Merge pull request #86 from quantified-uncertainty/new-questions

feat: firstSeen, new questions in graphql

prisma/migrations/20220519104640_new_timestamps/migration.sql

@@ -0,0 +1,28 @@
+-- questions
+ALTER TABLE "questions"
+  ADD COLUMN    "fetched" TIMESTAMP(6),
+  ADD COLUMN    "first_seen" TIMESTAMP(6);
+
+UPDATE "questions"
+  SET "fetched" = "timestamp", "first_seen" = "timestamp";
+
+ALTER TABLE "questions"
+  ALTER COLUMN "fetched" SET NOT NULL,
+  ALTER COLUMN "first_seen" SET NOT NULL;
+
+-- history
+ALTER TABLE "history"
+  ADD COLUMN    "fetched" TIMESTAMP(6);
+
+UPDATE "history" SET "fetched" = "timestamp";
+
+ALTER TABLE "history"
+  ALTER COLUMN "fetched" SET NOT NULL;
+
+-- populate first_seen
+UPDATE questions
+  SET "first_seen" = h.fs
+  FROM (
+    SELECT id, MIN(fetched) AS fs FROM history GROUP BY id
+  ) as h
+  WHERE questions.id = h.id;

prisma/migrations/20220520195517_indices/migration.sql

@@ -0,0 +1,14 @@
+-- CreateIndex
+CREATE INDEX "history_platform_idx" ON "history"("platform");
+
+-- CreateIndex
+CREATE INDEX "history_fetched_idx" ON "history"("fetched");
+
+-- CreateIndex
+CREATE INDEX "questions_platform_idx" ON "questions"("platform");
+
+-- CreateIndex
+CREATE INDEX "questions_fetched_idx" ON "questions"("fetched");
+
+-- CreateIndex
+CREATE INDEX "questions_first_seen_idx" ON "questions"("first_seen");

prisma/schema.prisma

@@ -24,24 +24,6 @@ model Dashboard {
   @@map("dashboards")
 }
 
-model History {
-  id                String
-  idref             String?
-  question          Question? @relation(fields: [idref], references: [id], onDelete: SetNull, onUpdate: Restrict)
-  title             String
-  url               String
-  platform          String
-  description       String
-  options           Json
-  timestamp         DateTime  @db.Timestamp(6)
-  qualityindicators Json
-  extra             Json
-  pk                Int       @id @default(autoincrement())
-
-  @@index([id])
-  @@map("history")
-}
-
 model Question {
   /// E.g. "fantasyscotus-580"
   id          String @id
@@ -68,7 +50,9 @@ model Question {
   //   }
   // ]
   options   Json
-  timestamp DateTime @db.Timestamp(6)
+  timestamp DateTime @db.Timestamp(6) // deprecated
+  fetched   DateTime @db.Timestamp(6)
+  firstSeen DateTime @map("first_seen") @db.Timestamp(6)
 
   // {
   //   "numforecasts": 120,
@@ -80,9 +64,33 @@ model Question {
   onFrontpage FrontpageId?
   history     History[]
 
+  @@index([platform])
+  @@index([fetched])
+  @@index([firstSeen])
   @@map("questions")
 }
 
+model History {
+  id                String
+  idref             String?
+  question          Question? @relation(fields: [idref], references: [id], onDelete: SetNull, onUpdate: Restrict)
+  title             String
+  url               String
+  platform          String
+  description       String
+  options           Json
+  timestamp         DateTime  @db.Timestamp(6) // deprecated
+  fetched           DateTime  @db.Timestamp(6)
+  qualityindicators Json
+  extra             Json
+  pk                Int       @id @default(autoincrement())
+
+  @@index([id])
+  @@index([platform])
+  @@index([fetched])
+  @@map("history")
+}
+
 model FrontpageId {
   question Question @relation(fields: [id], references: [id])
   id       String   @unique

src/backend/flow/doEverything.ts

@@ -1,9 +1,12 @@
-import { platforms } from "../platforms/registry";
+import { getPlatforms } from "../platforms/registry";
 import { executeJobByName } from "./jobs";
 
