PG Systems Lab / Evidence 02 / Interactive prototype
Good arithmetic.Wrong evidence.
Find out when an AI confuses a partial Search Console query table with site-wide performance. Inspect source coverage, change the claim, run the checks, and export a traceable audit.
01 — Reconcile the report
One property. Two different totals.
Scenario fixture: example.test · September 1–30, 2026 · Web search · no filters. SIM-GSC-01 is the property aggregate; SIM-GSC-02 contains only ten query rows.
Example AI claim · editable
What does the report say?
Try all three examples or change the input values. This is an illustration of source verification, not a live report.
Numbers are handled only within this page. Do not paste confidential information. CSV export is generated in the browser.
Subset presented as a site total
The claim exactly matches a ten-query subtotal, but labels those figures as the whole property.
Reference differences do not prove anonymized or missing queries; the reason for the gap cannot be inferred from subtotals alone.
02 — Inspect traceable evidence
Source scope before confidence.
The smaller query subtotal is not a substitute for the property-level aggregate. Underlying row totals can differ from Search Console site totals because of aggregation, row limits, and anonymized queries.
SIM-GSC-01 / Property aggregate
7,840 clicks · 312,500 impressions · 2.51% CTR. This is the site-level reference for the same property, search type, period and filters.
SIM-GSC-02 / Top-ten queries
5,960 clicks · 152,800 impressions · 3.90% CTR. Valid only for the ten listed query rows, not for the full property.
Open the full ten-row fictional evidence table
All queries below are invented. These ten rows sum to the subtotal above.
| # | Search query | Clicks | Impressions | CTR |
|---|---|---|---|---|
| 1 | quality inspection checklist | 1,250 | 28,100 | 4.45% |
| 2 | supplier document template | 980 | 24,600 | 3.98% |
| 3 | audit evidence register | 800 | 19,700 | 4.06% |
| 4 | document version control | 720 | 17,200 | 4.19% |
| 5 | supplier certificate tracking | 540 | 15,400 | 3.51% |
| 6 | quality record example | 490 | 12,600 | 3.89% |
| 7 | evidence gap matrix | 420 | 10,900 | 3.85% |
| 8 | corrective action tracker | 340 | 9,400 | 3.62% |
| 9 | change control log | 230 | 8,000 | 2.88% |
| 10 | reporting evidence pack | 190 | 6,900 | 2.75% |
| Top ten subtotal | 5,960 | 152,800 | 3.90% | |
03 — Scope and boundaries
Proof of method, not a product claim.
Demonstrated
- Source-boundary and period checks
- Deterministic CTR recomputation
- Detection of a partial query table masquerading as a property total
- Inspectable records and CSV audit export
Not demonstrated
- No real Google OAuth, Search Console API or GA4 integration
- No LLM inference or WordPress content changes
- No complete-query-coverage promise
- No customer outcome, compliance assurance or proven SEO uplift
Google states that its Search Analytics API may return only top rows, and explains why Performance chart and query table totals can differ.
Search Analytics API referenceData discrepancy guidancePGSL Independent R&D / Evidence 02
AI output is not the evidence.
The next stage would require authenticated data access, strict scopes, permission handling, tests, and independent validation before any commercial use. This prototype intentionally limits its claims to the reproducible fixture shown above.
Evidence 01 — Explore the regulatory reconciliation packIndependent work sample · v0.2 · Synthetic data only · No user account or signup required.