diff --git a/server/worldmonitor/intelligence/v1/deduct-situation.ts b/server/worldmonitor/intelligence/v1/deduct-situation.ts
index 20a87f021..eed19475c 100644
--- a/server/worldmonitor/intelligence/v1/deduct-situation.ts
+++ b/server/worldmonitor/intelligence/v1/deduct-situation.ts
@@ -7,6 +7,7 @@ import type {
 import { cachedFetchJson } from '../../../_shared/redis';
 import { sha256Hex } from './_shared';
 import { callLlm } from '../../../_shared/llm';
+import { buildDeductionPrompt, postProcessDeductionOutput } from './deduction-prompt';
 
 const DEDUCT_TIMEOUT_MS = 120_000;
 const DEDUCT_CACHE_TTL = 3600;
@@ -23,20 +24,9 @@ export async function deductSituation(
 
     if (!query) return { analysis: '', model: '', provider: 'skipped' };
 
-    const cacheKey = `deduct:situation:v1:${(await sha256Hex(query.toLowerCase() + '|' + geoContext.toLowerCase())).slice(0, 16)}`;
+    const cacheKey = `deduct:situation:v2:${(await sha256Hex(query.toLowerCase() + '|' + geoContext.toLowerCase())).slice(0, 16)}`;
 
-    const systemPrompt = `You are a senior geopolitical intelligence analyst and forecaster.
-Your task is to DEDUCT the situation in a near timeline (e.g. 24 hours to a few months) based on the user's query.
-- Use any provided geographic or intelligence context.
-- Be highly analytical, pragmatic, and objective.
-- Identify the most likely outcomes, timelines, and second-order impacts.
-- Do NOT use typical AI preambles (e.g., "Here is the deduction", "Let me see").
-- Format your response in clean markdown with concise bullet points where appropriate.`;
-
-    let userPrompt = query;
-    if (geoContext) {
-        userPrompt += `\n\n### Current Intelligence Context\n${geoContext}`;
-    }
+    const { mode, systemPrompt, userPrompt } = buildDeductionPrompt({ query, geoContext });
 
     const cached = await cachedFetchJson<{ analysis: string; model: string; provider: string }>(
         cacheKey,
@@ -53,7 +43,8 @@ Your task is to DEDUCT the situation in a near timeline (e.g. 24 hours to a few
             });
 
             if (!result) return null;
-            return { analysis: result.content, model: result.model, provider: result.provider };
+            const analysis = postProcessDeductionOutput(result.content, mode);
+            return { analysis, model: result.model, provider: result.provider };
         }
     );
 
