chore: release v1.1.0 — CHANGELOG trimmed, pre-1.0.0 history removed

- [1.1.0] released 2026-05-19: multi-GPU support, ROCm/XPU Dockerfiles,
  --device flag, REQ/TEST-OEA-023, scaffold fix, doc cross-check
- Pre-v1.0.0 history removed from CHANGELOG (kept in git history)
- Reference links updated: [Unreleased] now compares v1.1.0...HEAD

Co-Authored-By: Oz <oz-agent@warp.dev>

Author: tbitcs

CHANGELOG.md

@@ -7,6 +7,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.1.0] - 2026-05-19
+
 ### Added
 - `Dockerfile.cuda`: NVIDIA CUDA 12.1 GPU image (verified on RTX 4070 SUPER)
 - `Dockerfile.rocm`: AMD ROCm 6.x GPU image (community-tested; `rocm/dev-ubuntu-22.04:6.3` base)
@@ -36,9 +38,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - `docs/REQUIREMENTS.md`: REQ-OEA-020 updated to reference `Dockerfile.cuda` alongside
   `Dockerfile`
 - `docs/TESTS.md`: TEST-OEA-020 updated to reference `Dockerfile.cuda`
-- `scaffold.yml`: pinned `detected_type: aee-research` to suppress specsmith audit false-positive
-  (scanner infers `research-python` from file heuristics; `aee-research` is the intentional
-  governance type set at project bootstrap)
+- `AGENTS.md`: spec version updated `0.10.1` → `0.11.3.dev427`; type updated to `research-python`
 
 ## [1.0.0] - 2026-05-14
 
@@ -76,156 +76,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Markdown lint: MD010/MD040/MD060 violations resolved
 - Dependabot: all 5 GitHub Actions PRs merged (checkout v6, setup-python v6, etc.)
 
