How CommonQuant produces and measures research
CommonQuant Research Methodology, Evidence, and Limitations
CommonQuant separates sourced facts, computed evidence, AI-written interpretation, and deterministic trading rules. This page describes those boundaries so readers can judge what a public idea or Advanced Analysis report does—and does not—establish.
The methodology is designed for inspectability rather than certainty. Source links, reported periods, sample sizes, evidence tiers, public outcome tracking, and explicit invalidation conditions are surfaced when available. Missing evidence is omitted or identified; it is not replaced with invented figures.
Sources and data provenance
Market ideas begin with source-linked news or a user-supplied thesis. Suggested instruments must resolve against CommonQuant’s supported security universe. Company fundamentals come from reported SEC EDGAR XBRL facts and derived ratios; market-state and quantitative panels use observed market data. Each value remains subject to the coverage, timing, revisions, and definitions of its source.
A cited headline is evidence that an event was reported, not proof that the resulting market interpretation is correct. Readers should open primary filings and source articles when a decision depends on a claim, reporting period, or accounting definition.
What AI writes and what the system computes
Language models structure a thesis, summarize evidence, argue for and against it, and explain what could change the view. They do not supply the numerical values shown as computed evidence. Conviction components, XBRL-derived metrics, sector percentiles, correlation matrices, risk calculations, trigger distances, and backtest statistics are produced by application code from available data.
Every completed Advanced Analysis is projected once into a bounded public preview containing the first 35% of its narrative, selected non-actionable charts, and ticker-and-metric watch labels. The same persisted projection powers the anonymous React view, crawler HTML, Markdown mirror, structured data, and sitemap freshness. Paid trigger levels, directions, rationales, and the remainder of the report stay outside that projection.
Strategy validation and evidence tiers
A generated strategy is compiled into supported indicators, timeframes, instruments, and explicit entry, exit, and invalidation rules before it can be monitored. Backtests and stress tests are evidence about a rule set on a particular sample; they are not forecasts and do not remove execution, spread, liquidity, regime, or overfitting risk.
Reports distinguish realized, backtested, weak, pending, and unavailable evidence where the underlying pipeline can support that distinction. Inapplicable charts are omitted instead of being populated with synthetic data. Small trade counts, short histories, or missing observations should reduce confidence even when a displayed result is favorable.
Publication, revisions, and outcome measurement
Public ideas receive stable canonical URLs and appear in the sitemap. When Advanced Analysis finishes or its public projection changes, the publication transaction stores an immutable artifact, increments its public revision, updates sitemap freshness, and creates a durable indexing event. IndexNow accelerates discovery for participating search engines; standard search crawlers discover the same content through canonical links, internal links, Markdown alternates, and sitemaps.
Published idea outcomes are measured after publication using the platform’s documented observation windows and are shown whether the result is favorable or unfavorable. The public track record is intended to make selection effects visible; it is still historical evidence and does not guarantee that a future idea, strategy, or market regime will behave similarly.
Known limitations and responsible use
Source data can be delayed, revised, incomplete, or inconsistent across issuers and asset classes. News can be wrong or superseded. Model-written interpretation can omit context. Backtests can overfit, and live fills can differ from observed prices. A missing panel means the required evidence was not available or not applicable; it should not be read as a neutral result.
CommonQuant is an educational and informational research and alert platform, not a brokerage or financial adviser. It does not execute trades. Users are responsible for checking important claims, deciding whether evidence is sufficient, and making any investment decision through their own broker.
Frequently asked questions
Does CommonQuant let an AI invent financial figures?
No. Numerical evidence displayed as fundamentals, percentiles, risk metrics, trigger distances, or test results is computed from available source data. AI writes interpretation around that evidence and can still be incomplete or wrong.
What becomes public when Advanced Analysis finishes?
A persisted preview containing the first 35% of the narrative, selected sanitized charts, and non-actionable watch labels. The same projection is used for humans, crawler HTML, Markdown, and structured data; paid trigger details and the rest of the report remain private.
Does a backtest prove a strategy will work?
No. It describes historical behavior on a particular sample and is sensitive to data quality, parameters, market regime, liquidity, spread, and overfitting. It is evidence, not a forecast or guarantee.
How are completed research updates discovered by search engines?
A durable publication event submits the revised public URL to IndexNow, while canonical HTML, Markdown alternates, internal links, and updated sitemap timestamps support ordinary search-engine discovery.
Is CommonQuant financial advice or a brokerage?
No. CommonQuant provides educational research and alerts, does not execute trades, and does not guarantee returns. Users remain responsible for their decisions.
Continue exploring CommonQuant
Put your thesis through the research pipeline
Start free with no credit card. Review every strategy before activating it; CommonQuant sends signals and does not execute trades.
Create a free CommonQuant account