What Buteon does today, what recently shipped, and what is being explored next. Recorded the way Buteon records everything — with the limits stated.
Anyone can create a free account and open the product; no invite is required. The focus is unchanged: signal clarity, reliability, outcome memory, and feedback from the people using it — not growth for its own sake. Features keep improving, and some remain developing.
buteon.ai remains the public content and research hub. Start with the Intelligence Library, then see the method applied in Case Files and Research Notes. The product itself lives at app.buteon.ai, with a Free plan available to everyone and Plus and Pro for more depth.
Everything under The system today is live in the product or on the public site now, and each module states where its coverage ends. The release log records what was added or fixed. Roadmap items are directions being worked on, evaluated, or held — planned directions, not commitments, and not all features are available on every plan.
Buteon is built as a sequence of reads, not a feature list. Each layer answers one question, and each module below states its own limit.
The feed is deliberately quiet: across the names Buteon tracks, most filings never reach a card, and most cards carry no chip at all.
Ranked feed of insider and ownership signals, each carrying a verdict, confidence and freshness read.
Coverage is partial and scoped to the names Buteon currently tracks — an absent ticker is a coverage limit, not a read. There is no plan-based delay — every plan sees a read at the same time.
Filter the stream by verdict, cluster activity, freshness, priority, notable moves, watching or non-actionable.
Eight built-in lenses; custom saved views are kept separately.
Keep the companies you follow on one page, each with its current read and the alerts you chose for it.
How many companies you can follow depends on your plan; at the limit, unfollow one to make room. Following never changes the read.
A compact list of recent activity at the companies you follow.
Covers the last seven days and shows at most five items. Followed companies do not float to the top of the stream: relevance is only a tiebreaker after verdict, alert priority and freshness.
At most one chip per card, and only these five: High Signal Tension, Read capped, Evidence mixed, Big-player confluence, Notable activity.
A chip marks where to look inside the signal; it is never the explanation and never a recommendation. Risk Gate does not appear on stream cards.
Discovery grid over tracked tickers: fresh reads, measured history and actor activity.
Not a market-movers board, scanner or opportunity finder: everything shown is drawn from filed evidence.
In-app inbox of updates for the companies you follow, matched to the alerts you chose for each one.
Alerts cover activity at the companies you follow that Buteon picks up soon after it is filed; filings picked up late, and filings still awaiting analysis, do not create alerts. Daily and weekly email summaries are rolling out on every plan. Alerts reflect current signal coverage, which is partial.
One setting on the Following page for how alerts reach you, applied to every company you follow.
Daily and weekly email summaries are rolling out on every plan, and instant email on Plus and Pro; email stays off until it is available for your account and you choose it, and summaries can be late or skipped. Push notifications are not available yet.
On-demand in-app brief on recent reads — selected by evidence quality, not performance — and on what moved at the companies you follow.
Built from the most recent signals Buteon has loaded: not ranked, and not a complete view of the period. Generated on demand in the app, separate from any email summary, and on no schedule. Historical context, not a recommendation.
Open any signal and the reasoning is laid out — including what argues against it, what caps it, and where the evidence runs out.
The read inside every signal’s Full analysis: verdict, Read Confidence, horizon and the reasoning behind them.
A verdict is a read on disclosed activity, never a recommendation to buy or sell. Verdict, confidence and freshness are categorical labels, and horizon names the window the read is scoped to — not a price forecast and not a holding period.
Counts the references behind a verdict: supporting, weakening, contradicting, context, forward.
Counts only — the five groups are never blended into a single number. Some signals show no structured evidence items at all.
What runs counter to the verdict: evidence conflict, market non-confirmation, or a stale move.
Categorical only — the meter prints no number. When coverage is insufficient the panel says so instead of showing a level; when it is partial the level appears with the coverage limit stated beside it. It reads conflict; it does not forecast.
Names the ceiling holding a bullish-leaning read below a stronger grade.
Appears only on bullish-leaning reads — bearish verdicts are out of scope. It explains a ceiling the engine already recorded and never promises the read will upgrade. On stream cards it appears as the chip “Read capped”.
