Guide
Extracting treaty wording: attachment, hours clause, reinstatement
If a term cannot be pointed at on a page, it is not in the administration system yet. It is a gap, or it is a conflict with another page.
The wording is the system of record for the treaty. Not the broker summary. Not the pricing memo. Not the spreadsheet named Final_v7. If a term cannot be pointed at on a page, it is not in the administration system yet. It is a gap, or it is a conflict with another page.
This guide is how to extract the clauses that actually govern recoveries and cessions: attachment, hours clause, and reinstatement, plus the identity fields those clauses sit on. The field contract is source-grounded extraction. No span, no term. The parent hub is treaty reinsurance. The named worker is the Treaty Agent. Bordereaux that will be tested against these fields are the bordereaux automation guide.
Do not summarise a wording into a chat window and call it a treaty abstract. Summaries drop the clause that bites at the loss. A generic model will often fill a missing hours clause from market practice. That is completion, not extraction. See generic LLM versus IDP versus operations software if you need that argument at length.
What to extract before anyone posts a row
Start with identity, or attachment is an orphan.
- Treaty title, unique market reference if present, cedent legal name, reinsurer or following market as named.
- Inception and expiry, and the basis: losses occurring, claims made, clean-cut language, year of account.
- Class, occupancy, territory, and the exclusion list as cited snippets, not as a bullet titled "key exclusions."
- Structure type: quota share, surplus, per-risk excess, cat excess, aggregate excess. The type is a field. It is not inferred from a slide.
Then the three clauses this page exists for. Then commission basis, claims cooperation, and governing law if the template asks. Optional colour in a broker covering email is not a field.
Manuscript amendments and endorsements are documents in the pack. If endorsement 3 changes the hours clause, the live hours field cites endorsement 3. The original wording span is superseded with a record of that fact, or left as a conflict until someone confirms which text governs. "We always use the latest PDF" is not a span.
Renewal comparison is an ops job: last year's cited attachment versus this year's cited attachment, side by side, with both pages. A model that says "broadly similar" has not compared the treaty.
Attachment
Attachment is where cover starts. On excess of loss it is the retention the cedent keeps before the layer responds. On proportional business you still care about any inner deductible, per-risk cap, or event deductible the wording actually wrote. People say "attachment" and mean three different numbers. The pack may store more than one field. It must not blend them.
What is an attachment point in reinsurance? is the definition this site uses. The ops rule is narrower: the field cites the page of the signed wording, not the placing slip's shorthand if the signed wording differs, and not the pricing indication.
Failure modes are ordinary.
The slip said USD 10,000,000 any one risk. Endorsement 2 raised it. Extraction still points at the slip. Recoveries attach too early or too late.
The claims bordereau attaches a loss at USD 8,000,000 because someone used a different deductible than the wording. The recovery is wrong even if paid amounts are right.
Per-risk and cat attachments are not interchangeable because a slide said XL. A weather event that never reached the cat attachment may still hit a per-risk layer, or the reverse. If the pack has one attachment field for two clauses, you will book the wrong layer.
Aggregate attachment needs an aggregate definition: what adds, over what period, with what hours. A number without that definition is a gap pretending to be a limit.
Extract:
- The amount and currency.
- Any one risk, any one event, aggregate, or other basis, as written.
- Which layer, if the contract is a programme.
- The document and page, including the endorsement that last changed it.
If two pages disagree, store a conflict. Do not pick the lower attachment "to be conservative" unless the contract says that, and then cite that rule. Conservatism without a span is still a guess.
A facultative cousin exists on placing files. ACME Construction Ltd, acme.example, has an occurrence limit of USD 10,000,000 on slip page 2 in the sample pack. That is a facultative limit, not a treaty attachment. Do not copy it into a treaty abstract because the named insured appeared on a bordereau row. Different contracts. Different spans.
Hours clause
The hours clause is how an event is cut in time. Seventy-two hours, one hundred sixty-eight hours, or whatever the wording actually said. It decides whether two storms are one event or two, and whether a wildfire week is one occurrence. If you are administering cat excess, and the hours clause is missing from the extracted pack, you do not have a cat XL you can administer. You have a heading.
Hours language is often buried, amended, or present only in a general conditions booklet that was "as attached" and not attached. A broker summary that says "standard 72 hours" is not a clause. Chase the page.
