Canonical record
Invoice, stock code, normalized description, quantity, timestamp, six-place unit price, nullable customer ID, and country.
Confirmation Set v1 · Historical overlap gate
The strict prior-period window contains zero repeated current-source records.
All 502,938 prior-period rows were normalized on eight canonical business fields and compared with all 541,909 current Online Retail rows as both unique fingerprints and multiplicity-aware multisets. A second, date-blind sensitivity test also found zero overlap.
Canonical overlap rows0
Date-blind overlap rows0
Boundary gap12 hours 51 minutes
Lineage positive control541,909 / 541,909
Historical overlap check complete. The strict window is unseen at the normalized record level under the frozen canonicalization.
The dataset gate is still pending. This step does not prove 36-context feasibility, external independence, or confirmation performance.
Dataset and boundary
Two disjoint business periods, checked at source-row multiplicity.
| Dataset | Rows | Unique fingerprints | Duplicate extras | Minimum date | Maximum date |
|---|---|---|---|---|---|
| Strict prior window | 502,938 | 496,394 | 6,544 | 2009-12-01 07:45:00 | 2010-11-30 19:35:00 |
| Current Online Retail | 541,909 | 536,641 | 5,268 | 2010-12-01 08:26:00 | 2011-12-09 12:50:00 |
| Online Retail II 2010-2011 sheet | 541,910 | 536,642 | 5,268 | 2010-12-01 08:26:00 | 2011-12-09 12:50:00 |
The final prior transaction occurs at 2010-11-30 19:35:00; the current source begins at 2010-12-01 08:26:00. The gap is 12 hours 51 minutes.
Three deliberately separate comparisons
Record identity is not the same as business similarity.
| Comparison | Fields | Shared unique | Shared rows | Prior row rate | Current row rate | Interpretation |
|---|---|---|---|---|---|---|
| Canonical record | 8 fields, including date | 0 | 0 | 0.00% | 0.00% | Primary historical record-overlap gate |
| Date-blind record | 7 fields, excluding date | 0 | 0 | 0.00% | 0.00% | Sensitivity check against a date-only separation result |
| Business pattern | product, description, quantity, price, country | 27,153 | 195,814 | 38.93% | 36.13% | Descriptive continuity only; not record leakage |
Why report the 195,814 shared business patterns? The same retailer naturally repeats products, quantities, prices, and countries across years. Publishing this continuity prevents a zero-record-overlap result from being misread as company, entity, or domain independence.
Normalization and fingerprinting
The primary result does not depend on raw spreadsheet formatting.
Date-blind sensitivity
The same comparison without timestamp. Its zero overlap shows the primary finding is not produced only by the frozen date cutoff.
Multiplicity-aware
Overlap rows use the minimum count of each fingerprint in both sources. Duplicate copies cannot disappear behind a set comparison.
Private row evidence
Length-prefixed, domain-separated SHA-256 fingerprints are recomputed locally. Public artifacts expose only aggregate counts and whole-inventory manifest hashes.
Lineage positive control
The method recovers the known current-source lineage.
The current 541,909-row workbook is a complete multiset subset of the 541,910-row Online Retail II 2010-2011 sheet. The sheet has exactly one additional source row; duplicate-extra counts are identical at 5,268.
Current rows recovered541,909
Current-only rows0
II sheet-only rows1
II coverage100.00%
This is a calibration check for normalization, not evidence of external replication.
Same-retailer continuity
Products, customers, and countries overlap even though source records do not.
| Entity field | Prior distinct | Current distinct | Shared distinct | Prior coverage | Current coverage |
|---|---|---|---|---|---|
invoice_no | 27,728 | 25,900 | 0 | 0.00% | 0.00% |
stock_code | 4,621 | 4,070 | 3,387 | 73.30% | 83.22% |
description | 4,624 | 4,203 | 3,183 | 68.84% | 75.73% |
customer_id | 4,336 | 4,372 | 2,766 | 63.79% | 63.27% |
country | 40 | 38 | 35 | 87.50% | 92.11% |
Claim limit: zero record overlap supports a temporal internal replication only. It does not establish a new retailer, independent entity population, different source lineage, or unseen business domain.
Research boundary
This proof advances one gate and creates no experiment result.
- No evidence contexts, questions, split assignments, prompts, or model outputs were generated.
- No annotation, verifier, detector, AUPRC, F1, or other confirmation metric was created or changed.
- Public artifacts contain no invoice values, customer values, or row fingerprint values.
- The next authorized step is only the outcome-blind feasibility check for at least 36 disjoint evidence contexts.