codontrace 0.3.0b4.dev0__tar.gz → 0.3.0b5__tar.gz

{codontrace-0.3.0b4.dev0 → codontrace-0.3.0b5}/.github/workflows/ci.yml

@@ -34,7 +34,7 @@ jobs:
       - name: Install test dependencies
         run: |
           python -m pip install --upgrade pip setuptools wheel
-          python -m pip install -e . pytest
+          python -m pip install -e ".[dev]" pytest
 
       - name: Compile package, tests, examples, and tools
         run: python -m compileall -q src tests examples tools
@@ -98,7 +98,7 @@ jobs:
       - name: Install test dependencies
         run: |
           python -m pip install --upgrade pip setuptools wheel
-          python -m pip install -e . pytest
+          python -m pip install -e ".[dev]" pytest
 
       - name: Compile package, tests, examples, and tools
         run: python -m compileall -q src tests examples tools

{codontrace-0.3.0b4.dev0 → codontrace-0.3.0b5}/CHANGELOG.md

@@ -2,6 +2,129 @@
 
 ## Unreleased
 
+### HARD_EXPERIMENT_02 — honest pilot clear + research null
+
+CodonTrace Genesis. Identity `0.3.0b4.dev0`. ClaimGate not loosened beyond
+earned evidence. Phase A–E pins unchanged.
+
+- Wiring: harvest-at-nav-target for E1 selective value; shuffled-arm MI
+  records gated to `shuffle_mode=off` (adoption path matches emit).
+- Pilot seeds 1000–1009 clear assay / oracle / MI gates
+  (`cleared_for_research=true`).
+- Research 30 seeds (2000–2029) committed to
+  `docs/hard_experiment_02/results_v1.json` with Holm null — ceiling stays
+  `runtime_observation` (not `intervention_supported`).
+- Research ticks/pop locked to smoke-calibrated 8/8 after host OOM at 24/12
+  (prereg TBD-after-smoke).
+- HE03 research remains deferred.
+
+## [0.3.0b4] — 2026-09-17
+
+PyPI research-beta cut. Tag `v0.3.0b4`. Includes HE01 SCHEMA v7 (PR #35),
+discovery ClaimGate waves A–D, ESP32 engineering stub, and prior H–L work.
+Install: `pip install codontrace==0.3.0b4`.
+
+### HARD_EXPERIMENT_01 SCHEMA v7 — seed-contingent food amounts + lock digest
+
+CodonTrace Genesis. Identity `0.3.0b4`. Estimand unchanged. ClaimGate not
+loosened by the lock text alone. Phase A–E pins unchanged (HE01 overlay only).
+
+- Single prereg lock: `docs/HARD_EXPERIMENT_01_PREREG_AMENDMENT_LOCK.md` (no
+  sixth soft numbered amendment pile-on).
+- Every-cell food coverage retained (Amd 03 lesson); initial resource amounts
+  drawn per seed from discrete multipliers `{0.75, 1.0, 1.25}` × base
+  (`hard_experiment_01/food_amounts`) so analysis seeds are distinct environment
+  realizations (Avida spatial-heterogeneity / Taylor-style seed contingency).
+- Confirmatory negative control remains `capsules_content_null`; legacy
+  `capsules_shuffled` sensitivity-only (peer-rotation may beat off).
+- Assay gate: `assay_failed_treatment_seed_variance_zero` if treatment outcomes
+  are seed-invariant.
+- Artifacts: `docs/hard_experiment_01/pilot_v7.json`,
+  `docs/hard_experiment_01/results_v7.json`. Ceiling at most
+  `intervention_supported` when the full rule + healthy control hold; else
+  honest `assay_invalid` / zero / `runtime_observation`.
+- ESP32 Moj-ه demoted in CLAIMS/CHANGELOG/README to **engineering stub only**
+  (`SimEsp32Bridge`), not a scientific research surface.
+- Avida/MABE ClaimGate adapters remain **skeletons** until audited published
+  `.dat` / DataFile CSV campaigns exist (no literature-compatibility overclaim).
+- HE03 research results remain deferred (docs only); HE02 research landed with Holm null.
+
+### Discovery-wire + ESP32 Moj-ه thin bridge (engineering stub — not research surface)
+
+CodonTrace Genesis. Identity `0.3.0b4.dev0`. ClaimGate not loosened.
+Phase A–E pins unchanged when `discovery_wire` is default-off.
+ESP32 is an **engineering stub** (`SimEsp32Bridge`); not a scientific research surface.
+**Zero claim that physical robots ran.**
+
+- Optional `DiscoveryWireConfig` on `QDSearchConfig` (default OFF): every N
