GCVE Best Current Practice

GCVE BCP-05-X-04 - Security Vulnerability and Advisory Timeline

GCVE BCP-05-X-04: Security Vulnerability and Advisory Timeline

  • Version: 1.0
  • Status: Draft
  • Date: 2026-11-17
  • Authors: GCVE Working Group
  • BCP Extended ID: BCP-05
  • BCP ID: BCP-05-X-04

This guide is distributed and available under CC-BY-4.0.

Copyright (C) 2026 GCVE Initiative.

Abstract

This document defines an extension to GCVE BCP-05 for exchanging a chronological view of security vulnerability and advisory information collected from multiple sources. It preserves dated fields from vulnerability records and linked advisories, sightings, and known exploited vulnerability (KEV) assertions, together with their provenance and original date values.

The extension also defines optional duration metrics whose selected evidence, calculation time, precision, and limitations remain available to consumers. The payload follows the metadata, data, and metrics structure exported by the Vulnerability Lookup implementation.

Motivation and Scope

A vulnerability’s history can be spread across a CVE record, vendor advisories, CSAF documents, other vulnerability databases, sightings, and KEV catalogs. A combined timeline helps consumers inspect publication history, compare sources, and assess the intervals between discovery, reporting, disclosure, exploitation evidence, and patch evidence.

A date found in a linked record is not necessarily a fact about the requested vulnerability. An advisory can cover multiple vulnerabilities, acquire a new vulnerability in a later revision, or describe another issue altogether. A catalog addition date can reflect a backlog import rather than the beginning of exploitation. This extension therefore separates event collection from eligibility for metric calculations.

BCP-05-X-03 represents typed handling, coordination, disclosure, and regulatory events supplied by a publisher. X-04 represents an aggregated view of dated source information and optional derived measurements. X-03 events can be inputs to X-04. The extensions MAY coexist and MUST NOT be treated as interchangeable payloads. X-04 does not define current handling workflow state or establish regulatory compliance.

The key words MUST, MUST NOT, REQUIRED, SHOULD, SHOULD NOT, RECOMMENDED, MAY, and OPTIONAL in this document are to be interpreted as described in RFC 2119 and RFC 8174 when they appear in capitals.

Placement

The extension MUST be placed in the extensions object of a GCVE BCP-05 record using the key bcp-05-x-04. Its value MUST contain one x_timeline object:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
{
  "x_gcve": [
    {
      "vulnId": "GCVE-1-2026-12345",
      "recordType": "advisory",
      "extensions": {
        "bcp-05-x-04": {
          "x_timeline": {
            "metadata": {
              "vulnerability": "GCVE-1-2026-12345",
              "count": 0,
              "kinds": {"milestone": 0, "sighting": 0, "kev": 0, "other": 0},
              "linked_omitted": 0
            },
            "data": []
          }
        }
      }
    }
  ]
}

x_timeline is scoped by its extension key. In X-04 it contains metadata, data, and optionally metrics; in X-03 it contains handling events and optionally entities and regulatory information.

A standalone timeline export MAY contain the x_timeline value directly, without the GCVE wrapper. To embed a Vulnerability Lookup timeline export, a producer places that export under extensions.bcp-05-x-04.x_timeline. This does not replace the surrounding BCP-05 record or its required members.

Data Model

Member Type Required Description
metadata object yes Timeline subject, event counts, and collection limitations.
data array of objects yes Collected dated events, ordered oldest first. An empty array is allowed.
metrics object no Selected evidence anchors and calculated durations.

Metadata

Member Type Required Description
vulnerability string yes Identifier of the vulnerability or advisory whose timeline was requested.
count integer yes Number of events in data; MUST be non-negative.
kinds object yes Counts by event kind. Values MUST be non-negative integers.
linked_omitted integer no Number of linked records excluded from collection, for example because of a resource budget.

metadata.vulnerability MUST identify the containing record’s subject or a known identifier for the same subject. It MAY be a CVE identifier when the containing GCVE record describes that CVE. An arbitrary relationship to another record does not establish identifier equivalence.

count MUST equal the length of data. Producers MUST include the four defined kind counts, including zero values; any additional kinds present in data MUST also be counted. The sum of all kind counts MUST equal count.

linked_omitted, when present, MUST be non-negative. A positive value indicates that the collection excludes linked records and that metrics may change when those records are included. Zero indicates no known omission from that collection step; it does not establish complete knowledge of the vulnerability. Absence means the omission count was not supplied. Other collection limits SHOULD be documented by the producer.

