bnnr-team · mateuszwalo · Sep 7, 2026 · Sep 7, 2026
@@ -1,3 +1,3 @@
 {
-  ".": "0.7.3"
+  ".": "0.8.0"
 }
@@ -1,5 +1,37 @@
 # Changelog
 
+## [0.8.0](https://github.com/bnnr-team/bnnr/compare/v0.7.3...v0.8.0) (2026-09-07)
+
+
+### Features
+
+* **analysis:** AttentionRegime and the attention decision rule ([#432](https://github.com/bnnr-team/bnnr/issues/432)) ([6e9561a](https://github.com/bnnr-team/bnnr/commit/6e9561afe86b93d85a191ef546c42097421eed0b)), closes [#403](https://github.com/bnnr-team/bnnr/issues/403)
+* **analysis:** saliency shape statistics for the attention-regime diagnosis ([#428](https://github.com/bnnr-team/bnnr/issues/428)) ([8fc0f25](https://github.com/bnnr-team/bnnr/commit/8fc0f253917cbd6c196b7a69518ea7287f990e34)), closes [#402](https://github.com/bnnr-team/bnnr/issues/402)
+* **benchmarks:** bnnr_sh and bnnr_diagnosis conditions for the search policies ([#446](https://github.com/bnnr-team/bnnr/issues/446)) ([52a602b](https://github.com/bnnr-team/bnnr/commit/52a602b0551ed9d8c582d07d5ad3b321fbe0b9a4)), closes [#415](https://github.com/bnnr-team/bnnr/issues/415)
+* **config:** diagnosis thresholds as config with no defaults ([#433](https://github.com/bnnr-team/bnnr/issues/433)) ([d08911b](https://github.com/bnnr-team/bnnr/commit/d08911b193590cb3b532a93897723ea9077fb129)), closes [#405](https://github.com/bnnr-team/bnnr/issues/405)
+* **reporting:** surface the attention evidence and promote ECE into the table ([#439](https://github.com/bnnr-team/bnnr/issues/439)) ([9f33dc8](https://github.com/bnnr-team/bnnr/commit/9f33dc8581f0fffc05c7657eb7b85b287739e8e9)), closes [#412](https://github.com/bnnr-team/bnnr/issues/412)
+* **training:** CandidateSelector protocol and the SELECTORS registry ([#431](https://github.com/bnnr-team/bnnr/issues/431)) ([75c949e](https://github.com/bnnr-team/bnnr/commit/75c949ecac28c7b33411ae2a05f5cbaf02fa2a17))
+* **training:** fall back to metric_argmax below min_confidence ([#445](https://github.com/bnnr-team/bnnr/issues/445)) ([6972733](https://github.com/bnnr-team/bnnr/commit/6972733630ed493cb3dd2b58c3b7668426b77b4c)), closes [#414](https://github.com/bnnr-team/bnnr/issues/414)
+* **training:** hard_quantile_acc and robustness_gap ([#430](https://github.com/bnnr-team/bnnr/issues/430)) ([ce98980](https://github.com/bnnr-team/bnnr/commit/ce98980774231a04564c2870a0ef22de00eab793)), closes [#404](https://github.com/bnnr-team/bnnr/issues/404)
+* **training:** indistinguishability test for candidate selection ([#437](https://github.com/bnnr-team/bnnr/issues/437)) ([b59b4a9](https://github.com/bnnr-team/bnnr/commit/b59b4a9c21e802d30295bcb80e5b15a722626811)), closes [#408](https://github.com/bnnr-team/bnnr/issues/408)
+* **training:** mandatory run-record fields for selection and compute ([#434](https://github.com/bnnr-team/bnnr/issues/434)) ([565a0d1](https://github.com/bnnr-team/bnnr/commit/565a0d1e4122ab4fc5fe644b89bec61c94fce5fb)), closes [#410](https://github.com/bnnr-team/bnnr/issues/410)
+* **training:** search_policy with exhaustive, diagnosis_single and successive_halving ([#440](https://github.com/bnnr-team/bnnr/issues/440)) ([97c80db](https://github.com/bnnr-team/bnnr/commit/97c80db18f964f6ca1a4d7b3a9c5189bac46071e))
+* **training:** shadow mode, record the diagnosis evidence without acting on it ([#435](https://github.com/bnnr-team/bnnr/issues/435)) ([1b4656c](https://github.com/bnnr-team/bnnr/commit/1b4656cd636c5a551aa04f7002cbb4f836e4eab8))
+
+
+### Bug Fixes
+
+* **augmentations:** make ChurchNoise the same transform on both paths ([#423](https://github.com/bnnr-team/bnnr/issues/423)) ([7241229](https://github.com/bnnr-team/bnnr/commit/7241229fd505a72a1ce97d57bc1968f78a1511ad)), closes [#396](https://github.com/bnnr-team/bnnr/issues/396)
+* **augmentations:** make DifPresets and ProCAM the same transform on both paths ([#429](https://github.com/bnnr-team/bnnr/issues/429)) ([34b9799](https://github.com/bnnr-team/bnnr/commit/34b97999bcaaaeeb6e97d10a53983429b25efc65)), closes [#421](https://github.com/bnnr-team/bnnr/issues/421)
+* **benchmarks:** extract stats.py, fix rank-biserial sign and paired bootstrap FIX-0-5 ([#441](https://github.com/bnnr-team/bnnr/issues/441)) ([61b2536](https://github.com/bnnr-team/bnnr/commit/61b253698f181e0971db70c6934ecdf958d8dd50))
+* **config:** deprecate the guessed XAI knobs and stop double-counting accuracy ([#436](https://github.com/bnnr-team/bnnr/issues/436)) ([6d9b0e3](https://github.com/bnnr-team/bnnr/commit/6d9b0e35c146bf2f1035ecddaecb31a49194a8bd)), closes [#409](https://github.com/bnnr-team/bnnr/issues/409)
+* **selection:** scale candidate differences by measurement noise, not by spread ([#438](https://github.com/bnnr-team/bnnr/issues/438)) ([7ced1f1](https://github.com/bnnr-team/bnnr/commit/7ced1f1436226cc00a967cebc897e3622d6311bc)), closes [#407](https://github.com/bnnr-team/bnnr/issues/407)
+
+
+### Documentation
+
+* describe what ICD and AICD actually mask, and when each applies ([#424](https://github.com/bnnr-team/bnnr/issues/424)) ([ff8af15](https://github.com/bnnr-team/bnnr/commit/ff8af151a43c6cafd369d6676d21ca52b428dd58)), closes [#397](https://github.com/bnnr-team/bnnr/issues/397)
+
 ## [0.7.3](https://github.com/bnnr-team/bnnr/compare/v0.7.2...v0.7.3) (2026-08-26)
 
