Ink Foundry

MAND8 · Lloyd's Lab thesis · a Bankabil venture

Turn AI permission into underwritable evidence.

MAND8 issues Universal Insurance Receipts for AI-assisted insurance workflows: structured, signed evidence showing whether an action was authorized, bounded, reviewed, revocable, and reconstructable.

Built for delegated authority, underwriting, claims, monitoring, and renewal. MAND8 helps Lloyd's-market participants test whether AI-assisted actions can be converted from opaque system behavior into market-readable insurance evidence.

View the UIR workflowRead the Lloyd's Lab pitch
Shadow-mode only. No quoting, binding, declining, settling, or alteration of live insurance obligations.

Hierarchy: Bankabil parent trust substrate · MAND8 insurance product · UIR Lloyd's-facing artifact · Authority Receipt first UIR implementation profile.

The artifact

Universal Insurance Receipt

A UIR is an underwriter-readable evidence object for an AI-assisted insurance action. MAND8 emits one when a high-consequence action is proposed, reviewed, executed, blocked, escalated, overridden, suspended, or revoked.

Universal Insurance Receipt · schema v0.1uir/authority-receipt
Authority source
Binder, endorsement, policy, mandate, or delegated authority
Action class
Underwriting, referral, claims, monitoring, renewal, compliance
Boundary
Limit, territory, class, aggregate, exclusion, threshold
Review state
Required review, completed review, missing review, override
Status
Permit, refer, block, undecidable, invalid authority, missing evidence
Evidence binding
Hash, provenance, replay reference, receipt signature

The problem

AI insurance risk is not just a model-risk problem. It is an authority problem.

AI is entering underwriting, claims, delegated authority, compliance, and portfolio operations faster than institutions can prove whether a specific action was permitted. Model inventories show systems. Logs show activity. Point-in-time diligence shows review posture. None of those necessarily proves that a specific AI-assisted action was inside the governing binder, endorsement, mandate, limit, effective period, and required human-review boundary.

Authorization is fragmented across binders, endorsements, policy documents, tickets, model logs, approvals, role matrices, and human memory. The UIR is designed to close that gap for one action at a time.

The product flow

MAND8 sits inside the workflow.

AI-assisted action

Underwriting, referral, claims, renewal, monitoring.

UIR

Signed receipt: authority, boundary, review, status, evidence.

Evidence bundle

Provenance chain, hashes, replay reference, mandate binding.

Insurance review

Practitioner reconstructs whether the action stayed in authority.

The UIR does not claim the decision was correct. It answers the narrower institutional question: was the action authorized, what evidence supports reconstruction, and where did authority fail?

A valid Authority Receipt does not prove the underlying judgment was accurate, lawful, profitable, regulator-approved, or liability-free. It provides evidence of authority, boundaries, review / referral, and reconstructability.

Validation venue

Lloyd's Lab is the right place to pressure-test UIR as a market artifact.

The Lloyd's market is authority-dense: binders, endorsements, capacity, delegated authority, limits, exclusions, referrals, revocation, expiration, and renewal create the exact environment where UIRs should be tested. The goal is not to sell a generic AI-governance platform. The goal is to determine whether Lloyd's-market practitioners can use UIRs to reconstruct whether an AI-assisted insurance action stayed within authority.

Lloyd's Lab provides access to delegated-authority practitioners, workflow critique, market mentors, and a bounded ten-week calendar — the conditions under which the UIR can either survive practitioner scrutiny or fail informatively.

Ten-week programme

What we would do inside Cohort 17.

  1. Week 1Select one AI-assisted delegated-underwriting workflow and define shadow-mode boundary.
  2. Week 2Map authority chain: binder, endorsement, capacity, authority holder, dates, roles.
  3. Week 3Translate limits, exclusions, territories, referrals, and review requirements into controls.
  4. Week 4Finalize UIR schema v0.1 and identify evidence gaps.
  5. Week 5London midpoint review with underwriting, claims, compliance, legal, actuarial, and technical mentors.
  6. Week 6Connect permissioned or representative workflow events to receipt, hashing, provenance, and replay stack.
  7. Week 7Generate UIRs and structured PERMIT / REFER / BLOCK statuses.
  8. Week 8Run practitioner reconstruction tests.
  9. Week 9Define integration, security, buyer, pricing, and design-partner pilot.
  10. Week 10Deliver validation pack, reconstruction report, failure library, and proceed / narrow / stop decision.

Working team

Two people. Clear roles.

Meron Rudy Tewelde

Founder & Chief Architect

Leads MAND8 architecture, Holonom kernel translation, UIR schema design, receipt infrastructure, workflow modeling, and technical validation.

Sarem Hailu

Programme Support · Market & Investor Research

Supports Lloyd's Lab preparation, market research, application coordination, pitch logistics, and stakeholder mapping.

Product readiness

Not a concept deck. A receipt stack already exists.

MAND8 is built on a completed computational kernel and founder-developed infrastructure for canonical serialization, deterministic hashing, signed receipts, typed schemas, provenance chains, workflow replay, validation, and evidence-bundle generation.

The Lloyd's-specific work is vertical productization: insurance schemas, mandate ingestion, UIR profiles, PERMIT / REFER / BLOCK evaluators, developer documentation, security posture, and design-partner integration.

Kernel

Complete. Holonom computational engine for action paths, boundaries, exposure, reconstruction uncertainty.

Receipt infrastructure

Built. BLKBXS / Bankabil capture stack — canonical hashing, signed receipts, provenance chains, replay.

Insurance layer

Pre-pilot. UIR schema v0.1, Authority Receipt profile, PERMIT / REFER / BLOCK evaluators in design.

Shadow-mode guardrails

What MAND8 does not do.

  • Does not quote, bind, decline, or settle insurance business.
  • Does not alter live policy, binder, or claims obligations.
  • Does not claim Lloyd's approval, insurer deployment, or regulatory endorsement.
  • Does not certify the underwriting judgment was correct — only that authority is reconstructable.
  • Does not replace human review; it records whether required review occurred.
  • Does not require production integration for validation — shadow-mode receipts on representative data first.

Request materials

Three doors, depending on why you are here.

Lloyd's-market reviewers

Request the UIR validation brief.

Schema, reconstruction test protocol, shadow-mode boundary, and practitioner interview scope.

Request brief

Investors

Request the private capital memo.

Structure, milestones, and diligence pack shared privately — not published on this page.

Request memo

Technical reviewers

Request the schema and receipt demo.

UIR JSON schema, sample Authority Receipts, hashing / provenance walk-through.

Request demo
Thesis registryBLKBXS receipt infrastructureMethodologyResearch note