Updates
Completed engine improvements with verified evidence, from R50 onward. The date on each entry is the day the change was verified by local tests.
What this page counts
These entries are the engine’s own version history: detection rules, data matching, report language and tests. The entries come from a single evidence-backed history file. R50 is where this history opens; its date records the day those sources were taken into the engine.
Every entry carries the date, the version and title, the effect on you, and the coverage limit of that change. The date is the day the improvement was verified by local tests and recorded; live publication has to be verified separately.
For coverage figures see Capabilities and Score Model.
Boundary
An entry here does not show that the feature runs on your site, or that your site has been audited: what you read here is the engine’s own history.
LAST RECORDED RUN
Last verified research and test status
We took the values below from the engine’s own status output exactly as recorded in the run of 4 November 2026. Research time and test time are separate fields.
Research record
Last record: . We checked 20 of 24 recorded sources and 17 of 21 required sources. Four required sources stayed out of that run: we could not reach the Wordfence authorized feed, the canonical CISA KEV page or Exploit-DB, and we did not check HackerOne public reports. We recorded that, and we claim no coverage for those sources.
This record comes from a manually started run (trigger: manual request) and covers the 28 October – 4 November 2026 review window; it is not evidence of output from the scheduled 12:00 run.
Maintenance tests
Finished: . That run covered 326 daily maintenance tests; 76 related regressions also passed.
Execution mode
We run the engine on a local computer; there is no server set up.
This page is not a live status indicator: the values belong only to the moment of that run. The engine marks research or test results older than 48 hours “stale”; if the source files have changed, the test result becomes “source changed”. You read current status by running the command on a local installation.
Verified changes
Newest first. We use three status labels: BASELINE, the entry where the history opens; HISTORICAL RECORD, entries the source data classifies as historical, with archived evidence; VERIFIED, improvements verified by the current engine tests. We keep the evidence for every entry locally.
Security update for the PDF processing dependency
WTE’s local PDF processing dependency was updated to the vendor’s pypdf 6.21.4 security release, with matching installation minimum and lockfile. Maintenance and report regressions passed, along with Turkish/English Word/PDF generation, manifest, quality and read-only verification gates. This update does not modify customer servers or the website and does not guarantee the absence of all vulnerabilities. The date records local verification, not live deployment.
Priority context for conditional notices and a DNS security update
Configuration-dependent vendor notices now include optional CISA KEV records and dated FIRST EPSS estimates. Unknown applicability is not promoted to a version match or a confirmed target vulnerability. Turkish/English Word and PDF output for notice states passed verification. The local DNS dependency was updated to the vendor security fix. The core score is unchanged; this date records local verification, not deployment.
Advisory conditions and bilingual report accuracy
Permission and configuration requirements for supported vendor matches are now visible in reports and retained when provider aliases refer to the same advisory. Fixed source-link verification, file-path preservation, and missing identity-table rows in English PDFs. A sample Turkish/English Word and PDF delivery passed verification. This does not establish applicability on a target or coverage of every platform. The date records local verification, not live deployment.
Bilingual report generation and maintenance accuracy
Completed missing Turkish/English content that blocked report generation for partial measurements, alternative layouts, and the security advisory appendix. Shared content parity and a synthetic Word/PDF delivery passed verification. Changes to language resources now invalidate previous maintenance success. This is the local verification date, not live deployment or approval of every possible report input.
Report accuracy and PDF dependency security update
Fixed a language conversion that blocked native report generation. Reduced false findings in directory-listing and canonical checks while preserving missing evidence as unknown. Updated the PDF dependency to the vendor fix for three advisories. Regression tests and native Word/PDF verification passed. This is the local verification date, not live deployment or site approval.
Current advisory context and readable score guidance
Added optional KEV/EPSS context for CVE candidates, five Python dependency patches and native report guidance on score bands, meaning and next actions. Comparability, environment context and missing-measurement explanations passed integrated tests. No live website deployment.
Local preference components and report regression verification
Added consent and reading preference components, offline heatmap and portfolio views, Jira CSV and brand/citation analysis for supplied model answers. Fixed additional report regressions and expanded daily tests, with native Word/PDF verification. No live deployment.
The components are local: Jira CSV is an export, and brand/citation analysis examines only the model answers handed to it.
Content policies and evidence-based retesting
Added static content policies, team ownership, fix and regression tracking, CI budgets, a local site portfolio and aggregate customer analytics/heatmap-cell imports. This does not provide live visitor collection or hosted services.
Native maintenance tracking and research corroboration
Added evidence-bound local maintenance commands and capability records. Public security research joined the daily source registry, with guards against treating unverified or retracted signals as completed work. No server or dark-web connector was deployed.
Current security advisories and continuous maintenance
Added NVD/OSV version matching, source age and coverage reporting. Critical PHP advisories and conditional vendor notices are shown separately. Daily research records feed a deduplicated roadmap; only evidence-backed changes enter the public history.
Task-based usability assurance
Added local synthetic task evidence, regression comparisons, optional SafeFix task checks and expiring risk acceptance. Synthetic tasks do not replace real user research.
A synthetic task result does not replace real user research. Automated checks do not satisfy a human accessibility evaluation or any certification.
Browser evidence and product assurance
Expanded browser observations, offline field-data imports, signed decision records and controlled report publication.
