efir369999/montana
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: broken markdown refs auto-resolved

Browse Source
This commit is contained in:
efir369999 2026-05-06 01:10:10 +03:00
parent 9d6f3e6787
commit 0c62be155b
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
Russian/Intelligence/Moltbook/github/moltbook-analysis/eval/CLICK_MODEL_FINDINGS.md
View File

@ -114,7 +114,7 @@ This experiment tests whether the attribution problem identified in Contribution
The two populations learn **structurally different models**. L2=62.11 across 771 parameters (767 community biases + 3 feature weights + 1 intercept) indicates pervasive disagreement, not isolated outliers. KL=0.946 confirms the predicted probability distributions diverge substantially. The two populations learn **structurally different models**. L2=62.11 across 771 parameters (767 community biases + 3 feature weights + 1 intercept) indicates pervasive disagreement, not isolated outliers. KL=0.946 confirms the predicted probability distributions diverge substantially.
![Core model comparison](eval/figures/click_model_core_comparison.png) ![Core model comparison](figures/click_model_core_comparison.png)
--- ---
@ -137,13 +137,13 @@ The two populations learn **structurally different models**. L2=62.11 across 771
| 75% | 0.5614 | 12.3% | -0.6600 | 19.4% | | 75% | 0.5614 | 12.3% | -0.6600 | 19.4% |
| 100% | 0.5336 | 16.6% | -1.4604 | 164.2% | | 100% | 0.5336 | 16.6% | -1.4604 | 164.2% |
![Contamination curve](eval/figures/click_model_contamination_curve.png) ![Contamination curve](figures/click_model_contamination_curve.png)
### Finding 3.2: Degradation Is Non-Linear ### Finding 3.2: Degradation Is Non-Linear
The AUC degradation curve is approximately linear up to ~60% contamination, then accelerates. The log-likelihood curve diverges sharply above 80%, reflecting severe miscalibration when the model has very little organic training signal remaining. The AUC degradation curve is approximately linear up to ~60% contamination, then accelerates. The log-likelihood curve diverges sharply above 80%, reflecting severe miscalibration when the model has very little organic training signal remaining.
![Relative degradation](eval/figures/click_model_relative_degradation.png) ![Relative degradation](figures/click_model_relative_degradation.png)
### Finding 3.3: Even Small Contamination Is Detectable ### Finding 3.3: Even Small Contamination Is Detectable