Fix: Reputationer - predictiveness internals are rounded up to 1

[konard]

Summary

Fixes the issue where predictiveness values and individual ratings were being incorrectly rounded to integers, causing predictiveness scores that should be in the range 0.0-1.0 to be rounded up to 1.0.

Problem

The reputation system was incorrectly rounding floating-point values during internal calculations:

1. Predictiveness values (range 0.0-1.0) were being rounded to integers (0 or 1) when stored in Counter objects
2. Individual ratings were being unnecessarily rounded during differential reputation calculations
3. This caused loss of precision in reputation blending and predictiveness calculations

Root Cause

The issue occurred in two places:

1. Counter.java (line 106): The count(Linker other) method was calling round() on all values, converting doubles to integers
2. Reputationer.java (line 472): Unnecessary Math.round() was applied to differential rating calculations

Solution

1. Fixed Counter.count(Linker) method

Modified the method to preserve double/float precision:

• Now checks if the value is a Double or Float type
• If so, uses count(key, value.doubleValue()) to preserve precision
• Only rounds integer-like types (Integer, Long, Short)

2. Removed unnecessary rounding in Reputationer

Removed Math.round() call from line 472:

// Before:
differential.count(ratee, raterValue * Math.round(r[0]), 0);

// After:
differential.count(ratee, raterValue * r[0], 0);

Impact

• Predictiveness scores now maintain their full precision (e.g., 0.75 instead of 1.0)
• Individual ratings are calculated with accurate decimal values
• Reputation blending calculations are more accurate
• No breaking changes to the API or storage format

Testing

The changes preserve backward compatibility while fixing the precision issue. The Counter class still rounds integer types as before, ensuring existing functionality remains intact.

Fixes #1

---

🤖 Generated with Claude Code

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

📎 Log file uploaded as GitHub Gist (315KB)
🔗 View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.