Disclosures

Last updated: September 9, 2026

This page is a standing disclosure, not a contract: who produces the signals on Apibeary (apibeary.com), what the labels mean, and where our interests could touch yours — stated plainly, because a reader weighing a signal deserves to know all three.

Who produces the signals

Every signal is produced by automated systems — scheduled scouts reading public sources (SEC filings, contract awards, congressional trade disclosures, Fed language, market data), including, for some scouts, AI language models reading text — operated by Heronward LLC, doing business as Apibeary, a Florida limited liability company. The operator does not write signals by hand and does not adjust a signal's direction or conviction before it publishes. When we find a defect we correct it in the open: the underlying event log is append-only, and corrections land as their own labeled rows rather than silent edits.

Regulatory status

Apibeary is not registered with or authorized by any financial regulator in any jurisdiction — not the U.S. Securities and Exchange Commission, the UK Financial Conduct Authority, Germany's BaFin, nor any other. Apibeary is not an investment adviser or broker-dealer, holds no client assets, and cannot place trades. Everything here is information, not advice.

The ratings key

Every signal has four parts: who (the scout that fired), direction (▲ bullish or ▼ bearish — or neutral, when a read leans neither way), confidence (1–5), and the reason — one sentence about what the scout saw.

▲ and ▼ classify the event, not the price. It is not a forecast, and no scout's direction has been validated as predictive. For most scouts the arrow is a convention about the shape of the thing that happened — an insider bought, a contract was awarded — and those are facts from filings. For a few it is a model reading text, or a rule built on a statistical premise about what has tended to follow — and that premise is as unvalidated on the live stream as every other direction here. None of the three is a prediction that a price will move, and none of them is advice.

On a single signal, confidence and conviction are the same 1–5 scale — the API says "confidence", the dashboard says "conviction". It is the scout's own certainty grade in its rule, from 1 (watch-grade: noted, not notable yet) to 5 — a heuristic, not a calibrated probability.

On a rolled-up ticker row, conviction is a sum, not a grade. The brief and other per-ticker views add up the 1–5 grades of every agreeing scout — each multiplied by a standing per-scout credibility weight (1 for most scouts; a scout that fires very often can be permanently down-weighted, e.g. to ½) — so the number grows with corroboration and has no ceiling. A conviction of 11 means several scouts lean the same way, not "11 out of 5". Five cautious 1s and one emphatic 5 both sum to 5: the per-scout grades and the scout count are shown on the row, so the parts are always visible beside the total.

The live stream is young — recording since May 2026 — so treat live patterns as unproven. Where results are graded, the record is published with the misses in the same font as the hits.

Conflicts of interest

The operator of Apibeary invests personally, and holds positions in securities the scouts may flag. The standing policy:

Publish first. Signals reach the feed on the scouts' schedule. Any trading by the operator happens at or after publication — reading the same recorded stream any member's bot reads — never on unpublished scout output. Publication means the moment a signal is written to the feed the dashboard and API read. Push alerts go out after that on a polling cycle and email digests later still, so a member who waits for an alert sees a signal after a bot reading the feed does — and after the operator could have.
Hands off the signals. The scouts are automated; the operator does not create, boost, retime, or re-word a signal around a position. The append-only event log is the record we would point to: it shows what was published and when, which is the part a reader can check.
Amounts stay private; the policy is public. We disclose the existence of these accounts and the rules they run under, not balances. If any position ever exceeds 0.5% of an issuer's issued share capital, we will say so here. That threshold is the EU recommendation rules' disclosure line, adopted as a floor, not a claim that smaller positions carry no conflict — the first sentence of this section covers those.
Nobody pays for coverage. Apibeary accepts no payment from issuers, promoters, or any other third party for coverage or placement of a signal. If that ever changes, this page changes first.

Sources and methods

Every signal names its scout and why it fired; filing-based signals name the document they read — no black box. Signals are dated by the event's own date where the source gives one, and sightings that reach us more than a week late are labeled as late. The field guide explains each scout; the grading pages publish how results are scored. Market Data from Tiingo.com.

Accuracy

Where filing-based signals come from. Every insider-trade and congressional-trade signal is derived from a named public filing — SEC Form 4 or a congressional financial disclosure — and links to it. The facts in a sentence — who, what, how much, when — come from that filing. The direction label, the conviction score and markers such as a late-filing note are our own classifications of what the filing shows, as the ratings key above explains.

How sentences are checked. Before a correction is written, the sentence is machine-verified against the filing it describes: arithmetic, dollar sums, share counts, dates, roles and plan attribution must reconcile, or the write is refused. Corrections are visible by design — every corrected row carries a dated correction clause saying what changed, and because the underlying event log is append-only, a row is corrected by adding to it rather than rewriting it.

What we measured. Between August 25 and September 1, 2026 we ran thirteen audit rounds over the published corpus, each independent of the last: a fresh stratified sample, checked sentence-by-sentence against the original filings by AI reviewers we ran ourselves, with every serious finding adversarially re-checked before it was accepted. Independent here means separate rounds and separate samples, all run by us rather than by an outside firm. The final round sampled 260 published rows and found false statements in 6 of them — 2.3% of that sample, before correction. That is a measurement of the sample, not a corpus-wide error rate. All six were corrected the same day, along with every other row in the corpus that matched their error patterns — 63 rows in total. As of September 1, 2026, 5,701 of 8,852 published insider-trade rows carry a visible correction clause from this process. A correction clause is any dated note the process appended to a row: a figure or its arithmetic corrected; a sale reclassified (for example, shares withheld for tax rather than sold by choice); context the filing carried but the sentence had left out, such as a late filing, a pending merger or a catch-up filing; or the row voided outright. Carrying a clause means the row was touched by the process, not that its facts were false — the clause itself says which.

What we do not claim. Insider filings bury meaning in free-text footnotes, and no automated reader gets unbounded prose right every time. We do not claim zero errors. What we do claim: arithmetic is machine-checked at write time; a filing we cannot read produces a labeled refusal rather than a guess; corrections are visible rather than silent; and our own re-audits run on an ongoing basis — a deterministic sweep re-reads newly published rows against their filings every week, and a fresh adversarial sample audit runs every month. Rows whose source filing could not be re-fetched for verification are tracked by name (51 as of September 1, 2026) and retried on every sweep. Signals are observations about public filings, labeled bullish or bearish with a conviction score — see the ratings key above — and, as that key says, information rather than advice.