Step 2: Check and select posts
Purpose
Remove posts that fail the project's rules. Select a small, attributed sample.
Before you start
Step 1 must have saved a raw file with a time window. Read crypto-layer-2-reference for topic boundaries.
Input
data/raw/YYYY-MM-DD.json. Steps 1 and 2 run in the same collector program.
Procedure
- Reject posts with no handle or no timestamp.
- Reject timestamps outside the saved 24-hour window.
- Apply the price, promotion, giveaway, rumor and government-DOGE filters.
- Reject accounts below 500 followers and posts below 5 likes.
- Reject disallowed coin tickers and promotion terms in the account bio.
- Rank surviving posts by the engagement score.
- Select up to three posts per coin, with one post per account.
- Remove repeated selected claim text within each coin.
- Trim extra posts to fit eight middle cards, including quiet cards.
- Save
data/YYYY-MM-DD.jsonwith the selected items and rejection counts.
The score is likes + 2 ร reposts + 2 ร quotes + replies. It is a selection rule, not a truth test. A guest-chain card needs three distinct authors and at least 25% of the surviving posts for that coin.
Check and output
Each selected item has a coin, topic, quoted excerpt, handle, post URL and timestamp. Check the topic and source links before approving a sample. Continue to 03-make-video.
If the step fails
Stop if the saved window or required source fields are absent. If no posts pass for a coin, use a quiet card. This means no sampled posts passed, not that no news exists. Keyword matching can select unrelated content. The historical Lightning Lake card demonstrates this limit.