Semantic Debt Discovery Report #001 — How Many Meanings Does “Coverage” Have?

This Discovery Report examines the single most-used noun in insurance documentation: coverage. Across publicly available policy wordings, schedules of benefits, product disclosure statements and reinsurance treaties, the term resolves to at least six distinct governed meanings.

None of these meanings is wrong. Each is correct inside its own document section. The problem is that no enterprise registry records which meaning applies where — so every downstream consumer of those documents, human or AI, has to reconstruct the right meaning from context.

The accumulated reconstruction cost is Semantic Debt. This report makes it visible for one term.

Why can the same business term — coverage — produce different answers to the same question depending on which document, system or AI agent you ask?

The answer is not that one source is wrong. The answer is that the term carries multiple legitimate meanings that have never been formally separated, named or governed.

Across publicly available insurance documentation, the word coverage appears in patterns such as the following — each one resolving to a different concept:

“Your coverage is Tier 3.” — refers to a policy plan or product level chosen by the insured.

“Section 4 coverage applies only to listed perils.” — refers to the scope of what is insured within a section of the wording.

“Coverage is limited to the European Economic Area.” — refers to the territorial applicability of the policy.

“Coverage includes loss of earnings up to €50,000.” — refers to a benefit entitlement payable under the policy.

“Coverage must meet the statutory minimum required by national law.” — refers to a regulatory floor, not a product feature.

“Coverage commences on the inception date and ceases at expiry.” — refers to the temporal protection window.

Each sentence is well-formed insurance English. Each one teaches a downstream reader — or a downstream model — a slightly different definition of the same word.

From the evidence above, “coverage” resolves to at least six distinct Meaning Spaces:

Policy Tier — the product or plan level (Bronze / Silver / Gold; Tier 1 / 2 / 3).

Section Scope — what a specific section of the policy wording insures.

Territorial Applicability — the geographic region where the policy is valid.

Benefit Entitlement — the monetary or in-kind benefit the insured can claim.

Statutory Minimum — the regulator-mandated floor that the policy must meet.

Temporal Protection — the period during which the policy is in force.

Each of these is a legitimate, governable concept. None of them is “coverage” in the abstract. Treating them as one term is the root cause of contradictory AI outputs.

Semantic Drift occurs whenever the same surface term resolves to a different concept in a different context without that change being recorded. In the evidence above, drift is not occasional — it is continuous and structural.

Underwriting documents drift toward Policy Tier and Section Scope. Claims documents drift toward Benefit Entitlement and Temporal Protection. Regulatory filings drift toward Statutory Minimum and Territorial Applicability. The same word travels between meanings as it moves between departments.

The drift is invisible because every department is internally consistent. It becomes visible only when an AI agent — or an auditor — reads across them.

Every ungoverned meaning of coverage is one unit of Semantic Debt. Six meanings means six units, per term, per document family, per jurisdiction — multiplied by every other ungoverned term in the same wordings.

The cost is paid every time a claims handler reconciles two interpretations, every time a compliance officer reconstructs a definition for a regulator, and every time an AI assistant produces an answer that is locally correct but globally inconsistent.

The debt does not appear on any balance sheet. It appears in the time, rework and risk required to operate on top of it.

There is no canonical answer to the question “which meaning of coverage applies here?”. There is no registry that lists Policy Tier, Section Scope, Territorial Applicability, Benefit Entitlement, Statutory Minimum and Temporal Protection as six distinct, owner-accountable concepts under one surface term.

Until that registry exists, every reader — human or machine — has to infer the right meaning from context, every time. Inference at scale is exactly what produces drift.

The missing decision is not “what does coverage mean?”. The missing decision is “which six (or more) governed concepts does the word coverage resolve to in our organisation, and who owns each one?”.

Retrieval — A RAG system retrieves passages containing the word coverage and treats them as comparable. They are not; they describe different concepts. Retrieval quality looks fine; answer quality is silently degraded.

Summarisation — A model summarising a policy wording compresses six meanings of coverage into one, producing a summary that no domain expert would accept.

Classification — A model classifying claims by “coverage type” conflates Policy Tier with Benefit Entitlement, distorting any downstream analytics.

Agentic Workflows — An agent that decides whether a loss is covered must resolve to one specific meaning. Without governance, it picks whichever meaning is most strongly represented in its prompt — which is a coin toss, not a decision.

One word. Six meanings. Zero governance. That is the size of the Semantic Debt sitting inside a single insurance term — before counting claim, risk, exposure, deductible, sum insured or premium.

Semantic Drift is not a future risk introduced by AI. It is an existing condition that AI exposes. Semantic Governance is the discipline that retires it; WikiSure is the platform that makes that discipline operational.

Future Discovery Reports in this series will apply the same lens to “claim”, “risk”, and other high-frequency insurance terms.