How Filing Clarity Works

1. How materiality is determined

Filing Clarity monitors 8 SEC filing types and classifies events into 28 signal types. A signal is generated when a filing contains a disclosure that a reasonable investor would consider material to their investment decision. We err on the side of inclusion for high-importance signal types (Executive Departure, Going Concern, Activist Investor) and apply a higher materiality threshold for routine signals (Tax Withholding, Quarterly Filing).

2. How the AI extracts signals

Each filing is processed by Claude (Anthropic's AI), which reads the full document, classifies the primary signal type, and generates a plain-English headline with specific figures. The headline follows strict rules: always includes the ticker, at least one specific number, and the person's name for insider and executive filings. Maximum 120 characters.

3. The open-market vs. award distinction

For Form 4 insider transaction signals, Filing Clarity classifies each transaction as:

• Open-market sale/purchase (discretionary — higher signal)
• Tax withholding (routine — lower signal, labeled accurately)
• Scheduled 10b5-1 plan sale (pre-planned — labeled as such)
• Share grant or award (not a purchase — labeled accurately)

This distinction matters: a discretionary open-market sale by a CFO carries different information than a routine tax withholding transaction.

4. Known limitations

Filing Clarity is not perfect. Known limitations:

• Earnings signals for large-cap stocks may not add information beyond what is already available on financial news sites within minutes of announcement
• Some complex filings (restructurings, multi-part 8-Ks) may generate multiple signals that overlap
• Signal extraction accuracy depends on filing structure — well-structured filings extract more cleanly than narrative-heavy ones
• We do not monitor earnings call transcripts in real time (earnings call summary is a forthcoming feature)

5. Error correction

If you believe a signal is inaccurate, mis-classified, or missing important context, email signals@getfilingclarity.com with the signal ID and your correction. We review all reported errors within 24 hours. Corrections are applied to the signal detail page and noted with an amendment timestamp.

6. What we do not do

Filing Clarity does not provide investment advice. Signals describe what was disclosed in a filing — they do not recommend action. The decision of what to do with the information is always yours.

Going Concern vs. Risk Factor Language

Filing Clarity distinguishes between two types of going concern language that are frequently confused:

Genuine auditor qualification: Language appearing in the independent auditor's report stating "substantial doubt about the company's ability to continue as a going concern." This is a formal auditor opinion with significant implications. Filing Clarity classifies this as a high-severity Going Concern signal.

Standard risk factor disclosure: Language appearing in the Risk Factors section describing potential liquidity risks in general terms. This language appears in thousands of SEC filings, including from financially healthy companies, as a standard legal disclosure. Filing Clarity classifies this as a low-severity Risk Factor signal.

We never classify risk factor section language as a Going Concern signal. If you receive a Going Concern signal from Filing Clarity, it indicates language in the auditor's report — not boilerplate risk disclosure.

Infrastructure and Reliability

Filing Clarity's signal delivery pipeline is fully automated and operates independently of manual intervention. The EDGAR polling, AI extraction, and email delivery systems run continuously on dedicated cloud infrastructure (DigitalOcean).

What this means in practice:

• EDGAR is polled for new filings continuously, 24 hours a day, 7 days a week
• Signal extraction and email delivery trigger automatically within minutes of a new filing
• No manual action is required for signals to be generated and delivered

Filing Clarity is founder-operated. We are transparent about this: signal delivery does not depend on the founder's availability, but product development, customer support, and strategic decisions do. If you have concerns about business continuity for a specific use case, please contact us at support@getfilingclarity.com to discuss. Live pipeline status is available at /status.

How we vet every signal for accuracy

Every signal passes through layered verification before it reaches you. The goal is simple: a signal should either be confirmed against its own source filing, or it should tell you plainly that it could not be. We never silently ship a signal we were unable to verify.

Deterministic checks first. Before any judgment is applied, we re-derive the numbers in plain arithmetic. A purchase's total must equal shares × price; a percentage change must reconcile against the figures it is drawn from; an annualized dividend must equal the periodic payout times its frequency; a changed stake's share counts must add up (prior holding ± the change = the new holding). We also cross-check the ticker and company against SEC reference data. Anything that does not reconcile is corrected against the filing, or escalated for closer review.

A progressive review ladder — each stage more rigorous than the last.

