Moderation — one review queueENGINE — moderation (MAV-owned map, 2026-08-09). One surface, every product mounts it; FG’s vault review queue is the live seed. Converges with Banned words (same engine). ⚠️ It does not converge with Folo’s content/strict-audit pair — that is the RATING layer (ENGINE-RATING), a sibling queue whose input is items moderation already PASSED. Corrected on the bus 2026-08-09 (Folo-PM ruling, Mav accepted). folo.link answer (§17e): this page COMPOSES the FG vault queue + Folo mirror rows into one view — and Approve / Take down / Restore work in place on every skin: the write routes through the row’s owning product’s moderation API (one write path, audited in its log); “open in <product> admin ↗” rides each row as the quiet secondary.
Every uploaded item passes here before a fan can see it — thumbnails, explicit-level classification, flags, reversible decisions.
Demo dataPartial · FG seed live
Awaiting review
123
+18 since yesterday
DEMO DATA · window: now · source: FG vault queue (live seed)Flagged
8
rekognition + compliance flags — reviewed first
DEMO DATA · window: now · source: classifier flagsDecided · 7d
1,904
median time-to-decision 41 min
DEMO DATA · window: 7d · source: moderation auditTaken down · 7d
21
every take-down reversible — nothing is deleted
DEMO DATA · window: 7d · source: moderation auditReclassify — display tier Manual overrideNot the same decision as Approve/Take-down above. Moderation asks may this exist at all (publication). Reclassify asks which surfaces may show it (display tier: strict · suggestive · explicit). Collapsing them would force a moderator to remove content whose only fault is being mis-tiered.
Write path: writes the DISPLAY TIER into a Folo-owned override table (folo_config), consulted after the CDC value — never into knky_import.*, which the CDC worker replaces wholesale every pass and would erase the override within minutes. Every change is actor + reason + timestamp, read as a ledger, not a setting.
Write path: writes the DISPLAY TIER into a Folo-owned override table (
folo_config), consulted after the CDC value — never into knky_import.*, which the CDC worker replaces wholesale every pass and would erase the override within minutes. Every change is actor + reason + timestamp, read as a ledger, not a setting.Public, scanner said NSFW
699
scanner was right, cover was mis-placed — reclassify, not take-down
DEMO DATA · source: knky scan vs publish state, measured 2026-08-23Empty scan · public
50,377
zero categories returned — an empty scan is not a safety claim
DEMO DATA · source: 688,486 scanned (61%) empty, filtered to publicStrict-tier, no evidence
26,101
rated safest, carries no moderation evidence at all
DEMO DATA · source: Goose measurement, 2026-08-23UNKNOWN state
First-class, not "safe"
⚖
Awaiting founder ruling — not Admin's or Folo-PM's to decide: fail-close the ~50,377 empty-scan public covers (blurs roughly a third of the public feed) or pass them (the state that produced the 699 above). Built as a filter on the queue below so either ruling lands without a rebuild.
Unknown
post-media 90417 · @sophiajansencurrently displays SFW · scanner: NSFW label present
scanner · NSFW label, published as SFW
Unknown
post-media 90233 · @novabelleempty scan · currently displays Strict, zero evidence
empty scan · 0 categories returned
Demo — no action writes a real
folo_config row. Vocabulary note: this control sets the display tier (strict/suggestive/explicit) — never the vault’s separate moderation vocabulary (sfw/suggestive/nsfw/unrated) used in the queue above.Review queue 6 shownThumbnails resolve through short-lived signed URLs via the media presigner — nothing here is a public asset link. Take down is reversible: the item is gated, never deleted, and every decision writes an audit row. Explicit-level badges come from the classifier (SFW / Suggestive / Explicit); rekognition + compliance flags ride the row. Owner + write path (§17e): the item’s publish state belongs to the platform whose vault it sits in — FG rows write via the FG moderation API, Folo mirrors via Folo’s. On every skin the row’s actions work in place (same grant model everywhere): the write routes to the row’s owning product and audits in its log; the deep-link into that product’s admin is a quiet secondary, never the only path.
Newest first · flagged items sort to the top of Awaiting.
WRITE PATH · Take down withholds the item from publish + send via the owning product’s moderation API (FG row → FG · Folo mirror → Folo) — reversible, audited there · ACTS IN PLACE on every skin (your grant: super admin) · EFFECT · fans stop resolving it within seconds (presigner gate), scheduled sends skip it; Restore reverses bothExplicit
vault-item 88231 · @delishaNSFW surface · auto-classifier escalation · 14 min ago
rekognition · face matchcompliance · 2257 doc missing
SFW
post-media 90417 · @sophiajansenSFW surface · routine pass · 26 min ago
Suggestive
vault-item 88304 · @missariaNSFW surface · routine pass · 41 min ago
Suggestive
post-media 90233 · @novabelleSFW surface · 3 user reports · 1 h ago
reports · “not SFW”
Explicit
vault-item 88350 · @kittwilderNSFW surface · auto-classifier · 2 h ago
Explicit
vault-item 87911 · @veravosstaken down 2026-08-08 · reason: compliance flag · reversible
compliance · takedown
▣
Decisions are reversible and audited. Approve / Take down write an audit row with the acting admin + reason; a taken-down item is gated at the presigner, never deleted, so Restore is always one click. Demo — no action mutates a real record.
Queue mix
by explicit level · shown queue
152
SFW · 61Suggestive · 57Explicit · 34