diff --git a/server/worldmonitor/intelligence/v1/deduction-prompt.ts b/server/worldmonitor/intelligence/v1/deduction-prompt.ts
new file mode 100644
index 000000000..e6e0305a4
--- /dev/null
+++ b/server/worldmonitor/intelligence/v1/deduction-prompt.ts
@@ -0,0 +1,155 @@
+interface PromptContextParts {
+  primaryContext: string;
+  recentNews: string[];
+}
+
+export type DeductionMode = 'brief' | 'forecast';
+
+const BRIEF_MODE_PATTERNS = [
+  /\b2-3 sentences?\b/i,
+  /\bbrief\b/i,
+  /\bconvergence pattern\b/i,
+  /\bassess likelihood and potential implications\b/i,
+];
+
+function normalizeWhitespace(input: string): string {
+  return input.replace(/\r\n/g, '\n').replace(/[ \t]+\n/g, '\n').replace(/\n{3,}/g, '\n\n').trim();
+}
+
+function trimList(items: string[], maxItems: number, maxChars: number): string[] {
+  const out: string[] = [];
+  let total = 0;
+  for (const item of items) {
+    if (out.length >= maxItems) break;
+    const next = item.trim();
+    if (!next) continue;
+    if (total > 0 && total + next.length + 1 > maxChars) break;
+    out.push(next);
+    total += next.length + 1;
+  }
+  return out;
+}
+
+export function inferDeductionMode(query: string): DeductionMode {
+  return BRIEF_MODE_PATTERNS.some((pattern) => pattern.test(query)) ? 'brief' : 'forecast';
+}
+
+export function splitDeductionContext(geoContext: string): PromptContextParts {
+  const normalized = normalizeWhitespace(geoContext);
+  if (!normalized) {
+    return { primaryContext: '', recentNews: [] };
+  }
+
+  const headerMatch = /(?:^|\n\n)(Recent News[^\n]*)/.exec(normalized);
+  if (!headerMatch) {
+    return { primaryContext: normalized, recentNews: [] };
+  }
+
+  const primaryContext = normalized.slice(0, headerMatch.index).trim();
+  const afterHeader = normalized.slice(headerMatch.index + headerMatch[0].length);
+  const newsBlock = afterHeader.split('\n').filter(Boolean);
+  const recentNews = trimList(
+    newsBlock
+      .map((line) => line.replace(/^\s*[-*]\s*/, '').trim())
+      .filter(Boolean),
+    10,
+    1400,
+  );
+
+  return { primaryContext, recentNews };
+}
+
+export function inferProviderLabel(apiUrl: string): string {
+  try {
+    const host = new URL(apiUrl).hostname.toLowerCase();
+    if (host.includes('groq')) return 'groq';
+    if (host.includes('openrouter')) return 'openrouter';
+    if (host.includes('ollama')) return 'ollama';
+    if (host.includes('openai')) return 'openai-compatible';
+    return host.replace(/^api\./, '') || 'custom';
+  } catch {
+    return 'custom';
+  }
+}
+
+function buildSharedEvidencePrompt(primaryContext: string, recentNews: string[]): string {
+  const parts: string[] = [];
+  if (primaryContext) {
+    parts.push(`Context:\n${primaryContext}`);
+  }
+  if (recentNews.length > 0) {
+    parts.push(`Recent News Signals:\n${recentNews.map((line) => `- ${line}`).join('\n')}`);
+  }
+  if (parts.length === 0) {
+    parts.push('Context:\nNo additional context was provided.');
+  }
+  return parts.join('\n\n');
+}
+
+export function buildDeductionPrompt(input: {
+  query: string;
+  geoContext: string;
+  now?: Date;
+}): { mode: DeductionMode; systemPrompt: string; userPrompt: string } {
+  const now = input.now ?? new Date();
+  const today = now.toISOString().slice(0, 10);
+  const mode = inferDeductionMode(input.query);
+  const { primaryContext, recentNews } = splitDeductionContext(input.geoContext);
+  const evidence = buildSharedEvidencePrompt(primaryContext, recentNews);
+
+  if (mode === 'brief') {
+    return {
+      mode,
+      systemPrompt: `You are a concise forecasting analyst.
+Today is ${today} UTC.
+Use only the supplied evidence plus durable background knowledge.
+Do not invent current facts that are not supported by the evidence.
+Return plain text in exactly 2 or 3 sentences.
+- Sentence 1: core assessment and rough likelihood.
+- Sentence 2: primary drivers or constraints.
+- Optional sentence 3: the most important trigger to watch next.
+No markdown, no bullets, no headings, no preamble.`,
+      userPrompt: `Question:\n${input.query}\n\n${evidence}`,
+    };
+  }
+
+  return {
+    mode,
+    systemPrompt: `You are a senior geopolitical and market forecaster.
+Today is ${today} UTC.
+Your job is to produce a grounded near-term forecast from the supplied evidence.
+Rules:
+- Separate observed facts from forecasted outcomes.
+- Prefer the freshest and most specific evidence.
+- If evidence is thin or conflicting, say so explicitly.
+- Use rough probability ranges, not false precision.
+- Do not use AI preambles.
+- Keep the answer concise but structured.
+
+Return Markdown with exactly these sections in this order:
+**Bottom line**
+**What we know**
+**Most likely path (next 24-72h)**
+**Alternative paths**
+**Key drivers**
+**Signals to watch**
+**Confidence**
+
+Formatting rules:
+- Use short bullets under each section where useful.
+- In "Alternative paths", include 2 alternatives with rough likelihood bands.
+- In "Confidence", state High, Medium, or Low and explain why.
+- Ground claims in the supplied evidence by naming sources, dates, locations, or signal types when possible.`,
+    userPrompt: `Question:\n${input.query}\n\n${evidence}`,
+  };
+}
+
+export function postProcessDeductionOutput(raw: string, mode: DeductionMode): string {
+  const cleaned = normalizeWhitespace(
+    raw.replace(/<think>[\s\S]*?<\/think>/gi, '').replace(/<think>[\s\S]*/gi, ''),
+  );
+  if (mode === 'brief') {
+    return cleaned.replace(/\s+/g, ' ').trim();
+  }
+  return cleaned;
+}
diff --git a/src/utils/news-context.ts b/src/utils/news-context.ts
index d0f90e9f9..9c390b186 100644
--- a/src/utils/news-context.ts