Event Fields

Each member of data describes a date obtained from a source. Its presence states that the source contains that date or assertion; it does not independently verify the underlying claim.

Member Type Required Description
timestamp string yes Normalized RFC 3339 UTC date-time, using Z.
date_only boolean yes Whether the original value supplied a calendar date without a time.
value string yes Original source date value, before normalization.
kind string yes Broad event grouping: milestone, sighting, kev, or other.
label string yes Human-readable event description, such as Published or Vendor fix.
source string yes Source feed, publisher, catalog, or sighting origin. It need not be a URL.
record string yes Identifier of the source record, which can differ from the timeline subject.
field string yes Path to the date field within the source record.
linked_record boolean no true when the date was collected from a linked record.
uuid string no Identifier of the underlying sighting or KEV entry.
content string no Source text providing context.
text string no Source timeline entry text, for example a CVE free-text timeline description.
url string no HTTP or HTTPS reference to the source or supporting information.
status object no KEV assertion status, including exploited and status_reason when available.
metric_role string no Structured role for which this event is eligible as calculation evidence.
metric_basis string conditional Evidence basis; REQUIRED when metric_role is present.
batch_listing boolean no Whether the date appears to represent a catalog batch listing or backlog import.
publication_copy boolean no Whether an apparent KEV date appears to copy a vulnerability publication timestamp.
added_in_revision boolean no Whether the date identifies a revision that added the subject vulnerability to an advisory.

field uses dot-separated object member names and zero-based array indices, for example vulnerabilities.0.remediations.0.date. It is a descriptive source path, not a JSON Pointer or executable expression. Member names containing dots can be ambiguous; consumers MUST NOT execute or blindly evaluate this string.

source, record, field, and, where supplied, uuid SHOULD allow an analyst to trace an event to its input. source identifies provenance, not a trust level. linked_record: true alone does not make an event eligible for metrics.

Event Kinds

Kind Meaning
milestone Recognized record milestones, such as identifier reservation or publication.
sighting A dated sighting associated with the subject or an equivalent identifier.
kev A dated field from a KEV assertion or its evidence.
other Another collected date, including discovery, remediation, revision, or document generation dates.

Kinds support grouping and display. A milestone is not automatically eligible for calculation, and an other event can be eligible. Consumers MUST NOT infer a metric role from kind, label, content, or text alone. Unknown kinds MUST NOT cause rejection of the complete extension.

Dates, Ordering, and Duplicates

Producers MUST preserve the source value in value and normalize timestamp to UTC. A full date such as 2026-10-16 is represented as 2026-10-16T00:00:00Z with date_only: true. This midnight is a sorting convention, not evidence that the event occurred at midnight. A source date-time at midnight retains date_only: false even when the calculation treats it as day precision.

The reference implementation truncates sub-millisecond precision in timestamp while retaining the original value. It also accepts unambiguous year-first slash dates and treats offset-free source date-times as UTC. Producers using these compatibility rules SHOULD document them; consumers MUST NOT reinterpret normalized timestamps in their own local timezone. Invalid or ambiguous source dates MUST NOT be fabricated into events or anchors.

Events MUST be sorted by normalized timestamp. Producers SHOULD resolve ties deterministically, for example by label and source. Array order MUST NOT imply causality or an order within a date-only day.

Producers MAY collapse duplicate extractions of the same source assertion, but SHOULD retain distinct provenance and MUST NOT merge different metric roles merely because timestamps coincide. Coincident dates from independent sources are not duplicate evidence by themselves.

Metric Evidence

Metrics are OPTIONAL. An event without a metric role remains useful timeline information. Producers assigning roles MUST establish both the event’s meaning and its applicability to the requested subject.