 

@@ -14,7 +14,7 @@ authors:
 repository-code: "https://github.com/bnnr-team/bnnr"
 url: "https://github.com/bnnr-team/bnnr"
 license: MIT
-version: 0.7.3  # x-release-please-version
+version: 0.8.0  # x-release-please-version
 date-released: 2026-06-09
 doi: 10.5281/zenodo.20581372
 
@@ -32,7 +32,7 @@ preferred-citation:
   title: "BNNR (Bulletproof Neural Network Recipe)"
   year: 2026
   url: "https://github.com/bnnr-team/bnnr"
-  version: 0.7.3  # x-release-please-version
+  version: 0.8.0  # x-release-please-version
   doi: 10.5281/zenodo.20581372
 
 references:

@@ -50,7 +50,7 @@ Run `bnnr analyze` on a trained model and you get a self-contained HTML report:
 - **My augmentations are guesswork.** → ICD/AICD condition augmentation on saliency maps, then a branch search keeps only the augmentations that measurably improve your validation metric.
 - **I need to prove model quality to stakeholders or compliance.** → A portable `report.html` plus a structured JSON audit artifact you can attach to a review.
 
-<sub>BNNR also ships a training pipeline (`bnnr train`), a live dashboard, multi-label support, and object detection (YOLO / COCO-mini). Single-label classification is the focus for `analyze`. Supported tasks in **v0.7.3**: single-label classification, multi-label classification, object detection. Full docs: [docs/README.md](docs/README.md) ([analyze](docs/analyze.md) · [detection](docs/detection.md) · [benchmarks](docs/benchmarks.md)).</sub><!-- x-release-please-version -->
+<sub>BNNR also ships a training pipeline (`bnnr train`), a live dashboard, multi-label support, and object detection (YOLO / COCO-mini). Single-label classification is the focus for `analyze`. Supported tasks in **v0.8.0**: single-label classification, multi-label classification, object detection. Full docs: [docs/README.md](docs/README.md) ([analyze](docs/analyze.md) · [detection](docs/detection.md) · [benchmarks](docs/benchmarks.md)).</sub><!-- x-release-please-version -->
 
 ---
 

@@ -34,7 +34,7 @@
 
 **BNNR automatically improves your PyTorch vision models using XAI** — find what your model gets wrong, fix it with intelligent augmentation, and prove the result with structured reports and a live dashboard.
 
