How a trend qualifies early
There is one published object, the Trend. Most trends qualify from accumulated movement; a critical vulnerability, a binding decision or a major shutdown qualifies at once, from credible evidence that something important is happening now.
Methodology 2.1 · importance rules events-1.0 · importance-1.0 · updated 2026-11-21
Two ways to qualify
- From history: momentum, signal volume, growth and independent sources, measured over time (formulas and thresholds).
- Early, because of impact and urgency: a current development of high or critical importance, supported by a sufficiently credible source, published before any history exists. It is still a trend, labelled “Rapidly qualified” and “Published without historical trend data”.
Every trend records its qualification path (history, rapid importance, security alert, or an editor), the rule that qualified it and why.
Detection
Each retrieved signal of the last 14 days is checked for the kind of development it describes: vulnerability, security incident, outage, shutdown or deprecation, breaking change, regulation or enforcement, license or maintainer change, safety advisory, acquisition, major model or API release, research result. Signals about the same advisory identifier, or of the same kind within the same trend, form one importance assessment, attached to the trend those signals belong to (or to a new trend built from them). Low-importance items are not assessed further.
Importance: six dimensions
- Scope: How broadly people, organisations, systems, products or regions could be affected.
- Consequence: How significant the technical, economic, legal, operational, safety or social effects could be.
- Immediacy: How quickly people need to understand or act.
- Authority: How authoritative the evidence is for this specific claim.
- Actionability: Whether there is an update, migration, mitigation, deadline or decision to act on.
- Irreversibility: Whether delayed attention would make the consequences hard to avoid or reverse.
Each dimension is rated high, medium or low from retrieved evidence and configuration (for example a list of widely deployed software), or left unknown. Unknown never lowers importance. Coverage volume is not an input: popularity cannot make a trend important.
The rules, in order: an authoritative critical escalation (for example a KEV listing) → critical; scope, consequence and immediacy all high with at least medium authority → critical; any important escalation, or at least 2 high ratings among scope, consequence, immediacy and irreversibility with at least medium authority → important; any high rating or two medium ones → notable (held there when authority is low); otherwise low. An internal index (the sum of known ratings) is stored for audit only and is never presented as a score.
Evidence confidence
- Unverified: Reported by a source that is not authoritative for this claim, and not yet confirmed.
- Primary-source claim: Stated by the responsible organisation itself. Authentic, but not independently verified.
- Corroborated: Supported by at least two independent evidence groups, at least one of them not controlled by the subject.
- Confirmed: Confirmed by an authority competent for this claim (for example a CERT or regulator), or by an editor against retrieved evidence.
Authenticity and corroboration are separate questions. The organisation responsible is authoritative about whether it published, released or decided something. It never independently confirms its own claims about impact, performance, safety, adoption or benchmarks; copies of its announcement and outlets under the same owner do not either. Confidence is assessed per claim, and the trend shows the confidence of its central claim.
Claim labels: Verified fact, Primary-source or vendor statement, User testimonial, Hypothesis, Independent test result, Editorial inference.
The rapid path
These rules qualify a trend without history, velocity, persistence or broad media coverage (first match wins). Minimum observation periods, week-over-week comparisons, signal counts and multiple days of data never block them:
rapid.kev_listed: A vulnerability enters the CISA Known Exploited Vulnerabilities catalogue (retrieved from an authoritative source).rapid.exploitation_confirmed: An authority competent for exploitation (CERT, government agency, the vendor's advisory) confirms active exploitation.rapid.binding_regulation: A competent authority issues a binding decision, enforcement action or rule entering into force.rapid.severe_actionable_vulnerability: A severe vulnerability in widely deployed software with a fix or mitigation available.rapid.exploitation_reported: Active exploitation is reported by a non-authoritative source: escalated to an editor, unless the security alert rule (a trusted, current, specific report) publishes it.rapid.confirmed_outage: A major service outage confirmed by the operator's own status channel or an authority.rapid.major_vendor_change: The vendor or project itself announces a major shutdown, breaking change, release, incident or safety advisory affecting many users.rapid.independent_authorities: Several genuinely independent authoritative sources report a consequential event.rapid.broad_deadline: A dated deadline within the configured window affects a broad population.
