Outreach Observatory

Documentation outreach,
recorded in public time

CROVIA offers optional documentation enhancement packages to AI maintainers. We record public replies and timeline changes without judging intent or compliance. Everything shown here is tied to verifiable, public records.

What this is

  • Neutral documentation suggestions, delivered publicly.
  • Timestamped outreach records with traceable links.
  • Continuous monitoring for visible documentation changes.

What this is not

  • No accusations or judgments about compliance.
  • No private messages or undisclosed sources.
  • No claims of causality — only temporal sequencing.

Safety controls

  • Strict rate limits, cooldown windows, and dedupe.
  • 1 target per org in GitHub outreach list.
  • All activity logged and reproducible.
--
Suggestions Offered
--
Public Replies
--
Closed
--
With TPA Evidence
--
Avg Days Pending
--%
Avg Completeness

How the outreach record works

01 — OBSERVE
Read public documentation

We analyze model cards and public metadata to compute a completeness score and identify missing disclosure sections.

02 — OFFER
Send a neutral suggestion

We open a public GitHub issue (or historical HF discussion) with optional enhancements, citations, and checklists.

03 — RECORD
Track visible replies

We log public replies and closures as status updates. This is a timeline record, not a verdict.

Platform transition notice

Outreach channel shift: HuggingFace → GitHub

In early 2026, CROVIA's outreach shifted from HuggingFace Discussions to GitHub Issues. This change was necessitated by access limitations encountered on the HuggingFace platform for this account. This is a factual observation — platform access policies are the prerogative of each provider. We document this shift transparently because it affects the channel distribution visible in our records.

HUGGINGFACE DISCUSSIONS
--
Historical records (access limited since Jan 2026)
GITHUB ISSUES
--
Current active outreach channel

Both channels are included in all statistics. Every record links to its original public thread for independent verification.

Observatory (post-outreach drift)

DDF — DRIFT
Documentation change monitor

We continuously scan for model card disclosure drift. When changes are observed after outreach, we record a temporal sequence (not causality).

SAFE FILTERS
False-positive controls

Correlations require a time gap and specific disclosure fields (training section / declared datasets). This is conservative by design.

PUBLIC OUTPUT
Reproducible evidence

All observatory entries are backed by public links and DDF hash changes. If no correlations are shown, monitoring is still active.

--
Drift Events (30d)
--
Outreach Targets With Drift
--
Post-Outreach Correlations

Recent suggestions

TargetCompletenessProvenanceDocumentationOfferedDaysStatus
Loading...
PendingWe opened an outreach record. No public response yet.
AcknowledgedA maintainer publicly replied. We do not infer agreement or action.
ClosedThe thread was closed publicly. We record closure without interpretation.

Methodology & neutrality

Outreach entries are generated from real logs and public links. Each suggestion includes a completeness_score derived from documentation analysis. We separate provenance_status (explicit data provenance) from documentation_status (overall completeness). We track days_pending to show time since the offer. We do not claim causality; we only record what was publicly visible and when it happened.