-## [0.4.0] - 2026-05-13
-
-### Added
-- `experiments/data/scientific_corpus.txt` — 50-sentence scientific/natural-philosophy public
-  domain corpus (Newton, Feynman, Sagan); second independent domain for credibility suite
-- Two-model real LLM validation: `--model` CLI arg to `real_lm_experiment.py`;
-  `results/real_lm/distilgpt2/` and `results/real_lm/gpt2/` committed
-  - distilgpt2 (82M): `oea_anchored` log-prob +1.14 nats; `oea_miscalibrated` -0.82 nats
-  - gpt2 (124M): `oea_anchored` log-prob +1.61 nats; `oea_miscalibrated` -0.80 nats
-  - Causal mechanism model-size independent; effect strengthens with capacity
-- 5 new verified citations: drayson2025detection (EMNLP 2025), zhu2025synthesize (ICML 2025),
-  kovac2025recursive, keisha2025knowledge (NeurIPS 2025 workshop), abbasiyadkori2024believe
-- Differentiation paragraphs in §2.1 (OEA vs. Drayson/Zhu) and §2.3 (Abbasi Yadkori)
-- UNK-002 resolved in UNCERTAINTY-MAP.md; stable two-corpus setup documented
-
-### Changed
-- `experiments/data/public_domain_corpus.txt`: expanded from 18 lines to 62 lines spanning
-  Carroll, Austen, Melville, Hume, Darwin (~1600 words; 5 domains)
-- `experiments/credibility_suite.py`: corpus plan v2 uses `[public_domain_snippets,
-  scientific_snippets]` — removes `arxiv/main.tex` self-reference (UNK-002 fix)
-- `experiments/config/credibility_plan.json`: study_name → oea_credibility_suite_v2;
-  corpora → `[public_domain_snippets, scientific_snippets]`
-- `real_lm_experiment.py`: N_SEEDS 5→10, N_ITERATIONS 5→10; results dir per model
-- Table 2: refreshed with 2-domain values; Cohen d: 3.10→4.56, p<0.001
-- Table 3: restructured as two-model comparison; log-probability as primary metric;
-  JSD-anchoring interaction finding documented in dedicated subsection
-- Abstract, conclusion updated with two-model results and new statistics
-- CITATION.cff: version 0.4.0, date 2026-05-13, abstract updated
-- references.bib: citation audit cycle 3 — all 13/13 VERIFIED
-
-## [0.3.2] - 2026-05-12
-
-### Added
-- `results/real_lm/`: real LLM experiment artifacts (distilgpt2, BM25 RAG, 5 seeds x 5 iter)
-  - `oea_anchored`: JSD=0.088 (41% less drift), mean log-prob +0.574 nats vs control
-  - `oea_miscalibrated`: mean log-prob −0.387 vs control — causal proof of mechanism
-  - `oea_rag_only`: RAG without filter degrades log-prob; epistemic filter is operative
-- Table 3 (Real LLM results) in `arxiv/main.tex` with actual numbers
-- Saturation note for TRR metric in manuscript Limitations
-- SEAL-0008: manuscript lock milestone
-
-### Changed
-- `experiments/credibility_suite.py`: annotation on `oea_full` CQ=0.83 noting provisional
-  status pending threshold recalibration (TRR saturated at 1.0 in real LLM run)
-
-### Fixed
-- `arxiv/main.tex`: removed `\citet{}` (natbib not loaded); replaced with `\cite{}`
-- `.github/workflows/ci.yml`: added `-bibtex` flag to latexmk args
-
-## [0.3.1]
-
-### Added
-- `BM25Retriever` class in `real_lm_experiment.py`: corpus-grounded token-overlap RAG
-  implementing OEA Layer 1 (Ontological Anchoring). Not a log-probability proxy.
-- `oea_rag_only` variant: retrieval without epistemic filtering; isolates RAG contribution
-- REQ-OEA-010/011/012 and TEST-OEA-010/011/012 (RAG spec, manuscript results, CQ chain)
-- `\section{Real LLM Validation}` in `arxiv/main.tex`: design, frozen-weights scope as
-  necessary-condition framing, results placeholder (pending `real_lm_experiment.py` run)
-- DEC-004 in `docs/ARCHITECTURE.md`: explicit frozen-weights scope decision
-- CQ Measurement output in `real_lm_experiment.py main()`: derives `_CALIBRATION_QUALITY`
-  updates from measured `true_reject_rate` (closes evidence chain: REQ-OEA-012)
-
-### Changed
-- `real_lm_experiment.py`: N_SEEDS 3 → 5; RAG added to all non-control variants;
-  `oea_anchored` = RAG + K=3 + highest log-prob + vocab anchoring;
-  `oea_miscalibrated` = RAG + K=3 + lowest log-prob (anti-calibrated falsification control)
-- `docs/ARCHITECTURE.md`: updated component table, data flow, key decisions
-
-## [0.3.0]
-
-### Added
-- `experiments/real_lm_experiment.py` — distilgpt2 (82M) recursive stability experiment
-  with genuine neural log-probability epistemic filter; no hardcoded constants.
-  Variants: `control`, `oea_anchored` (keep highest log-prob), `oea_miscalibrated`
-  (keep lowest, anti-calibrated). Proves the mechanism is causal, not definitional.
-- `requirements-experiments.txt` — torch (CPU) + transformers install spec
-- `experiments/config/credibility_plan_fast.json` — minimal plan for CI validation
-  (10 runs vs 7,128; covers key variants including `ablation_miscalibrated`)
-- `scripts/run-experiments.sh` / `scripts/run-experiments.cmd` — one-command experiment
-  runner; `--all` flag includes real LLM experiment
