Decks
One per week, built every Monday morning for the 11:00 meeting.
← → to move between slides
Test with your cases
Runs the same pipeline as the weekly deck: you get result.json, with the items for Maya's deck and every other input with the reason it was skipped. No keys needed, nothing is stored.
From a fresh clone: ./run --input cases.json --output result.json
Or try a sample
Results
Errors
Items
Skipped
How it decides: the first rule that applies wins
- Unsafe or unusable (instructions aimed at AI, a name collision, no text) → skipped
- Outside the week → skipped
- Not from the competitor (press, Reddit, ads, rumors) → "verify on the official page"
- A change on its own pricing page or homepage → item
- Its own announcement elsewhere → item, "check the page"; features go to Noa, ads to Daniel
| id | reason |
|---|
Setup
What the daily capture looks at. Only the competitors' own pages count.
Competitors
Two pages each, captured every morning.
| Competitor | Pricing page | Homepage | Last capture |
|---|
Schedule
- Capture
- Every night after midnight, Israel time (ready by 06:30; backups until 04:30 UTC)
- Deck
- Mondays, from the week before
- Timezone
- Asia/Jerusalem
- Last run
- See the run log
Connectors
GitHubRuns the daily capture and the Monday deck
Connected
Run log
Every run, newest first. Each dot is one page: Lovable, Replit, Bolt · pricing, then homepage.
Changed
No change
First capture
Not captured or failed
| When (Israel time) | Run | For | Pages | Found | Problems |
|---|