Survivorship & Staleness Map: Is Your AI Memory System Still Maintained?

How rows are classified

Signal source (in priority order): latest-release cell → code-release "last commit YYYY-MM" → created cell (as a soft signal, when nothing else is available).
Thresholds: Active ≤ 12mo · Stale 12–24mo · Abandoned > 24mo, measured against a fixed snapshot date (2026-05-12) so this view doesn't drift over time.
Research split: Tier 3–5 rows route to the Research bucket regardless of date. "Maintained" is the wrong axis for a published paper — but if a research row has a stale public repo AND inbound citations, it surfaces in the "Abandoned but cited" card below.
Unknown subdivision: The Unknown bucket is split four ways so its structure is visible: closed-source · OSS with weak signal · newly-created · N/A. Most Unknown rows are closed-source products that don't publish changelogs — this is a limit of the catalog signal, not evidence of abandonment.

What each bucket means for a reader

Active

166

A release or commit in the last 12 months. Use confidently — the project is shipping. The Active cohort is the working answer to "what is alive in this space right now".

Stale

Last signal 12–24 months ago. Re-verify the README, changelog, and recent issues before recommending. A stale row is often a project on a slow cadence rather than a dead one, but the burden of proof is on the catalog user.

Abandoned

No signal in 24+ months. Treat as historical: it shipped, it influenced things, but production deployments should not depend on it being maintained. The "Abandoned but cited" card below surfaces the abandoned rows the rest of the catalog still builds on.

Unknown

364

No parseable freshness signal in our cells — most often because the record is a closed-source product that does not publish a changelog. This is a limit of our signal, not a property of the system. The breakdown below shows the structure.

Research

369

Tier 3–5: papers, techniques, theoretical proposals. "Active development" is the wrong frame — these contribute ideas, not maintained codebases. Where a research record has a stale public repo AND is still cited, it surfaces in the "Abandoned but cited" card.

Unknown bucket — what's actually in there?

The Unknown bucket dominates the catalog (364 / 912 rows). Most of that is structural — closed-source products with no public release cadence — not evidence of abandonment. The breakdown below makes that structure visible.

Closed-source · 323OSS weak-signal · 41

Closed-source, internal status unknown 323 · 89% of Unknown
Proprietary products with no public release cadence. Common for T1 commercial vendors. The catalog cannot observe their internal status — this is a limit of our signal, not a property of the system.
OSS but signal-too-weak 41 · 11% of Unknown
Has a GitHub presence but we did not capture a parseable "last commit" date in the cell. Fixable by deeper data collection on these specific rows.
Newly created (< 6mo) 0 · 0% of Unknown
Created in the last 6 months. Too new to assess freshness — re-check at the next snapshot.
N/A (research/benchmark-only) 0 · 0% of Unknown
Genuinely not applicable: benchmark or evaluation harness mislabelled as a product, or a record that explicitly carries no operational signal.

Abandoned but still cited

The most actionable finding of this view. These rows show as Abandoned (or as research with a stale public repo) yet remain load-bearing in the catalog — other systems still cite them, build on them, or integrate against them. Re-verify before any new work depends on them; consider the lineage and the active siblings. Score = inbound integrations × 3 + inbound citations.

#1 Compressive Transformer T3