Structural tags — role, transaction type, isolation, size, freshness — characterizing the filing.
Descriptive tags, not a score or a ranking. Some dimensions are omitted when the underlying data is unknown, so the tag set varies between signals.
Flags what could weaken or cap a read, within available Form 4 and engine signals.
Coverage is partial by design: Form 4 filings plus existing engine signals only. Broader regulatory filings are outside its scope. “None detected” is limited scope, never an all-clear. Display-only — it does not change verdicts, scoring or alerts, and it is not a sell or short read. It appears only inside the signal dossier, not on stream cards or the company page.
Recent market context around a signal: trend, backdrop, volume, short-window moves.
Informational only, never a direction claim. Market data loads only when you click “Load market context”, and it may be delayed, partial or unavailable — the panel says which. It contains no beta, no sector-relative measure and no volume multiple.
Categorical read on the acting insider: reliability band, provenance, same-company outcome history.
No numeric actor score and no standalone alignment percentage. Outcome history is same-company scope and shown as counts rather than rates, with sample size stated: historical context, not a performance claim. Identity matching is conservative and carries an explicit caveat when verified history is thin.
Public reference for how to read a card: verdict, confidence, freshness, and what Buteon will not claim.
Readable without an account.
Context is aggregated only as far as the filings support it, and never presented as complete. Each company page states which coverage tier it is showing.
Ticker-level command page: current read, active signals, actor activity, market context, outcome history, audit.
You reach it by clicking any ticker; the page is titled with the ticker itself. Depth varies by coverage tier: actor activity, big-player context, outcome memory and historical analogs appear only for companies marked Covered, while Limited coverage and Directory only companies show a reduced page. There is no plan-based delay — every plan sees a read at the same time.
Click a marker on the price chart to open the read behind it, across selectable timeframes.
It plots signal markers on a price series and measures no statistical relationship between reads and price movement. It sits on the company page inside Ticker Pulse and appears only after you click “Load market context”; nothing loads before then. Unavailable for directory-only companies, and price history can be delayed or missing for smaller and non-US symbols. Informational context, not a forecast.
Same-company view of one actor on one ticker: read, pattern, filings, connected signals.
Same-company scope: every row is one actor on one ticker, never a cross-company identity. No numeric actor score, no ranking, no leaderboard. The page states that market-regime and cross-company views are not available yet, and identity states whether a CIK is on record, with a caveat on how firmly it is established. Its roster is drawn from the signals already loaded and shows at most two hundred rows — not a complete actor directory.
Categorical reliability read for this actor on this ticker, with its reasons.
Never a numeric score. It requires a signal where this actor is the primary insider among those loaded; otherwise it states that the read is not available rather than inventing one. The source may be inferred from role or profile rather than aggregated history.
Filing structure: open-market versus routine, planned share, cluster participation and direction.
Same-company scope, read from a fixed history summary attached to the signal. Direction appears as raw counts, never as a ratio or percentage. It shows an empty card when no history summary is attached, and it describes past filing structure only — not future filings.
Signals where this actor appears on this ticker; each row opens the signal.
Same-company scope: the count line names the one ticker it covers. Drawn from the signals already loaded, capped at twelve rows, and empty when nothing matches. Not a cross-company signal history.
This actor’s Form 4 filing references with accession numbers, identity resolution and read provenance, in expandable rows.
Contents are Form 4 filing references and identity and provenance details for one actor on one company only — not a broader regulatory filing archive.
What Buteon flagged, and what unfolded after, at the T+5 and T+30 horizons.
Historical context — not a forecast, not a performance claim, not a track record. Counts are measured cases, never a rate. T+30 cohorts are developing and coverage is partial; tickers with no matured outcomes show an empty state. On the company page it appears only for Covered companies.
Past signals with the same structural shape, and what followed them.
