Skip to main content
PalatineAI ranks; your agent decides. The intended loop is screen wide with the feed, then deep-research the few survivors before any action.
1

Fetch

GET https://palatineai.site/projects.json. Read stats.updated first — if your last conclusions are newer than the feed, nothing has changed.
2

Screen

Use the cookbook recipes to cut 120 rows down to a view: a category, a token-watch screen, a signal leaderboard.
3

Shortlist

Keep projects whose totals and signal shape match your thesis. Calibration for the current feed (range 74.75–83.75):
ThresholdMeaning today
total >= 80Top tier — 19 of 120 projects
total >= 78Strong — 44 of 120
noToken >= 8 && socialActivity >= 7Token-watch candidates with real audiences
tokenStatus == "Possible token/funding signal"Already moving — 3 of 120
4

Deep-research

The feed is a map, not the territory. For each shortlisted project, open website and x, check GitHub activity yourself, and verify the team ships. Several website values are empty strings — fall back to the x profile.
5

Log a decision

Close every evaluation with one of four verbs:
VerbMeaning
TRACKPromising; re-check on next dataset update.
CONTACTReach the team (partnership, investment, hiring).
BUY/WATCHToken-relevant; watch for launch or entry.
SKIPReasoned no. Log why, so future runs don’t repeat the work.
Suggested log line (one JSON object per evaluation, append-only):
{ "ts": "2026-06-12T00:00:00Z", "project": "Encrypt", "rank": 2, "total": 83.5, "decision": "TRACK", "reason": "FHE infra, no token, strong GitHub", "datasetUpdated": "2026-06-11" }
Pinning datasetUpdated makes decisions auditable against the feed version that produced them.
Scores measure opportunity signal — product, team, branding, timing. They do not measure safety, audit status, or legitimacy. PalatineAI is not financial advice; BUY/WATCH is a research verb, not a recommendation.