Trust & Data

The questions people actually ask before they trust a numbers-based site — answered plainly, up front.

One promise governs this page: a stranger should be able to visit ThePickLog and verify that every claim is true — without taking our word for anything. That's why the entire scored record rebuilds from two public files, and why we tell you exactly where each number comes from below.
Where does ThePickLog's data come from?

Four sources, each doing one job:

SourceWhat it providesNotes
SEC EDGAR public domain Company fundamentals, filings, and the dilution/catalyst read behind the Quality Lens; also the free-float proxy. U.S.-government public information, explicitly redistributable. The verifiable core of the site.
Yahoo (via yfinance) Prices, previous close, volume, average volume, short interest, and daily OHLC used to grade picks. Runs server-side only. It never streams to your browser — it's baked into our own published picks.csv / paths.csv / outcomes.csv.
Alpaca (paper) The owner's simulated (paper) trading account. Compete uses play-money via our own database. No real money, no order routing to any real venue.
Anthropic Powers the site assistant. Language model, not a market-data feed.

We do not scrape exchanges, and we do not resell anyone's live feed. What's public on this site is our own derived record — scores and returns we computed — not a re-broadcast market data feed.

Is any of this a live "buy now" signal?

No. Everything shown — the screen, the graded picks, the hypothesis results — is published as general-interest research after the fact, not as a real-time alert. Market data here may be delayed, derived, or estimated. Nothing on the site is a set of trades to place. See the Disclaimer for the full statement.

Do you execute trades, hold money, or connect to my brokerage?

Never. ThePickLog does not execute trades, route orders, hold funds or assets, or connect to any brokerage account. Every trade you see is simulated. We're the scoreboard and the referee — not the place you place a trade. That boundary is deliberate: it's what lets the record stay independent and checkable.

Why might a price here differ from my broker or another site?

Data providers use different venues, timestamps, consolidation rules, and split/dividend adjustments, so two "prices" for the same stock rarely match to the penny. Some values here are delayed, derived, or estimated rather than an official exchange last-trade. Treat prices on ThePickLog as indicative research data, not execution data.

Microcaps especially. Low-float names are thinly traded and can move sharply on tiny volume. Yahoo is also known to silently revise microcap daily lows after the fact — so we capture the path of each pick forward-only in paths.csv at grade time, and the record is never rewritten with a later, prettier number.

How fresh is the data, and how are returns measured?

Picks are graded five trading days out on the regular-session open → close return — enter at the day's open, exit at that day's close — net of a 2% round-trip cost haircut for wide low-float spreads. We show the worst drawdown you'd have sat through (MAE) next to every result, and we label best-case intraday spikes (MFE) as not an achievable return. The full scoring rules are on the Method page.

Can a stranger really verify every number?

Yes — that's the whole design. Every screened name is written to an append-only, tamper-evident picks.csv before its outcome is known; grading writes to a separate outcomes.csv. A single open-source evaluator (hypo_eval.py, standard-library only, deterministic) scores house rules and rules you register through the same code, and the build aborts if its numbers ever drift from the published report. Download the files and recompute anything:

⬇ picks.csv  ·  ⬇ outcomes.csv  ·  Validation dashboard ↗

Where the tamper-evidence is weaker than it sounds

Two limits we'd rather state than have someone discover:

The hash chain starts after the first picks. The log opens on 2026-06-09; the chain's genesis block is 2026-07-06. That means 253 picks predate the seal. Everything from the genesis block onward is chained — any edit to a sealed row is provable — but for those first four weeks you are trusting the git history and the Vercel deploy record rather than a cryptographic chain. We can't retrospectively seal them without defeating the point of sealing.

The external archive starts later still. Anchoring the ledger to the Internet Archive — a store we don't control, so the seal can be checked against a copy we can't touch — only began in late July 2026. Before that, the chain existed but lived entirely on infrastructure we control.

Practically: the further back you go, the more you are trusting us rather than checking us. That is the honest shape of any record that started before its own audit trail did, and it is why the strongest evidence here is always the forward record — the part written after the rules were frozen and the seal was running. See the audit log for corrections, including the seven cohorts we logged after the opening bell.