 /* Do everything */
 export async function doEverything() {
-  let jobNames = [...platforms.map((platform) => platform.name), "algolia"];
+  let jobNames = [
+    ...getPlatforms().map((platform) => platform.name),
+    "algolia",
+  ];
 
   console.log("");
   console.log("");

src/backend/flow/jobs.ts

@@ -1,7 +1,7 @@
 import { doEverything } from "../flow/doEverything";
 import { rebuildFrontpage } from "../frontpage";
 import { processPlatform } from "../platforms";
-import { platforms } from "../platforms/registry";
+import { getPlatforms } from "../platforms/registry";
 import { rebuildAlgoliaDatabase } from "../utils/algolia";
 import { sleep } from "../utils/sleep";
 
@@ -14,7 +14,7 @@ interface Job<ArgNames extends string = ""> {
 }
 
 export const jobs: Job<string>[] = [
-  ...platforms.map((platform) => ({
+  ...getPlatforms().map((platform) => ({
     name: platform.name,
     message: `Download predictions from ${platform.name}`,
     ...(platform.version === "v2" ? { args: platform.fetcherArgs } : {}),

src/backend/frontpage.ts

@@ -26,7 +26,7 @@ export async function rebuildFrontpage() {
         AND questions.description != ''
         AND JSONB_ARRAY_LENGTH(questions.options) > 0
       GROUP BY questions.id
-      HAVING COUNT(DISTINCT history.timestamp) >= 7
+      HAVING COUNT(DISTINCT history.fetched) >= 7
       ORDER BY RANDOM() LIMIT 50
     `;
 

src/backend/platforms/givewellopenphil.ts

@@ -79,7 +79,7 @@ export const givewellopenphil: Platform = {
     const dataWithDate = data.map((datum: any) => ({
       ...datum,
       platform: platformName,
-      timestamp: new Date("2021-02-23"),
+      // timestamp: new Date("2021-02-23"),
     }));
     return dataWithDate;
   },

src/backend/platforms/guesstimate.ts

@@ -2,9 +2,8 @@ import axios from "axios";
 
 import { Question } from "@prisma/client";
 
-import { prisma } from "../database/prisma";
-import { AlgoliaQuestion } from "../utils/algolia";
-import { FetchedQuestion, Platform, prepareQuestion } from "./";
+import { AlgoliaQuestion, questionToAlgoliaQuestion } from "../utils/algolia";
+import { FetchedQuestion, Platform, prepareQuestion, upsertSingleQuestion } from "./";
 
 /* Definitions */
 const searchEndpoint =
@@ -55,10 +54,12 @@ async function search(query: string): Promise<AlgoliaQuestion[]> {
   const models: any[] = response.data.hits;
   const mappedModels: AlgoliaQuestion[] = models.map((model) => {
     const q = modelToQuestion(model);
-    return {
+    return questionToAlgoliaQuestion({
       ...q,
-      timestamp: String(q.timestamp),
-    };
+      fetched: new Date(),
+      timestamp: new Date(),
+      firstSeen: new Date(),
+    });
   });
 
   // filter for duplicates. Surprisingly common.
@@ -76,12 +77,8 @@ async function search(query: string): Promise<AlgoliaQuestion[]> {
 
 const fetchQuestion = async (id: number): Promise<Question> => {
   const response = await axios({ url: `${apiEndpoint}/spaces/${id}` });
-  let q = modelToQuestion(response.data);
-  return await prisma.question.upsert({
-    where: { id: q.id },
-    create: q,
-    update: q,
-  });
+  const q = modelToQuestion(response.data);
+  return await upsertSingleQuestion(q);
 };
 
