The System
The architecture, and the All-Weather Run.
Two engines, a multi-model council, and 118 scheduled workflows that run across the full trading day and after-hours. This is the machine, laid out end to end.
Research & education only. Not investment advice. · WATCHLIST_ONLY
Architecture
A text-version diagram of the pipeline.
No single model is allowed to be right alone — every path below ends in either agreement or an explicit, surfaced disagreement.
12+ DATA SOURCES (prices · news · filings · social sentiment · insider flow ·
supply-chain mapping · market-structure signals)
│
▼
┌───────────────────────────┐ ┌───────────────────────────┐
│ ENGINE 1 — Hermes Agent │ │ ENGINE 2 — Codex │
│ Orchestrator + analyst │──────▶│ Independent checker │
│ Fundamentals · Macro · │ │ Technicals · re-derives │
│ Sentiment · schedules │◀──────│ Engine 1's key numbers │
│ the nightly pipeline │ │ (valuation, growth, etc.) │
└───────────────────────────┘ └───────────────────────────┘
│ agree? │
▼ no ──────────────────────────────────▶│
┌─────────────────────────────────────────────────────────────────┐
│ MULTI-MODEL COUNCIL │
│ R1 DeepSeek V4 Pro proposes → R2 GLM-5.2 reviews → │
│ R3 GPT-5.6 breaks the tie → scenario label + confidence /100 │
└─────────────────────────────────────────────────────────────────┘
│
▼
FALLBACK CHAIN (if a model is unavailable):
GLM-5.2 → GPT-5.6 → DeepSeek V4 Pro → Qwen3 14B (local, free)
│
▼
DELIVERY: English → Telegram · Chinese → Feishu
Every report carries a compliance line. WATCHLIST_ONLY by architecture.
No broker API. No order execution. Research & education only.
"One model can be confidently wrong. Two engines catch it. A council of three decides."
The All-Weather Run
One full cycle, end to end.
The system runs across the full trading day and after-hours — collecting, reasoning, cross-verifying, and voting — so the verdict is ready when you are. Times follow each market session; the flagship tasks are highlighted.
Evening · Pre-market build-up
Mon–Fri·ET
Pre-market Deep Dive
GLM-5.2 · 14 sectionsThe most detailed scheduled report of the night: watchlist panorama, ticker-by-ticker deep dive, macro overview, full-list coverage, news & impact, valuation & technicals, a scoring panel, catalysts, and a data-quality check — an illustration of the workflow's report structure.
ET
Scoring Engine
A worked example of turning a model's conviction into a numeric score — then a plain scenario label (e.g. Watch / Hold / Review) per ticker. Illustrative of the scoring workflow, not a recommendation.
ET
Deep Analysis / Ticker Analyst
GLM-5.2 + CodexMulti-engine deep analysis per ticker: Engine 1 (fundamentals + valuation + insider) → Engine 2 (technicals + cross-verify) → a co-authored Synthesis. 8 sections, ≤1,200 words, ending in a scenario label and an illustrative valuation range. A worked example of the analysis workflow. See a sample →
Mon–Fri·ET
Screening Radar
A full-market screening example. Scans 8 sources and hard-filters out OTC, meme, penny, and micro-cap names. Five-factor scoring — Surge Intensity, Catalyst Backing, Supply-Chain Chokepoint, Market Blindness, Sector Alignment — sorts results into three illustrative score bands.
Night · Deep reasoning & the vote
Mon/Thu·ET
Supply Chain Onion
GPT-5.6A five-layer peel across 8 sectors — upstream (raw materials → components → equipment) and downstream (contract manufacturing → end brands). Each layer gets a bottleneck story ("who's got who by the neck") and a company deep-dive: moat, financials, valuation, and a scenario label. Plain language, no jargon.
ET
Bull/Bear
GLM-5.2Per-holding bull case vs bear case, ≤3 points each, scored X/10 — the tension made explicit.
ET
Scenario Explorer
DeepSeek V4 ProAn educational walk-through of common trade-offs — how a workflow can frame different market scenarios and outcomes side by side. Illustrative only; nothing here is a strategy recommendation.
Mon–Fri·ET
Council War Room
The multi-model vote for contested outputs: R1 DeepSeek proposes → R2 GLM-5.2 reviews → R3 GPT-5.6 breaks the tie → a single scenario label with a confidence score out of 100. An illustration of the verification workflow.
Overnight · Intraday watch (scheduled)
ET
Intraday Snapshots
DeepSeek V4 Pro · 11 sectionsFour passes — 10:30 AM, 12:00 PM, 1:30 PM, 3:00 PM ET — tracking the latest available watchlist prices, news, sentiment, and a 6-factor scoring panel. Sends a data alert the moment a tracked ticker moves ±4% / ±6%.
Dawn · The wrap-up
Mon–Fri·ET
After-Hours Daily
GPT-5.6 · 12 sectionsThe most comprehensive daily read: watchlist panorama, the full-list coverage (49 tickers), sentiment divergence, oversold/overbought score examples, next-session watch items, and a data-quality report — an illustration of the workflow's daily output.
ET
Market Sentiment Pulse
GPT-5.6Breadth metrics, the Fear & Greed index, and sector sentiment — the market-mood side of the morning brief.
The supporting cast — the other jobs that run through the night
- T17 — Earnings calendar & surprise tracking across the watchlist.
- T18 — Insider activity from SEC Form 4 filings — who's buying, who's selling.
- T15 — Macro sweep — rates, CPI, jobs, and the calendar that moves everything.
- T12 — Weekly synthesis — the week compressed into a single review.
- Backtest — Strategy backtesting against historical windows.
- DCF — Discounted-cash-flow valuation for the core holdings.
TMC v2.0 · times shown per market session · For research only · WATCHLIST_ONLY
Want a pipeline like this on your own watchlist?
The tutorials and engineering patterns teach the exact architecture above — schedule, fallback chain, and council included.