Fix UK→GB geo normalisation + clear Stage 8 too-few-videos error

The Dove2 run on prod failed because every hashtag/search seed 400'd
with "Field input.proxyCountryCode must be equal to one of the allowed
values" — Apify uses ISO codes ("GB"), not the colloquial "UK" stored
on the brief. Only profile scrapes (which don't pass proxyCountryCode)
got through, leaving 24 videos and a 16% manifest gate.

Two layers of fix:
- Brief Zod schema transforms geo: trims, uppercases, maps "UK" → "GB".
  All briefs created or edited from now on are normalised at the form
  boundary.
- Stage 2 also normalises at actor-input time, as belt-and-braces for
  briefs already in the DB with "UK" written before this commit.

Plus a clear pre-flight error in Stage 8: when fewer than 5 videos made
it through analysis the trends schema literally can't be satisfied (each
trend needs ≥5 supporting_video_ids). Previously Claude tried, Zod
rejected with a 50-line "too_small" wall, and the operator was left
guessing. Now we throw a single sentence pointing at the actual cause:
the dataset is too small — adjust the brief and force re-run.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

v2/pipeline/stages/stage_2_pass1_scrape.ts

@@ -208,6 +208,15 @@ export async function runStage2Pass1Scrape(args: Stage2Args): Promise<{ ok: true
   // overshoot it slightly. Worth ~$3 of overshoot to save 10+ minutes.
   const PASS1_PARALLEL = 4;
 
+  // Apify rejects "UK" — its allowed-list uses ISO codes, where Britain is "GB".
+  // Briefs created before the schema normalisation in v2/server/schemas/brief.ts
+  // (or any other "colloquial" alpha-2 collision) need this safety net or every
+  // hashtag/search seed 400s with "Field input.proxyCountryCode must be equal..."
+  const proxyCountry = (() => {
+    const upper = (brief.geo || '').trim().toUpperCase();
+    return upper === 'UK' ? 'GB' : upper;
+  })();
+
   function buildActorInput(job: ScrapeJob): { actor: string; input: Record<string, unknown>; label: string } {
     const label = job.kind === 'hashtag'
       ? `hashtag:${job.tag} (${job.tier})`
@@ -223,7 +232,7 @@ export async function runStage2Pass1Scrape(args: Stage2Args): Promise<{ ok: true
           resultsPerPage: limits.resultsPerPage,
           shouldDownloadVideos: process.env.DISABLE_VIDEO_DOWNLOADS !== 'true',
           shouldDownloadCovers: true,
-          proxyCountryCode: brief.geo,
+          proxyCountryCode: proxyCountry,
           minPlayCount: brief.min_plays,
         },
       };
@@ -248,7 +257,7 @@ export async function runStage2Pass1Scrape(args: Stage2Args): Promise<{ ok: true
         resultsPerPage: limits.resultsPerPage,
         shouldDownloadVideos: process.env.DISABLE_VIDEO_DOWNLOADS !== 'true',
         shouldDownloadCovers: true,
-        proxyCountryCode: brief.geo,
+        proxyCountryCode: proxyCountry,
         minPlayCount: brief.min_plays,
       },
     };

v2/pipeline/stages/stage_8_trends.ts

@@ -223,6 +223,21 @@ export async function runStage8Trends(reportId: string, brief: BriefInput): Prom
 
   // Load per-video analyses + pass1 for KPI computation
   const analyses = loadAnalyses(reportId);
+
+  // Trend synthesis schema requires ≥5 supporting videos per trend. If the
+  // analysis pool is smaller than that, the rubric literally cannot satisfy
+  // the schema — Claude will return whatever it can and Zod throws an
+  // unactionable wall of "Array must contain at least 5 element(s)" errors
+  // that points the user at "fix the data shape" rather than "your dataset
+  // is too small". Fail loudly and clearly here.
+  if (analyses.size < MIN_SUPPORTING) {
+    throw new Error(
+      `Stage 8: only ${analyses.size} videos analysed but trend synthesis needs at least ${MIN_SUPPORTING} per trend. ` +
+      `This usually means Stage 2 (broad scrape) returned too few videos — check the brief: ` +
+      `lower min_likes/min_plays, broaden seed hashtags, raise budget_usd, or verify geo is a valid ISO code (e.g. "GB" not "UK"). ` +
+      `Then click Force re-run.`,
+    );
+  }
   const pass1Path = PATHS.pass1Videos(reportId);
   type LiteMeta = { plays: number; likes: number; saves: number; comments_count: number; shares: number; stl_pct: number; handle: string };
   const pass1Lite = new Map<string, LiteMeta>();

v2/server/schemas/brief.ts

@@ -27,7 +27,14 @@ export const BRIEF_INPUT = z.object({
   }),
   competitors: z.array(COMPETITOR).min(3).max(15),
   audience: AUDIENCE,
-  geo: z.string().min(2),
+  // Apify's TikTok scraper requires ISO 3166-1 alpha-2 codes for proxyCountryCode.
+  // The most common gotcha is "UK" — colloquial but invalid; the ISO code is "GB".
+  // We normalise on input so the brief form is forgiving but the wire payload is correct.
+  geo: z.string().min(2).transform((v) => {
+    const upper = v.trim().toUpperCase();
+    if (upper === 'UK') return 'GB';
+    return upper;
+  }),
   language: z.string().default('en'),
   business_question: z.string().refine(
     (v) => v.split(/\s+/).filter(Boolean).length >= 8,