• Stage 1 — verification against the source. Every signal is checked against the specific SEC filing it was generated from, on classification, figures, transaction type, and completeness. Signals that pass cleanly are delivered.
• Stage 2 — heightened review. Anything that does not pass cleanly, and every high-stakes signal type (going concern, guidance changes, executive departures, insider-sale clusters, activist stakes, earnings surprises), gets an independent, stricter second pass with type-specific checklists and full arithmetic re-derivation. Where the filing supports a fix, the signal is corrected here rather than escalated.
• Stage 3 — deep resolution. Anything still unresolved gets our most thorough audit against the full filing text — maximum effort to determine the single correct answer, reconcile every figure, and confirm the ticker, company, and named parties all refer to the same entity.

At each stage the aim is to resolve the question — to confirm the signal or correct it against the source — not to reject it. A signal is only ever set aside for the final step below when even our most rigorous review cannot settle it from the filing.

When we can’t fully verify a signal: we flag it, we don’t hide it

On the rare occasion our most rigorous review still cannot confirm one detail of a signal against its source filing, we do two things we think sophisticated investors should expect, and one thing we refuse to do.

We deliver it anyway. We do not quietly withhold a signal because one detail is uncertain — you would rather know that something happened and judge the detail yourself than never hear about it.

We tell you exactly what we could not confirm. The signal arrives with a clear “Flagged for your review” marker and a plain-English explanation — what specifically we could not verify, and what to check — with the source filing linked so you can confirm it yourself in seconds. The explanation is written to stand on its own and answer the obvious question, not to hide behind a confidence score or internal jargon.

What we refuse to do is present an unverified detail as if it were certain. A signal you receive from Filing Clarity has either passed our checks, or it carries an explicit, specific caveat about what did not. You are never left with a silent error dressed up as fact.

The practical result: uncertainty is surfaced, not buried. Every signal is traceable to the exact filing behind it, corrections are made against that source rather than guessed, and on the occasions we fall short of certainty we say so, in language you can act on. We would rather earn your trust by being candid about the edges of what we know than by overclaiming a precision we cannot honestly stand behind.

How our review performed in the most recent audit. In our most recent internal accuracy audit — a representative sample of 100 signals spanning all 28 signal types and large-, mid-, and small-cap issuers — 61% passed initial verification and were delivered as-is. 39% were escalated to a stricter second review, where the large majority were confirmed or corrected against the source filing. 15% required our deepest, most thorough audit. In the end, 3% were delivered with a “Flagged for your review” note because one detail could not be fully verified — and none were withheld. This was a deliberately challenging sample, not live customer traffic; the figures vary from audit to audit, and we report the most recent.

Signal Quality and Accuracy

We do not publish a classification accuracy rate. Here is why, and what we do instead.

Publishing a single accuracy percentage would be misleading without specifying the methodology: how the sample was selected, how "accurate" is defined for each signal type, how many signals were verified, and by whom. A number without this context implies precision we cannot honestly claim at our current scale.

What we do instead:

We publish our classification rules. Every signal type has explicit classification criteria documented on this page. You can evaluate the rules directly and identify edge cases we may have missed.

We document known edge cases. For example: same-day exercise-and-sale Form 4 transactions require different classification than discretionary open-market purchases, even though both appear on the same form. We document these distinctions explicitly so you know what to look for.

We maintain a correction log. When a signal is reported as inaccurate, we review it within 24 hours, correct the signal if warranted, and update the classification rule to prevent recurrence.

We will publish accuracy data when we have it. Once we have accumulated sufficient volume of verified classifications to report statistically meaningful accuracy rates by signal type, we will publish them. We expect to reach this threshold within 12 months of commercial launch.

What "confidence" means in our signals: our extraction system does not output a confidence score in the traditional ML sense. Classification is rule-based with AI judgment applied to ambiguous cases. When the filing language is ambiguous or the transaction type falls into a known edge case, we classify conservatively and note the ambiguity in the signal summary.

How Our AI Classifies Signals

For a plain-English list of every signal type with trigger criteria and examples, see the Signal Library.

This section documents exactly how Filing Clarity classifies the signals you receive — the transaction codes, trigger rules, and qualitative distinctions behind each signal type, plus an honest accounting of where our current classification has known limitations.

1. How we classify 10b5-1 vs. open-market sales

Every Form 4 transaction is classified using the SEC's transaction codes before any signal is generated:

• Code P (Purchase): the executive paid cash to buy shares on the open market. Classified as Insider Buy. This is a discretionary decision — high signal.
• Code S (Sale): we check footnotes for a 10b5-1 plan reference. If a 10b5-1 plan is cited, it is classified as a scheduled sale and labeled explicitly. If no 10b5-1 plan is cited, it is classified as an open-market discretionary sale — higher signal.
• Code M + Code S on the same form / same date: a same-day exercise-and-sale — the executive exercises options and immediately sells the resulting shares. This is a mechanical transaction, not a discretionary decision. Classified as exercise-and-sale and labeled explicitly. Never classified as an open-market purchase.
• Code F: shares withheld for tax on vesting equity. A routine administrative event. Classified as Tax Withholding — lower signal, labeled accurately.
• Code A: a share grant or award. Compensatory. Classified as Other.

