• 

  Mockup for Bangermeter

• 

  Mockup for Bangermeter

• 

  The Extension in Action

Bangermeter ⚡

Score any tweet with X's own open-sourced ranking fundamentals.

A Chrome extension that scores tweets on x.com using the exact weighted-sum formula from twitter/the-algorithm (NaviModelScorer) with the last officially published weight set — validated against the Phoenix-era xai-org/x-algorithm release and everything else learned since January 2026. Full source-traced findings: RESEARCH.md.

Install

1. Open chrome://extensions
2. Enable Developer mode (top right)
3. Click Load unpacked → select the extension/ folder
4. Browse x.com — every timeline tweet gets a ⚡ badge; click it for the full breakdown. The compose box gets a live draft meter as you type.

What the scores mean

Every tweet gets up to two 0–100 scores (50 = median baseline tweet):

• C — Content score (prospective). Estimates how the pre-Phoenix pipeline would value this tweet's content: baseline engagement probabilities per head, adjusted by detected content signals (question, thread, video, bare link, hashtag spam, engagement bait), run through the exact repo formula.
• E — Engagement score (retrospective). Applies the published weights to the tweet's actual engagement rates (likes/reposts/replies ÷ views): "how the algorithm's weighted sum values what this tweet earned." Requires a visible view count.

Click the badge for the breakdown: per-head contributions (weight × P), detected signals with direction and provenance, rescoring factors (reply ×0.75, optional out-of-network ×0.75), unweighted signals (e.g. bookmarks), and age-decay context.

Methodology and honesty

• Formula: exact NaviModelScorer.computeWeightedModelScore math — weighted sum over engagement heads, ε = 0.001, negative-sum squashing, good-click v1/v2 max-combine — plus the HeuristicScorer rescoring chain. This score skeleton is confirmed to survive in X's 2026 production code (xai-org/x-algorithm, weighted_scorer.rs).
• Weights: the published 2023 set (fav 0.5, retweet 1.0, reply 13.5, profile-click 12, conversation-click 11/10, video-50% 0.005, author-engages-replier 75, negative feedback −74, report −369) — confirmed by 2026 research as the only sourced values ever released. Specifically the April 5, 2023 snapshot (commit b85210863f; the original March 31 table had reply=27 before X retuned it within five days — weights are "periodically adjusted" by X's own admission). Heads that never had a published value (bookmark, share, dwell, …) are excluded, not guessed, and shown as sourced signals.
• Two archival-sourced factors (leak-hunt research, Aug 2026): an OPT-IN toggle applies the 2023-era ×4 in-network / ×2 out-of-network multiplier recovered from the archived initial-release commit (ec83d01dca, removed from X's code Sept 2025 — off by default since it floors every verified author near the top of the scale), and posts with a displayed Community Note get ×0.5 on the prospective score (sourced range 0.4–0.55 from X's own A/B test, Nature Communications, and PNAS causal studies).
• No folklore numbers. "Links −30–50%", "3+ hashtags −40%", "bookmark 20×", "retweet 20×" all fail source-tracing. Link/hashtag signals appear only as mild, labeled, directional estimator adjustments.
• Two separate layers. The weight layer (published values) never mixes with the estimator layer (how we approximate P(engagement) from what a browser can see). Every number in the UI is tagged with its provenance.
• Relative score, not predicted reach. The repo's heartbeat-optimizer code reveals production weights were per-user-bucket and time-varying; any static set is one point in a moving distribution. Since ~Nov 2025 production ranking is Phoenix (Grok transformer), so this is a historical-fundamentals lens, not a reach predictor.

Known limitations

• "Replying to" detection and count parsing assume an English X locale.
• The E score needs a visible view count (hidden on some surfaces).
• Viewer-specific factors (in-network status, feedback fatigue, Control AI) can't be observed; out-of-network ×0.75 is available as a popup toggle instead.
• X changes its DOM without notice; selectors have documented fallbacks but may need updating.

Project files

The scoring formulas were extracted from the September 2025 re-release of twitter/the-algorithm (commit c54bec0d — NaviModelScorer.scala, PredictedScoreFeature.scala, HeuristicScorer.scala, RescoringFactorProvider.scala, and the earlybird AgeDecay sigmoid), cross-checked against the current production structure in xai-org/x-algorithm (weighted_scorer.rs, oon_scorer.rs, author_diversity_scorer.rs).

Verification status

• Engine math: 33/33 self-tests pass (test.html), including small-sample shrinkage regressions and the verified-author / Community-Note factors.
• Content script (badges, panel, meter, virtualized-list handling, quote-tweet scoping): verified against the DOM fixture (fixture.html).
• Live x.com: verified in a logged-in session (Aug 2026). X changes its DOM without notice; selectors target article[data-testid="tweet"], the action-bar role="group" aria-label, and tweetTextarea_* with per-button fallbacks.

Built With

• css
• html
• javascript
• powershell