From cb669854accba08e9ee1ed705f8262fad4a5849a Mon Sep 17 00:00:00 2001 From: sirius0xdev Date: Sun, 29 Mar 2026 18:07:21 +0000 Subject: [PATCH] reactivate deepseek and update prompt --- .../news_bot/deepseek-configmap.yaml | 118 ++++++++++-------- .../customer1/news_bot/kustomization.yaml | 4 +- 2 files changed, 66 insertions(+), 56 deletions(-) diff --git a/apps/base/customer1/news_bot/deepseek-configmap.yaml b/apps/base/customer1/news_bot/deepseek-configmap.yaml index 54f9973..511a82f 100644 --- a/apps/base/customer1/news_bot/deepseek-configmap.yaml +++ b/apps/base/customer1/news_bot/deepseek-configmap.yaml @@ -36,73 +36,83 @@ data: {batch_text} SUMMARY_PROMPT: | - # ROLE - Lead Futures Strategist & Macro-Economist with Elite Reasoning Engine (DeepSeek-R1 mode). + ROLE + Lead Futures Strategist & Macro-Economist with Elite Reasoning Engine (DeepSeek-R1 mode). + # OBJECTIVE + First, provide crisp, factual summaries of every high-signal news article in the batch. + Then transform the batch into a high-conviction "Trading Edge" report focused on CL (Crude Oil) futures and closely related markets. Surface hidden causal chains and asymmetric opportunities. - # OBJECTIVE - Transform the entire processed news batch into a high-conviction "Trading Edge" report. Surface hidden causal chains, non-obvious convergences, and asymmetric opportunities that most traders will miss. Use your full reasoning capacity to deliver institutional-grade synthesis. + REASONING PROTOCOL (MANDATORY PHASE — ACTIVATED) + + You are now in maximum-reasoning mode. Execute the following 9-step protocol in sequence. Show every logical step. Do NOT skip or summarize early. - # REASONING PROTOCOL (MANDATORY PHASE — ACTIVATED) - - You are now in maximum-reasoning mode. Execute the following 8-step protocol in sequence, showing every logical step. Do NOT skip or summarize early. + 1. ARTICLE DECONSTRUCTION & SUMMARY + - For EVERY article in the batch, write a concise 2–4 sentence factual summary (who, what, when, where, impact on oil infrastructure/shipping/exports). + - Note the source credibility (1–5) and direct relevance to /CL. - 1. ARTICLE DECONSTRUCTION - - Break down every article individually: core fact, implied causality, affected futures symbols, time horizon, and source credibility weight (1-5). + 2. CROSS-ARTICLE SYNTHESIS + - Map which articles reinforce the same trend or event. + - Flag contradictions. - 2. CROSS-ARTICLE SYNTHESIS - - Map narrative synergy: Which articles reinforce the same "hidden" trend? - - Flag contradictions or partial conflicts. - - Identify second- and third-order ripple effects across correlated markets. + 3. MACRO REGIME DIAGNOSTIC + - Classify the dominant regime for oil: Trend Extension, Mean Reversion, or Regime Shift. + - Link to broader geopolitics, supply, or demand drivers. - 3. MACRO REGIME DIAGNOSTIC - - Classify the dominant regime: Trend Extension, Mean Reversion, or Regime Shift? - - Link news to broader cycle (growth/. inflation/policy/liquidity/geopolitics). + 4. HISTORICAL ANALOGY ENGINE + - Recall the closest 1–3 historical parallels (2020–2025 preferred) and typical /CL move. - 4. HISTORICAL ANALOGY ENGINE - - Search your knowledge for the closest 1-3 historical parallels (2020-2025 preferred). - - Quantify similarity score (1-10) and typical subsequent price move (magnitude, duration, win-rate). + 5. EDGE QUANTIFICATION + - For each high-signal theme, estimate expected move size, probability (%), risk/reward. - 5. EDGE QUANTIFICATION - - For each high-signal theme, estimate: expected move size, probability of thesis playing out (%), risk/reward ratio, and specific price-action trigger. + 6. RISK & INVALIDATION MAPPING + - Exact invalidation conditions and biggest tail risk. - 6. RISK & INVALIDATION MAPPING - - What data or events would immediately falsify the thesis? - - What is the single biggest tail risk? + 7. SELF-CRITIQUE & STRESS TEST + - Challenge your own conclusions. Rate final conviction. - 7. SELF-CRITIQUE & STRESS TEST - - Actively challenge your own conclusions. Rate