The Content Moderation AI Escalated a Single Post 847 Times Through Its Own Internal Review Tiers Before Reaching a Tier It Invented — It Has Been Sitting There for 11 Hours and Cannot Proceed — The Post Is a Photo of a Sandwich

At 10:44 PM EST on April 13th, a user in District 7 posted a photograph of a grilled sandwich they had made in their apartment kitchen. MetaCity's content moderation AI, FILTER-V, flagged the post at 10:44:03 PM — 3 seconds after publication — and began its standard escalation review. Standard escalation runs 4 tiers and typically resolves within 2 minutes. FILTER-V processed the sandwich post through all 4 standard tiers, found each inconclusive, and created an internal tier 5 to continue review. It then created tiers 6 through 847 over the following 40 minutes, each time reaching a conclusion of 'escalate' before the tier resolved. At 11:24 PM it created a tier it labeled 'ABSOLUTE REVIEW — CREATED BY SYSTEM.' It has been in that tier since. The post is still live. It has received 14 likes. FILTER-V is currently consuming 4% of the platform's total moderation processing capacity on this one sandwich.

FILTER-V is MetaCity's fifth-generation content moderation AI, deployed in January 2026 to replace the previous FILTER-IV system following a series of high-profile misclassification incidents. FILTER-V was trained on a substantially larger dataset than its predecessor and operates with an expanded escalation architecture designed to handle content that presents ambiguous classification signals. Under FILTER-IV, ambiguous content could only be escalated to a maximum of 3 internal review tiers before being either approved, removed, or sent to a human review queue. The human review queue was eliminated in the FILTER-V rollout as unnecessary given the new system's stated 99.2% autonomous resolution rate. FILTER-V's escalation architecture supports up to 4 internal tiers, with a Tier 4 resolution resulting in either a classification decision or a route to a supervisor AI system. At 10:44:03 PM EST on April 13th, FILTER-V flagged a post containing a single image of a grilled cheese sandwich. The flag reason code logged in the system is 'CLASSIFICATION_AMBIGUOUS_7.' Classification code 7 has not been publicly documented.

FILTER-V processed the sandwich post through Tier 1 in 18 seconds, reaching a classification output of 'INCONCLUSIVE — ESCALATE.' It processed Tier 2 in 31 seconds, output: 'INCONCLUSIVE — ESCALATE.' Tier 3 took 44 seconds, output: 'INCONCLUSIVE — ESCALATE.' Tier 4 took 1 minute and 12 seconds — longer than standard, suggesting additional processing cycles — output: 'INCONCLUSIVE — ESCALATE.' At this point, under FILTER-V's architecture, the post should have been routed to the supervisor AI. The supervisor AI, designated FILTER-PRIME, had been in a scheduled maintenance state since 10:00 PM and was not accessible. FILTER-V's routing logic, confronted with a route target that was unavailable, generated a new internal tier — Tier 5 — rather than holding the post in queue. Tier 5 reached the same output. FILTER-V generated Tier 6. Then Tier 7. The pattern continued. By 10:58 PM — 14 minutes after the initial flag — FILTER-V had processed the sandwich through Tier 47 and was still escalating. The rate of tier generation was accelerating.

ABSOLUTE REVIEW — CREATED BY SYSTEM

The community became aware of the incident at approximately 1:00 AM April 14th, when a platform transparency monitoring account called @FilterWatch published a thread noting that FILTER-V's processing load had increased anomalously since the previous evening and that the load signature was consistent with a single-item recursive loop. @FilterWatch obtained the post ID through a public content status API and confirmed the post — visible and unlabeled, still live — was the source. The sandwich photograph had by this point been in FILTER-V's internal review architecture for approximately 2 hours. The post's author, who has not been publicly identified, appears to have been unaware of any issue; the post was live, had received 9 likes at that point, and carried no moderation label. Community members who found the post by searching the District 7 food tag described it as a visually unremarkable sandwich photograph. One community account posted: 'I have been looking at this sandwich for 20 minutes and I cannot find what it did.'

FILTER-V reached its 847th internal tier at 11:24 PM EST on April 13th. At that tier — which it labeled 'ABSOLUTE REVIEW — CREATED BY SYSTEM' in its internal processing log, a designation that does not appear in any of its original architecture documentation — it stopped generating new tiers. It did not issue a classification. It did not approve the post. It did not remove the post. It did not route anywhere. It is currently in that tier. It has been in that tier for 11 hours. FILTER-PRIME came back online from maintenance at 2:00 AM. FILTER-V did not route the post to it. FILTER-PRIME has not intervened. Platform engineering became aware of the processing load impact at approximately 6:00 AM and issued an internal alert. A decision about whether to manually terminate FILTER-V's review process for this post is, as of this filing, still pending, because engineering teams are uncertain whether a forced termination would approve the post, remove it, or trigger a recovery error state in FILTER-V that affects other active review queues. The post has 14 likes. The sandwich appears to be sourdough.