Structural matches, distinct from the Outcome Engine’s measured cases. Matching is strict, so “no identical analogs” is a common result. Analog outcomes appear only after you ask for them and are raw thirty-calendar-day price moves, not benchmark-adjusted and not the Outcome Engine’s T+30 horizon. Samples are typically small, and analogs are not available for older signals. Historical context, not a prediction.
Disclosed Section 16 insider and ten-percent-owner activity, with per-ticker ownership dossiers.
Ranked by transaction materiality only — never by reliability or outcome, and never a buy list. The ten-percent-owner and beneficial lane is Section 16 only and its coverage there is thin; Schedule 13D and 13G large-holder disclosures sit outside this lane and appear in Live Wire, with limited reviewed coverage. Multi-lane activity is ticker-level, not cross-company actor linkage.
Which managers disclosed holdings in selected names at quarter-end, from 13F filings.
A static register compiled by hand, covering fifteen tickers as of the quarter-end it was last compiled for. It changes only when it is compiled again; there is no refresh schedule. 13F is quarterly disclosed posture filed up to forty-five days late — stale context, not recent movement. No deltas, no ownership percentages, no rankings. Absence from the register is a coverage statement, not a read on a company.
These tools let you stress a read and set it beside others. Where the sample is too thin to read anything into, they say so.
Re-run a signal under stricter assumptions and watch verdict, confidence and constraints move.
Scenario output is exploratory, nothing in the lab is saved, and it never revises the recorded verdict. Not a recommendation.
Put two signals side by side and see where their reads diverge.
Exactly two signals at a time. Findings are exploratory research context, not a recommendation.
Build and track theses on companies, linking signals as evidence over time.
Theses, their status labels and notes are authored by you — Buteon does not score, rank or evaluate them. Pinning a signal can also follow its company, and how many companies you can follow depends on your plan.
What unfolded after past reads: measured cases and developing signals.
Historical context only — not a prediction, not a track record, not a performance claim and not financial advice. T+30 windows are maturing, and no measured outcome is fed back into scoring. How many measured cases you can page through depends on your plan, and Cohort Memory is a Plus and Pro tab.
The public side of Buteon is method, not marketing. It grows as work is published; this page does not keep count.
Open, source-led guides on reading insider activity and Form 4 mechanics — from how to read a purchase to why one can mislead.
Educational reference, never investment advice and never a recommendation. Each guide carries its primary SEC sources and a review date, and is revised as filing forms and SEC rules change. It covers insider and ownership reporting, not the full range of SEC filings.
Worked examples: one real filing set at a time, separating the filing facts, the read recorded at the time, and later qualitative follow-up.
Cases are chosen for what they teach, not sampled — the collection is illustrative, never representative, and never a track record. Qualitative follow-up is context, not a performance claim.
Evidence synthesis on questions the filings alone do not settle, with every figure labeled as survey, actual, estimate, forecast or projection.
Evidence synthesis, not reads Buteon produced at the time, and not a recommendation. Each note is dated to the evidence available when it was written and is not maintained as live coverage.
Free transaction-code lookup, with a dedicated page for each code it covers.
Covers the codes it publishes pages for, not the full Form 4 code table. A code describes how a transaction happened; it never says whether the transaction matters.
Month by month, newest first: the larger changes to the product, then the work published on buteon.ai. Smaller fixes ship in between and are not listed one by one, and plan changes and account mechanics are left out.
Buteon is evolving quickly, so this page may lag what is live in the product. When the two differ, the product is the current record.
The Outcome Engine opens on Measured cases, and each case opens a case story: the read Buteon recorded at the time and what happened 5 and 30 trading days later. A window that has not closed says “Not closed yet”, and a read in its first five trading days says so. Historical context, not a performance claim.
The Actor Dossier opens on a roster of the insiders in the loaded signals. Each dossier lays out one insider’s filings on one company: the name as filed, the rhythm of their filings, plain-language terms with a glossary, and past outcomes shown as counts out of the measured cases, never as a rate. Same-company scope, and no actor score.
Buteon 50 opens with a short briefing: up to two recent insider reads, how many signals were added over the last day and the last two weeks, and a few facts about insider buying in recent signals. The briefing points to reads; it never reorders the board.