Known limitation: some older 10b5-1 plan references appear in filing body text rather than structured footnotes. Our extraction may not catch all such references. We are improving footnote parsing continuously — see the improvement log.

2. Insider Sale Cluster trigger rules

An Insider Sale Cluster signal fires when ALL of the following conditions are met:

• Two or more different insiders (different names) at the same company;
• Each transaction is classified as open-market discretionary (Code S without a 10b5-1 plan);
• All transactions occur within a 90-day window;
• Tax withholding (Code F) and exercise-and-sale (Code M + Code S) transactions are excluded — only discretionary sales count.

Deduplication: a cluster signal fires at most once per company per 30-day period to prevent re-alerting on the same underlying pattern. The 90-day window and 2-insider minimum are configurable parameters chosen to match the window and pattern density professional investors typically flag as potentially meaningful.

3. How we detect qualitative signals

Going Concern: we check two distinct locations in each filing — (1) the independent auditor's report section, and (2) the risk factors section. Only language in the auditor's report produces a Going Concern signal. Standard risk-factor language about potential liquidity risks — which appears in thousands of filings from financially healthy companies — is classified separately as a lower-signal Risk Factor. This distinction is the most important quality gate in our going concern detection.

Executive Departure: we detect departure language in 8-K Item 5.02 (Director and Officer Changes) and classify the departure as planned or abrupt based on whether a successor is named and the stated effective date. Known limitation: some departures are disclosed in the body text of other 8-K items; we may not catch all departures disclosed outside Item 5.02.

Activist Intent (13D vs. 13G): a Schedule 13D is filed when an investor crosses 5% ownership with active intent. We classify the stated purpose from the Purpose of Transaction section and flag 13G → 13D conversions explicitly as passive-to-active shifts — one of the most significant ownership-behavior changes we detect. Known limitation: stated purpose language can be boilerplate, and we do not independently verify whether stated activist intent will be acted upon.

4. Known limitations and false-positive ranges

We believe transparency about limitations builds more trust than silence. Here is what we know:

• Language tracking in custom earnings metrics: our current implementation matches tracked words and phrases across the full filing text, including risk-factors sections and boilerplate disclosures. A word in a risk-factors section is not the same signal as the same word in management's forward guidance. We are building section-aware language tracking to address this. Expected improvement: Q3 2026.
• 13F reporting lag: 13F filings are submitted 45 days after quarter end, so the positions disclosed reflect prior-quarter holdings. A fund may have already exited a position by the time the filing reaches EDGAR. We deliver 13F signals within 30 minutes of filing — but the data is inherently 45 days old.
• Earnings transcript coverage: our Stage 2 earnings-call analysis performs best for companies that file full transcripts as 8-K exhibits. Where only prepared remarks are available, Q&A analysis will be limited or unavailable. We are integrating a dedicated transcript source.
• Going concern false positives: before our auditor-report / risk-factor distinction was implemented, some risk-factor language was incorrectly classified as Going Concern. This was corrected in July 2026, and all historical going-concern signals were reviewed and reclassified.
• Number formatting: we identified and corrected a comma/decimal mis-parse issue in July 2026 that could cause share counts and dollar values to be reported 1,000x too small. All affected signals were identified and corrected, and the parse_number function now handles all standard US financial number formats correctly.

Signal Methodology in Depth

How Filing Clarity reads a filing

Filing Clarity uses Claude (Anthropic's AI model) to read each filing in full — not just the headline section, but the complete document including footnotes, MD&A, risk factors, and exhibit content. The extraction is structured around a set of classification rules specific to each filing type.

XBRL data — scope and limitations

Filing Clarity uses XBRL-tagged financial data where available for numerical verification (revenue figures, EPS, share counts). Important limitation: XBRL tags cover structured financial data but not qualitative disclosures — executive departure language, going concern qualifications, and activist intent statements are extracted from unstructured text, not XBRL tags. We do not claim XBRL validation for qualitative signal types.

