Skip to content

Conversation

infinitumsimias
Copy link

@infinitumsimias infinitumsimias commented Sep 19, 2025

Changes

  • Introduced maxDailyNegDelta, normalCap, and attackCap constants.
  • Added an entropy/diversity guard: if fewer than 50% of issuers are unique, classify the event as a coordinated brigade.
  • Implemented dynamic clamping: normal cap = –100, brigade cap = –20.
  • Included telemetry stubs for brigade detection and clamp activation.

Impact

  • Prevents a small group of accounts from collapsing a user’s Tweepcred in a single batch run.
  • Protects small and new creators from brigading while still applying legitimate, organic penalties.
  • Preserves organic feedback fidelity (diverse, independent block signals continue to count).
  • Provides clear telemetry for monitoring and future adjustments.

Notes

  • Backward-compatible: retains the original one-line clamp.
  • New constants are scoped private with safe defaults.
  • This patch only affects Tweepcred batch reputation adjustment; no changes to visibilitylib or ranking layers.

…ades.

### Summary
This patch hardens Tweepcred against coordinated blocklist brigades.

### Changes
- Added `maxDailyNegDelta`, `normalCap`, and `attackCap` constants.
- Added entropy/diversity guard: if <50% of issuers are unique, treat as a coordinated brigade.
- Dynamic clamping: normal cap = -100, brigade cap = -20.
- Added telemetry stubs: counters for brigade detection and clamp activations.

### Impact
- Prevents a small group of accounts from collapsing a user’s Tweepcred in one batch run.
- Protects small/new creators from coordinated brigading while still applying organic penalties.
- Maintains organic feedback fidelity (diverse, independent blocks still count).
- Provides clear telemetry for monitoring and future tuning.

### Notes
- Backward-compatible: legacy one-line clamp retained.
- All new constants are scoped private, safe defaults.
- No changes to visibilitylib or ranking layers in this patch; this only touches Tweepcred batch reputation adjustment.
@CLAassistant
Copy link

CLAassistant commented Sep 19, 2025

CLA assistant check
All committers have signed the CLA.

@kehlsteinhaus
Copy link

Did you really use ChatGPT to make this pull request?

@infinitumsimias
Copy link
Author

Did you really use ChatGPT to make this pull request?

My original idea with a ChatGPT-5 + Grok-4 collab, lots of iterations until I was happy with it.

Looking into taking it further with retroactive reach bumps and tracking.

What do you think?

@infinitumsimias infinitumsimias changed the title Twepcred: Clamp + Entropy Guard to prevent coordinated lists & brigades Tweepcred: Clamp + Entropy Guard to prevent coordinated lists & brigades Sep 22, 2025
Adjusted reputation adjustment logic

- Fix penalty direction: use (adjusted - baseMass) instead of (mass - adjusted) so brigades are penalized.
- Replace log(log(numFollowings)) with log(log1p(numFollowings)) to avoid invalid math for small values.
- Add input sanitization: guard against NaN/Infinity and negative follower/following counts; clamp mass to non-negative before scaling.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants