Summary

  • The duty to describe a certificate failure in public is a Common CA Database guideline made binding by private root-programme policy: 72 hours from awareness, a Full Incident Report within 14 days, a prescribed template that requires a named root cause, public unless the filer ticks a security box.
  • Every clock in that instrument starts with an act of the certificate authority, which is why the 109 days between the close of the access window in bug 2084460 and the determination that made it reportable sit outside all of them — and why the published record ends at the filing, with no programme's binding text obliging anyone to publish the basis on which a report was judged.

At 17:31 GMT on 18 July 2025, DigiCert finished revoking 1,834 certificates it had issued for alias domains after the WHOIS deprecation date of 15 July. Three hours and thirty minutes later, at 21:01 GMT, it opened the public file on the failure. Hold that order: the revocation ended before the paperwork began, and the paperwork names the cause — validation data for an alias domain is stored separately from the Authorized Domain Name's data, so the script that invalidated WHOIS-based validations never cleared the alias. The report's own words: "an oversight in writing the script" (bug 1978163).

Fifteen months later the same company filed again. Bug 2084460, "DigiCert: Unauthorized access to business system", carries two adjacent lines in its own timeline: "2026-10-19 | Determination of reportable event under Browser root policy program requirements" and "2026-10-20 | Preliminary Incident Report filed for this event" (bug 2084460). The data-access window recorded in the first of those lines ends on 2 July 2026. Between the window closing and the determination: 109 days.

That interval falls outside every deadline the instrument sets, and the reason is structural rather than evasive. The 72 hours run from the moment a CA Owner becomes aware. The record prints a determination date, not an awareness date, and says nothing about when awareness arrived. A regime that anchors its clocks to the filer's own act does not measure the time before the filer acts; it measures only what happens afterwards.

The duty is not where the forum put it

The operative text is the Common CA Database's Incident Reporting Guidelines, whose version 3.2 is presented as current and effective 20 March 2026, following version 3.1 of 15 July 2025 and version 3.0 of 1 March 2025 (CCADB Incident Reporting Guidelines). The publisher describes itself as a Linux Foundation project.

Within 72 hours of a CA Owner becoming aware of an incident or an undisclosed audit finding, the CA Owner must disclose a Preliminary or Full Incident Report, or respond to a Preliminary Incident Report filed by a third party. A Full Incident Report must be posted within 14 days of that initial disclosure. Filing happens in Bugzilla on prescribed Markdown templates, with every applicable field completed and inapplicable ones marked "N/A"; the Summary line must open with the CA Owner's name, a colon and a brief title.

The Full template asks for more than a narrative. Its sections are Summary, Impact, Timeline, Related Incidents, Root Cause Analysis, Lessons Learned, Action Items and Appendix, and the guidelines expect a structured method, naming Five Whys and fishbone. Action Items are tabulated. A report built to this template is not a press release with a date on it; it is an engineering document that has been made compulsory.

Publication is the default rather than the exception. Only ticking the "CA Program Security" checkbox makes a bug security-sensitive, and the guidelines spell out the consequence: "leaving all security boxes unchecked makes it a public bug". Security-sensitive incidents must later be converted into, or re-filed as, a public report once the risk is mitigated. Anyone may file on credible and well-substantiated evidence, and everyone is encouraged to comment on existing reports.

Open reports carry a "Next update" date in the Bugzilla Whiteboard field — a date CA Owners are told to ask a Root Store Operator to set — with a 7-day fallback where none is recorded, and questions must be answered within 7 days.

Closure is the CA Owner's move. It requests closure in a comment using the prescribed template, ticks "Request information from" and adds the CCADB incident-reporting address; "a final call for comments will be made by a Bugzilla moderator, and the report will be closed accordingly". The accompanying Incident Closure Summary carries fields for incident description, root cause or causes, remediation description and commitment summary.

Auditors are encouraged to opine on incidents open during an audit period, including confirming that scope, impact and root cause are "accurately and fairly stated in the publicly-disclosed incident reports". The guidelines also offer their own reassurance: a report's existence "does not generally indicate serious problems with a CA".

What turns a database convention into an obligation is private policy. Mozilla Root Store Policy v3.1, effective 1 July 2026, treats any failure to comply — "whether it be a misissuance, a procedural or operational issue, or any other variety of non-compliance" — as an incident, and requires CA operators to comply with the CCADB Policy and these guidelines for disclosure and handling; Mozilla may additionally require a plan of action with milestones, or further audits (Mozilla Root Store Policy). Chrome Root Program Policy v1.8, last updated 5 February 2026, requires public disclosure and response in Bugzilla on the then-current CCADB format and timelines, and states that "Chrome uses the information in the public disclosure as the basis for evaluating incidents" (Chrome Root Program Policy).

The CA/Browser Forum document my reservation expected to carry the duty does not appear to, on the evidence available. Its TLS Baseline Requirements v2.3.1, dated 4 October 2026, list §4.9 "Certificate revocation and suspension" and §5.7.1 "Incident and compromise handling procedures" — with subsections for incident response and disaster recovery plans, and for mass revocation plans — in the tables of contents (requirements page, v2.3.1 PDF). Those contents label no section a public incident-reporting duty and carry no template. The bodies of those sections were not read, and nothing here says what they require.