Roles and Evidence Bases

metric_role Meaning
discovery Recorded discovery of the vulnerability.
report Report to, receipt by, or notification of a vendor or coordinator.
acknowledgement Acknowledgement of the report.
disclosure Public disclosure or publication of information covering the vulnerability.
exploit_publication Report of a published proof of concept or exploit.
exploitation Recorded exploitation observation or qualifying exploitation evidence.
exploitation_report A report or catalog listing asserting exploitation, without establishing its actual first occurrence.
patch Recorded evidence that a corrective measure is available.
patch_observed A sighting reporting a patched system, providing an upper bound on availability.
metric_basis Meaning
recorded_event Structured event type or recognized source field.
sighting_report A typed sighting; its report date may lag the underlying activity.
kev_first_seen KEV evidence treated as a first-seen observation.
kev_report KEV assertion, catalog listing, import, or reported patch date.

Neither roles nor bases guarantee source accuracy. A proof of concept does not establish observed exploitation. A patched sighting does not establish deployment across all affected systems. A vendor fix date does not establish that every affected product or version has a fix.

Eligibility and Source Mapping

The reference implementation uses the following mappings where subject scope is established:

Input Role Basis
X-03 discovered discovery recorded_event
X-03 reported, received, affected-party-notified report recorded_event
X-03 acknowledged acknowledgement recorded_event
X-03 public-disclosure, advisory-published disclosure recorded_event
X-03 exploit-observed exploitation recorded_event
X-03 fix-available patch recorded_event
Recognized discovery, report, acknowledgement, publication, exploitation, or fix date fields Corresponding role recorded_event
CSAF vulnerability release_date disclosure recorded_event
CSAF remediation date with category: vendor_fix patch recorded_event
Sighting exploited exploitation sighting_report
Sighting patched patch_observed sighting_report
Sighting published-proof-of-concept exploit_publication sighting_report
Eligible KEV first-seen evidence exploitation kev_first_seen
Eligible KEV assertion, listing, or import date exploitation_report kev_report
KEV evidence details.patchedSince patch kev_report

Known field names qualify only in documented semantic locations, such as the record root, cveMetadata, containers.cna, document.tracking, or the matching CSAF vulnerability entry. A publication date nested in a reference, tool, generator, provider, or regulatory notification MUST NOT automatically become the subject’s disclosure date. A CVE free-text timeline entry has no structured event type and MUST NOT acquire a role merely from its wording.

A linked record MUST explicitly describe the requested vulnerability or a known equivalent identifier before supplying its vulnerability-specific anchors. References, correlations, and generic GCVE relationships alone are insufficient. For a multi-vulnerability CSAF advisory, per-vulnerability dates MUST be taken from the matching vulnerability entry. When the requested subject is itself an advisory covering several vulnerabilities, producers MUST NOT combine different vulnerabilities’ dates into a single vulnerability lifecycle.

Advisory publication can provide disclosure evidence for a vulnerability covered by that publication. If a later revision added the vulnerability, producers SHOULD use the date of the revision introducing it and set added_in_revision: true. The reference implementation recognizes selected revision descriptions for this purpose; incomplete revision histories and unrecognized wording remain limitations.

Only KEV entries currently asserting exploitation qualify as exploitation inputs. In the reference implementation, status.exploited must be true and status reasons suspected, disputed, and withdrawn are excluded. Missing status reasons or confirmed, historical, and unknown are accepted; the original status is retained so consumers can apply stricter criteria.

The reference implementation treats first-hand KEV evidence, or a first-seen date earlier than its assertion date, as kev_first_seen. Otherwise the listing, assertion, or import supplies kev_report. These are evidence classifications, not independent verification of the catalog’s dates.

Catalog backlog dates SHOULD carry batch_listing: true. The reference implementation flags days with at least 25 exploited entries from the same catalog, separately considering catalog listing days and instance import days. This threshold is an implementation heuristic, not proof of when exploitation began. Its publication-copy heuristic flags KEV exploitation dates less than one second from a sub-day disclosure timestamp as publication_copy: true; such dates remain visible but MUST NOT supply exploitation anchors under that profile.

Reservation, ordinary record updates, EPSS predictions, remediation deadlines, fix-developed, mitigation-available, and planned disclosure MUST NOT substitute for the defined anchors. In particular, neither mitigation nor fix development establishes fix availability.

Duration Metrics

When metrics is present, it MUST include as_of and anchors. Producers MAY omit individual metrics; an omitted metric means it was not supplied.

as_of MUST be an RFC 3339 UTC timestamp identifying the calculation time. Future events MAY remain in data but MUST NOT become observed anchors for that calculation. Metrics describe the collected evidence set and SHOULD NOT change merely because a user filters or sorts a displayed timeline.

