Before calling a lower warranty claim rate an improvement, give both shipment groups the same observation time. Then check whether their claims were counted under the same rules and with the same processing allowance. A newer group can look better simply because its record stops earlier.
The decision here is which shipment group or unresolved record deserves investigation. Our metric is the percentage of eligible original units with at least one eligible accepted claim received within a fixed period after shipment. It is not the number of tickets, repair visits or product failures. The worked example measures recorded claims, not legal coverage, a reserve or future performance.
Define what enters the fraction
Keep the denominator tied to eligible original shipments. A rejected claim or a still-pending decision does not remove the underlying eligible unit from that shipment count. Freeze the product, revision, channel, shipment eligibility and internal acceptance rule before comparing groups.
For the numerator, link claim records to stable unit IDs. Count each eligible original unit once, using its earliest-received eligible claim among those recorded as accepted before the extraction timestamp. Preserve the receipt, decision and recorded-status timestamps. Receipt determines whether the event falls inside the window. The extraction date determines which accepted decisions are known in that version.
A repeat ticket or another accepted claim on a unit already counted adds no unit. Unmatched serials, conflicting timestamps and uncertain replacement-to-original links belong on an exception list until resolved. Keep known exclusions visible too. A smaller unresolved file is not automatically a better result.
Use the same age and the same snapshot rule
In a fixed-duration reliability test, the NIST/SEMATECH handbook explains that failure times beyond the observation cutoff are missing. The same distinction helps here: a shipment observed for 60 days cannot settle a 90-day claims result. NIST/SEMATECH Engineering Statistics Handbook, Censoring
For this fictional exercise, use Toronto local calendar days. The receipt window starts at shipment, inclusive, and ends just before that shipment timestamp plus the chosen number of calendar days. Do not replace calendar days with a multiple of 24 hours. A receipt exactly at the ending boundary falls outside that window.
Also choose an administrative lag: how long after the receipt window ends to take the extract. “90+7” below means a 90-day receipt window followed by seven calendar days before extraction. The seven days allow a later status snapshot. They do not extend the receipt window or guarantee that every claim has been resolved.
Open the two 200-unit files
All figures and records in this example are invented. Northbank Products shipped 200 eligible original units in the older group on January 1, 2026, and 200 in the newer group on April 1, 2026. Both groups use the same stipulated product revision, channel and acceptance rule. Their shipment and claim records are comparable except for the timing and later information described below.
Every shipment, boundary and extraction time in this example is midnight in Toronto. The dates below are all in 2026. The 200 units in each group stay in their original denominator throughout.
The tempting comparison on May 31
The older group’s 90-day receipt window ended just before April 1. Its April 1 extract, with no administrative lag, counted 12 distinct units: 12 ÷ 200 × 100 = 6%.
The newer group’s May 31 extract counted four distinct units: 4 ÷ 200 × 100 = 2%. But April 1 to May 31 is only 60 calendar days. The newer group’s 90-day window ends just before June 30. Its 90-day result is incomplete on May 31.
Comparing 6% with 2% produces an apparent four-percentage-point fall while mixing two different ages. Do not label that an improvement in the 90-day result.
Rebuild the comparison at 60+0
For the older group, January 1 plus 60 calendar days reaches March 2. Its frozen March 2 extract counted four accepted distinct units received before that boundary. For the newer group, April 1 plus 60 days reaches May 31. Its frozen May 31 extract also counted four.
The matched comparison is therefore 4 ÷ 200 = 2% against 4 ÷ 200 = 2%. Both use a 60-day window and zero administrative lag. At that age and those frozen versions, the displayed difference disappears. Neither row supplies the newer group’s future 90-day count.
Keep a late decision in a new version
One newer-group claim arrived on May 29, age day 58. It was pending at the May 31 extraction. The claim belongs to a distinct eligible original unit, and its acceptance was recorded on June 12.
A June 13 extract retains the original May 31 receipt boundary but knows more about accepted status. It now counts five units: 5 ÷ 200 × 100 = 2.5%. The receipt still belongs inside the 60-day window. The later acceptance changes the known count, not the receipt date.
Keep the original 60+0 row at four units. Add a newer-group 60+13 restatement at five units, identifying the claim, unit, decision date and reason for the change. Do not place this later-vintage 2.5% beside the older frozen 60+0 figure and call it a matched improvement comparison.
Change one input for an alternative outcome: suppose that accepted ticket belongs to one of the four units already counted. The restatement remains 4 ÷ 200 = 2%. It adds a claim record, not another unique unit. This alternative does not supply a replacement 90-day result.
If the ticket looks eligible but its unit link is unresolved, leave it on the exception list instead of guessing a fifth unit. A known ineligible replacement shipment can sit outside the original cohort under the defined convention, with its original-unit crosswalk retained. Neither treatment silently changes the 200 eligible original shipments.
Compare the mature 90+7 records
The older group’s 90-day receipt boundary is April 1, and its seven-day-lag extract is April 8. That snapshot counts 12 units: 6%. The newer group’s corresponding boundary is June 30, and its extract is July 7. It counts ten units: 10 ÷ 200 × 100 = 5%.
The late-accepted distinct unit described above is already included in the newer group’s ten. Do not add it again. Under the stipulated comparable records, the newer 90+7 share is one percentage point lower: 6% − 5% = 1 percentage point.
That is a descriptive difference in these records. It does not establish a product-quality change or its cause. If pending or unmatched records could change the conclusion, keep the result provisional and resolve them. Equal elapsed time alone cannot repair different intake coverage or acceptance practices.
For groups shipped across several dates, test eligibility from each unit’s own shipment timestamp. Wait until the full defined cohort reaches the chosen age, or disclose an equivalently selected subset and its actual denominator. Do not quietly give later shipments a shorter window inside a supposedly mature group.
Turn the comparison into an owner decision
Mira, Northbank’s fictional quality lead, prepares Review v1 on July 8. She attaches SHIP-OLD-v1 and SHIP-NEW-v1, each containing the 200 eligible unit IDs and shipment timestamps. Her definition sheet fixes the product revision, channel, acceptance rule, replacement convention and Toronto calendar-day boundaries.
Her snapshot register keeps OLD-60-v1 and NEW-60-v1 as the original 2% pair. NEW-60-v2 carries the June 13 restatement, the distinct unit’s May 29 receipt and June 12 acceptance. OLD-90-v1 and NEW-90-v1 carry the April 8 and July 7 extracts and the 6% versus 5% result. Each row points to its intake extract and status-history version.
Mira’s next action depends on the open issue. If IDs or timing are unresolved, she asks service lead Noor for the serial crosswalk, intake record or recorded decision history, with a July 14 checkpoint. The row stays provisional. If those records reconcile, she asks operations lead Leo to examine claim-reason coding and the relevant shipment batch and production records. The lower percentage identifies work to investigate, not a cause to announce.
For that upstream check, use our rework and scrap guide to organise manufacturing records after identifying a batch or revision. It covers production evidence, not field warranty claims.
Copyable blank worksheet
Scope: Product/SKU ______. Revision ______. Channel ______. Cohort ______. Original-unit eligibility ______. Acceptance rule ______. Replacement and serial-crosswalk convention ______.
Shipment basis: Source/version ______. Unit ID ______. Shipment timestamp and timezone ______. Eligible denominator ______. Excluded or unresolved rows and reasons ______.
Claim record — repeat per claim: Intake source/version ______. Claim ID ______. Linked original unit ______. Receipt timestamp ______. Eligibility ______. Status ______. Decision timestamp ______. Recorded as-of state ______. Earliest eligible accepted claim for this unit? ______. Duplicate/crosswalk reference ______.
Snapshot: Horizon ______. Inclusive start and exclusive end for each unit ______. Extraction timestamp ______. Fixed lag ______. Accepted distinct units ______. Eligible shipped units ______. Percentage or held ______. Pending/unmatched exceptions ______.
Version and action: Previous snapshot/date ______. Changed unit or decision ______. Restatement reason/effect ______. Matched comparison row ______. Next evidence or investigation ______. Responsible person ______. Next checkpoint ______.
Keep the frozen comparison, the later restatement and the investigation decision together. A new extract should explain what changed in the record. It should not erase the earlier snapshot or turn a short observation period into a completed result.
For general information and education, not legal, tax, investment or valuation advice. Examples are illustrative and do not predict your business’s value, financing terms or sale outcome. Consult qualified advisers about your situation.
Subscribe to Deal Flow Canada for practical owner guides and Canadian deal updates.