The engine reads each transaction by how it happened, using the filing’s own footnotes. A sale the footnotes describe as required to cover tax carries no directional weight, unless they also cite a trading plan, and shares disposed of back to the company — in a merger cash-out or a tender — are kept as context rather than counted as selling. A filing disclosed late never rewrites an earlier read, and each read identifies the analysis version used, with results from different versions kept separate.
New accounts open on a three-step introduction: find a company you already own, or explore Buteon 50; understand its current read; and follow what matters. Search also finds companies Buteon does not actively cover yet, and their pages say so.
A product-wide pass for phones and assistive technology: larger touch targets, more reading room inside a signal on small screens, stronger contrast in light and dark themes, reduced motion respected, and focus returned where it started. Terms such as Rule 10b5-1, T+5 and T+30 are explained where they appear, and error and loading states say plainly what happened.
A free research tool on buteon.ai that replays real 13F disclosures with every date visible — when the positions were held, when they became public, and when a reader could have acted — to show what the timing changes. A companion Research Note asks whether a famous investor’s 13F can be copied, and quarterly reports replay each newly reported stock position in Berkshire Hathaway’s and Scion’s Q3 2025 filings, with every other position listed separately.
A Case File on an Oracle director’s open-market purchase, which claims only what its one filing shows; Research Notes on Oracle’s insider buying and cash flow and on Alphabet’s position in AI; and plain-language holdings reports on who reported holding NVIDIA and what Berkshire Hathaway reported holding.
Press Follow on a company and choose Essential or All updates. Supported insider-activity updates for the companies you follow arrive in Alerts inside the app, and Following keeps every followed company and its settings in one place. Following never changes the read. Daily and weekly email summaries are rolling out on every plan; until email reaches an account, alerts stay in-app.
Reads carry a Read Confidence level: how strongly the evidence supports the read. It sits beside the verdict and freshness, and it is not certainty or a probability of profit.
Each read draws its filing evidence from the insider filings made within a few days of it, while the insider’s longer history is still checked. A published read is never rewritten afterwards, and while a new Form 4 is still being analyzed, the company page says so instead of showing a read early.
Reported 13F positions for selected institutions and companies, each traced to its SEC filing: compare portfolios, see which managers reported holding a company, and follow the filing calendar. 13F positions are quarter-end snapshots filed weeks later — context, not recent trading.
Research Notes on a SpaceX executive’s planned share sale and what its Form 144 does and does not show, on Leopold Aschenbrenner’s reported holdings, on AI security companies, and on AMD.
Buteon opened as an open beta. Anyone can create a free account and open the product, with no invite required; features keep improving, and some remain developing.
A board of companies where Buteon has meaningful current evidence or an established history now opens the app. Its views separate the current board, what recently changed, the deepest history and emerging evidence, and each company opens on its current read. It is not a buy list, a recommendation or a performance ranking, and it holds fewer names when the evidence does not support more.
Every signal opens at two depths: Summary, the read at a glance, and Full analysis, the complete investigation with the evidence, what weakens it and the source filings.
Conviction gave way to evidence, confidence, case and read across the product: the ceiling chip now reads Read capped, and notability is stated as notability rather than as a strong read. A missing outcome says it is missing instead of promising it will mature, and threshold-based selections state that they are neither ranked nor a complete view of the period.
A redesign across every page of buteon.ai, a homepage that starts from one real filing, and an Insider Trading Tracker page that sets out what a Form 4 reports, what weakens a read and what comparable cases did next.
A Case File on the Jeff Bezos Amazon sale headline and what the SEC filings actually show, and Research Notes on SpaceX stock and on a planned AI-company IPO.
The stream now explains the threshold and points you to Signal Radar, your Watchlist and the Field Guide. Signal detail opens on the transaction facts, and the company page and the signal agree on the current read.
Alerts are on for new accounts and respect per-plan daily and per-ticker limits, grouping anything above the limit rather than dropping it. Your saved preferences stay as you set them.