 export const guesstimate: Platform & {

src/backend/platforms/index.ts

@@ -28,9 +28,14 @@ export interface QualityIndicators {
 
 export type FetchedQuestion = Omit<
   Question,
-  "extra" | "qualityindicators" | "timestamp" | "platform" | "options"
+  | "extra"
+  | "qualityindicators"
+  | "fetched"
+  | "firstSeen"
+  | "timestamp"
+  | "platform"
+  | "options"
 > & {
-  timestamp?: Date;
   extra?: object; // required in DB but annoying to return empty; also this is slightly stricter than Prisma's JsonValue
   options: QuestionOption[]; // stronger type than Prisma's JsonValue
   qualityindicators: Omit<QualityIndicators, "stars">; // slightly stronger type than Prisma's JsonValue
@@ -78,8 +83,14 @@ export type Platform<ArgNames extends string = ""> = {
 // So here we build a new type which should be ok to use both in place of prisma's Question type and as an input to its update or create methods.
 type PreparedQuestion = Omit<
   Question,
-  "extra" | "qualityindicators" | "options"
+  | "extra"
+  | "qualityindicators"
+  | "options"
+  | "fetched"
+  | "firstSeen"
+  | "timestamp"
 > & {
+  fetched: Date;
   extra: NonNullable<Question["extra"]>;
   qualityindicators: NonNullable<Question["qualityindicators"]>;
   options: NonNullable<Question["options"]>;
@@ -91,8 +102,8 @@ export const prepareQuestion = (
 ): PreparedQuestion => {
   return {
     extra: {},
-    timestamp: new Date(),
     ...q,
+    fetched: new Date(),
     platform: platform.name,
     qualityindicators: {
       ...q.qualityindicators,
@@ -101,6 +112,21 @@ export const prepareQuestion = (
   };
 };
 
+export const upsertSingleQuestion = async (
+  q: PreparedQuestion
+): Promise<Question> => {
+  return await prisma.question.upsert({
+    where: { id: q.id },
+    create: {
+      ...q,
+      firstSeen: new Date(),
+      timestamp: q.fetched, // deprecated
+    },
+    update: q,
+  });
+  // TODO - update history?
+};
+
 export const processPlatform = async <T extends string = "">(
   platform: Platform<T>,
   args?: { [k in T]: string }
@@ -144,9 +170,9 @@ export const processPlatform = async <T extends string = "">(
 
   for (const q of fetchedQuestions.map((q) => prepareQuestion(q, platform))) {
     if (oldIdsSet.has(q.id)) {
+      // TODO - check if question has changed for better performance
       updatedQuestions.push(q);
     } else {
-      // TODO - check if question has changed for better performance
       createdQuestions.push(q);
     }
   }
@@ -154,7 +180,11 @@ export const processPlatform = async <T extends string = "">(
   const stats: { created?: number; updated?: number; deleted?: number } = {};
 
   await prisma.question.createMany({
-    data: createdQuestions,
+    data: createdQuestions.map((q) => ({
+      ...q,
+      firstSeen: new Date(),
+      timestamp: q.fetched, // deprecated
+    })),
   });
   stats.created = createdQuestions.length;
 
@@ -181,6 +211,7 @@ export const processPlatform = async <T extends string = "">(
   await prisma.history.createMany({
     data: [...createdQuestions, ...updatedQuestions].map((q) => ({
       ...q,
+      timestamp: q.fetched, // deprecated
       idref: q.id,
     })),
   });

src/backend/platforms/registry.ts

@@ -17,30 +17,43 @@ import { smarkets } from "./smarkets";
 import { wildeford } from "./wildeford";
 import { xrisk } from "./xrisk";
 
-export const platforms: Platform<string>[] = [
-  betfair,
-  fantasyscotus,
-  foretold,
-  givewellopenphil,
-  goodjudgment,
-  goodjudgmentopen,
-  guesstimate,
-  infer,
-  kalshi,
-  manifold,
-  metaculus,
-  polymarket,
-  predictit,
-  rootclaim,
-  smarkets,
-  wildeford,
-  xrisk,
-];
+// function instead of const array, this helps to fight circular dependencies
+export const getPlatforms = (): Platform<string>[] => {
+  return [
+    betfair,
+    fantasyscotus,
+    foretold,
+    givewellopenphil,
+    goodjudgment,
+    goodjudgmentopen,
+    guesstimate,
+    infer,
+    kalshi,
+    manifold,
+    metaculus,
+    polymarket,
+    predictit,
+    rootclaim,
+    smarkets,
+    wildeford,
+    xrisk,
+  ];
+};
+
+let _nameToLabelCache: { [k: string]: string } | undefined;
+export function platformNameToLabel(name: string): string {
+  if (!_nameToLabelCache) {
+    _nameToLabelCache = Object.fromEntries(
+      getPlatforms().map((platform) => [platform.name, platform.label])
+    );
+  }
+  return _nameToLabelCache[name] || name;
+}
 