Anchor Selection

anchors MUST contain the following members. Each value is a complete selected event object or null when no eligible event is available.

Symbol Anchor Selection
V discovery Earliest eligible discovery evidence.
R report Earliest eligible vendor or coordinator report evidence.
A acknowledgement Earliest eligible acknowledgement evidence.
D disclosure Earliest eligible disclosure evidence.
X exploit_publication Earliest eligible public exploit evidence.
E exploitation Earliest eligible exploitation or exploitation_report evidence.
P patch Earliest eligible patch or patch_observed evidence.
O origin Earliest of V, R, and D.
N Calculation clock metrics.as_of; not an anchors member.

Selection MUST compare normalized timestamps across eligible sources. At equal timestamps, recorded_event and kev_first_seen SHOULD take precedence over report bases; remaining ties SHOULD be resolved by source and field path. The selected event MUST retain its provenance, original date, precision, role, basis, and flags.

The earliest eligible evidence bounds what is recorded in the collection. It does not necessarily identify the actual first occurrence. For example, a distributor’s discovery date may record when that distributor learned of an already public issue, and the earliest recorded fix may be downstream of an earlier upstream fix.

Defined Metrics

For the following formulas, Q is P when known, otherwise N. Only the three formulas explicitly using Q permit this calculation-clock fallback.

Metric Formula Interpretation
vendor_response_time A − R Report to acknowledgement.
time_to_patch P − R Report to patch evidence.
total_time_to_patch P − V Discovery to patch evidence.
disclosure_to_patch P − D Public disclosure to patch evidence.
time_unpatched max(0, Q − O) Recorded waiting interval, open if no patch is recorded.
time_to_exploit E − O Origin to exploitation evidence.
disclosure_to_exploit E − D Disclosure to exploitation evidence.
exploitation_to_patch Q − E Exploitation evidence to patch evidence, or an open interval.
disclosure_to_exploit_publication X − D Disclosure to public exploit evidence.
exploit_publication_to_patch Q − X Public exploit evidence to patch evidence, or an open interval.

The start of time_to_patch MUST NOT silently fall back from report to disclosure or discovery. Public exploit publication MUST NOT substitute for exploitation. Negative signed intervals MUST be retained except for time_unpatched, whose nominal value and bounds are clamped to zero. A negative exploitation-to-patch interval can indicate that patch evidence predates exploitation evidence; its precision and evidence bases must still be considered.

Metric Fields

Member Type Required Description
available boolean yes Whether both endpoints required by the formula are available.
missing array of strings when unavailable Missing endpoint names: start, end, or both.
seconds number when available Nominal signed interval in seconds, subject to the clamping rule.
hours number when available seconds / 3600.
days number when available seconds / 86400.
min_seconds number when available Lower bound accounting for endpoint precision.
max_seconds number when available Upper bound accounting for endpoint precision.
approximate boolean when available Whether either evidence endpoint is treated as day precision.
ongoing boolean when available Whether the calculation clock substitutes for missing patch evidence.
order_known boolean when available Whether endpoint precision establishes their temporal order.
end_before_start boolean when available Whether the end certainly precedes the start, before clamping.
batch_listing boolean no Whether either selected endpoint carries the batch-listing flag.
discovery_after_disclosure boolean no On total_time_to_patch, whether discovery certainly follows the disclosure day or instant.
start object when available Complete selected start event.
end object when available Complete selected end event, or the calculation-clock object.

An unavailable metric MUST omit numeric durations and endpoint objects rather than supplying fabricated zeros. For example:

1
2
3
4
{
  "vendor_response_time": {"available": false, "missing": ["start", "end"]},
  "time_to_patch": {"available": false, "missing": ["start"]}
}

When P is absent, the three Q-based metrics use ongoing: true, provided their start exists. Their end MUST be a synthetic clock object containing timestamp equal to as_of, date_only: false, label: As of, and source: Calculation clock. This clock object is not a source event and is not included in data or its counts. Other patch metrics remain unavailable.

No recorded patch does not establish that no patch exists. No exploitation evidence does not establish that exploitation has never occurred. These public evidence intervals MUST NOT be presented as an organization’s actual exposure period without deployment evidence for that organization.