Award and grant transaction codes stopped counting as bullish directional evidence: forty-two reads were corrected, all landing on Watching or Non-actionable, and none introduced a new bearish read. Separately, the cluster definition used in scoring was corrected and applied forward only, so historical reads reflect what the engine believed at the time they were produced.
The Outcome Engine now shows how many signals are eligible for measurement, how many have been measured at each horizon and when the next ones mature — labeled developing, and carrying no performance claim.
Signal cards, top reads, brief rows and analog rows open on the distinctive fact, so a read scans as ticker, verdict, what happened, what weakens it. A quiet section and an opt-in lens show the events where notability differs from the strength of the read, with the engine verdict unchanged beside the notability label. An execution-type chip appears only when the underlying fact is unambiguous, verdict, confidence and freshness chips explain themselves on hover, and a bearish current read no longer appears in green with an approving check mark.
A ticker-level discovery grid shipped with a Daily Hunt of names to investigate — never buy ideas — plus a coverage map and a ticker inspector. It reads from data Buteon already holds; market context loads only when you ask for it.
Power Actors added an activity board of case-file cards, a context panel and a per-ticker ownership dossier, all drawn from data Buteon already holds. The Institutional Register shipped inside it: a quarter-end 13F register compiled by hand, with filing-verified rows and copy stating that it is not a coverage claim.
Outcomes are measured against a market benchmark rather than raw price, answering what happened after comparable signals as historical context, and stating when a sample is insufficient.
A single intelligence layer drives the feed, signal detail, alerts, watchlist, compare and chart markers, carrying verdict, confidence and freshness with the labels Watching and Non-actionable.
A signal reads in one order: Verdict Brief, Evidence Stack, Why Not Green, Signal Tension, Signal Signature, Outcome Memory and Risk Gate. Audit & Source links to the source filings with a coverage caveat.
The company page shipped as a ticker-level command page with a Ticker Pulse sparkline drawn from live price history, short-window tiles and a categorical actor reliability ring. Buteon publishes no company score. The Actor Dossier shipped with Actor Read, Behavior Pattern, Connected Signals and a collapsible Audit & Source section.
Scenario re-runs in Signal Lab, two-signal comparison, theses tracked in the Research Workspace and a price chart with clickable signal markers shipped together.
The What Is Buteon page now sets out where AI is and is not used: an optional, on-demand brief narrates a read Buteon has already produced. It never collects filings, sets verdicts or offers investment recommendations.
After independent review, the number of sale transactions in one case file was restated against the source filing, gift-recipient wording was narrowed to what the filing names, and an ownership claim in a research note was narrowed to the coverage the filings actually support. Corrections are made against the source document, not around it.
Three lanes: in progress, being evaluated, and held for later. Order within a lane is not a promise of sequence — and none of this is a delivery guarantee.
Buteon’s sharpest cut: one unusual move, investigated in full — the evidence behind the read, the context around it, and where that evidence stops. Selected deliberately and published selectively, never on a cadence that would force a pick.
Research on the large ownership disclosures in Live Wire. The Live Wire feed itself is already in the app: Schedule 13D and 13G filings carried as filed, with no verdict and no inferred intent, and with limited reviewed coverage.
Daily and weekly email summaries for the companies you follow, rolling out on every plan, and instant email alerts after them. Until email reaches an account, alerts arrive in the app.
Extending what-happened-after context to more of the product as measured history matures.
A research direction for framing how long a read stays timely, and stating plainly when it no longer is.
Roadmap items describe research lanes and planned directions under evaluation. Scope and timing change as the evidence model matures, and nothing here should be read as a commitment or as a claim about results.
Every read starts from public filings and measured facts, with provenance kept visible.
Where available, an optional AI brief narrates an existing structured read. It does not change the evidence, score, or verdict, and it does not pick stocks.
Thin samples are called thin. Developing reads are called developing. Nothing is dressed up.
Buteon describes evidence. It does not tell anyone what to buy, sell, or hold.
Create a free account to start reading, or open Buteon if you already have one.
Buteon is a research tool for informational and educational purposes. It is not investment advice.