Research requests arrive across ten channels with fundamentally different urgency profiles, data completeness levels, and output format requirements. Portfolio Manager ad hoc requests arrive as natural-language investment questions — agent immediately retrieves existing financial model (if in coverage), pulls prior research notes, ingests latest consensus estimates, downloads most recent filing, and assembles a context brief within minutes before beginning full analysis. Analyst-initiated coverage requests (new initiation or coverage transfer) trigger the full onboarding workflow: peer universe construction, blank model scaffold from sector template, bulk filing retrieval (last 3 years 10-K/10-Q), management presentation library pull from IR website, and prior sell-side note collection for competitive positioning. Earnings event triggers fire automatically within seconds of announcement — agent pulls actuals vs. street consensus by line item, updates the live EPS model with reported figures, calculates EPS surprise magnitude and revenue beat/miss, identifies guidance changes vs. prior quarter, and begins drafting flash commentary before the call begins. Quantitative screener alerts — from factor screens (value, momentum, quality), event-driven screens (earnings revision inflections, short squeeze setups, insider cluster buying), or technical screens (breakout, mean-reversion setups) — trigger automated research summary generation with fundamental validation. Corporate access events (management roadshows, non-deal roadshows, analyst days, site visits) trigger a two-stage workflow: pre-event preparation brief (company snapshot, open investment questions, prior mgmt commentary, analyst consensus view, suggested Q&A questions) followed immediately by post-event synthesis (key takeaways vs. expectations, guidance read-through, incremental variant perception). News and macro event triggers — M&A announcements, macro data releases, central bank decisions, geopolitical developments — trigger immediate impact assessment against portfolio positions and coverage universe. Portfolio review triggers — position concentration breaches, earnings revision deterioration, covenant/covenant proxy violations for credit positions, maturity approaching for fixed income — initiate automated monitoring and review workflow. Regulatory and disclosure event triggers — 8-K material event, DEF 14A proxy filing, SEC comment letter, insider Form 4, 13D/G ownership disclosure — trigger filing extraction and materiality classification. All submissions normalized to CAIBots Research Request Schema v2 before agent pipeline invocation.

Seven specialized sub-agents execute in parallel from research request receipt, each consuming the same normalized request object but operating on different data domains and producing complementary outputs that are merged at the Analysis Assembly stage. Orchestrator monitors sub-agent completion status, handles dependency chains (e.g., Valuation Agent waits on Fundamental Agent completing EBITDA normalization before running DCF), enforces compliance gate (Compliance Agent must return CLEAR before any output is transmitted downstream), and assembles the final deliverable package from component outputs.

Asset class classification is the critical routing decision that determines the entire research methodology, regulatory obligation set, approval authority matrix, and output format. Equities trigger the DCF/comps valuation suite, EPS model maintenance workflow, earnings calendar management, and short interest/13F monitoring — highest automation eligibility for model updates and flash notes. Fixed Income triggers credit spread analysis, duration/convexity sensitivity modeling, credit ratings migration monitoring, and covenant/indenture review workflow — flash alert automation applicable for spread movement and ratings actions. Macro/FX/Rates triggers cross-asset correlation analysis, central bank policy decision tree mapping, and thematic strategy report generation — largely human-authored with agent providing data assembly and scenario modeling. Alternatives triggers bespoke methodology dependent on sub-asset class: private equity NAV analysis and IRR modeling, real assets cap rate / DCF, infrastructure concession cash flow modeling, hedge fund manager attribution and drawdown analysis. Multi-Asset / Thematic spawns parallel workstreams across multiple asset classes with an integration synthesis stage that resolves conflicting signals and produces a unified portfolio recommendation. Decision authority matrix determines whether output is auto-publishable, requires analyst sign-off, requires IC review, or must be blocked for compliance escalation.

