightly undercounts revise calls because the prompt includes the original score JSON (~1-3K extra tokens).
+  const revisionCost = gateEvaluations * callCost(TOKEN_ESTIMATES.critic) + revisionRounds * callCost(TOKEN_ESTIMATES.creativeDirector);
   const ttsPerChar = PRICING.ttsPerChar[ttsProvider];
   const ttsCost = ttsCharacters * ttsPerChar;
   const perImage = imageProvider === "openai" ? PRICING.openaiPerImage : PRICING.geminiPerImage;
@@ -161,7 +163,7 @@ export function estimateCost(
 
   return {
     llmCost, revisionCost, ttsCost, imageCost, videoCost, musicCost, totalCost,
-    details: { llmCalls, revisionRounds, ttsCharacters, aiImages, aiVideos: aiVideoScenes },
+    details: { llmCalls, gateEvaluations, revisionRounds, ttsCharacters, aiImages, aiVideos: aiVideoScenes },
     perScene,
   };
 }
@@ -177,7 +179,10 @@ export function formatCostEstimate(
     `  LLM:    $${breakdown.llmCost.toFixed(4)} (${breakdown.details.llmCalls} calls)`,
   ];
   if (breakdown.revisionCost > 0) {
-    lines.push(`  Gate:   $${breakdown.revisionCost.toFixed(4)} (${breakdown.details.revisionRounds} quality gate eval${breakdown.details.revisionRounds !== 1 ? "s" : ""})`);
+    const evals = breakdown.details.gateEvaluations;
+    const revs = breakdown.details.revisionRounds;
+    const detail = revs > 0 ? `${evals} eval${evals !== 1 ? "s" : ""}, ${revs} revision${revs !== 1 ? "s" : ""}` : `${evals} eval${evals !== 1 ? "s" : ""}`;
+    lines.push(`  Gate:   $${breakdown.revisionCost.toFixed(4)} (${detail})`);
   }
   lines.push(
     `  TTS:    $${breakdown.ttsCost.toFixed(4)} (${breakdown.details.ttsCharacters} chars)`,

src/pipeline/orchestrator-callbacks.test.ts

@@ -82,7 +82,7 @@ describe("createCliCallbacks", () => {
         totalCost: 0.1,
         videoCost: 0,
         musicCost: 0,
-        details: { llmCalls: 3, revisionRounds: 0, ttsCharacters: 500, aiImages: 2, aiVideos: 0 },
+        details: { llmCalls: 3, gateEvaluations: 0, revisionRounds: 0, ttsCharacters: 500, aiImages: 2, aiVideos: 0 },
       },
       "gemini",
     );

src/pipeline/orchestrator.ts

@@ -510,8 +510,9 @@ function buildPipelineWorkflow(
       }
 
       // Cost estimation (uses the final revised score for accurate scene counts)
-      // Pass evaluationsCompleted so the cost includes the gate evaluation + any revision rounds
-      const costBreakdown = estimateCost(score, opts.imageProvider, opts.ttsProvider, opts.videoProvider, opts.llm.id, opts.musicProviderKey, evaluationsCompleted);
+      // Pass evaluations and revisions separately: evaluations count critic calls,
+      // revisions count director calls (evaluations >= revisions since gate always evaluates)
+      const costBreakdown = estimateCost(score, opts.imageProvider, opts.ttsProvider, opts.videoProvider, opts.llm.id, opts.musicProviderKey, evaluationsCompleted, revisionRoundsCompleted);
       directorResult.costBreakdown = costBreakdown;
       log.totalCost = { estimated: costBreakdown.totalCost };