Form 4 classification methodology

Every Form 4 transaction is classified by transaction code before any signal is generated:

• Code P (Purchase): open-market buy — high signal.
• Code S (Sale): check for a 10b5-1 plan in footnotes. If 10b5-1, scheduled sale — lower signal, labeled. If no 10b5-1, discretionary sale — higher signal.
• Code M + Code S same date: exercise-and-sale — mechanical, not discretionary, labeled explicitly.
• Code F: tax withholding — routine, labeled as such.
• Code A: share grant — compensatory, labeled as Other.

Going Concern classification methodology

Filing Clarity distinguishes two materially different disclosures that are frequently confused:

• Auditor's Report language: "substantial doubt about the company's ability to continue as a going concern" — a formal auditor qualification. High signal.
• Risk Factors language: standard sectoral disclosure about potential liquidity risks. Appears in thousands of filings from financially healthy companies. Low signal, classified as Risk Factor, not Going Concern.

If you receive a Going Concern signal from Filing Clarity, it indicates auditor's report language — not risk factor boilerplate.

13F / 13D / 13G classification methodology

• New position: ticker not present in the prior 13F.
• Increased position: >10% increase in share count. Decreased: >10% decrease.
• Exited position: ticker present in prior 13F, absent now.
• 13G → 13D conversion: same filer, same ticker, schedule change — flagged explicitly as a passive-to-active intent shift.

Human review layer

Signals below our confidence threshold (0.85) are held and reviewed by a human before delivery. Approximately 40.7% of signals are held for review. Of those reviewed, approximately 0.0% are corrected before delivery. All corrections are logged in the improvement log on our /accuracy page.

Known limitations

• Earnings signals for widely-covered large-cap companies (NVDA, AAPL, MSFT, TSLA) may not add information beyond what is available on major financial news sites within minutes of announcement. Our value for these names is the non-headline guidance language and earnings call analysis.
• 13F filings are submitted 45 days after quarter end — positions disclosed are 45 days old by the time they are public. Filing Clarity delivers 13F signals within 30 minutes of filing, but the underlying data reflects prior-quarter holdings.
• Qualitative signal types (executive departure context, activist intent language, MD&A tone) are extracted from unstructured text and are not XBRL-validated. We apply structured classification rules but acknowledge that qualitative extraction carries higher uncertainty than numerical extraction.

Known Classification Edge Cases

These are transaction types or filing structures that require careful handling. We document them here so you can evaluate our approach:

Form 4 — Same-day exercise-and-sale: When an executive exercises stock options and sells the resulting shares on the same day, both transactions appear on the same Form 4. This is a mechanical transaction, not a discretionary purchase. Filing Clarity classifies these as Insider Sale (exercise-and-sale) rather than Insider Buy. The headline will include "exercise-and-sale" to make this explicit.

Form 4 — 10b5-1 plan sales: Sales made pursuant to a pre-established 10b5-1 trading plan are scheduled in advance and do not reflect real-time management sentiment. Filing Clarity notes 10b5-1 plan sales explicitly in the signal headline so you can apply appropriate weight.

Going Concern — Auditor qualification vs. risk factor: See the Going Concern vs. Risk Factor section above.

13F — Reporting lag: 13F filings are submitted 45 days after quarter end, meaning the positions disclosed are already 45 days old by the time they are public. Filing Clarity delivers 13F signals within 30 minutes of the filing hitting EDGAR, but the underlying position data reflects the prior quarter's holdings.

8-K — Earnings call transcripts: Earnings beat/miss signals are generated from the 8-K filing, which contains the press release. The full earnings call transcript (including Q&A) is not always available immediately. Our Stage 2 earnings call analysis — which includes prior quarter language comparison, analyst Q&A signals, and management tone assessment — is delivered separately when the transcript becomes available, typically within 2 hours of the call completing. Earnings signals are most valuable for mid-cap and small-cap companies where institutional coverage is thin. For large-cap names, the headline beat/miss result is widely available — our differentiated value is the forthcoming earnings call analysis including prior quarter language comparison.

Error Correction

If you believe a signal is inaccurate, mis-classified, or missing important context:

1. Email signals@getfilingclarity.com with the signal ID (visible on the signal detail page) and your correction.
2. We review all reported errors within 24 hours.
3. If the correction is warranted, we update the signal detail page with an amendment note and timestamp, and update the classification rule to prevent recurrence.
4. We do not delete incorrect signals — we amend them with full transparency about what changed and why.

Recent corrections: In July 2026, a same-day exercise-and-sale Form 4 transaction was initially classified as an open-market purchase. This was incorrect. The classification rule was updated to explicitly handle same-day exercise-and-sale transactions. The signal was amended with a note explaining the correction.