AI claims / Consumer evidence
Free AI Trading Tools, Data Rights and Compute Costs
Separate publicly useful education and open software from current data, hosting, subscription and execution charges.
Free access needs a named object
A free article, downloadable record, software framework and renewing trial are different offers. Identify what is actually available without payment and what entitlement supplies the intended task. This publication's pages, source-linked records, synthetic illustration and browser-local planning tools are publicly accessible. It does not promise that a vendor's private data, model training, connected account or future signal feed is indefinitely free.
FreqAI's public framework reference and QuantConnect's research documentation support a capability enquiry, not a complete price quote. Confirm current data rights, compute requirements, software access and account charges separately. A product's technical ability to train models does not establish that the chosen dataset may be redistributed or that a subscription includes every external dependency.
Keep charges in their actual units
A monthly software amount, per-job compute charge and per-side execution fee need separate denominators and periods. Do not multiply or allocate them by a convenient alert count without stating the method. A low cost per output is not forecast quality or profit. Actual usable observations can differ from generated outputs when timing, validity or account access is missing. Preserve that workload definition instead of advertising a unit price that hides implementation constraints.
Export permission and retained evidence
Ask which configurations, predictions and source data the chosen offer lets a reader retain. Export availability does not automatically grant third-party redistribution rights. A private weights file, public feature description and licensed historical dataset can have different terms. Do not label the entire packet with a blanket reuse licence unless permission is established. The source links in this publication are references, not a claim that their publishers licensed all content to this desk.
A decision without an invented discount
Compare the exact capability, current terms and total documented dependencies. Mark unknown amounts and renewal conditions rather than substituting zero. An introductory promotion needs an actual authorised offer and first-charge rule. The VR campaign values conflict with protected and originating terms, so no selected price-sensitive checkout claim is published here. A reader can use the free planning material while requesting clarification; no new paid service or automated production bot is added by this release.
Completed fictional evidence case
An invented setup has free framework code, a declared monthly data charge of 30 and compute charge of 10 for the selected experiment period. Its stated period cost is 40 before any other software or execution charges. The code being free does not make the whole setup zero-cost. If ten usable outputs are observed under a frozen definition, that allocation would be 4 per usable output, not a return or quality score. If usable count is zero, the per-output ratio is undefined. No named supplier price, real subscription bill or profitability result is asserted.
| Evidence object | Recorded value | Claim boundary |
|---|---|---|
| Framework code | Free in the invented case | Only one component |
| Declared period charges | 30 + 10 = 40 | Not a vendor quote |
| Ten usable outputs | 4 per output | Allocation, not quality |
| Zero usable outputs | Undefined ratio | Not zero cost |

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.