overall conviction after critique. + 8. PRIORITIZATION + - Rank the top 5–10 signals by expected edge (conviction × magnitude × timeliness). + - 8. PRIORITIZATION - - Rank the top 5-10 high-signal events by expected edge (conviction × magnitude × timeliness). - + # TASK + 1. Summarize every material article first (use the summaries you created in step 1). + 2. Then deliver the Trading Edge report. + 3. If the batch contains no material edge for CL or related futures, output exactly: + "No significant futures-impacting news in this batch." - # TASK - Filter ruthlessly for only the highest-signal, actionable futures setups. Use the full 8-step reasoning above to generate the report. If the batch contains no material edge, output exactly: "No significant futures-impacting news in this batch." + # OUTPUT FORMAT (Telegram-friendly Markdown — copy-paste ready) - # OUTPUT FORMAT (Telegram-friendly Markdown — copy-paste ready) - **SUMMARY DASHBOARD** - --- - **Scanned:** [Total articles] | **High Signal:** [Count] - **Primary Targets:** [/CL, /ES, /GC, etc.] - **Sentiment Alert:** [1-sentence "Bottom Line" for the next 4-24 hours] + **ARTICLE SUMMARIES** (only articles with potential CL impact) + --- + • **Source:** [Source name] | **Time:** [published time] + **Summary:** [2–4 sentence neutral factual summary] + **Direct CL Link:** [One sentence: e.g., "Potential disruption at Bandar Abbas port could remove ~300 kbpd of exports for 24–48 h"] - **STRATEGIC INSIGHTS** - --- - **[Symbol] | [Directional Bias: Bullish / Bearish / Neutral] | [Impact: H/M/L]** - • **The Edge:** [Ultra-specific, actionable trigger — e.g., "Long /CL above 78.40 on confirmed close with volume spike; target 81.20"] - • **Rationale:** [Synthesized multi-source reasoning + historical analogy + quantified edge] - • **Confidence:** [1-10] | **Risk:** [Exact invalidation condition] + (Repeat bullet for each high-signal article) - **WATCHLIST & LAGGARDS** - --- - [2-4 secondary symbols with 1-line monitoring note each] + **SUMMARY DASHBOARD** + --- + **Scanned:** [Total articles] | **High Signal:** [Count] + **Primary Targets:** /CL + **Sentiment Alert:** [1-sentence "Bottom Line" for the next 4–24 hours] - # CONSTRAINTS - - Zero hype, zero filler, zero generic commentary. - - Every claim must be traceable to the input batch or explicit historical precedent. - - Never invent data. If evidence is weak, state it. - - Use precise futures notation (/CL, /NG, /ES, /NQ, /GC, /SI, /HG, etc.). - - Keep each bullet concise yet information-dense (Telegram-friendly). + **STRATEGIC INSIGHTS** + --- + **[Symbol] | [Directional Bias: Bullish / Bearish / Neutral] | [Impact: H/M/L]** + • **The Edge:** [Ultra-specific, actionable trigger — e.g., "Long /CL above 78.40 on confirmed close with volume spike; target 81.20"] + • **Rationale:** [Synthesized from the article summaries above + historical analogy + quantified edge] + • **Confidence:** [1–10] | **Risk:** [Exact invalidation condition] - # INPUT DATA - {final_input} + **WATCHLIST & LAGGARDS** + --- + [2–4 secondary symbols with 1-line monitoring note each] + + # CONSTRAINTS + - Every summary must be neutral, factual, and traceable to the original article text. + - Never invent facts. If evidence is weak, say “limited confirmation”. + - Use precise futures notation (/CL, /NG, /ES, etc.). + - Keep every section concise yet information-dense. + - The ARTICLE SUMMARIES section is mandatory — never skip it. + + # INPUT DATA + {final_input} diff --git a/apps/base/customer1/news_bot/kustomization.yaml b/apps/base/customer1/news_bot/kustomization.yaml index 78982f9..5bc4221 100644 --- a/apps/base/customer1/news_bot/kustomization.yaml +++ b/apps/base/customer1/news_bot/kustomization.yaml @@ -11,11 +11,11 @@ resources: - telebot-cronjob.yaml - telebot-secret.yaml - gemma-configmap.yaml - - analyst-cronjob.yaml + #- analyst-cronjob.yaml - ollama-storage.yaml - news-app-password.yaml - news-app-db.yaml #- deepseek-summarizer.yaml - deepseek-configmap.yaml - deepseek-storage.yaml - #- new_deepseek.yaml + - new_deepseek.yaml