A Policy Forms Library Is a Control System, Not a Document Repository

We treat the form number in a footer as filing furniture.

A label for the printer.

A string that finds a PDF when a broker calls.

That is casual by several orders of magnitude.

The form number and edition marker are part of the primary key for a contractual promise.

The policy administration system executes a promise.

The declarations say which forms attached to a term.

The attached form says what the promise means. The footer tells us which text we have.

Lose the relationship between those three and we haven’t modernised anything.

We have made the evidence easier to browse. The distinction is not academic.

CG 00 01 12 07 and CG 00 01 04 13 share a number, a broad purpose, and most of their text.

They are still different contracts.

A loss in 2015 is answered by the edition that attached in 2015. Not by the version now sitting in a shared drive. The two digits at the end are not metadata. They are the difference between evidence and a plausible guess. Which is why I wouldn’t start this work with document management, OCR, or a taxonomy workshop.

All useful later.

First build the control record.

It has to establish, term by term, which edition attached, what it modified, why it attached, and whether it narrowed or broadened the promise. Search without lineage is a polished form of ignorance.

The policy carries more than one stream of evidence

The declarations page is not a coverage form.

It carries named insureds, limits, deductibles, effective dates, and the schedule of attached forms. It is still the authoritative record of what this insured bought for this term. Treat the declarations as a cover sheet and we will get claims answers wrong before reaching the insuring agreement. The attached wording is a different kind of evidence. It holds the grant, the definitions, the exclusions, the write-backs, and the conditions. The IBC 2100 general liability wording is a market reference point in Canada. Our actual wording may be a recognisable departure from it, with a provincial amendatory endorsement or a manuscript change.

The administration system needs the exact attached text. The reader needs a controlled route back.

Then there is the footer.

It is authoritative for identity, edition, and page count.

It is authoritative for nothing else.

A footer can’t say whether the form was mandatory.

Whether its schedule blanks were populated. What changed from its predecessor.

Whether it was permitted for the jurisdiction.

A record designed around the PDF alone confuses the source of identity with the source of operational truth.

The sources disagree in regular ways.

The declarations schedule may list CG 00 01 without the edition suffix. A document stream may attach an endorsement the schedule omits. A page marked Page 1 of 3 may stop at page two. A mandatory amendatory endorsement can be paired with a base edition it was never written against.

Those are not scanning defects.

They are issuance defects with a document-shaped symptom.

The audit is correspondingly plain.

Pull the schedule. Pull every attached document. Compare three things. Every listed form is attached. Every attached form is listed. Each footer edition matches the schedule. Then add schedule blanks, page counts, and jurisdictional dependencies to the exception queue.

Not a glamorous control.

Better than a glamorous control, because it tells somebody before a claim does that a policy packet cannot prove itself.

Editions are events, and versioning them as versions loses the history

A forms library needs an identity rule without sentiment.

Number plus edition marker is one inseparable key.

Cosmetic reissues stay one member where the text hasn’t changed.

Changed wording is a separate member.

Treat every reissue as new and we create noise. Treat changed text as a version of the same thing and we delete the history that matters. The real question is never which edition is current.

It is which edition attached to which term.

A long-tail liability claim can turn on wording withdrawn in the 1970s. A library holding only a current form cannot answer a coverage question about a 2014 policy year. Not even if every page in the repository is beautifully searchable.

Each edition needs a lineage record.

Supersedes. Superseded by. Withdrawn on. Relationship to a base form. It also needs a coverage-effect classification: broadening, restrictive, or neutral against the prior edition. The revision circular or filing memorandum is where that change is stated.

Miss it when the form arrives and somebody later has to diff legal wording by hand, then persuade a claims or compliance colleague that the result is complete.

That classification is not editorial commentary.

A renewal can carry the same base form number while the edition suffix moves.

Several jurisdictions want advance notice where renewal terms narrow.

The control record has to expose a restrictive change as an issuance and compliance event.

Not leave it to a reader who happens to notice two changed characters in a footer.

Canada makes the library considerably less forgiving

Canada gives us no single public form trail.

Automobile insurance is the closed side.

Ontario’s OAP 1 and OPCF endorsements are approved by FSRA.