We import field data from outside; live field measurement is outside the scope of this entry. A signed decision record is an internal record discipline, not an independent certification.
Evidence-linked catalogue and decision reporting
Catalogue records were classified by execution and evidence type, preserving unknown results and traceable report figures.
A catalogue row is neither a feature nor a pass mark. Not having been measured does not mean a score of zero.
R50 baseline recorded
Historical R50 sources were archived separately from the active engine. This is the source import date, not the original product release date.
How daily research and maintenance run
We schedule research and maintenance work daily at 12:00 Türkiye time. The arrangement is local: the computer and Codex have to be running. If the computer is off, that day’s work does not happen, and we do not count a missed run as a successful one. If an urgent advisory appears after 12:00, it is handled in the next local run — this is not a real-time alerting service.
The subject of this maintenance is the engine itself: detection rules, product and version matching, data adapters, report wording and tests. WebTrustEngine runs it in its own local environment; we install nothing in a customer environment.
Which source layers are consulted
- Vendor and advisory data — PHP, WordPress and other vendors, NVD, OSV, GitHub Advisories: product identity, affected and patched versions, fix conditions.
- Coordination and exploitation data — CERT/CC and CISA KEV: independent corroboration and priority based on observed active exploitation.
- First-hand research — research teams’ own technical publications: new bug classes and examples of incomplete patches.
- Publicly disclosed findings — publicly disclosed reports only.
- Public exploit references — the existence of a PoC and its product/version relationship; we never execute a PoC.
- Probability-based priority — EPSS: a dated estimate signal.
In the last recorded run we could not check all of these layers. We could not reach NVD or CISA KEV; we could read only KEV’s non-required official mirror (the catalogue publication it returned is dated 21 October 2026). In the first-hand research layer we could not read PortSwigger Research. We saw the Cisco Talos index only as a cached crawl five days old, and did not verify freshness through 23 October.
The publicly disclosed findings and public exploit reference layers had no successful check at all. We put on record which source we read in which run.
Dark-web and closed threat feeds are out of scope: there is no provider, membership or adapter in place. A source being on the list does not mean all of its content is scanned every day; we record the actual access result in every run.
What a match says
A version match is a candidate finding: it is not evidence that the vulnerability is exploitable on the target. We verify the running version, the distributor’s backported patch and the configuration conditions separately. KEV is a record that a vulnerability has been exploited in the wild; EPSS is a 30-day probability estimate. Both are optional context, and both stay outside the report’s core score.
We read the layers separately: the PHP runtime, the WordPress core and plugins/themes are different scopes. A PHP advisory may not be closed by a WordPress update; the right layer has to be patched. We produce no definite match for a component whose identity or version cannot be verified, and we record the missing input as open work.
Where these queries go
An offline scan uses no network: refreshing the version query is off by default. The KEV/EPSS priority context is a separate option, and it is off by default as well. When the operator explicitly enables one of them, the request goes to the data source rather than to the target site: the component name and version to OSV, the full CPE identifier to NVD, and only public CVE identifiers to FIRST EPSS. We send nothing to CISA KEV: we download the public catalogue and match it locally.
We send no customer source file, target address or password. The requests come out of a hard budget: 20 requests and 120 seconds by default, at most 200 CVEs, and groups of at most 50 identifiers for EPSS. Identifiers beyond those limits stay recorded as not queried; we record the query, the response, the source address and the retrieval time.
What is not promised
Finding every vulnerability, covering unknown infrastructure, a zero-day guarantee, covering all WordPress plugins and themes, uninterrupted monitoring and automatic patching: none of these exists in this product.
For the whole security boundary see Security Boundary, and for what the score bands mean see Score Model.
TEN QUESTIONS
About updates and daily maintenance
Q: Does a WordPress update close a new PHP vulnerability?
A: Not always. The layer the advisory affects — the PHP runtime, the WordPress core or a plugin — is assessed separately, and the right layer has to be patched.
Q: Are all vulnerabilities in every stack found?
A: No. Coverage is limited to components whose identity and version can be verified, and to the data provider’s own scope. For concealed versions, bespoke software and vulnerabilities not yet known, there is no completeness guarantee.
Q: Are WordPress plugins and themes covered too?
A: We can take an inventory, but any scope whose current provider connection is unfinished stays unverified. Core version matching runs separately.
Q: Is a match a definitely exploitable vulnerability?
A: No. We verify the running version, the patch and the configuration conditions separately.
Q: What happens if a source service cannot be reached?
A: We record the outage or the stale data explicitly. Failing to obtain information never produces a “no vulnerabilities” result.
Q: Is a new feature added every day?
A: No. Daily maintenance may produce no new product feature at all. This page holds only completed, verified changes.
Q: Does it run while the computer is off?
A: The current local arrangement does not. Server and cloud execution are out of scope for now.
Q: Is a patch installed automatically on a customer site?
A: No. This maintenance task improves WebTrustEngine’s own detection rules and engine; it installs no patch in a customer environment.
Q: Is the date in the history the live publication date?
A: Unless stated otherwise it is the day the product change was verified. Live publication has to be verified separately.
Q: How is an incorrect or retracted advisory handled?
A: We re-examine the source and correct the affected rule and wording. We keep the earlier evidence and put the correction on record.
For the wider question set see the FAQ.
Seeing what a change means for your own site takes an evidenced snapshot of it.
Prepare an audit request