Is the AI assistant reliable? Is it investment advice?

The assistant reads the material already on the site and helps explain it — what a metric means, what a hypothesis is claiming. It is AI-generated content: it can be incomplete or wrong, so verify anything important against the primary source (the CSVs, the filing, the Method page). It is not investment advice and makes no recommendation to buy or sell.

What's licensed vs. free — and are you affiliated with the SEC?

EDGAR data is U.S.-government public information and freely redistributable; we use it within the SEC's fair-access limits and identify our crawler as required. We are not affiliated with, endorsed by, or connected to the SEC or EDGAR — "sourced from SEC EDGAR" is an attribution, nothing more. The Yahoo-derived price layer appears only inside our own committed record, never as a live feed. As the project matures we plan to move the backend price layer onto a fully licensed feed before any paid tier.

Who runs this, and how do I reach a human?

ThePickLog is a transparency-first research project operated by AMD Ventures, LLC (Florida, USA). Missing a question? Send it through the contact form — we read every message.

The Verified-by-ThePickLog Standard

An open standard for a track record a stranger can trust. Free to read, free to adopt, free to hold us to — including ThePickLog itself.

The North Star: can a stranger visit a record and verify that every claim is true — from public data, trusting no one? If any claim rests on taking the operator's word for it, the record is not verified. Everything below serves that one test.

Almost every trading track record you'll see is unverifiable by design: winners posted, losers deleted, "wins" scored at prices nobody could get, rules invented after the fact. Independent trackers routinely find advertised numbers running 10–20 points above reality. This standard defines the opposite — and we grade ourselves against it in public, including the parts that make us look bad.

The seven principles
PrincipleWhat it requires
1 · Pre-registrationEvery prediction is published before its outcome can be known, with a timestamp that provably precedes the result. Nothing is added to the record after the fact.
2 · Full distributionEvery pick is graded and shown — winners and losers. The complete set is published, not a selection from it.
3 · Realizable gradingOutcomes are scored at prices a normal participant could actually transact — a real entry, a real exit, net of realistic costs and slippage. "Ever touched +X%" is not a result.
4 · Rules frozen in advanceAny decision rule is registered with a date and judged only on picks logged after it. Backtests may inform, but only out-of-sample counts as the record.
5 · Sufficient sampleHeadline statistics are claimed only once enough graded, out-of-sample picks exist to be meaningful — a working floor of n ≥ 200. Smaller samples are labeled provisional.
6 · ImmutabilityThe log is append-only: picks can be added, never edited, deleted, or backdated. Grading runs as an automated, scheduled job — no human touches an outcome after the fact.
7 · Open verifiabilityThe underlying data is published so anyone can re-derive every headline figure. Nothing that matters is asserted; everything that matters is checkable.

What it is not. Meeting the standard says a record is honest, not profitable — an honestly-published losing record fully qualifies (ours currently is one). It is not investment advice, and it is not a seal you can buy: adoption is the badge, and the audit trail is the proof.

Our own compliance (living disclosure)

The standard, applied to its author. This can change as the project matures — if any cell says "Met" and you can prove otherwise from the public data, that's a bug in our honesty and we want to know.

PrincipleStatusWhere to check
1 · Pre-registrationMetpicks logged pre-open — Track record
2 · Full distributionMetevery pick graded, losers shown — Track record
3 · Realizable gradingMetopen→close, −2% cost haircut, drawdown shown — Method
4 · Rules frozen in advanceMethypotheses dated, out-of-sample only — Validation
5 · Sufficient sampleProvisionalsample still filling toward n ≥ 200 — Validation
6 · ImmutabilityMetappend-only log, scheduled grading — Track record
7 · Open verifiabilityMetfigures re-derivable from public CSVs — Validation

New here? The 7-check vetting guide is the plain-English version — how to run this same standard on any alert service, including us.

Prototype — sample copy pending legal review. Research and educational information only, not investment advice. Past performance does not predict future results.
Disclaimer · Method · Privacy
Build 2026-08-04 01:43 ET · fa46a8a