Disclosures

The natural person producing this content is named, with a job title. This is the disclosure MAR Article 2 asks for, and it is published on every page rather than on a legal notice nobody reads.

The identity of the producer: name and job title of all natural persons involved in producing the recommendation, and the name of the legal person responsible.

Facts must be clearly distinguished from interpretations, estimates and opinions; all substantially material sources must be indicated; projections and forecasts labelled.

• Every AI-generated passage carries a banner stating that it is model output, and states how many of its quantitative claims traced back to source data.
• Every figure the site renders from its own payload carries the licence key of the source it came from; there is no rendering path for such a figure without provenance.
• One class of number sits outside that gate and is stated here rather than left to be discovered: the model's own prose, published under a model-output banner, quotes fact-pack figures as text. Those quotations pass no provenance check and carry no licence key of their own. Every fact pack is assembled solely from sources this site may republish, so nothing quoted this way is unlicensed in substance — but the enforcement is on the payload and not on the prose, and this disclosure says so rather than implying otherwise.
• The hash of the complete fact pack given to the model for each decision is committed to the audit chain before the model is called, and is published on every decision card; direct download of the pack itself is specified and not yet available.

A summary of any basis of valuation or methodology and the underlying assumptions; an indication and summary of any change to them; the meaning of any recommendation such as buy, sell or hold; and an appropriate sensitivity analysis.

• Every mandate is published verbatim as the machine-readable YAML the rule engine actually executes — not a prose summary of it.
• The execution cost model is published verbatim, with each assumption either sourced or explicitly flagged unsourced.
• Changes to any rulebook are RULEBOOK_CHANGE events in the same hash chain as trades, with a full diff and a reason. A limit cannot be quietly loosened after a bad quarter.
• The meaning of each permitted action is published: the AI may only express direction and conviction, never a price, quantity or weight.

Disclose relationships and circumstances that may reasonably impair objectivity. Experts additionally disclose net long or short positions exceeding 0.5% of an issuer's total issued share capital, holdings by the issuer above 5%, market-making and any service agreement with the issuer.

• No client money is managed, held or received. There are no clients, no mandates and no power of disposal over anyone's assets.
• No order is routed to any venue. All trading is on paper.
• No affiliate link, issuer sponsorship, revenue share, paid placement or payment-for-order-flow arrangement exists with any person connected to any instrument named on this site.
• No market-making activity and no service agreement with any issuer named on this site.
• No net position above 0.5% of any issuer's issued share capital is held. The portfolios are hypothetical and hold no instruments in fact.

A description of the effective internal organisational and administrative arrangements and of any information barriers set up to prevent and avoid conflicts of interest.

• The AI layer ranks and explains. Its output schema has no field in which a price, a quantity, a notional or a risk limit could be expressed; the single sizing input it may offer is a requested portfolio weight, bounded by the schema, accepted only on the mandates that declare it, and clamped to that mandate's own caps by deterministic code. A test fails the build if that schema is widened.
• All sizing, pricing and risk enforcement is deterministic code the model cannot see or influence.
• The rule engine applies three gates the AI cannot bypass, and every refusal is published with a stable reason code.
• The decision, the model identity and the hash of its complete input are committed to an append-only hash chain BEFORE the model is called and before any fill price is known.

Investment firms and credit institutions must publish quarterly the proportion of recommendations that are buy, hold or sell over the previous twelve months.

This site currently carries no "trade this at broker X" button and no brokerage integration of any kind. That is a product decision, not a permanent structural one: an uncompensated link or a public model portfolio at a broker is under consideration, and it was moved out of the permanent list rather than left there to be quietly broken. What does not change is the bullet above it — no broker, issuer or platform pays to appear on this site, and no arrangement that pays us will ever be presented as a recommendation. Any change here is recorded as a rulebook change in the same audit chain as every trade.