+++ b/src/utils/news-context.ts
@@ -5,3 +5,20 @@ export function buildNewsContext(getLatestNews: () => NewsItem[], limit = 15): s
   if (news.length === 0) return '';
   return 'Recent News:\n' + news.map(n => `- ${n.title} (${n.source})`).join('\n');
 }
+
+export function buildNewsContextFromItems(items: NewsItem[], limit = 15): string {
+  const seen = new Set<string>();
+  const lines: string[] = [];
+  for (const item of items) {
+    if (lines.length >= limit) break;
+    const key = item.title.toLowerCase().trim();
+    if (seen.has(key)) continue;
+    seen.add(key);
+    const ts = item.pubDate instanceof Date ? item.pubDate.toISOString() : String(item.pubDate);
+    const tier = item.tier != null ? ` | tier-${item.tier}` : '';
+    const loc = item.locationName ? ` | ${item.locationName}` : '';
+    lines.push(`- ${ts} | ${item.source}${tier} | ${item.title}${loc}`);
+  }
+  if (lines.length === 0) return '';
+  return 'Recent News Signal Snapshot:\n' + lines.join('\n');
+}
diff --git a/tests/deduction-prompt.test.mjs b/tests/deduction-prompt.test.mjs
new file mode 100644
index 000000000..5a0980751
--- /dev/null
+++ b/tests/deduction-prompt.test.mjs
@@ -0,0 +1,118 @@
+import assert from 'node:assert/strict';
+import { describe, it } from 'node:test';
+
+import {
+  buildDeductionPrompt,
+  inferDeductionMode,
+  inferProviderLabel,
+  postProcessDeductionOutput,
+  splitDeductionContext,
+} from '../server/worldmonitor/intelligence/v1/deduction-prompt.ts';
+import { buildNewsContextFromItems } from '../src/utils/news-context.ts';
+
+describe('inferDeductionMode', () => {
+  it('selects brief mode for short convergence assessments', () => {
+    assert.equal(
+      inferDeductionMode('Analyze this convergence pattern and assess likelihood in 2-3 sentences.'),
+      'brief',
+    );
+  });
+
+  it('selects forecast mode for open-ended user forecasting', () => {
+    assert.equal(
+      inferDeductionMode('What will possibly happen in the next 72 hours in the Taiwan Strait?'),
+      'forecast',
+    );
+  });
+
+  it('ignores trigger phrases in geoContext — mode is query-only', () => {
+    assert.equal(
+      inferDeductionMode('What is the strategic outlook for the Gulf theater?'),
+      'forecast',
+    );
+  });
+});
+
+describe('splitDeductionContext', () => {
+  it('separates primary context from recent news lines', () => {
+    const result = splitDeductionContext(
+      'Theater: Levant.\n\nRecent News Signal Snapshot:\n- 2026-03-15T10:00:00.000Z | Reuters | tier-1 | Israel mobilizes reserves\n- 2026-03-15T09:00:00.000Z | AP | ceasefire talks stall',
+    );
+
+    assert.equal(result.primaryContext, 'Theater: Levant.');
+    assert.equal(result.recentNews.length, 2);
+    assert.match(result.recentNews[0], /Reuters/);
+  });
+});
+
+describe('buildDeductionPrompt', () => {
+  it('builds a structured forecast prompt for panel usage', () => {
+    const { mode, systemPrompt, userPrompt } = buildDeductionPrompt({
+      query: 'What is the expected strategic impact of the current military posture in the Gulf theater?',
+      geoContext: 'Theater: Gulf.\n\nRecent News Signal Snapshot:\n- 2026-03-15T08:00:00.000Z | Reuters | naval deployment increases',
+      now: new Date('2026-03-15T12:00:00Z'),
+    });
+
+    assert.equal(mode, 'forecast');
+    assert.match(systemPrompt, /\*\*Most likely path \(next 24-72h\)\*\*/);
+    assert.match(systemPrompt, /2026-03-15 UTC/);
+    assert.match(userPrompt, /Recent News Signals/);
+  });
+
+  it('builds a terse brief prompt for correlation-card usage', () => {
+    const { mode, systemPrompt } = buildDeductionPrompt({
+      query: 'Assess likelihood and potential implications in 2-3 sentences.',
+      geoContext: 'Countries: Taiwan, China',
+      now: new Date('2026-03-15T12:00:00Z'),
+    });
+
+    assert.equal(mode, 'brief');
+    assert.match(systemPrompt, /exactly 2 or 3 sentences/);
+    assert.doesNotMatch(systemPrompt, /\*\*Bottom line\*\*/);
+  });
+});
+
+describe('postProcessDeductionOutput', () => {
+  it('removes think tags and flattens brief responses', () => {
+    const output = postProcessDeductionOutput('<think>hidden</think> First line.\n\nSecond line.', 'brief');
+    assert.equal(output, 'First line. Second line.');
+  });
+});
+
+describe('inferProviderLabel', () => {
+  it('maps known providers and falls back to hostname', () => {
+    assert.equal(inferProviderLabel('https://api.groq.com/openai/v1/chat/completions'), 'groq');
+    assert.equal(inferProviderLabel('https://example.internal/v1/chat/completions'), 'example.internal');
+  });
+});
+
+describe('buildNewsContextFromItems', () => {
+  it('deduplicates duplicate headlines and includes metadata', () => {
+    const now = new Date('2026-03-15T12:00:00Z');
+    const context = buildNewsContextFromItems([
+      {
+        source: 'Reuters',
+        title: 'Markets fall after new tariff threat',
+        link: 'https://example.com/1',
+        pubDate: now,
+        isAlert: true,
+        tier: 1,
+        locationName: 'Washington',
+        threat: { level: 'high', category: 'economic', confidence: 0.9, source: 'ml' },
+      },
+      {
+        source: 'AP',
+        title: 'Markets fall after new tariff threat',
+        link: 'https://example.com/2',
+        pubDate: new Date('2026-03-15T11:30:00Z'),
+        isAlert: false,
+      },
+    ]);
+
+    assert.match(context, /Recent News Signal Snapshot/);
+    assert.match(context, /Reuters/);
+    assert.match(context, /tier-1/);
+    assert.match(context, /Washington/);
+    assert.equal((context.match(/Markets fall after new tariff threat/g) || []).length, 1);
+  });
+});