Website change detection: our controlled test results
Controlled cases for our extraction, condition, screenshot-comparison and weekly visual-event deduplication modules. Inspect what was tested and download the inputs and results.
ChangeEagle ·
These are local module tests, run without network requests, AI calls or message delivery. They do not measure production uptime, scheduler delay, email delivery, whole-system duplicate suppression or competitor performance. The fictional fixtures are deliberately small and readable; they do not represent the range of real websites.
Observed results
- Cases run
- 25
- Matched expected outcome
- 25
- Missed expected changes
- 0
- Unexpected change decisions
- 0
Report generated: . Module source hashes are included in the download so this result can be tied to the tested implementation.
| Case | Method | Expected | Observed |
|---|---|---|---|
| Price: above, equal, below, repeat, reset, return · 1 | HTML extraction + condition | No change / no alert | No change / no alert |
| Price: above, equal, below, repeat, reset, return · 2 | HTML extraction + condition | No change / no alert | No change / no alert |
| Price: above, equal, below, repeat, reset, return · 3 | HTML extraction + condition | Change / alert decision | Change / alert decision |
| Price: above, equal, below, repeat, reset, return · 4 | HTML extraction + condition | No change / no alert | No change / no alert |
| Price: above, equal, below, repeat, reset, return · 5 | HTML extraction + condition | No change / no alert | No change / no alert |
| Price: above, equal, below, repeat, reset, return · 6 | HTML extraction + condition | Change / alert decision | Change / alert decision |
| Stock: unavailable, available, repeat, reset, return · 1 | HTML extraction + condition | No change / no alert | No change / no alert |
| Stock: unavailable, available, repeat, reset, return · 2 | HTML extraction + condition | Change / alert decision | Change / alert decision |
| Stock: unavailable, available, repeat, reset, return · 3 | HTML extraction + condition | No change / no alert | No change / no alert |
| Stock: unavailable, available, repeat, reset, return · 4 | HTML extraction + condition | No change / no alert | No change / no alert |
| Stock: unavailable, available, repeat, reset, return · 5 | HTML extraction + condition | Change / alert decision | Change / alert decision |
| Text: absent, case-insensitive match, repeat · 1 | HTML extraction + condition | No change / no alert | No change / no alert |
| Text: absent, case-insensitive match, repeat · 2 | HTML extraction + condition | Change / alert decision | Change / alert decision |
| Text: absent, case-insensitive match, repeat · 3 | HTML extraction + condition | No change / no alert | No change / no alert |
| Selected text: baseline, footer edit, allowance edit, repeat · 1 | HTML extraction + condition | No change / no alert | No change / no alert |
| Selected text: baseline, footer edit, allowance edit, repeat · 2 | HTML extraction + condition | No change / no alert | No change / no alert |
| Selected text: baseline, footer edit, allowance edit, repeat · 3 | HTML extraction + condition | Change / alert decision | Change / alert decision |
| Selected text: baseline, footer edit, allowance edit, repeat · 4 | HTML extraction + condition | No change / no alert | No change / no alert |
| Identical screenshots | Screenshot comparison | No change / no alert | No change / no alert |
| Changed price and stock badge | Screenshot comparison | Change / alert decision | Change / alert decision |
| Repeat latest screenshot | Screenshot comparison | No change / no alert | No change / no alert |
| Unreadable latest PNG | Screenshot comparison | Unreadable image | Unreadable image |
| Weekly: three copies of one visual event | Weekly visual-event deduplication | 1 events retained | 1 events retained |
| Weekly: event already in prior period | Weekly visual-event deduplication | 0 events retained | 0 events retained |
| Weekly: change, reversal and return | Weekly visual-event deduplication | 3 events retained | 3 events retained |
How to interpret the report
Text and product cases feed controlled HTML into the real extractor, then evaluate the configured condition while carrying forward prior state. Visual cases compare controlled PNGs with automatic full-page noise masking disabled, as appropriate for a cropped region. Identical images, an altered image and invalid image data are included. An unchanged repeat tests the decision module only. Three additional cases exercise the weekly visual-event deduplication module: repeated evidence, an event from the prior period, and a reversal followed by a recurrence. Sending notifications and generating summaries involve additional code outside this report.
What these results do not establish
A passing fixture is evidence about that input, not a population accuracy rate. Real sites can change layout, hide information behind scripts, block requests or show different variants. Monitoring can miss changes between checks. AI explanations and third-party delivery can fail independently. Start with a known page, inspect the baseline and review early results.
Inspect the inputs and method
The download includes each HTML input, condition, prior state, observed result, screenshot references and source hashes. The application source is private, so this is not a standalone detection engine. We rerun the cases with our internal generator and check that the published result matches the tested source. You can use the disclosed inputs to design your own comparison.