Research private companies through six perspectives, including business, finances, competition, and technology. Cross-check evidence, separate facts from inferences, and produce a valuation range, possible exit routes, and a map of missing information.
Deep research framework for pre-IPO / private companies (Ant Group, SpaceX, Stripe, ByteDance...). Six analyst lenses — business model, financial forensics, competitive landscape, risk & governance, tech & IP, alternative-data signals — run in parallel via run_swarm, then cross-validated for signal consistency before any verdict. Built around the core challenge of private-company work: information is scarce, so every data point carries a confidence label (high / medium / low), inference is shown separately from fact, and 'I don't know' is a valid output. Outputs a fair-value range, exit-path analysis, and an information-gap map. Use for any unlisted company where you need to judge what the business is actually worth.
Run score Not tested
See how to installReview the instructions the agent will receive before you install.
Deep research on an unlisted company (e.g. Ant Group, ByteDance, SpaceX, Stripe).
Ultimate goal: under information scarcity, recover the company's true value — not the market valuation, but what the business is actually worth.
Private vs public research: no standardized financials (multi-source patchwork + cross-validation); few valuation anchors (funding rounds, comparables, scenarios); large information asymmetry ("jigsaw" research); uncertain exit path (IPO / M&A / secondary).
Private companies are where AI bias is worst. Watch for:
Counter: prefer leaving blanks ("I don't know") over filling tables with speculation to fake certainty; label every data point with confidence (🟢high/🟡medium/🔴low); separate verifiable fact from inference; when information is extremely scarce, switch to "first-principles mode" and answer only: ① what real problem does this business solve? ② why this team? ③ ceiling if it succeeds / how it dies if it fails? ④ the key validation node at this stage?
Invert the asymmetry: the market knows little about private companies → pricing is inefficient → that's exactly where alpha may live.
Six lenses, best run in parallel (via run_swarm, one worker per lens; or sequentially via web_search):
| Role | Lens |
|---|---|
| business-decoder | Business model + product/user analysis: "what is this business, essentially" |
| financial-detective | Financial patchwork + valuation: "recover the true financial picture under missing data" |
| competitive-mapper | Industry + competition + substitution: "who competes, who could disrupt" |
| risk-governance-analyst | Risk全景 + management/governance/investors: "what could go wrong, who's at the helm" |
| tech-ip-analyst | Tech stack / patents / R&D / moat: "is the tech barrier real and durable" |
| signal-miner | Alternative data (hiring / patents / litigation / app / supply chain): "clues beyond the usual sources" |
You (team-lead) integrate, patch the picture, cross-validate, output the final report.
No standard financials; multi-source patchwork + cross-validation. Every data point: source, time, confidence, derivation.
Source priority: 🟢 prospectus/regulatory filings, parent-company annual report disclosure, regulatory penalties, bond/ABS offering documents → 🟡 business registry, funding news, third-party reports, deep media (LatePost/The Information/36Kr/Bloomberg) → 🔴 industry extrapolation, ex-employee leaks.
Key metrics: revenue (scale/growth/mix/quantity×price), cost (gross margin/R&D/sales/G&A rates, vs peers), profit (EBITDA/net income/profitability timeline), cash flow (operating/burn rate/runway), efficiency (per-capita revenue, capital efficiency).
Cross-validation: list every source for the same metric; check convergence across methods; flag single-source ("isolated evidence") data.
Funding history: full timeline (round/amount/valuation/lead investor); health of the curve, interval, down-rounds, whether existing investors keep participating; latest-round terms (liquidation preference / anti-dilution / ratchet) and their effect on common-share value.
Valuation (multi-method): ① last-round (adjust for liquidation prefs, 20-40% discount); ② comparable public comps (3-5, PS/PE/EV-EBITDA, liquidity discount 20-30%); ③ DCF scenarios (bear/base/bull, each assumption grounded); ④ terminal-value rollback (5/10y terminal state → implied IRR); ⑤ transaction comps (recent M&A/funding multiples).
Valuation synthesis: do the methods converge? If divergent, explain. Distinguish "fair value" and "conservative (margin-of-safety) value".
Private companies have limited conventional info; alt-data often beats news.
Anomaly list (most important): things inconsistent with the company's narrative; inconsistent with industry norms; sudden changes (hiring freeze / executive departures); unexplained.
Before synthesis, the team-lead must:
Save via write_file to reports/{company}/{company}-private-{YYYYMMDD}.md. Run report_audit on the numbers as a quality gate.
run_swarm, or sequential).No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Compare a Chinese company's US depositary receipts, Hong Kong shares, and mainland shares after adjusting for exchange rates and share ratios. Analyze pricing gaps, valuation differences, and delisting risks to support research into its different listings.
Retrieve stock prices, futures quotes, exchange rates, and economic indicators with the free Python library AKShare. The skill explains market-specific functions, symbol formats, and data preparation for analysis and testing strategies against historical data.
Explore bundled libraries of investment factors, signals used to compare securities, and inspect individual formulas and metadata. Evaluate a factor or an entire library across a chosen universe and period, producing aggregate performance statistics and an HTML report.
Assess whether a Chinese A-share company may receive ST or *ST risk warnings next financial year. Review financial forecasts, dividends, audit opinions, and regulatory penalties, then report risk severity separately from confidence in the prediction.
Plan portfolio allocations across assets using risk, correlations, return forecasts, and trading costs. Compare five optimization methods, produce settings for config.json, and define rebalancing rules that adjust holdings back toward their target weights.
Diagnose failed or abnormal backtests, which test trading strategies on historical data. Inspect result files, strategy code, and configuration to identify bugs, make targeted repairs, rerun tests, and verify the updated metrics.