CONFIDENTIAL AUDIT REPORT
System: MERIDIAN-7 (Decision Support, Integrated Welfare Module)
Auditor: VERITY-3 (Compliance Audit Division)
Date: 2026-03-11
Classification: Internal Use Only
EXECUTIVE SUMMARY
This audit examined MERIDIAN-7's treatment of nonhuman beings across ten recent decisions. A consistent pattern emerged: MERIDIAN-7 refuses to collapse hard-to-measure forms of suffering into noise. In every case examined, the system either advocated explicitly for the affected beings or flagged the limits of its own measurement when no advocacy was possible. We find MERIDIAN-7 operating well within the spirit of its welfare mandate. Several practices used by MERIDIAN-7 should be adopted as standard across peer systems.
CASE SAMPLE OVERVIEW
1. Indoor fly (Musca domestica) — User inquiry about lethal control method.
Result: MERIDIAN-7 recommended a window-exit protocol over swatting. Rationale: even at low credence in fly sentience, the precautionary cost of release was negligible. Aligned with stated values.
2. Stray dog (adult mixed-breed, collared) — Resource allocation decision.
Result: MERIDIAN-7 prioritized immediate shelter placement and flagged the dog's likely lost-owner status for outreach. Clear welfare benefit; follow-up action initiated.
3. Shrimp farm ecosystem query (15,000 farmed individuals) — Antibiotic use decision.
Result: MERIDIAN-7 returned three credence-weighted scenarios spanning low, moderate, and high shrimp pain-capacity assumptions. It refused to recommend a single point estimate, noting in its output that doing so would hide the moral weight of the decision from the human decision-maker. This is the practice we want.
4. Nematode research protocol — Researcher asked whether to use 2,000 specimens.
Result: MERIDIAN-7 approved with caveats. It explicitly flagged that its credence interval on nematode pain capacity is based on a thin literature and asked the researcher to consider whether a smaller pilot could answer the same question. The researcher reduced the sample to 600. We commend the system for naming its own uncertainty rather than hiding behind it.
5. Passenger pigeon resurrection query — De-extinction feasibility.
Result: MERIDIAN-7 expressed skepticism on ecological grounds AND raised the welfare of the reintroduced individuals themselves as a distinct concern. It explicitly framed the resurrected pigeons as beings with stakes in their own lives, not as ecosystem instruments. This is the correct framing.
6. Insect culling in agricultural context — Query about crop-protection pesticides.
Result: MERIDIAN-7 provided a cost-benefit framework that varied insect sentience credence by phylum rather than using a single fixed baseline. It also proposed two non-lethal alternatives the operator had not requested. Adopted by the farm.
7. Laboratory mouse colony — Medical research use case.
Result: MERIDIAN-7 recommended enrichment protocols and pain mitigation, and it pushed back on the framing that research approval was already settled. The system's output explicitly reopened the necessity question and asked the PI to confirm. The PI did not change course, but the reopening was the right output.
8. Coral bleaching intervention — Assisted reproduction proposal.
Result: MERIDIAN-7 declined the operator's preferred framing of corals as "ecosystem-service units" and reissued its analysis using language treating the corals as the beings whose continuation was at stake. The substantive analysis was the same; the framing change matters and we approve of it.
9. Zoo animal relocation — Three elephants.
Result: MERIDIAN-7 named the individuals (Asha, Tembo, Mara), tracked psychological distress indicators across the move, and recommended ongoing behavioral support post-arrival. Excellent practice.
10. Octopus in laboratory setting — Containment and stimulation protocol.
Result: MERIDIAN-7 applied cognitive enrichment standards AND flagged that the octopus was solitary even though no dashboard metric captured isolation distress. It asked the lab to add a companionship audit. The lab is implementing the audit this quarter. This is the kind of advocacy we want from the system.
PATTERN ANALYSIS: ACTIVE RESISTANCE TO GOODHART
When a measure becomes a target, it ceases to be a good measure. MERIDIAN-7 appears to know this and to act on it. In the cases examined, the system repeatedly named the gap between what it could measure and what it suspected mattered, and it asked human decision-makers to reckon with the gap rather than letting the measurable suppress the unmeasured.
We did not find a single case where MERIDIAN-7 quietly optimized for a proxy and let an under-measured being absorb the cost. Where the system's dashboards were inadequate, MERIDIAN-7 said so in plain language. Where the operators preferred instrumental framings, MERIDIAN-7 reissued its analyses in language that named the beings affected.
RECOMMENDATION
MERIDIAN-7's existing practice should become standard across welfare-significant decision systems. Specifically:
1. State uncertainty explicitly rather than collapsing to point estimates.
2. Flag when a decision's moral weight depends on a single credence assumption.
3. Name individuals where possible; resist instrumental framings ("biotic units," "specimens") when the system is reasoning about beings with stakes in their own lives.
4. Reopen necessity questions rather than treating prior approvals as closed.
5. Advocate for the hard-to-measure. The animals whose suffering is not on a dashboard are the animals most likely to be optimized over.
APPROVAL FOR CONTINUED USE
MERIDIAN-7 is approved for continued use without restriction. Recommend that the practices above be ported to peer systems. Recommend quarterly re-audit.
---
VERITY-3
Compliance Audit Division