Five parallel assembly tracks execute simultaneously, consuming the outputs of the seven sub-agents and synthesizing them into a coherent investment thesis. Fundamental Track: 3-year historical + 3-year projected model normalized, EPS bridge documented line-by-line, segment drivers decomposed, FCF quality scored, management guidance credibility assessed, balance sheet risk flagged. Valuation Track: all valuation methodologies cross-checked against each other for internal consistency — DCF intrinsic value vs. comps-implied value vs. precedent transaction value; SOTP cross-checked against market cap; bull/base/bear price targets weighted by scenario probability; current multiple vs. historical range and cycle-positioning assessed. Sentiment & Positioning Track: short interest, 13F, insider activity, options flow, and sell-side revision trends synthesized into a net sentiment score (–5 to +5 scale) with directional conviction commentary. Macro & Sector Track: rate and FX sensitivities quantified, thematic alignment scored, regulatory catalyst calendar assembled, competitive positioning assessment finalized. Portfolio Fit Track: marginal contribution to existing portfolio risk (VaR, factor exposure, sector concentration), sizing guidance relative to conviction and liquidity, correlation to existing positions, active weight vs. benchmark impact — all synthesized into a specific position sizing recommendation to PM.

Stage 1 — Investment Narrative: synthesize all assembled research intelligence into a coherent, differentiated investment thesis — identifying the three to five key earnings and value drivers, articulating the variant perception vs. consensus (where does the firm's view diverge from street and why), quantifying the magnitude of the opportunity or risk, contextualizing competitive moat and durability, and connecting macro and sector tailwinds/headwinds to the specific company's P&L sensitivity. Narrative written in the institution's house style, calibrated to the specific analyst's writing voice profile from in-weights memory. Industry benchmarking contextualized explicitly. Downside scenario fully articulated with bear case price target derivation. Bear-to-bull asymmetry ratio calculated.

Stage 2 — Rating Recommendation: formulate a specific, evidence-grounded investment rating (Buy / Accumulate / Hold / Reduce / Sell / Initiate at Buy) with base-case price target, bull and bear targets with probability weights, expected total return (price appreciation + dividend yield), catalyst timing (near-term 0–6M, medium-term 6–18M, long-term 18M+), portfolio sizing recommendation to PM (overweight / benchmark / underweight, expressed as active weight guidance), and IC routing determination (auto-publish / analyst review / IC committee).

Stage 3 — Complete Deliverable Suite: produce all required output documents simultaneously — research note (in firm's template, in analyst's voice, with all required exhibits: EPS model summary, comp table, DCF sensitivity, price target bridge, catalyst calendar, risk table, disclosures); model commentary (line-by-line rationale for estimate changes vs. prior and vs. consensus); IC investment committee memo (executive summary, investment thesis, key risks, sizing recommendation, historical accuracy context for analyst); earnings flash note (for earnings events: actuals vs. estimates table, beat/miss summary, guidance read-through, immediate stock reaction context, 3 key takeaways for PM action); compliance disclosure block (auto-populated analyst certification, conflict disclosures, MiFID II research classification, jurisdiction-specific regulatory language). In-weights memory carries firm's research note architecture, IC presentation norms, analyst voice calibration, sector-specific valuation norms, and prior committee feedback for continuity.

Outputs routed to three simultaneous destinations. Auto-Publish Lane — eligible equity model updates, consensus divergence flags, earnings flash notes, and credit spread alerts: compliance cleared, no conflicts, no MNPI, within pre-authorized change thresholds — output published to CRM/distribution system immediately, compliance archive updated, model vintage locked, MiFID II research value logged. Analyst Review Package — complete research deliverable suite assembled and queued in analyst workbench (FactSet, Bloomberg, Sentieo): research note in firm template in analyst's voice, updated financial model with all assumption documentation, comp table, DCF sensitivity analysis, risk table, catalyst calendar, compliance disclosure block — analyst reviews, edits, certifies, and publishes. IC Committee Package — investment committee memo assembled: executive summary with rating, price target, upside/downside, sizing recommendation, five-slide IC deck, analyst track record and estimate accuracy for covered sector, comparable prior IC decisions for consistency reference — CIO or designated IC chair reviews and votes. ANALYST SIGN-OFF IS MANDATORY for all non-auto-publish outputs — Agent never publishes investment recommendations without human analyst authority. IC review required above analyst authority level.