Add AI scoring fields to Application and Review models, and update grading logic

Author: rdesai14

services/registration/src/models/application.ts

@@ -88,6 +88,7 @@ export interface Application extends mongoose.Document {
   createdAt: Date;
   updatedAt: Date;
   finalScore: number;
+  finalAiScore?: number;
 }
 
 const applicationSchema = new Schema<Application>(
@@ -309,6 +310,11 @@ const applicationSchema = new Schema<Application>(
     finalScore: {
       type: Number,
     },
+    // Average of the graders' AI likelihood ratings across this application's reviews.
+    // Recorded for data collection only; nothing currently acts on it.
+    finalAiScore: {
+      type: Number,
+    },
   },
   {
     timestamps: true,

services/registration/src/models/review.ts

@@ -8,6 +8,7 @@ export interface Review extends mongoose.Document {
   essayId: Types.ObjectId;
   score: number;
   adjustedScore: number;
+  aiScore?: number;
   timestamp: Date;
 }
 
@@ -40,6 +41,13 @@ const reviewSchema = new Schema<Review>({
     type: Number,
     required: true,
   },
+  // Grader's rating of how likely the essay was written by AI. Optional so that reviews
+  // submitted before this field existed (and any client not yet sending it) remain valid.
+  aiScore: {
+    type: Number,
+    min: 1,
+    max: 3,
+  },
   timestamp: {
     type: Date,
     required: true,

services/registration/src/routes/grading.ts

@@ -15,6 +15,11 @@ import { getCalibrationMapping } from "../common/adjustScores";
 const MAX_REVIEWS_PER_ESSAY = 2;
 // NOTE: No. of essays for each application. As such, will need to be updated whenever we add/remove essays.
 const ESSAY_COUNT = 4;
+// Graders rate how likely an essay was written by AI alongside the normal score, so that they can
+// grade the essay on its merits and record the suspicion separately.
+// 1 = no chance of AI, 2 = some hints and sounds kinda like AI, 3 = fairly confident this is AI.
+const AI_SCORE_MIN = 1;
+const AI_SCORE_MAX = 3;
 
 type AggregatedEssay = {
   applicationId: string;
@@ -224,6 +229,18 @@ gradingRouter.route("/actions/submit-review").post(
       throw new BadRequestError("One or more of the method body arguments is missing.");
     }
 
+    // Optional so clients that haven't been updated yet can still submit reviews.
+    const { aiScore } = req.body;
+    if (
+      aiScore !== undefined &&
+      aiScore !== null &&
+      (!Number.isInteger(aiScore) || aiScore < AI_SCORE_MIN || aiScore > AI_SCORE_MAX)
+    ) {
+      throw new BadRequestError(
+        `AI score must be an integer between ${AI_SCORE_MIN} and ${AI_SCORE_MAX}`
+      );
+    }
+
     const gradingGroup = req.body.gradingGroup as GradingGroupType;
 
     let criteriaScores = grader.calibrationScores.find(
@@ -294,6 +311,7 @@ gradingRouter.route("/actions/submit-review").post(
         score: req.body.score,
         timestamp: new Date(),
         adjustedScore,
+        aiScore,
       });
 
       const allEssayReviews = await ReviewModel.find({
@@ -305,6 +323,17 @@ gradingRouter.route("/actions/submit-review").post(
         application.gradingComplete = true;
         const sumScores = allEssayReviews.reduce((prev, review) => prev + review.adjustedScore, 0);
         application.finalScore = sumScores / allEssayReviews.length;
+
+        // Average only over the reviews that actually submitted an AI score, so reviews from
+        // before this field existed don't pull the average toward zero.
+        const aiScores = allEssayReviews
+          .map(review => review.aiScore)
+          .filter((score): score is number => score !== undefined && score !== null);
+        if (aiScores.length > 0) {
+          application.finalAiScore =
+            aiScores.reduce((prev, score) => prev + score, 0) / aiScores.length;
+        }
+
         await application.save();
       }