tweet-price-charts

Correlates $THREE token price action with three.ws’s X (Twitter) posts, and renders an interactive TradingView-style chart with every post plotted as an avatar bubble on the candles — inspired by rohunvora/tweet-price-charts.

What’s here

generate.mjs        # the analyzer + chart generator (Node, zero deps)
data/               # scraped tweets (trythreews + nichxbt), input
assets/             # real X avatars, embedded into the chart as base64
out/
  chart.html        # ← open this. self-contained interactive chart
  chart.json        # correlation stats (original posts only)
  chart.csv         # per-post detail: 1h/4h/24h returns
  *-with-replies.*  # same, but including replies/retweets

The chart (out/chart.html)

Self-contained — just open it in a browser (loads TradingView lightweight-charts from CDN; all data + avatars are baked in).

• $THREE/SOL candles with 15m / 1h / 1d toggle
• Every post as an avatar bubble on the candle it landed on (cyan border = @trythreews, magenta = @nichxbt, ▲ = announcement)
• Nearby posts cluster with a count badge so they don’t overlap; clusters split as you zoom
• Hover a bubble → tweet text, views, and the actual +1h / +4h / +24h price return
• Click → opens the original tweet
• Right sidebar lists every post by 24h return; click to jump the chart there

How returns are measured

For each post at time T, the entry price is the $THREE hourly candle close at-or-just-before T. Then +Nh = (close N hours after T − entry) / entry × 100. Price history is hourly (and 15-minute for the chart) OHLCV from GeckoTerminal for the $THREE/SOL pump.swap pool.

Key findings (original posts only, Apr 29 – Jun 15 2026, n=405)

Using median returns (a memecoin’s mean is dominated by fat-tail outliers):

Window	Post median	Random-hour baseline	Edge	Win rate
+1h	−1.25%	−0.59%	−0.66pp	46%
+4h	−1.30%	−1.61%	+0.32pp	45%
+24h	+8.18%	+1.87%	+6.31pp	59%

• No instant pump — the median post is flat-to-down at 1h; the large mean is a few outliers.
• Real 24h signal — posts precede +8% vs +1.9% baseline, up 59% of the time.
• Announcements drive it — announcement posts: +35% avg / 24h vs +19% for routine posts.
• Reach > frequency — views↔24h return r≈0.15; posts-per-day↔daily return r≈0.04 (none).

Caveats: overlapping 24h windows inflate significance; reverse causality (we post more around events) can’t be ruled out; single token in an uptrend regime. Suggestive, not causal.

Regenerate

node generate.mjs --accounts trythreews,nichxbt --chart \
  data/trythreews_tweets_2026-06-15.json \
  "data/nichxbt_tweets_2026-06-15.json" \
  "data/nichxbt_tweets_2026-06-15 (1).json" \
  "data/nichxbt_tweets_2026-06-15 (2).json" \
  "data/i_tweets_2026-06-15.json" \
  "data/i_tweets_2026-06-15 (1).json" \
  --out out/chart

Flags: --accounts a,b restrict to owned handles · --own original posts only (drop replies/RTs) · --chart also emit the HTML · --out <base> output path prefix.

Price data: GeckoTerminal (free, no key). No mocks, no synthetic data.