Skip to the evidence
AI Trading SignalsDefinition / Population / Proof

AI claims / Consumer evidence

QuantConnect, FreqAI and TrendSpider: AI Research Tasks

Compare a programmable research environment, an ML framework extension and a model-building interface without inventing a common profit league table.

Choose the required research task first

QuantConnect documents model-training and optimisation workflows. FreqAI describes an ML extension within Freqtrade. TrendSpider's ML Quant Lab and support pages describe tools for constructing trained strategies. Those are concrete product-function references, not proof that a selected user strategy is causal or profitable. This comparison used public documents and did not install a bot, train a market model, open paid accounts or operate an external broker route.

A user seeking a programmable experiment, a configured framework extension and a guided model-building workflow may need different artefacts. Preserve the chosen product and version before comparing feature exports, sample boundaries and operating ownership. None of the three platform names alone supplies an untouched evaluation. Documentation supporting a capability also does not mean every algorithm or alert produced through the platform uses that capability.

Ask for evidence at the chosen boundary

For a training claim, retain model, features, labels, transformations and sample membership. For a forecast claim, add original outputs and their effective version binding. For an account claim, add actual orders, quantities, fills, costs and residual exposure. A platform can support several layers, but each needs its own records. Do not mark an analytical experiment a failed execution service when no execution was requested, or upgrade simulation to live profit because routing is documented elsewhere.

Operating and purchase responsibilities

Confirm current access, data rights, compute resources and what component owns evaluation and delivery. Open-source availability does not mean a complete account workflow is cost-free or safely configured. A paid interface is not a substitute for a frozen experiment history. This publication creates no local production bot, workstation fallback or paid capacity. The free date checker illustrates one declared calendar relation; it cannot certify a framework's full data pipeline or selected strategy.

Write a task-fit rather than a universal verdict

Compare documented input control, retained records, workflow scope and unanswered questions. Keep the exact supplier reference and observation date. A missing private test is a gap in this review, not a measured low score for the product. Later independent evidence can change a narrow assessment without transforming today's documentation review into a past hands-on trial. The expanded provider profiles give each product an original fictional packet and identify which stronger claims still need records.

Completed fictional evidence case

A fictional buyer asks for a frozen training run, a prospective output ledger and actual subscriber fills. Packet A supplies a research configuration, packet B a framework definition and packet C a model-building description. They establish different documented functions; none establishes all three requested results. The reviewer lists the missing run, prediction and account artefacts rather than assigning invented financial scores. No real product is claimed to have supplied these packets. The comparison can still guide a focused enquiry without pretending an investment winner has been independently validated.

Evidence objectRecorded valueClaim boundary
Training functionPublic documentationNot a tested run
Forecast historyOriginal outputs requiredNot supplied by a feature page
Subscriber resultActual fills and cashDifferent evidence layer
An original claim, its available references and the bounded review
Illustrative record flow; no vendor experiment or customer account is pictured.

Direct sources and reviewed scope

Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement.

Related evidence questions