How we verify —
and when we can’t.

The standard: accuracy over completeness.

Every number, chart, or claim we show must be defensible from the underlying data. When the data doesn’t support a number, we don’t show it — we show what we have and explain the gap. A wrong number is worse than no number. Everything below exists to enforce that one sentence.

Where the data comes from.

Every analysis surface shows the as-of time of the data it was built from, and the date range it actually covers.

How sources check each other.

• Prices are verified across two independent sources. When they agree within 1%, you’ll see a Verified · 2 sources mark on the ticker page. When they diverge, we flag the drift instead of quietly picking a side.
• Fundamentals anchor to filings.Where possible, valuation figures are recomputed from what the company actually filed with the SEC. Vendor numbers that can’t be anchored need independent agreement; when sources disagree beyond tolerance, the figure is stored as low-confidence and treated accordingly — not averaged into false precision.
• Analysis is cross-examined, not single-sourced. Three independent lenses — Foundation, Expansion, Climate — each built on a different frontier model family, read the same verified evidence in isolation. A supervisor reviews all three and surfaces genuine disagreement instead of smoothing it over.
• Citations are checked by a deterministic verifier. Sentences that cite a figure are re-checked against the source data mechanically — a claim that cites nothing, or cites a number that isn’t in the evidence, is rejected before you ever see it.
• Stored claims are re-verified on a schedule.A weekly drift check re-tests previously shown claims against their sources, so an error that slips through doesn’t quietly persist.

When we won’t show you a number.

• Missing data renders as a dash, never a zero. A blank is honest; a fabricated 0 is not.
• No interpolation.If we don’t have an observation for a date, we don’t draw a line through it.
• Charts never mix resolutions. A portfolio chart covers only the range where every account has real history.
• Historical signal studies have hard floors. A signal study renders only with at least 120 trading days of history and at least 5measurable historical occurrences. Below the floor you get an honest “not enough data” — never a thin sample dressed up as a distribution. And what it shows is a historical record, not a forecast.
• Stale data says so.If a “today” figure is built on data older than the last market day, the surface says “as of” that date prominently — or declines to make the claim.
• No silent downgrades.If a primary source is unavailable, we label the fallback (“as of yesterday,” “estimated”) instead of pretending.

How we rank sources.

Not every source carries equal weight. When accounts conflict, our analysis leans on the more authoritative one — a filing over a report about it, the wire over a blog — and says so. The ordering, most to least authoritative:

1. 1.primary source — Regulatory filings and central-bank data — the document itself, not a report about it.
2. 2.major press — Established financial press and wire services with editorial standards.
3. 3.independent commentary — Independent analysts and respected commentators — opinion, weighed as opinion.
4. 4.community chatter — Social and forum chatter — an attention signal, never a verdict input.

And when two independent sources disagree on the same figure, we show both rather than average them into false certainty.

Where receipts live today.

Different surfaces carry their evidence in different ways. So this page doesn’t overclaim, here is the current state:

• Trade coaching, watch alerts, and outlook insights write per-claim records to a claim ledger — each carries a “Why?” trail listing the exact data points, their sources, and dates.
• Research verdicts cite their sources inline; every cited figure is mechanically checked against the evidence block the analysis ran on, and the verdict records the data as-of time.
• Price-move and portfolio alertsare computed deterministically from the same nightly-refreshed data. Extending full claim-ledger trails to every alert type is in progress — where a trail doesn’t exist yet, the alert still states what triggered it and when.
• Ask (Q&A about a saved analysis)answers only from that archived analysis, cites the sections it drew on, and rejects any figure that isn’t present in it. It works at section granularity rather than per-number claim-ledger trails, so treat it as an explainer of an existing verdict, not a fresh-verified source.

What this is, and what it is not.

ClearPath is a research desk. It does not execute trades, does not manage money, and does not tell you what to do — the verbs stay in your hands. Everything we publish is informational and educational; nothing here is investment, tax, or legal advice, and our analyses can be incomplete or wrong. Where our methods fall short of the standard above, we’d rather say so on this page than pretend otherwise. Full legal terms live in our disclosures.