 // get frontend-safe version of platforms data
 
 export const getPlatformsConfig = (): PlatformConfig[] => {
-  const platformsConfig = platforms.map((platform) => ({
+  const platformsConfig = getPlatforms().map((platform) => ({
     name: platform.name,
     label: platform.label,
     color: platform.color,

src/backend/platforms/smarkets.ts

@@ -162,7 +162,6 @@ async function processEventMarkets(event: any, ctx: Context) {
       url: "https://smarkets.com/event/" + market.event_id + market.slug,
       description: market.description,
       options,
-      timestamp: new Date(),
       qualityindicators: {},
       extra: {
         contracts,

src/backend/platforms/wildeford.ts

@@ -96,7 +96,8 @@ async function processPredictions(
       url: prediction["url"],
       description: prediction["Notes"] || "",
       options,
-      timestamp: new Date(Date.parse(prediction["Prediction Date"] + "Z")),
+      //// TODO - use `created` field for this
+      // timestamp: new Date(Date.parse(prediction["Prediction Date"] + "Z")),
       qualityindicators: {},
     };
     return result;

src/backend/utils/algolia.ts

@@ -3,16 +3,23 @@ import algoliasearch from "algoliasearch";
 import { Question } from "@prisma/client";
 
 import { prisma } from "../database/prisma";
-import { platforms } from "../platforms/registry";
+import { platformNameToLabel } from "../platforms/registry";
 
-let cookie = process.env.ALGOLIA_MASTER_API_KEY || "";
+const cookie = process.env.ALGOLIA_MASTER_API_KEY || "";
 const algoliaAppId = process.env.NEXT_PUBLIC_ALGOLIA_APP_ID || "";
 const client = algoliasearch(algoliaAppId, cookie);
 const index = client.initIndex("metaforecast");
 
-export type AlgoliaQuestion = Omit<Question, "timestamp"> & {
-  timestamp: string;
+export type AlgoliaQuestion = Omit<
+  Question,
+  "fetched" | "firstSeen" | "timestamp"
+> & {
+  timestamp?: string; // deprecated
+  fetched?: string;
+  firstSeen?: string;
   optionsstringforsearch?: string;
+  platformLabel: string;
+  objectID: string;
 };
 
 const getoptionsstringforsearch = (record: Question): string => {
@@ -26,23 +33,24 @@ const getoptionsstringforsearch = (record: Question): string => {
   return result;
 };
 
+export const questionToAlgoliaQuestion = (
+  question: Question
+): AlgoliaQuestion => {
+  return {
+    ...question,
+    fetched: question.fetched.toISOString(),
+    timestamp: question.timestamp.toISOString(), // deprecated
+    firstSeen: question.firstSeen.toISOString(),
+    platformLabel: platformNameToLabel(question.platform),
+    objectID: question.id,
+    optionsstringforsearch: getoptionsstringforsearch(question),
+  };
+};
+
 export async function rebuildAlgoliaDatabase() {
   const questions = await prisma.question.findMany();
 
-  const platformNameToLabel = Object.fromEntries(
-    platforms.map((platform) => [platform.name, platform.label])
-  );
-
-  const records: AlgoliaQuestion[] = questions.map(
-    (question, index: number) => ({
-      ...question,
-      timestamp: `${question.timestamp}`,
-      platformLabel:
-        platformNameToLabel[question.platform] || question.platform,
-      objectID: index,
-      optionsstringforsearch: getoptionsstringforsearch(question),
-    })
-  );
+  const records: AlgoliaQuestion[] = questions.map(questionToAlgoliaQuestion);
 
   if (await index.exists()) {
     console.log("Index exists");

src/graphql/schema/platforms.ts

@@ -1,5 +1,5 @@
 import { prisma } from "../../backend/database/prisma";
-import { platforms } from "../../backend/platforms/registry";
+import { getPlatforms } from "../../backend/platforms/registry";
 import { builder } from "../builder";
 