Extract:
- The number of hours, or the fact that the clause is occurrence-worded without a clock, if that is truly what the text does.
- What it applies to: wind, flood, riot, all perils, named perils only.
- Whether the cedent chooses the start, and any non-overlap rules for two events.
- The page, and the endorsement if later.
If the clause is absent, the field is empty. The chase list says hours clause not found in Wording.pdf or endorsements. Filling 72 because "that is market" is how you discover the 168-hour manuscript at the loss.
Claims files without event dates cannot apply the clause. That is a bordereaux gap, not a wording gap. Do not invent event dates to make hours look operational. See the bordereaux automation guide for event date versus report date.
Unverifiable scans of a hours clause are refusals. Chase a readable copy. Do not let a model complete the number from the blur.
Reinstatement
Reinstatement is whether capacity comes back after exhaustion, free or paid, automatic or not, once or more. What is a reinstatement clause in reinsurance? is the short answer. The ops answer is a formula with a span. "Two reinstatements" in a summary is not a formula. The premium for a paid reinstatement is in the wording or it is a gap.
Extract:
- Count: none, one, two, unlimited, or as written.
- Free versus paid, and the rate or fraction of original premium if paid.
- Automatic versus at cedent request.
- What exhausts: the limit on an event, the aggregate, a layer.
- Pro rata for time or for amount, if the clause says so.
- Any hours or event definition that gates whether one event can exhaust and reinstate.
A late claims bordereau changes reinstatement before anyone notices. A catastrophe sitting in an unsent file does not stop the clock. The wording field has to exist before the accounts team can tell whether a reinstatement premium was due. Empty is honest. A chatbot sentence that says "typically one free reinstatement" is not.
If original premium for the layer is not itself a traced field, a paid reinstatement percentage has nothing lawful to multiply. Chase the premium basis. Do not pull "original premium" from a pricing spreadsheet that never bound.
Exhaustion is an accounting event. If paid plus outstanding on an event exceeds the cited limit, the next question is reinstatement, not a silent extra recovery. Store the limit span and the reinstatement span as separate fields. Blending them into "USD 50 million with reinstatements" is how programmes get posted as a single pot.
How extraction must behave
Source-grounded extraction is not a brand line. It is the stored record: field name, value, document, page, span, status.
Traced: the hours clause is on wording page 12, endorsement 3 did not touch it.
Gap: no hours clause in the zip. Empty value. Chase item.
Conflict: page 12 says 72 hours for wind; endorsement 3 says 168 hours for all perils including wind. Both spans stored. No winner.
Unverifiable: the clause is a manuscript scribble. Chase a typed endorsement.
A weaker system completes the hours clause, picks the lower attachment, and writes "two reinstatements" without a formula. At loss, nobody can show what was signed. The product test is the same test as the facultative pack: ACME Construction Ltd has no hours clause in the sample placing file, and TIV conflicts. A treaty wording zip with a missing hours clause and two attachments must behave the same way. Anything that completes those fields for you has failed.
Operators still decide which endorsement governs, whether a special acceptance exists, and whether to book. Software lists evidence. It does not bind a layer because a model was confident.
Keep files in the tenant. This website does not take live wordings. Walk the behaviour on a sample, then ingest from a read-only mailbox in a workspace. The pack inspector is the placing sample, not a treaty admin. The rule you are buying is still visible there: empty when missing, two numbers when they disagree.
Questions
- Which treaty wording clauses must be extracted before posting bordereaux?
- Identity and basis first, then attachment, hours clause, and reinstatement as separate fields with page spans. Class, territory, and exclusions are cited snippets, not a summary titled key terms. Endorsements that change a clause must become the live source or a conflict.
- What should happen when the hours clause is missing?
- Leave the field empty and chase the page. Do not fill 72 hours from market practice. A cat XL pack without a hours span is a heading, not a contract you can administer. Unreadable scans are unverifiable, not a licence to guess.
- Why is 'two reinstatements' not enough to store?
- Reinstatement is a formula: count, free or paid, rate, automatic or not, and what exhausts. A paid reinstatement needs a traced original premium to multiply. A summary phrase cannot compute a reinstatement premium or stop a silent extra recovery after exhaustion.
- Can a broker summary or a chatbot abstract replace the wording?
- No. The signed wording and its endorsements are the system of record. Summaries drop the clause that bites at the loss. Source-grounded extraction stores document, page, and span, or it stores a gap.