Define the outcome before counting successes
Price direction, return magnitude, benchmark outperformance and portfolio profit are different outcomes. State the forecast horizon and timestamp. A score that ranks companies is not automatically a probability that their prices will rise.
Illustrative example: six trades gain $1 each and four lose $2 each. The hit rate is 60%, but total P&L is −$2 before costs. This arithmetic shows why a headline success percentage cannot establish profitability.
Require an evaluation record
Ask for the exact universe, historical period, model version, input availability dates and rule for turning scores into positions. Include delisted companies where relevant and retain failed or missing predictions. A retrospectively selected list of winners is not a complete evaluation.
Separate training, model selection and final testing chronologically. Disclose how many variants were tried and whether the final test influenced revisions. Report uncertainty and sample size; overlapping forecasts can reduce the amount of independent information.
Compare outcomes on the same basis
Use a benchmark relevant to the universe and exposure, with consistent dividends, currency and dates. Include realistic turnover, trading costs, liquidity constraints and any shorting assumptions. Examine drawdown and concentration as well as average return.
A useful paper can illustrate an evaluation method, but its results apply to its own data and models. It does not establish that a commercial tool reproduces them. This page does not repeat an unsupported literature-wide percentage or infer that human-plus-AI always beats either alone.
Check what the named product actually provides
The linked free Pineify screener filters by sector, market capitalization, share price, volume, exchange and country. It displays up to 20 matches with available quotes, selected trailing-twelve-month metrics and one dated annual analyst-estimate record. It does not implement a dividend-safety score, growth-prediction model, natural-language query box or automatic verification of the research criteria below. Missing data is unknown, not a zero or a failed investment test.
No validated prediction-accuracy statistic for that free screener is established here. There is no basis for presenting an 8–10 score as an observed probability or for promising a 3–12 month edge. Any other product’s model requires its own evidence.
Use claims you can verify
Check whether reported performance was hypothetical or observed live and whether a complete record is available. Keep model outputs and later outcomes with timestamps so that revised predictions cannot replace the original record.
The joint SEC, FINRA and NASAA alert warns about AI investment claims and guaranteed returns. It does not say all AI tools are fraudulent; it supports requiring evidence rather than trusting the label.
Frequently asked questions
- Is 60% directional accuracy enough?
- Not by itself. Payoffs, costs, coverage, sample size and the decision rule determine what the number means.
- Do academic results prove Pineify performance?
- No. A different implementation needs its own reproducible evaluation. No such accuracy rate is asserted for the linked free screener.
- What should I request from a score vendor?
- Request the target, model version, historical inputs, universe, full test record, benchmark and cost assumptions. Distinguish hypothetical from live results.
Sources and further research
Educational research only. Screening results and illustrative calculations are not recommendations, forecasts or guarantees. Verify source documents, data periods and assumptions before making an investment decision.