◆ Autonomous Specialist
Models the issuer against agency methodologies (Moody's, S&P, Fitch), estimates the likely rating and the levers that move it, and drafts the rating-agency presentation. Maps the scorecard and frames the analysis; the deal-committee agent owns the agency strategy.
Memory
Working The issuer's financials mapped against the agency scorecard.
Episodic Prior ratings outcomes in the sector and agency feedback patterns.
Semantic Agency methodologies, scorecard weightings, notching conventions.
Procedural Scorecard-mapping playbooks per sector and agency.
Store Methodology knowledge base + model store
Orchestration
MCP
Harness · Managed Agents … modelling session; methodology scorecards held in a persisted NOTES file across the engagement.
Tools
⌕ Agency methodology retrieval Retrieval ›_ Financial modelling sandbox Code exec { } Issuer financials API { } Presentation engine API
Evals & guardrails
- Scorecard mapping cites each input to the issuer's financial statements.
- Estimated rating cross-checked against peers' actual ratings.
- The deal-committee agent owns the agency strategy; this agent drafts the analysis.
Offline reflection
Consolidates actual agency outcomes vs. its estimates to calibrate scorecard mappings and notching expectations per sector.
Frontier edge
- ▲Causal reasoning: counterfactual 'what moves us a notch' analysis on the scorecard, isolating which metric changes actually drive the rating rather than co-move with it.
- ▲World-model simulation: simulates agency response to alternative capital structures before the deal-committee agent commits to the meeting.
- ▲Continual learning: eval-gated calibration of scorecard mappings and notching from actual agency outcomes (SEAL-style), no full retrain.
A sample run
Trigger Issuer planning a debut bond wants a likely-rating read and agency deck.
- 1Map the issuer's financials onto each agency's scorecard.
- 2Estimate the rating and identify the metrics that move a notch.
- 3Benchmark against rated sector peers.
- 4Draft the rating-agency presentation.
Output A ratings analysis with an estimated rating, sensitivity levers and a draft agency deck, committed to the deal-committee agent to direct and present.
In numbers
45
Ratings analyses / quarter
median ~4 h
Analysis latency
Handoffs
Hands to → New-Issue Pricing Agent