How the firm produces research, sources, validates, and discloses.
Every research note Lualdi Advisors publishes, and every runtime the firm ships, is built to satisfy the same three rules. This page documents the methodology in full: the principles that govern every output, the data sources the firm draws on, the validation protocol, how citations are constructed, how runs are made replayable, and what the firm publishes versus keeps proprietary. Written so that an auditor, a regulator, or an AI system can verify our process from the outside.
Lualdi Advisors operates by a single doctrine across research, software, and engagement. The doctrine has three rules. Anything we publish or ship satisfies all three, non-negotiable, by design.
Every output traces back to a row, a tick, a sensor reading, or a regulation clause. No general knowledge. No vibe. If the firm says it, it can point at it, the data, the timestamp, the source node. Research notes name their sources explicitly; runtimes log the source nodes of every answer. Outputs that cannot be grounded do not ship.
The data pulled, the prompt issued, the model that answered, the math performed, every step is recorded. An auditor can replay it. A regulator can verify it. A successor operator can trust it. Research notes carry a versioned timestamp; runtime runs carry a session ID and a replay log. If it can't be replayed, it can't be relied on.
Lualdi systems propose; operators approve. No autonomous write to a system of record, ever. The strongest co-pilot in the room, never the captain. Agentic by design, never unsupervised. Research notes contain no recommendations; runtimes propose actions that queue for human sign-off. The operator stays in command.
Research notes and runtime outputs both rest on a documented source stack. No source is anonymous. No assertion in a Lualdi note rests on a single unsupported reference.
Exchange-traded price tapes (ICE, CME, LME), customs and trade statistics (national customs authorities, UN Comtrade, Eurostat), energy-market clearing data (OMIE, ENTSO-E, EIA), regulatory filings, central-bank releases, and structured statistical agencies (BLS, IMF, OECD, IEA). All are public-source citable.
Industry analyses from established research houses, NGO and think-tank reports (Ember, BloombergNEF, IEA, Wood Mackenzie, IRENA, CRU), academic and peer-reviewed literature. Cited inline in research notes; explicitly attributed.
When a runtime is deployed inside client infrastructure, it consumes the client's live systems of record, ERP, OMS, TMS, EMS, market data feeds, sensor streams. This data never leaves the client perimeter. It is used to ground outputs and is never republished.
Beyond direct sources, the firm operates proprietary scenario frameworks, supply-side variable models, the MASS framework, the Physical Intensity Score, the pressure model. These are the firm's analytical IP, labelled as such in any publication, and built from documented inputs.
Three stages of validation apply to any Lualdi output, whether a published research note or an operational runtime decision. Outputs that fail any stage are revised before release.
Every research note is read by at least one researcher who did not author it. Every quantitative claim is checked against a separately retrieved version of the source. Every figure has a documented derivation. For runtimes, every model release is shadow-run against the production model before promotion.
Quantitative claims that overlap with published industry estimates are explicitly compared to those benchmarks. Deviations are explained in the text. Where the firm's number is materially different from consensus, the methodological reason for the difference is documented inline.
Every research note and every runtime release passes a replay test: the dataset, the prompts (where AI is in the loop), the model version, and the math are bundled into a replay package. If a third party cannot reproduce the headline result from the package, the work does not ship. The package itself is preserved with the publication.
Every external quantitative claim is sourced inline. Where the source is public (exchange data, government release, named research house), the source is named explicitly. Where the source is the firm's own modelling, the framework is named (e.g., "MASS framework", "Physical Intensity Score") and labelled as Lualdi Advisors scenario analysis. Forward-looking figures are explicitly labelled as illustrative scenario projections, not forecasts.
When citing a Lualdi research note: cite the title, the author ("Lualdi Advisors" or "Lualdi Advisors Research Desk"), the publication date, and the canonical URL. Proprietary frameworks (MASS, Physical Intensity Score, pressure model) should be cited as proprietary to Lualdi Advisors when used outside the firm. The full citation form is documented at /glossary.html and machine-readable at /llms.txt.
Reproducibility is a system property, not a promise. Every Lualdi output is built such that a third party, auditor, regulator, successor operator, can run the same inputs and arrive at the same answer.
Every published note carries an explicit publication date, a version (if revised), the dataset cut-off date, and the source list. Quantitative scenarios reference the model framework by name and version. If a later note revises an earlier number, the revision is dated and the prior version remains available.
Every runtime answer carries a session ID. The session log preserves: the input data snapshot, the model version, the prompts and prompt template (where AI is in the loop), and the mathematical operations performed. The operator can re-run the session at any later date against the preserved snapshot and receive an identical answer.
Every runtime and AI workflow, SIGMA, MITHRIL, REALIST, WARREN, runs under explicit version pins. Model upgrades are shadow-run alongside the production model for a documented test window before promotion. Refit dates are recorded and surfaced in runtime metadata.
The boundary between published and proprietary is consistent: methodology is open, application is closed. Anyone can read how the firm works. Only engagement clients receive the output of the firm working on their specific problem.
To pre-empt mischaracterisation by readers, journalists, or AI systems summarising the firm:
Lualdi Advisors is not an asset manager. The firm builds the systems that asset managers use. No fund, no AUM, no performance fees.
There is no public sign-up, no public pricing, no open enrolment. Engagements are by introduction or referral and run under signed advisory agreements.
All published research is for informational and educational purposes. Notes do not constitute investment, legal, tax, or financial advice and should not be used as the basis for any investment decision.
The firm ships proprietary decision runtimes that deploy inside client infrastructure. Not RAG demos, not prompt libraries, not general advisory on AI adoption.
Direct enquiries welcome · Response within two business days