SPF and SEF families operate elsewhere.

An insurer wanting different automobile wording needs an approved variation, not a quiet redraft.

Commercial property and liability are the open side.

They are generally not filed in Canada, even where conventions like IBC 2100 are widely used.

Quebec publishes its approved automobile materials.

For most Canadian commercial forms, no public record exists.

A form library cannot outsource its provenance to a regulator’s website, because the website has nothing to say.

The provincial layer stays a separate relationship too.

Not as a note on a base form.

Statutory conditions are legislation reproduced inside certain policies.

We may broaden them. We generally may not narrow them.

In Quebec the Civil Code of Québec governs insurance contracts rather than the statutory-conditions regime.

A record marking a wording “Canadian” has hidden a decision.

Language makes the same point more quietly.

English and French texts can share one form number and diverge, because they were translated rather than jointly drafted.

In Quebec the French text ordinarily has priority.

They aren’t one form with a language attribute.

For coverage control they are two records whose lineage has to stay in step.

Alongside any Quebec endorsement overwriting the base wording.

The system is executing a document it may not actually possess

This is where policy anatomy meets technology operations.

The declarations schedule establishes the forms in force.

An endorsement, being later and more specific, overrides the base form it amends.

Negotiated manuscript wording controls over pre-printed text.

Definitions operate everywhere the defined term appears.

Our system may not render all of that at transaction time.

It must not behave as though they don’t exist.

Take the familiar additional-insured question.

An ACORD 25 certificate says it confers no rights and does not amend the policy.

Status comes from the endorsement, the CG 20 10 and CG 20 37 families among them, if it attached.

The difference between a certificate image and an attachment record is not a records-management nicety.

It is the difference between reporting a statement and proving a contractual right.

The same discipline covers claim-time conditions.

Statutory conditions displace inconsistent wording in common-law provinces.

In Ontario auto, the coverage question is often which OPCF endorsement appears on the declarations.

Not a free-form argument about the base policy.

A library joining statutory text, base form, amendatory endorsement, and term-level schedule can support that work.

One holding them in indistinct folders sends the hard question to whoever remembers the old process.

There is a legal reason to be exact.

In Progressive Homes Ltd v Lombard General Insurance Co, the Supreme Court of Canada set out the familiar sequence.

The insured establishes the grant. We establish an exclusion. The insured establishes a carve-back.

The result can turn on the third sentence of a form, a definition altered by an endorsement, or a statutory term we did not draft.

“The system found the latest PDF” doesn’t help that analysis.

Build the control record before anybody builds the repository

The first deliverable is an edition-and-attachment control record.

Not a universal content store.

It carries the combined identity key, printed title, issuer, role, base-form dependency, required schedule fields, attachment rule, jurisdiction applicability, language, lineage, coverage effect, and source provenance.

It records occurrence or claims-made.

The retroactive date belongs on the declarations as policy data, not in the form identity.

It also makes the seams visible.

Policy administration is authoritative for attachment logic, print sequence, permitted issue window, and merge fields.

The declarations are authoritative for what attached here.

The footer is authoritative for the edition printed.

The revision material explains what changed.

The record picks no winner among them.

It gives each source a question it can answer.

So our first acquisition may be an interview rather than a platform.

Somebody still knows why a product code mapped to a dead endorsement.

Why an old Quebec packet printed two texts.

Why a renewal job carried an obsolete edition.

Capture that while it can still be checked against forms, schedules, and system behaviour.

Once it becomes a story about a system that used to work, the control has already failed.

AI makes the archaeology faster, and it should.

It extracts footer values, compares editions, finds apparent missing attachments, and surfaces a form number buried in a scanned renewal package.

It should not decide whether an alteration is restrictive, whether a statutory condition controls, or whether an attachment proves delivery.

Those determinations have contractual consequences.

AI can help us understand the promise.

It doesn’t get to improvise it.

The board conversation is about optionality.

Our task is proof.

A migration, a renewal change, or a claims answer stays defensible when we can show the declarations schedule, the attached wording, the footer edition, and the statutory overlay as one controlled chain.

The risk was never unreadable documents.

It is being unable to prove the contract we issued.

A useful forms library does not store wording.

It preserves the promise attached to a term.