 export const PlatformObj = builder.objectRef<string>("Platform").implement({
@@ -20,7 +20,7 @@ export const PlatformObj = builder.objectRef<string>("Platform").implement({
           return "Guesstimate";
         }
         // kinda slow and repetitive, TODO - store a map {name => platform} somewhere and `getPlatform` util function?
-        const platform = platforms.find((p) => p.name === platformName);
+        const platform = getPlatforms().find((p) => p.name === platformName);
         if (!platform) {
           throw new Error(`Unknown platform ${platformName}`);
         }
@@ -36,10 +36,10 @@ export const PlatformObj = builder.objectRef<string>("Platform").implement({
             platform: platformName,
           },
           _max: {
-            timestamp: true,
+            fetched: true,
           },
         });
-        return res._max.timestamp;
+        return res._max.fetched;
       },
     }),
   }),
@@ -49,7 +49,7 @@ builder.queryField("platforms", (t) =>
   t.field({
     type: [PlatformObj],
     resolve: async (parent, args) => {
-      return platforms.map((platform) => platform.name);
+      return getPlatforms().map((platform) => platform.name);
     },
   })
 );

src/graphql/schema/questions.ts

@@ -70,8 +70,21 @@ const QuestionShapeInterface = builder
       }),
       timestamp: t.field({
         type: "Date",
-        description: "Timestamp at which metaforecast fetched the question",
-        resolve: (parent) => parent.timestamp,
+        description:
+          "Last timestamp at which metaforecast fetched the question",
+        deprecationReason: "Renamed to `fetched`",
+        resolve: (parent) => parent.fetched,
+      }),
+      fetched: t.field({
+        type: "Date",
+        description:
+          "Last timestamp at which metaforecast fetched the question",
+        resolve: (parent) => parent.fetched,
+      }),
+      fetchedStr: t.string({
+        description:
+          "Last timestamp at which metaforecast fetched the question, in ISO 8601 format",
+        resolve: (parent) => parent.fetched.toISOString(),
       }),
       qualityIndicators: t.field({
         type: QualityIndicatorsObj,
@@ -114,10 +127,20 @@ export const QuestionObj = builder.prismaObject("Question", {
       resolve: (parent) => (parent.extra as any)?.visualization, // used for guesstimate only, see searchGuesstimate.ts
       nullable: true,
     }),
+    firstSeen: t.field({
+      type: "Date",
+      description: "First timestamp at which metaforecast fetched the question",
+      resolve: (parent) => parent.firstSeen,
+    }),
+    firstSeenStr: t.string({
+      description:
+        "First timestamp at which metaforecast fetched the question, in ISO 8601 format",
+      resolve: (parent) => parent.firstSeen.toISOString(),
+    }),
     history: t.relation("history", {
       query: () => ({
         orderBy: {
-          timestamp: "asc",
+          fetched: "asc",
         },
       }),
     }),
@@ -130,7 +153,21 @@ builder.queryField("questions", (t) =>
       type: "Question",
       cursor: "id",
       maxSize: 1000,
-      resolve: (query) => prisma.question.findMany({ ...query }),
+      args: {
+        orderBy: t.arg({
+          type: builder.enumType("QuestionsOrderBy", {
+            values: ["FIRST_SEEN_DESC"] as const,
+          }),
+        }),
+      },
+      resolve: (query, parent, args) => {
+        return prisma.question.findMany({
+          ...query,
+          ...(args.orderBy === "FIRST_SEEN_DESC"
+            ? { orderBy: [{ firstSeen: "desc" }, { id: "asc" }] }
+            : {}), // TODO - explicit default order?
+        });
+      },
     },
     {},
     {}

src/graphql/schema/search.ts

@@ -63,7 +63,15 @@ builder.queryField("searchQuestions", (t) =>
 
       return results.map((q) => ({
         ...q,
-        timestamp: new Date(q.timestamp),
+        fetched: new Date(
+          q.fetched || q.timestamp || new Date().toISOString() // q.timestamp is deprecated, TODO - just use `q.fetched`
+        ),
+        timestamp: new Date(
+          q.fetched || q.timestamp || new Date().toISOString()
+        ),
+        firstSeen: new Date(
+          q.firstSeen || new Date().toISOString() // TODO - q.firstSeen is not yet populated in algolia
+        ),
       }));
     },
   })

src/pages/index.tsx

@@ -1,7 +1,7 @@
 import { GetServerSideProps, NextPage } from "next";
 import React from "react";
 