Precision and Bounds

For compatibility with the reference implementation, a calculation evidence endpoint has day precision if date_only is true or its normalized timestamp is exactly midnight UTC. This conservative midnight rule also applies when date_only is false. It MUST NOT be applied to the synthetic calculation clock.

Let s and e be the normalized endpoints. Let s_upper be s plus one day for a day-precision start, otherwise s; define e_upper similarly. Before clamping:

1
2
3
seconds     = e - s
min_seconds = e - s_upper
max_seconds = e_upper - s

All differences are in seconds. A day is 86400 seconds for these calculations. The bounds conservatively cover uncertainty within the UTC day, not uncertainty about the truth of the source assertion. Both exact endpoints produce identical nominal, minimum, and maximum values. Units SHOULD be exported without presentation rounding; displayed values MAY be rounded.

end_before_start is true when an exact end is earlier than s, or when a day-precision end’s e_upper is at or before s. order_known is true when that condition holds or e is at or after s_upper. Otherwise, consumers MUST NOT assert an order based solely on the nominal duration. Flags are calculated before time_unpatched clamping.

For example, exploitation at 2026-10-05T10:00:00Z and a patch dated 2026-10-05 yield a nominal exploitation-to-patch interval of −10 hours with bounds of −10 to +14 hours. The metric is approximate and its order is unknown.

If the selected discovery certainly follows public disclosure, total_time_to_patch SHOULD carry discovery_after_disclosure: true. This warns that the recorded discovery may describe a publisher learning of the issue. It does not change the selected origin or silently replace the discovery anchor.

Example Timeline

The following synthetic standalone payload illustrates publication, a date-only fix, a linked advisory, and a sighting. Identifiers and sources are illustrative; they do not assert a real vulnerability history.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
{
  "metadata": {
    "vulnerability": "GCVE-1-2026-12345",
    "count": 3,
    "kinds": {"milestone": 1, "sighting": 1, "kev": 0, "other": 1},
    "linked_omitted": 0
  },
  "data": [
    {
      "timestamp": "2026-10-03T09:00:00Z",
      "date_only": false,
      "value": "2026-10-03T11:00:00+02:00",
      "kind": "milestone",
      "label": "Published",
      "source": "Example GNA",
      "record": "GCVE-1-2026-12345",
      "field": "cveMetadata.datePublished",
      "metric_role": "disclosure",
      "metric_basis": "recorded_event"
    },
    {
      "timestamp": "2026-10-05T00:00:00Z",
      "date_only": true,
      "value": "2026-10-05",
      "kind": "other",
      "label": "Vendor fix",
      "source": "Example vendor CSAF",
      "record": "EXAMPLE-SA-2026-001",
      "field": "vulnerabilities.0.remediations.0.date",
      "linked_record": true,
      "metric_role": "patch",
      "metric_basis": "recorded_event"
    },
    {
      "timestamp": "2026-10-05T10:00:00Z",
      "date_only": false,
      "value": "2026-10-05T10:00:00Z",
      "kind": "sighting",
      "label": "Exploited",
      "source": "https://example.org/observations/1",
      "record": "GCVE-1-2026-12345",
      "field": "creation_timestamp",
      "uuid": "b49cb7c6-40a5-4e8f-847e-51e833c979d2",
      "metric_role": "exploitation",
      "metric_basis": "sighting_report"
    }
  ]
}

This payload omits optional metrics. At a calculation time of 2026-10-06T09:00:00Z, its anchors would have V, R, A, and X set to null, D and O equal to the publication event, P equal to the fix event, and E equal to the sighting. Selected results are:

Metric Nominal result Precision or availability
disclosure_to_exploit 49 hours Exact timestamp difference; the sighting remains report-based evidence.
disclosure_to_patch 39 hours Approximate; bounds 39 to 63 hours.
time_unpatched 39 hours Approximate; bounds 39 to 63 hours, ongoing: false.
exploitation_to_patch −10 hours Approximate; bounds −10 to +14 hours, order_known: false.
vendor_response_time Unavailable Both report and acknowledgement are missing.
time_to_patch Unavailable Report is missing.
total_time_to_patch Unavailable Discovery is missing.

An available metric retains the full selected events. For example, a calculation of exploitation_to_patch on this payload MUST copy data[2] into start and data[1] into end, and include the numeric values and flags described above.