A rapid match also sets a floor on importance (important, or critical for a critical rule), and says so in the reasons. It qualifies the trend, not every claim about it. A rapidly qualified trend is published by an editor, unless automatic publication is switched on; then only trends supported by authoritative evidence are published without review. Potentially critical classifications without an authoritative rule, conflicting reports, high importance with unverified evidence and single-source items go to an editor, except under the security rule below. The first published text is written by rules from retrieved evidence only, and a full analysis replaces it later.
Security alerts from a single trusted source
One exception lets a single, not yet corroborated signal publish a trend at once. It applies to security issues only (vulnerabilities, incidents, safety advisories), and every condition must hold:
- a trusted source: an authority for security advisories or exploitation (a CERT, CISA, a national cybersecurity agency, a CVE record), the affected vendor's own official channel, or a publication Trendoviz rates tier 1;
- a concrete issue, not commentary, rumour, speculation, general advice or marketing;
- a stated publication time within the last 72 hours (old incidents being recirculated do not qualify);
- specific facts: an identifier, an affected product, a version, a severity or exploitation;
- material impact: severe consequences, exploitation, a breach or broad exposure; routine and low-severity fixes do not qualify;
- no open contradiction between sources.
Such a trend keeps its labels: primary-source claim or unverified, provisional, not independently corroborated, developing. The labels never block publication under this rule; corroboration, when it arrives, upgrades the same trend. Automatic publication must be switched on. Every other kind of trend needs the general importance and evidence rules above.
Cybersecurity evaluator
- Reads CVE and GitHub advisory identifiers, CVSS scores and vectors, EPSS (only from structured data), exploitation statements, KEV listings, public exploits, attack vector, authentication and user interaction, affected and fixed versions, patches and mitigations, supply-chain reach and ransomware use.
- CVSS rates technical severity (consequence: high from 9, medium from 7); it never decides real-world importance on its own.
- EPSS is a predicted probability of exploitation (≥ 0.5 or percentile ≥ 0.95): it raises immediacy to medium at most. Confirmed exploitation always takes precedence.
- Exploitation confirmed by an authority, or a KEV listing, escalates to critical at once.
- Versions, exploitation, patches, mitigations and actions are shown only when a retrieved item states them. Missing data is listed as an open question.
Status and priority of an early trend
Critical · Important · Developing · Resolved · Established. Status is separate from confidence, importance, the trend lifecycle, attention and substance. Every change records the previous and new status, the time, the evidence, the rule or editor, the rule version and an explanation. Invalid changes are refused; lowering or resolving a critical trend is reserved to editors. Early trends stay provisional for at least 48 hours and until their central claims are corroborated. A non-critical one with no new signal for 14 days is resolved by rule.
Important and critical trends are shown first in the Trends feed for 48 hours (96 when critical) after their last signal, and no longer once resolved or established.
From early to established
Later signals about the same issue join the same trend, and a cluster detected from history later is kept on it rather than published again. When the trend reaches a measured stage (emerging, rising, breakout, established) with at least 3 independent publishers, 3 days or more after detection, it is shown as established, and its score and lifecycle take over. Many early trends are resolved before that, and that is normal.
Example assessments
- KEV listing, one source, no history. CISA adds a PHP flaw to its catalogue. Authority high, immediacy high (exploitation confirmed), escalation
cyber.kev_listed→ critical, confidence confirmed, rapid rulerapid.kev_listed. Published as a trend at once when automatic publication is on. - A WhatsApp flaw reported by a trusted security publication. One source, exploitation reported and a fixed version named: the security rule publishes it, labelled provisional and not independently corroborated. A rumour from a forum, or a routine low-severity patch, is not published.
- CVSS 9.8 in a little-known library. Consequence high, scope unknown, no exploitation data → notable; no rapid rule. Severity alone does not make it important.
- NGINX advisory, then a fix. After “fixed in 1.27.4”: actionability high, rapid rule severe-actionable → important, and followers get a “fix available” alert.
- A regulator's binding fine. Rapid rule binding regulation → important, confidence confirmed (a competent authority stating its own decision).
- Twelve outlets citing one post. One piece of underlying evidence, no authoritative source: unverified, at most notable, however loud.
These examples are executable tests in the code base. Back to the methodology.