-import { getPlatformsConfig, platforms } from "../backend/platforms/registry";
+import { getPlatforms, getPlatformsConfig } from "../backend/platforms/registry";
 import { Layout } from "../web/common/Layout";
 import { Props, QueryParameters, SearchScreen } from "../web/search/components/SearchScreen";
 import { FrontpageDocument, SearchDocument } from "../web/search/queries.generated";
@@ -19,7 +19,7 @@ export const getServerSideProps: GetServerSideProps<Props> = async (
     query: "",
     starsThreshold: 2,
     forecastsThreshold: 0,
-    forecastingPlatforms: platforms.map((platform) => platform.name),
+    forecastingPlatforms: getPlatforms().map((platform) => platform.name),
   };
 
   const initialQueryParameters: QueryParameters = {

src/pages/secretEmbed.tsx

@@ -3,7 +3,7 @@
 import { GetServerSideProps, NextPage } from "next";
 import React from "react";
 
-import { platforms } from "../backend/platforms/registry";
+import { getPlatforms } from "../backend/platforms/registry";
 import { QuestionFragment } from "../web/fragments.generated";
 import { QuestionCard } from "../web/questions/components/QuestionCard";
 import { SearchDocument } from "../web/search/queries.generated";
@@ -23,7 +23,7 @@ export const getServerSideProps: GetServerSideProps<Props> = async (
     query: "",
     starsThreshold: 2,
     forecastsThreshold: 0,
-    forecastingPlatforms: platforms.map((platform) => platform.name),
+    forecastingPlatforms: getPlatforms().map((platform) => platform.name),
     ...urlQuery,
   };
 

src/web/worker/searchWithAlgolia.ts

@@ -96,6 +96,7 @@ export default async function searchWithAlgolia({
         title: "No search results match your query",
         url: "https://metaforecast.org",
         platform: "metaforecast",
+        platformLabel: "metaforecast",
         description: "Maybe try a broader query?",
         options: [
           {
@@ -109,7 +110,7 @@ export default async function searchWithAlgolia({
             type: "PROBABILITY",
           },
         ],
-        timestamp: `${new Date().toISOString().slice(0, 10)}`,
+        fetched: new Date().toISOString(),
         qualityindicators: {
           numforecasts: 1,
           numforecasters: 1,
@@ -126,8 +127,10 @@ export default async function searchWithAlgolia({
         title: `Did you mean: ${queryString}?`,
         url: "https://metaforecast.org/recursion?bypassEasterEgg=true",
         platform: "metaforecast",
+        platformLabel: "metaforecast",
         description:
           "Fatal error: Too much recursion. Click to proceed anyways",
+        fetched: new Date().toISOString(),
         options: [
           {
             name: "Yes",
@@ -140,7 +143,6 @@ export default async function searchWithAlgolia({
             type: "PROBABILITY",
           },
         ],
-        timestamp: `${new Date().toISOString().slice(0, 10)}`,
         qualityindicators: {
           numforecasts: 1,
           numforecasters: 1,
@@ -161,6 +163,7 @@ export default async function searchWithAlgolia({
       title: "No search results appear to match your query",
       url: "https://metaforecast.org",
       platform: "metaforecast",
+      platformLabel: "metaforecast",
       description: "Maybe try a broader query? That said, we could be wrong.",
       options: [
         {
@@ -174,7 +177,7 @@ export default async function searchWithAlgolia({
           type: "PROBABILITY",
         },
       ],
-      timestamp: `${new Date().toISOString().slice(0, 10)}`,
+      fetched: new Date().toISOString(),
       qualityindicators: {
         numforecasts: 1,
         numforecasters: 1,