Sentimentor uses optional analytics cookies to understand visits and improve the product. Essential cookies remain enabled either way.
Privacy policySentimentor is a research platform built around a simple rule: every market idea is a hypothesis, not a conclusion. Research may be manual or accelerated by AI; the useful question is whether the evidence, rules, assumptions and results can be inspected and challenged.
Augur gathers the requested market, company, filings, flow and sentiment evidence through explicit research operations. Tool results remain visible so a reader can distinguish a source observation from the agent’s synthesis. Missing or stale evidence is treated as a limitation rather than silently filled with invented certainty.
A thesis only becomes testable when entry, exit, sizing, timeframe and risk rules are explicit. PineNode expresses those rules as readable JavaScript-based strategy code. The generated code is editable; the user can inspect assumptions and change them before a run.
A standard backtest is the first check, not the final verdict. Sentimentor screens the frozen configuration on discovery bars, tests it across three chronological validation folds, inserts embargo gaps to reduce overlap leakage, and opens a server-held final lockbox once only after the earlier stages pass. The freeze covers code, parameters, universe, timeframe, features, data snapshot, sizing, costs and, when supplied, hashed AI-generation provenance. Deflated Sharpe uses account-wide server-recorded validation attempts so starting a new project cannot erase selection-bias history. The same protocol applies to manually written, imported and AI-assisted code.
Full Forward is the validated track: it runs a server-verified, lockbox-passed strategy against newly arriving bars without rewriting historical decisions. Exploratory runners can also collect new observations but never inherit the Validated label. Either evidence mode may use modeled Sentimentor execution or eligible Alpaca Paper broker execution. Backtest, simulated-forward and broker-paper results stay separately labeled. Sentimentor is not a broker. Alpaca Paper uses simulated funds and fills that can differ from live-market execution.
Research Evals test claims that are not naturally expressed as a trading strategy. A sealed evaluation records the question, model lineage, universe, evidence cutoff, ranked picks, probabilities, benchmark, horizon, costs and server-captured entry prices before future returns are available. Scheduled scoring appends benchmark-relative performance and probability calibration without rewriting the original prediction. Historical replays remain labeled declared or unverified unless their point-in-time evidence can be independently audited.
Use this methodology as a checklist when reviewing any result in Sentimentor. Start with the dated evidence, identify which statements are observations and which are interpretations, then inspect the strategy rules and execution assumptions. Compare historical results with an appropriate benchmark and pay attention to out-of-sample behavior, drawdown and the number of trades—not only headline return. A result with incomplete evidence, a tiny sample or unstable parameters should remain a hypothesis until independent data and forward observations support it.