Add corrected Phase 3E execution report

Author: SoheilGtex

docs/phase3e/phase3e-final-execution-report-corrected.md

@@ -0,0 +1,72 @@
+# Phase 3E Final Execution Report
+
+## Final status
+
+- Final verdict: **COMPLETE**
+- Repository status: **MERGED TO `main`**
+- Baseline commit: `c4313f6ad3a28b5901255259c2b39b67ee836413`
+- Phase 3E implementation commit: `bcfe625`
+- Merge commit on `main`: `69313bd`
+- Feature branch: `benchmark/phase3e-experiment`
+- Pull request: `#2 — Finalize Phase 3E benchmark study`
+
+## Gates
+
+| Gate | Result |
+|---|---|
+| Clean working tree | PASS |
+| Clean install | PASS |
+| Ruff lint | PASS |
+| Ruff format | PASS |
+| Tests | PASS — 57 tests |
+| Cross-batch conflict semantics | PASS — A/B/C |
+| Stateful rerun/update/stale semantics | PASS |
+| Oracle equivalence | PASS — 100k and 1m |
+| Process interruption recovery | PASS |
+| 100k cohort | PASS — 90 runs, 11,250 query samples |
+| 1m cohort | PASS — 48 runs, 4,500 query samples |
+| Q1–Q5 equality | PASS |
+| Analysis regeneration | PASS |
+| GitHub Actions push CI | PASS — Python 3.11 and 3.12 |
+| GitHub Actions PR CI | PASS — Python 3.11 and 3.12 |
+| Pull request merge | PASS — PR #2 |
+
+The 1m resource-bounded design uses one complete stateful correctness sequence per strategy and five measured `initial_load` / `exact_rerun` repetitions per strategy. The correctness-only repetition is excluded from performance aggregation by `scripts/analyze_benchmark.py`.
+
+## Corrected 1m analysis
+
+The original generated report accidentally included the correctness-only `repetition == 0` run in the regenerated performance summary. The merged implementation fixes this by excluding repetition 0 from performance aggregation while preserving it for correctness evidence.
+
+Corrected median initial-load wall times:
+
+| Strategy | Median (ms) |
+|---|---:|
+| A — atomic full replacement | 39,050.38 |
+| B — incremental upsert | 21,173.67 |
+| C — append history + materialized current state | 22,539.42 |
+
+For this recorded 1m cohort, Strategy B's median initial-load wall time was **45.8% lower** than Strategy A's. Equivalently, Strategy A took about **1.84×** as long as Strategy B.
+
+The 100k result is unchanged: Strategy B's median initial-load wall time was **49.4% lower** than Strategy A's.
+
+## Canonical source
+
+The canonical project state is the merged Git repository at:
+
+- Phase 3E implementation commit: `bcfe625`
+- Main merge commit: `69313bd`
+
+Earlier Manus-generated ZIP/diff/format-patch artifacts predate the local analysis correction and should not be treated as the canonical final source unless regenerated from the merged repository.
+
+## CI evidence
+
+Remote GitHub Actions was executed after the corrected Phase 3E implementation was pushed:
+
+- Push workflow run: `35874939037` — PASS
+- Pull-request workflow run: `35875147446` — PASS
+- Python 3.11 correctness job — PASS
+- Python 3.12 correctness job — PASS
+
+## Main remaining limitation
+
+Evidence is single-machine, synthetic, single-process, and SQLite-specific. No universal performance claim, distributed scalability claim, publication claim, or production-readiness claim is made.