Four clocks, one anchor

Count the intervals the instrument actually creates. Seventy-two hours, from the CA Owner becoming aware. Fourteen days, from the initial disclosure the CA Owner makes. A "Next update" date, set in a field the CA Owner asks a Root Store Operator to fill, with 7 days as the fallback. Seven days to answer a question. Then closure, when the CA Owner decides its Action Items are done and asks for it.

Every one of those runs from something the certificate authority does or asks for. None runs from the start of the incident or its end. None runs from filing to disposition. That is why the two DigiCert filings point in opposite directions while neither is anomalous. In July 2025 the CA moved first and reported after: revocations complete at 17:31, bug filed at 21:01. In the 2026 filing the determination arrived 109 days after the access window closed, and the preliminary report followed the next day. One instrument, one company, both sequences — and no deadline missed in either.

The identifier does not help. Both filings cite CCADB unique ID A000021, the CA Owner's number rather than an incident number. It does not distinguish the 2025 misissuance from the 2026 data-access event, and neither filing carries any number but its Bugzilla issue.

Where the record stops

The confession is public in detail. The judgement is not public at all.

Chrome's binding policy lists what it weighs: demonstrated understanding of the root cause, a substantive commitment and timeline, past incident-handling history, severity. It commits to evaluating "the broader context of an individual incident" case by case, and it requires a secondary incident report when prescribed follow-up actions such as revocation deadlines are missed. Then §1.5.2 draws the curtain in plain language — as standard practice Chrome does not discuss ongoing public reports privately, does not advise on or approve a CA Owner's planned response, and offers no guarantees of outcome.

What the same policy holds in reserve is removal: a participant's failure to follow the minimum requirements "may result in the corresponding certificate's removal from the Chrome Root Store, limitations on Chrome's acceptance of the certificates they issue, or other technical or policy restrictions".

Mozilla's binding text is thinner still: it may require a plan of action with milestones, or one or more additional audits. The harshest words in the material I read are not in a binding policy at all. They sit on MozillaWiki, in guidance whose own page characterises the response list as non-binding — active distrust of a small, enumerable misissued certificate set in Thunderbird, with intermediate or root distrust as a further consideration for larger, unenumerated cases (Maintenance and Enforcement) — and, for repeated delayed-revocation incidents without sufficient justification, "heightened scrutiny and sanctions, including removal of the CA from the Mozilla Root Store" (Responding To An Incident). Mozilla's Incident Dashboard is where such filings are tracked, and it points back to Bugzilla for the report itself.

Across the eleven documents whose text I read on 19 November 2026 — the CCADB guidelines, two root-programme policies, two DigiCert filings, two Baseline Requirements tables of contents, three MozillaWiki pages and one mailing-list record — no published evaluation criteria and no written procedure governing a root-store operator's decision on an individual report were located. In the text of bug 1978163 that was readable, no Incident Closure Summary and no closing moderator comment appeared; the closure evidence was the bug's own RESOLVED FIXED status.

No CA/Browser Forum-hosted index of filed reports was read; the filings discussed here are Bugzilla issues.

The one document that says the quiet part out loud is a discussion record, not a procedure. A Mozilla dev-security-policy roundtable reports that the CCADB Steering Committee takes turns reviewing incident closure summaries during two-week on-duty assignments, that participants asked for the criteria and root-programme commitments used to evaluate reports to be published, and that they described written procedures for handling a report after its closure summary as non-existent (roundtable record). That is a room full of people asking for the missing piece.

What this does and does not say

Nothing in the material read states that any certificate authority missed a reporting deadline, misstated a root cause, or concealed an incident. The 109 days are not a breach; they are an interval the instrument does not measure. The claim here is narrower and harder to answer: this regime publishes an unusually complete account of what went wrong, in a structured form that forces the author to say why, and publishes nothing that tells a relying party what the trust-store operator concluded about it.

The control that would close the loop fits in a comment. A dated disposition note attached to each closed report, naming the basis on which the programme accepted, conditioned or declined the CA's account, and the next milestone. That is a design recommendation, not an allegation that any programme has acted arbitrarily. It would give the checking end of a published self-report a published counterpart.

What to hold loosely

The bodies of the CA/Browser Forum TLS Baseline Requirements sections were not read — only front matter and tables of contents, truncated before §4.9 and §5.7.1 — so this article takes no position on what they require. Chrome Root Program Policy v1.8 post-dates both filings, so the enforcement wording quoted is current policy rather than the text in force in July 2025. The guidelines in force at that filing were v3.1; the text read is v3.2, and the closure language was not compared line by line. The exact closure dates of both bugs were unreadable in the retrieved text and are not printed here.

The roundtable record is an informal discussion with anonymised speakers, describing practice rather than setting it.

For the directory context on the filer, see DigiCert, Inc. in the BTW directory. Prior BTW coverage of this company read a revocation clock, a certificate-authority migration and the 2024 domain-validation underscore error; none of the three read the reporting obligation, its prescribed fields or what follows a filing, which is the delta this article claims.