removed extra docs

Author: sjarmak

docs/codescalebench_blog_v1.md

@@ -109,15 +109,17 @@ Curated GT deltas (`MCP - baseline`, combined):
 
 ### 2) Gains persist across size bins, with strongest lift in 1M-5M proxy bucket
 
-Curated GT deltas (`MCP - baseline`):
-- `<1M`: F1@10 +0.1007, Total +0.1318
-- `1M-5M`: F1@10 +0.2680, Total +0.2392
-- `5M-20M`: F1@10 +0.0648, Total +0.0565
-- `>20M`: F1@10 +0.1247, Total +0.1075
+Curated GT deltas (`MCP - baseline`) by revised LOC size bands:
+- `<400K` (n=15): F1@10 +0.2503, Total +0.2780
+- `400K-2M` (n=31): F1@10 +0.2618, Total +0.2424
+- `2M-8M` (n=143): F1@10 +0.1796, Total +0.1622
+- `8M-40M` (n=74): F1@10 +0.0719, Total +0.0590
+- `>40M` (n=3): F1@10 +0.0242, Total +0.0667
+- `unknown` (n=63): F1@10 +0.0992, Total +0.1601
 
 Interpretation: retrieval lift is not uniform, but MCP shows clear upside where task context is more distributed and retrieval-heavy.
 
-Method note: I corrected an Org path-normalization bug in an earlier draft where some baseline paths were mismatched due to path shape differences (for example `repo/repo/path` vs `repo/path`).
+Method note: I corrected an Org path-normalization bug in an earlier draft where some baseline paths were mismatched due to path shape differences (for example `repo/repo/path` vs `repo/path`). I also replaced SDLC size proxies with non-proxy repository size mapping for the size-bin slice in this version.
 
 ## Cost and Speed
 

docs/ops/SCRIPT_INDEX.md

@@ -209,6 +209,7 @@ Generated from `scripts/registry.json` by `scripts/generate_script_index.py`.
 - `scripts/plot_build_diary.py` - Utility script for plot build diary.
 - `scripts/plot_build_diary_supplementary.py` - Utility script for plot build diary supplementary.
 - `scripts/plot_build_narrative.py` - Utility script for plot build narrative.
+- `scripts/plot_csb_mcp_blog_figures.py` - Utility script for plot csb mcp blog figures.
 - `scripts/prepare_analysis_runs.py` - Utility script for prepare analysis runs.
 - `scripts/promote_agent_oracles.py` - Utility script for promote agent oracles.
 - `scripts/push_base_images_ghcr.sh` - Utility script for push base images ghcr.

docs/technical_reports/TECHNICAL_REPORT_V2.md

@@ -992,18 +992,20 @@ On curated ground truth (`MCP - Baseline`, combined):
 | single_repo | 158 | +0.0853 | +0.1119 |
 | multi_repo | 171 | +0.2089 | +0.1862 |
 
-Curated size-bin deltas (`MCP - Baseline`):
+Curated size-bin deltas (`MCP - Baseline`), using the revised LOC bands shared with prior repo-size analysis:
 
-| Size Bin (proxy) | n | Δ F1@10 | Δ Total File Recall |
+| Size Bin | n | Δ F1@10 | Δ Total File Recall |
 |------------------|---:|--------:|--------------------:|
-| <1M | 139 | +0.1007 | +0.1318 |
-| 1M-5M | 104 | +0.2680 | +0.2392 |
-| 5M-20M | 57 | +0.0648 | +0.0565 |
-| >20M | 29 | +0.1247 | +0.1075 |
+| <400K | 15 | +0.2503 | +0.2780 |
+| 400K-2M | 31 | +0.2618 | +0.2424 |
+| 2M-8M | 143 | +0.1796 | +0.1622 |
+| 8M-40M | 74 | +0.0719 | +0.0590 |
+| >40M | 3 | +0.0242 | +0.0667 |
+| unknown | 63 | +0.0992 | +0.1601 |
 
 These slices indicate MCP retrieval gains are larger on multi-repo tasks than single-repo tasks in this snapshot.
 
-Methodology note: size bins here are metadata-driven proxies (`repo_set` fixture LOC totals for Org; `context_length` proxy for SDLC with fallback), so they should be interpreted as directional rather than exact physical repository size measurements.
+Methodology note: size bins here are no longer `context_length` proxies. Org tasks use fixture `loc_estimate` totals; SDLC tasks use repository size from GitHub metadata mapped into the same LOC bands (`docs/analysis/repo_size_bins_revised_20260303.json` conventions). `unknown` indicates tasks without resolved size metadata in this pass.
 
 ### 11.6 Correlation Analysis