-- `.venv` bootstrap: `scripts/setup.sh --experiments` / `scripts/setup.cmd --experiments`
-
-### Changed
-- `experiments/credibility_suite.py`: replaced hardcoded `p_reject_false`/`p_reject_true`
-  constants with `_CALIBRATION_QUALITY` dict + `_rejection_rates(cq)` formula. Rejection
-  rates are now derived from calibration quality via linear interpolation between random
-  baseline (CQ=0.5) and perfect discrimination (CQ=1.0). Addresses "simulation fallacy".
-- `experiments/credibility_suite.py`: added `ablation_miscalibrated` variant (CQ=0.22)
-  demonstrating that anti-calibrated selection degrades faster than control.
-- `experiments/config/credibility_plan.json`: added `ablation_miscalibrated` to variant list
-- `scripts/setup.sh` / `scripts/setup.cmd`: full venv bootstrap with dependency tiers
-
-## [0.2.2]
-
-### Fixed
-- 26 markdownlint errors eliminated across governance, docs, and CHANGELOG files
-  (MD009 trailing spaces, MD010 hard tabs, MD012 blank lines, MD031/MD040 code
-  fence language/spacing, MD037 emphasis markers, MD001 heading levels)
-- `.markdownlint.json`: added `siblings_only` for MD024, disabled MD026/MD034
-- `pytest.ini`: added `pythonpath = .` so `tests/` can import `experiments.*`
-- `.github/workflows/ci.yml`: removed `cm-super` apk (not available in Alpine TeX image)
-
-### Added
-- CI: `python-tests` job (pytest + pip-audit security scan)
-- `.github/dependabot.yml`: weekly pip and GitHub Actions dependency updates
-
-## [0.2.1] - 2026-05-12
-
-### Fixed
-- `arxiv/references.bib`: all 8/8 citations VERIFIED (citation lock closed)
-  - `fu2025selfverification`: NeurIPS 2025 poster confirmed (Accept, submission 12388)
-  - `roumeliotis2025trust`: arXiv:2507.10571 v3 confirmed; trailing comma removed
-- Trace vault SEAL-0007: citation lock audit-gate sealed
-- REQ-OEA-006 submission guardrail: citation lock now satisfied
-
-## [0.2.0] - 2026-05-12
-
-### Added
-- `arxiv/main.tex`: `\section{Conclusion}` with OEA hypothesis restatement, scope-bounded
-  evidence summary, stability/epistemic orthogonality finding, 4-item future-work agenda
-- `arxiv/main.tex`: Table 2 — full ablation study (11 variants × 648 runs) with stability,
-  true-reject, false-reject means + 95% CI, Cohen's d vs baselines (sourced from artifacts)
-- `arxiv/references.bib`: full citation audit; 6/8 VERIFIED, 2 flagged for human check
-- REQ-OEA-007/008/009 and TEST-OEA-007/008/009 added to belief artifact registry
-- Trace vault: SEAL-0004 (cycle 2 architecture), SEAL-0005 (verification), SEAL-0006 (release)
-
-## [0.1.0] - 2026-05-12
-
-### Added
-- specsmith 0.10.1 governance overlay (`aee-research` type, `enable_epistemic=true`)
-- `AGENTS.md` — agent governance hub with OEA protocol and H13 epistemic boundary rules
-- `docs/ARCHITECTURE.md` — OEA tri-layer architecture, experiment harness components, data flow
-- `REQUIREMENTS.md` / `docs/REQUIREMENTS.md` — 6 REQ-OEA-\* belief artifacts (all P1, Accepted)
-- `docs/TESTS.md` — 6 TEST-OEA-\* specifications with 100% REQ coverage
-- `docs/governance/` — 11 governance files including full epistemic layer (EPISTEMIC-AXIOMS,
-  BELIEF-REGISTRY, FAILURE-MODES, UNCERTAINTY-MAP)
-- `scaffold.yml` — aee-research project type, specsmith 0.10.1
-- `arxiv/main.tex` — manuscript scaffold with pilot results and ablation tables
-- `experiments/credibility_suite.py` — bigram-proxy ablation harness (11 variants)
-- `results/summary_metrics.json` — pilot: stability delta +0.121, true-reject delta +0.232
-- Trace vault: SEAL-0001 (architecture), SEAL-0002 (verification), SEAL-0003 (v0.1.0 release)
-- Community files: `CODE_OF_CONDUCT.md`, `.github/ISSUE_TEMPLATE/`, `.editorconfig`
-[Unreleased]: https://github.com/BitConcepts/oea-framework-paper/compare/v1.0.0...HEAD
-[1.0.0]: https://github.com/BitConcepts/oea-framework-paper/compare/v0.4.0...v1.0.0
-[0.4.0]: https://github.com/BitConcepts/oea-framework-paper/compare/v0.3.2...v0.4.0
-[0.3.2]: https://github.com/BitConcepts/oea-framework-paper/compare/v0.3.1...v0.3.2
-[0.3.1]:
-[0.3.1]: https://github.com/BitConcepts/oea-framework-paper/compare/v0.3.0...v0.3.1
-[0.3.0]: https://github.com/BitConcepts/oea-framework-paper/compare/v0.2.2...v0.3.0
-[0.2.2]: https://github.com/BitConcepts/oea-framework-paper/compare/v0.2.1...v0.2.2
-[0.2.1]: https://github.com/BitConcepts/oea-framework-paper/compare/v0.2.0...v0.2.1
-[0.2.0]: https://github.com/BitConcepts/oea-framework-paper/compare/v0.1.0...v0.2.0
-[0.1.0]: https://github.com/BitConcepts/oea-framework-paper/releases/tag/v0.1.0
+[Unreleased]: https://github.com/BitConcepts/oea-framework-paper/compare/v1.1.0...HEAD
+[1.1.0]: https://github.com/BitConcepts/oea-framework-paper/compare/v1.0.0...v1.1.0
+[1.0.0]: https://github.com/BitConcepts/oea-framework-paper/releases/tag/v1.0.0