Processing and Validation Requirements

  1. Producers MUST validate required members, types, date normalization, and count consistency before publication.
  2. Producers MUST keep event provenance and MUST distinguish displayed source dates from eligible metric inputs.
  3. Consumers MUST NOT assume that every event in a linked advisory concerns the requested vulnerability.
  4. Metric endpoints MUST follow the defined formulas and eligibility rules. Source endpoints MUST correspond to events in data; clock endpoints are the stated exception. Copies in anchors and metrics MUST remain consistent with those events.
  5. All numeric durations MUST be finite. min_seconds MUST be no greater than seconds, which MUST be no greater than max_seconds after applicable clamping.
  6. Unknown members, kinds, roles, or evidence bases MUST NOT cause rejection of the complete extension. Consumers MUST NOT use unknown roles or bases in calculations unless their semantics are understood.
  7. Consumers SHOULD preserve unknown members when forwarding the extension.
  8. Consumers MUST distinguish omission, null anchors, unavailable metrics, and measured zero durations.
  9. A producer recomputing metrics MUST update as_of, anchor selection, endpoint copies, and durations consistently. UI filtering MUST NOT silently change the collection over which exported metrics were calculated.

Updates and Interoperability

An aggregate timeline is a snapshot. New sources, corrected dates, withdrawn assertions, changed links, and collection limits can change both events and selected anchors. Producers SHOULD retain sufficient source history to explain material changes. An earlier snapshot’s metrics MUST NOT be combined with a later snapshot’s event set as if they were calculated together.

When collecting records that already contain X-04, producers SHOULD avoid recursively harvesting the embedded aggregate’s event copies and calculated endpoints as new source assertions. This prevents feedback and duplicate evidence.

This extension complements:

  • BCP-02, for vulnerability handling and disclosure practice.
  • BCP-03, for publication and synchronization. Historical event timestamps and metrics.as_of do not replace synchronization timestamps.
  • BCP-05, for the containing GCVE record.
  • BCP-05-X-03, for typed handling events used as inputs.
  • BCP-07, for the full KEV assertions underlying timeline dates.
  • BCP-12, for the full sighting records underlying timeline dates.

Metrics borrow lifecycle concepts used by the Vulnerability Timeline and Exposure Metrics (VTEM) framework, as described in the reference implementation’s documentation. This extension does not claim to implement every VTEM date or an organization-specific deployment or exposure model. The event projection does not replace the original vulnerability, advisory, KEV, or sighting records.

Privacy, Security, and Trust Considerations

Aggregated timelines can expose confidential coordination dates, personal data, internal identifiers, and source content. Publishers MUST review the payload for its intended distribution and MUST remove secrets and inappropriate private data. Optional content and text SHOULD be omitted when unnecessary.

Consumers MUST treat source strings and quoted content as untrusted data and MUST NOT execute instructions embedded in them. Renderers SHOULD escape source text and MUST validate URLs before making them actionable. A URL MUST NOT trigger an automatic fetch without appropriate controls against unsafe destinations.

An event count is not an independent-source count: many sightings can repeat one claim. Neither a large timeline nor precise arithmetic establishes source reliability. Consumers SHOULD display provenance, evidence bases, missing data, approximation, unknown ordering, backlog flags, and collection omissions alongside derived measurements. Vendor or product comparisons require comparable collection coverage and evidence scope.

Example JSON Paths

The aggregate timeline in a CVE record carrying a GCVE extension is located at:

1
containers.cna.x_gcve[0].extensions.bcp-05-x-04.x_timeline

Its event list and optional calculated durations are located at:

1
2
containers.cna.x_gcve[0].extensions.bcp-05-x-04.x_timeline.data
containers.cna.x_gcve[0].extensions.bcp-05-x-04.x_timeline.metrics

Reference Implementation

This draft is based on a sample timeline JSON export and the following Vulnerability Lookup sources, pinned to commit 038227307a41574d05ad9e48ea0a84f2110ff33c:

The implementation exposes standalone timelines at GET /api/vulnerability/<id>/timeline and /vuln/<id>/timeline.json. Its collection budgets, extraction rules, and heuristics can evolve; consumers requiring reproducible studies SHOULD retain the exported snapshot and the generator version used to produce it.