+  generations, `NoveltyProposer.propose` → cheap S1 `run_discovery_pipeline` →
+  admit to QD archive / innovation_protection only when ClaimGate ceiling is
+  `discovery_witness_candidate`. Omitted from digests when disabled.
+- ESP32 Moj-ه thin adapter `claimgate/adapters/esp32_bridge.py`:
+  `move` / `read_sensor`, Pydantic untrusted boundary, `SimEsp32Bridge`,
+  Serial/MQTT stubs fail closed (no secrets), STR-disparity helper + §11 stop
+  criterion. Firmware sketches under `firmware/esp32/` with safety comments.
+- Docs: `docs/design/ESP32_BRIDGE_v0.1.md` (Koos/Mouret/Doncieux 2013;
+  OMNI-EPIC arXiv:2405.15568; ARE 2024 STR-disparity honesty).
+- Replay digest policy registration for `DiscoveryWireEvent` + `STRDisparityResult`.
+- No AGI / collective-intelligence / fabricated hardware-campaign claims.
+
+### Open-ended discovery ClaimGate pipeline (waves الف–د)
+
+CodonTrace Genesis. Identity `0.3.0b4.dev0`. ClaimGate not loosened.
+Phase A–E pins unchanged. Zero ESP32 / Moj-ه code.
+
+- Goal doc: `docs/design/OPEN_ENDED_DISCOVERY_CLAIMGATE_PIPELINE_v2.md`
+  (ASAL Kumar et al. arXiv:2412.17799; Flageat et al. IEEE TEVC 2026 /
+  arXiv:2409.13315; MAP-Elites; QD; OMNI-EPIC; HE01 three-negative lesson).
+- Wave الف: `run_discovery_pipeline(candidate, *, scale=...) -> DiscoveryAuditReport`
+  with digest-bearing accept/reject and unit tests on ≥3 hand-crafted candidates.
+- Wave ب: `NoveltyProposer` Protocol + `RandomProposer` + `ExternalModelStubProposer`
+  (callable without paid API) + Pydantic boundary `parse_external_proposal`.
+- Wave ج: `CandidateSpec` / `DiscoveryAuditReport` / `NegativeControlResult` as
+  `@dataclass(frozen=True, slots=True)`; witness sha256 matches content.
+- Wave د: campaign 3–5 proposals with all three negatives
+  (`proposer_random`, `proposer_shuffled_archive`, `archive_no_op`); ceiling
+  above `runtime_observation` only if meaningfully better than all three;
+  honest accept/reject rates (100% reject is valid data).
+- Hypothesis: `paired_effect_size` never returns NaN when variance is 0.
+- Replay digest policy registration for new public digest dataclasses.
+- No AGI / collective-intelligence claims.
+
+### HARD_EXPERIMENT_03 — E3 TaskSwitchCost + gorelick_nmi + IsolationAssay
+
+CodonTrace Genesis. Identity `0.3.0b4.dev0`. ClaimGate not loosened.
+Phase A–E pins unchanged.
+
+- Default-off `TaskSwitchCostConfig` (Goldsby 0 / ~25 / ~50 cycle → ATP
+  analogues 0 / 0.25 / 0.50) with conditional `to_dict`, digests,
+  ConfigurationError validation, and pin+runtime tests.
+- Confirmatory `gorelick_nmi` at `codontrace.genesis.metrics.division_of_labor`
+  (Gorelick NMI/NME); non-NMI DoL proxies stay **legacy**.
+- `IsolationAssay` secondary solo-vs-group drop (not collective intelligence).
+- Campaign module `hard_experiment_03` with prereg arms, honest
+  `runtime_observation` ceiling, pilot seeds 1000–1009.
+- ClaimGate adapter `bundle_from_hard_experiment_03`; replay digest policy
+  registration for HE02/HE03 public digest dataclasses.
+- Docs: `HARD_EXPERIMENT_03.md`; CLAIMS ceiling starts at
+  `runtime_observation`. No fabricated research `results_v1.json`.
+
+### HARD_EXPERIMENT_02 — E1+E2+E5 after prereg
+
+CodonTrace Genesis. Identity `0.3.0b4.dev0`. ClaimGate not loosened.
+Phase A–E pins unchanged.
+
+- Default-off knobs: `FoodPatchSignalConfig`, `DemeSelectionConfig`,
+  `SteppingStoneRewardConfig` with conditional `to_dict`, digests,
+  ConfigurationError validation, and pin+runtime tests.
+- `MOVE_TOWARD_CAPSULE_TARGET` on HE02 action registries only.
+- Campaign module `hard_experiment_02` with prereg arms, manipulation
+  checks (`assay_invalid` when not realized), pilot seeds 1000–1009,
+  research 30-seed path, ClaimGate adapter.
+- Docs: `HARD_EXPERIMENT_02.md`; CLAIMS ceiling starts at
+  `runtime_observation`.
+
+
 ### Wave 2 — simulator-agnostic ClaimGate auditor
 