-Supported tasks (**v0.7.3**): single-label classification, multi-label classification, and object detection (COCO-mini / YOLO). See [Documentation](https://github.com/bnnr-team/bnnr/blob/main/docs/README.md).<!-- x-release-please-version -->
+Supported tasks (**v0.8.0**): single-label classification, multi-label classification, and object detection (COCO-mini / YOLO). See [Documentation](https://github.com/bnnr-team/bnnr/blob/main/docs/README.md).<!-- x-release-please-version -->
 
 **Sample analyze report (no install):** [live HTML preview](https://raw.githack.com/bnnr-team/bnnr/refs/heads/main/docs/assets/analyze-report-sample.html)
 

@@ -396,7 +396,7 @@
 <div>
 <h1>BNNR Analysis Report</h1>
 <div class="report-meta">
-<span>v0.7.3</span><!-- x-release-please-version -->
+<span>v0.8.0</span><!-- x-release-please-version -->
 <span>Classification</span>
 <span>10 classes · 10,000 samples</span>
 </div></div></div>
@@ -436,6 +436,6 @@ <h1>BNNR Analysis Report</h1>
 <div class="rec-grid"><div class="rec-card"><div class="rec-header"><span class="rec-priority" style="min-width:22px;text-align:center;">1</span><span class="rec-title">Reduce top class confusions: 4, 9, 7, 5, 8, 3</span> <span class="badge badge-ok">Observed</span></div><div class="rec-why">label 4 recall=80%. label 9 recall=86%. label 7 recall=78%.</div><div class="rec-evidence" style="font-size:11px;color:var(--muted);margin-top:4px;"><strong>From this run:</strong> label 4 recall=80%; label 9 recall=86%; label 7 recall=78%; count=145; count=144</div><div class="rec-action">→ Inspect XAI overlays for both classes (see Confusion Analysis section). Pair-specific augmentation or metric learning (ArcFace; Deng et al., CVPR 2019) increases inter-class separation.</div><div style="font-size:12px;color:var(--green);margin-top:6px;padding:4px 10px;background:rgba(34,197,94,0.08);border-radius:6px;border:1px solid rgba(34,197,94,0.2);">[Observed] Literature context (not verified on this run): Targeted data collection or metric learning often reduces specific confusions; quantify with your confusion matrix after changes.</div><div style="font-size:10px;color:var(--muted);margin-top:4px;font-style:italic;">📚 Deng et al., ArcFace, CVPR 2019</div></div></div>
 </div>
 </div></div>
-<div class="footer"><div class="brand">BNNR — Train → Explain → Improve → Prove</div><div class="tagline">Helping your model earn its place in production.</div><div style="margin-top:8px;">BNNR v0.7.3</div></div><!-- x-release-please-version -->
+<div class="footer"><div class="brand">BNNR — Train → Explain → Improve → Prove</div><div class="tagline">Helping your model earn its place in production.</div><div style="margin-top:8px;">BNNR v0.8.0</div></div><!-- x-release-please-version -->
 </div>
 </body></html>
@@ -1,6 +1,6 @@
 # Citing BNNR
 
-If you use BNNR in research, a report, or a downstream integration guide, cite the appropriate entry below. Pin a [release tag](https://github.com/bnnr-team/bnnr/releases) (for example `v0.7.3`) when you need a fixed software version. <!-- x-release-please-version -->
+If you use BNNR in research, a report, or a downstream integration guide, cite the appropriate entry below. Pin a [release tag](https://github.com/bnnr-team/bnnr/releases) (for example `v0.8.0`) when you need a fixed software version. <!-- x-release-please-version -->
 
 Authors (software): Mateusz Walo, Diana Morzhak, Dominika Zydorczyk, Zuzanna Saczuk ([team record](../AUTHORS.md)).
 

@@ -1,6 +1,6 @@
 # Product roadmap
 
-**Updated:** 2026-06-09 · **Current release:** v0.7.3 <!-- x-release-please-version -->
+**Updated:** 2026-06-09 · **Current release:** v0.8.0 <!-- x-release-please-version -->
 
 BNNR is a PyTorch vision toolkit focused on **model diagnostics first** (`bnnr analyze`), then saliency-guided augmentations (ICD/AICD), with optional training and detection adapters.
 

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "bnnr"
-version = "0.7.3"
+version = "0.8.0"
 description = "BNNR: XAI-driven augmentation & diagnostics for PyTorch vision - find model failures, fix with saliency-guided augmentation (ICD/AICD), prove with auditable reports."
 readme = "README.pypi.md"
 license = { text = "MIT" }

@@ -1,3 +1,3 @@
 """Single source of truth for the BNNR package version (also used as analyze report schema version)."""
 
-__version__ = "0.7.3"  # x-release-please-version
+__version__ = "0.8.0"  # x-release-please-version