| Internet-Draft | Repository Availability Reporting | November 2026 |
| Arsentev | Expires 20 May 2027 | [Page] |
Papers, catalogues, model cards and tool lists point to code and model repositories by URL. Such repositories can be deleted, made private, blocked or simply stop changing, and reports of "link rot" or "project survival" often mix these different events into one number. This document defines how to classify each observation of a referenced repository, when a repository counts as unavailable, how to report inactivity separately from unavailability, and a small JSON report object. It also recommends citing a persistent identifier or an archived copy next to the repository URL.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 20 May 2027.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document.¶
Lists of software and models, including curated catalogues of AI tools, cite repositories on hosting platforms such as GitHub and Hugging Face by URL. A URL can stop resolving for several reasons, and a project can remain reachable while nobody changes it any more.¶
Platforms answer a missing repository in different ways. A platform may return 404 (Section 15.5.5 of [RFC9110]) for one missing repository and 401 (Section 15.5.2 of [RFC9110]) for another. The status 451 ([RFC7725]) is used when access is blocked for legal reasons. Single checks are noisy, because a request can fail for reasons unrelated to the repository.¶
The goal of this document is to make the definition behind every reported availability number explicit, so that numbers from different studies can be compared. This document does not define how to archive repositories, does not rank projects, and does not say which projects should be cited.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
This document uses the following terms:¶
Each snapshot records, for every referenced repository, one of the following states:¶
The report MUST list, per platform, which status codes it treats as missing. A repository MUST NOT be counted as unavailable after a single missing observation. The report MUST state how many consecutive missing snapshots it requires.¶
A repository that becomes available again after being counted as unavailable is reported as a return. It is not deleted from the data.¶
Inactivity means no change to the repository for a stated number of days, for example no push for 90 days. Inactivity and archiving by the owner are reported separately from unavailability.¶
A composite count that adds inactivity to unavailability MUST be labelled as composite.¶
A fixed inactivity threshold produces a step in a composite curve once observed projects pass that threshold. Such a step is a property of the rule, not evidence that projects disappeared.¶
Availability is reported at stated horizons counted from each repository's selection date, not from the first snapshot.¶
The report gives the estimator used (for example Kaplan-Meier) and the number of repositories still under observation at each horizon. The report SHOULD NOT report horizons where very few repositories remain under observation.¶
The report states the selection rule of the list, because a result about one list does not describe a whole platform.¶
The results of a study can be carried in a JSON object ([RFC8259]). The object carries the observation window, the per-platform missing status codes, the consecutive-snapshot rule, and per-horizon availability with the number of repositories under observation. The example in the appendix uses illustrative numbers, not measurements.¶
The report is a JSON object [RFC8259] with the following members.¶
Members not defined here MAY be added; consumers MUST ignore members they do not understand.¶
Documents that cite a repository SHOULD also give a persistent identifier (for example a DOI of an archived release) or the URI of an archived copy, such as a Memento ([RFC7089]), next to the repository URL.¶
The repository URL alone tells a later reader where the project was, not what was cited.¶
One curated radar of AI tools selected 757 entries between 11 June and 31 October 2026. 733 GitHub and Hugging Face projects were observed in 50 daily snapshots from 12 September to 31 October 2026. The metric was fixed on 5 October 2026, before analysis.¶
Unavailability was defined as 404 or 451 on GitHub and 401 or 404 on Hugging Face, two snapshots in a row. 23 projects became unavailable and 5 of them later came back.¶
Kaplan-Meier availability was 96.7% at week 4, 94.6% at week 8 and 93.0% at week 12 after selection, roughly one project in 14 lost by week 12.¶
Inactivity was more common than removal: 67 projects had no push for 90 days, against 23 that became unavailable. A composite curve that included inactivity dropped sharply after week 12, and that drop came from the 90-day rule.¶
These numbers come from one radar with one selection rule and do not describe GitHub or Hugging Face as a whole. The study is published with DOI 10.5281/zenodo.24101933 [SURVIVAL].¶
Repeated automated checks of many repositories can trigger platform rate limits. Checks SHOULD respect the platform's published limits.¶
A repository URL that resolves again after a gap may now hold different content under the same name. Reports that record returns SHOULD note that content was not compared unless it was.¶
The report carries counts and status codes only. It SHOULD NOT name individual repository owners.¶
Small counts per platform or per month can still point to a specific project. Operators SHOULD consider this before publishing per-cell numbers.¶
This document has no IANA actions.¶
The numbers in this example are illustrative and are not measurements.¶
{
"list": {"name": "example-tool-catalogue",
"selection_rule": "weekly editorial pick",
"entries": 500},
"window": {"first_snapshot": "2026-09-01",
"last_snapshot": "2026-10-31",
"snapshots": 61},
"platforms": [
{"platform": "example-code-host", "observed": 400,
"missing_status": [404, 451]},
{"platform": "example-model-host", "observed": 100,
"missing_status": [401, 404]}
],
"consecutive_missing": 2,
"unavailable": 20,
"returned": 4,
"inactive": {"rule": "no push for 90 days", "count": 50},
"estimator": "kaplan-meier",
"horizons": [
{"weeks": 4, "available": 0.98, "at_risk": 450},
{"weeks": 8, "available": 0.96, "at_risk": 300}
]
}
¶
In this example 20 repositories were missing in two snapshots in a row and 4 of them later answered again. The 50 inactive repositories are reported in their own member and are not part of the availability figures.¶