 CodonTrace Genesis standalone evidence auditor. **Does not** claim

{codontrace-0.3.0b4.dev0 → codontrace-0.3.0b5}/CITATION.cff

@@ -1,14 +1,14 @@
 cff-version: 1.2.0
 message: "If you use CodonTrace Genesis, please cite the versioned software release."
 title: "CodonTrace Genesis"
-version: "0.3.0b4.dev0"
+version: "0.3.0b4"
 authors:
   - family-names: "Jamei"
     given-names: "Parvaz"
 license: "AGPL-3.0-or-later"
 repository-code: "https://github.com/Parvaz-Jamei/codontrace-genesis"
 doi: "10.5281/zenodo.20337435"
-date-released: "2026-09-11"
+date-released: "2026-09-17"
 abstract: "CodonTrace Genesis is a deterministic, library-first research-beta toolkit for digital-evolution experiments, causal mechanism auditing, capsule-mediated transfer, skill-compression inheritance, role and collective-behavior instrumentation, and replayable evidence artifacts."
 keywords:
   - digital evolution

{codontrace-0.3.0b4.dev0 → codontrace-0.3.0b5}/MANIFEST.in

@@ -8,6 +8,8 @@ include CITATION.cff
 include pyproject.toml
 recursive-include docs *.md
 recursive-include docs/hard_experiment_01 *.json
+recursive-include docs/hard_experiment_02 *.json
+recursive-include docs/hard_experiment_03 *.json
 recursive-include src/codontrace/genesis/rag/corpus *.jsonl
 recursive-include src/codontrace/genesis/rag/corpus *.md
 include src/codontrace/genesis/ilw/INTEGRATION_DAG.json
@@ -30,3 +32,5 @@ global-exclude *.html
 global-exclude *.patch
 global-exclude .grok_write_probe*.txt
 global-exclude .size_test_*.txt
+
+recursive-include firmware *

{codontrace-0.3.0b4.dev0/src/codontrace.egg-info → codontrace-0.3.0b5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: codontrace
-Version: 0.3.0b4.dev0
+Version: 0.3.0b5
 Summary: CodonTrace Genesis: deterministic research-beta engine for digital evolution, causal mechanism auditing, capsule-mediated transfer, skill compression, role emergence, and replayable evidence artifacts.
 Author: Parvaz Jamei
 License-Expression: AGPL-3.0-or-later
@@ -32,6 +32,8 @@ Requires-Dist: mypy<3,>=2.0; extra == "dev"
 Requires-Dist: build<2,>=1.5; extra == "dev"
 Requires-Dist: twine<7,>=6.2; extra == "dev"
 Requires-Dist: pip-audit<3,>=2.10; extra == "dev"
+Requires-Dist: hypothesis<7,>=6.0; extra == "dev"
+Requires-Dist: pydantic<3,>=2.0; extra == "dev"
 Provides-Extra: science
 Requires-Dist: numpy<3,>=2.0; extra == "science"
 Requires-Dist: scipy<2,>=1.13; extra == "science"
@@ -87,6 +89,8 @@ CodonTrace Genesis is **not** currently presented as:
 - proof of open-ended intelligence, or a Tokyo Type 1 *pass* (Channon 2024)
 - a biological evolution simulator or wet-lab chemistry engine
 - a replacement for Avida, MABE, DEAP, QDax, pyribs, or similar tools
+- a physical-robot research platform (ESP32 Moj-ه is an engineering stub / `SimEsp32Bridge` only)
+- an Avida/MABE literature-compatible campaign runner (ClaimGate adapters are skeletons until audited published `.dat`/CSV exist)
 
 The project is ambitious. Claims must pass evidence gates. See [`CLAIMS.md`](CLAIMS.md) and [`docs/WHY_NOT_INTELLIGENCE_YET.md`](docs/WHY_NOT_INTELLIGENCE_YET.md).
 
@@ -97,16 +101,20 @@ The project is ambitious. Claims must pass evidence gates. See [`CLAIMS.md`](CLA
 | Field | Current status |
 |---|---|
 | Package | `codontrace` |
-| Public PyPI tip | `0.3.0b3` — Phase A–G (life-loop + materials). Confirm on [PyPI](https://pypi.org/project/codontrace/). |
-| GitHub `main` / this PR | Development identity `0.3.0b4.dev0` (Phases H–L plus hygiene / hard experiment 01). |
+| Public PyPI tip | `0.3.0b4` — Phases H–L + HE01 SCHEMA v7 + discovery ClaimGate pipeline. Confirm on [PyPI](https://pypi.org/project/codontrace/). |
+| GitHub `main` | Development identity `0.3.0b4.dev1` (ahead of the published cut). |
 | Python | `3.11–3.14` |
 | DOI | `10.5281/zenodo.20337435` |
 | License | `AGPL-3.0-or-later` |
+| Official GitHub release | [`v0.3.0b4`](https://github.com/Parvaz-Jamei/codontrace-genesis/releases/tag/v0.3.0b4) (set older handoff tags to pre-release so Latest stays here) |
+| HE01 | SCHEMA v7 locked; ceiling at most `intervention_supported` when the full rule holds |
+| HE02 | Research null after analysis v1b; ceiling stays `runtime_observation` |
+| HE03 | Code + prereg present; research `results_v1` is absent on purpose |
 | Claim ceiling | Research software and evidence infrastructure. Not a proof of AGI, consciousness, collective intelligence, