Record the phase 5 ScoreSaber merge and map OST hashes in the sync

ScoreSaber names official levels "ost_<LevelId>" where BeatLeader uses
"<levelid>"; ssMapHash applies that for leaderboard snapshots and maps.
The plan records the merge results and where it differs from the steps.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/plan.md

@@ -733,6 +733,57 @@ profile and scores, and a replay opens in ArcViewer.
 
 ### Phase 5: Convert ScoreSaber scores and merge
 
+**Built and checked on a scratch copy 2026-11-09; committed, deploy
+pending.** Server `50f7df6` (`Tools/ArchiveImporter/ScoreMerge.cs`, rules in
+the fork's `MIRROR.md`), website `5522f245` (SS badge, `/watch/` for `.dat`),
+exporter `ssMapHash` in this repo. All pushed. Result on the local archive: 3,710 leaderboards with one score
+each (3,566 BL + 141 SS-only BL leaderboards + 3 DLC levels BL lacks). Winners: 3,564 BL main,
+1 BL legacy, 139 SS legacy, 6 SS main; 2,022 SS scores were the same play as
+a BL score and dropped; 20 BL scores got an improvement recomputed against an
+earlier SS play. An unchanged rerun writes 0 rows. What differs from the
+steps below:
+
+- Step 1: **no new columns.** Provenance goes in `Platform`
+  (`scoresaber,<SS player id>`), and `Id` ≥ 1e9 marks SS scores. Columns
+  would have needed a models-fork change, a migration, and edits to the ~90
+  score projections before the website saw them; `Platform` already reaches
+  every score response.
+- Step 2: OST levels are `ost_<LevelId>` on SS and `<levelid>` on BL; 5 of 8
+  resolve that way. Me Porto Bonito, Nuthin' But A "G" Thang and Stressed Out
+  (3 scores) aren't on BL (404), so the importer builds a song and leaderboard
+  from the SS item (`<LevelId><diff><mode>` ids, no ratings, accuracy 0).
+  The exporter now maps `ost_` hashes too (`scoresaber.ts` `ssMapHash`).
+  HMD and controller come from SS `deviceHmd`/`deviceControllerLeft`. SS
+  replays are served from `storage/ssreplays` → `archive/scoresaber`, so the
+  URL is `/cdn/ssreplays/<player>/replays/<id>.dat`.
+- **Pp isn't upstream's.** Neither the fork's `rebalanced-pp` formula
+  (median 0.94x, range 0.69–1.16x) nor upstream `master`'s (0.81x)
+  reproduces the pp on archived BL scores; live BL's code is private. So the
+  fold compares every candidate by **locally computed** pp, which keeps the
+  order right, and BL winners keep upstream's pp. The 70 SS winners on
+  ranked maps show local pp, and improvement pp deltas are local on both
+  sides. Matching pp first sent two BL scores (KillerToy, Make Me) to worse
+  SS plays.
+- Step 3: same-play duplicates (equal base score, ≤ 2 min apart; observed
+  −30 s…+9 s) keep the BL copy. Each upstream `scoreImprovement` is turned
+  back into the best it replaced and added as a candidate, so upstream's
+  improvement stays exactly when that rebuilt best is also the fold's previous
+  best. Older archived BL bests are candidates too.
+- Step 5: the badge sits next to the rank in both `ScoreRank.svelte` (rank 0
+  already renders "-"). `/watch/` takes the map hash from the score
+  (`/score/{id}` → `song.hash`) for `.dat` replays, from `?scoreId=` or
+  1e9 + the id in the filename. Expanding an SS row shows its leaderboard and
+  no stats tab; the page logs a JSON error for the 404 score stats, harmless.
+- Check: The Goddess Figure (Normal) is the case where a 2022 legacy SS
+  play beats the BL score; it shows with the SS badge, and the BL Hard score
+  there now reads "+76,821 vs 4 years ago" (vs the legacy SS play). An SS
+  replay redirects to ArcViewer with the `.dat` and local zip and loads the map;
+  watching it play still needs a real browser.
+
+To deploy: nix-pleb bumps `plebleader-server`, `plebleader-website` and
+`plebsaber-stream` (sync CLI); switch. The next `plebleader-sync`
+run re-imports with the merge.
+
 ScoreSaber scores become BeatLeader `Score` rows, and each leaderboard keeps
 exactly one score. BeatLeader already shows how a score compares with the
 previous best (`ScoreImprovement`), so after the merge that comparison covers

scripts/archive/cli.ts

@@ -21,7 +21,7 @@ import * as path from 'node:path';
 import { ARCHIVE_DIR, BL_API, DEFAULT_MAPS_SINCE, LEGACY_PLAYER_ID, MAIN_PLAYER_ID, PLAYER_IDS, TOKENS_FILE, USER_AGENT, loadDotEnv, parseSince } from './config.ts';
 import { getAccessToken, login, readTokens, refresh } from './auth.ts';
 import { archivedScores, blHashes, exportLeaderboards, exportProfile, exportReplays, exportScoreStats, exportScores, newestScoreTime } from './beatleader.ts';
-import { archivedSSScores, exportSSPlayer, exportSSReplays, exportSSScores, newestSSScoreTime, ssHashes } from './scoresaber.ts';
+import { archivedSSScores, exportSSPlayer, exportSSReplays, exportSSScores, newestSSScoreTime, ssHashes, ssMapHash } from './scoresaber.ts';
 import { exportMaps } from './beatsaver.ts';
 import { archivePath, exists, listFiles, readState, updateState } from './store.ts';
 
@@ -83,7 +83,7 @@ function playedHashes(since: number): string[] {
 function mapHashes(mapsSince: number): string[] {
 	const legacyWithReplay = archivedSSScores(LEGACY_PLAYER_ID, 0)
 		.filter((it) => exists(archivePath('scoresaber', LEGACY_PLAYER_ID, 'replays', `${it.score.id}.dat`)))
-		.map((it) => it.leaderboard.songHash.toLowerCase());
+		.map(ssMapHash);
 	return [...new Set([...playedHashes(mapsSince), ...legacyWithReplay])].sort();
 }
 
@@ -137,7 +137,7 @@ async function sync(opts: { ratings: boolean; ssReplays: boolean }): Promise<num
 	await exportSSPlayer(MAIN_PLAYER_ID);
 	const ssSince = Math.max(0, newestSSScoreTime(MAIN_PLAYER_ID) - SYNC_OVERLAP_S);
 	const newSS = await exportSSScores(MAIN_PLAYER_ID, ssSince);
-	const newSSHashes = [...new Set(newSS.map((it) => it.leaderboard.songHash.toLowerCase()))].sort();
+	const newSSHashes = [...new Set(newSS.map(ssMapHash))].sort();
 	await exportLeaderboards(newSSHashes.filter((h) => !newBLHashes.includes(h)));
 	if (opts.ssReplays) await exportSSReplays(MAIN_PLAYER_ID, ssSince);
 

scripts/archive/scoresaber.ts

@@ -102,6 +102,14 @@ function seedFromScraper(playerId: string, items: SSItem[]): void {
 	if (copied) console.log(`  copied ${copied} replays from ${dir}`);
 }
 
+/**
+ * The map hash BeatLeader and the archive use for a ScoreSaber item: lowercase, and for official
+ * levels the level id without ScoreSaber's "ost_" prefix (BeatLeader: "beatsaber", SS: "ost_BeatSaber").
+ */
+export function ssMapHash(it: SSItem): string {
+	return it.leaderboard.songHash.toLowerCase().replace(/^ost_/, '');
+}
+
 export function ssHashes(playerId: string, sinceUnix = 0): string[] {
-	return [...new Set(archivedSSScores(playerId, sinceUnix).map((it) => it.leaderboard.songHash.toLowerCase()))].sort();
+	return [...new Set(archivedSSScores(playerId, sinceUnix).map(ssMapHash))].sort();
 }