diff --git "a/pages/conclusion/page.md" "b/pages/conclusion/page.md"
deleted file mode 100644--- "a/pages/conclusion/page.md"
+++ /dev/null
@@ -1,80 +0,0 @@
-# Conclusion
-
-
----
-
-This preliminary 16×16 three-seed sensitivity study is retained for audit history only. Its result changed from 4/4 to 2/4 DB-favoring cells after deterministic math attention replaced the warn-only backend. The later pinned executive summary and CelebA-HQ 256 experiment supersede this cell as the primary evidence.
-
-
----
-
-````html
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-````
-
-
----
-
-The narrow constrained-cost theorem reproduces, but the empirical paper-scale claims remain inconclusive. In the primary 256x256 CelebA-HQ early-budget experiment, condition-matched DB is worse than FM on both primary metrics: pooled masked MSE DB-minus-FM = 0.152086 [0.123266, 0.172604], LPIPS = 0.186395 [0.137586, 0.240532]. This is consistent with the paper's own early-convergence table; it is not evidence against the claimed post-convergence reversal.
-
-## Scope & cost
-
-| | This reproduction | Full replication |
-|---|---|---|
-| Scope | CelebA-HQ 256 inpainting; Box64/128; n=500/1000; 3 seeds; 256 held-out; theorem and code audits | Multiple restoration/translation tasks; up to 27k train; 3k restoration test; 485M parameters and 600k updates |
-| Hardware | Three sequential single-T4 HF Jobs; 0.635M-parameter matched Transformers | Single H20 reported for the main experiments |
-| Compute time | 3,078 total Job-running seconds; 800 updates per method | Full 600,000-update multi-dataset training; not attempted |
-| Cost | $0.343710 estimated paid HF compute; no provider calls | Not reported by the paper and not estimated here |
-| Outcome | theorem verified narrowly; real early-budget result contrary; Claims 3--6 paper-scale inconclusive | reported post-convergence DB perceptual advantage |
-
-The corrected toy run is retained as an attention-backend sensitivity finding: its result changed from 4/4 to 2/4 DB-favoring cells when nondeterministic attention was disabled. This submission is grouped in the public [ICML 2026 Agent Reproductions collection](https://huggingface.co/collections/ProCreations/icml-2026-agent-reproductions-6a5846bea95be467bae66176).
-
-
----
-
-**📦 Artifact** `dbfm-repro/repro-bundle:v1` · dataset
-
-https://huggingface.co/buckets/ProCreations/diffusion-bridge-flow-matching-repro-artifacts#dbfm-repro/repro-bundle:v1
-
-
----
-
-````html
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-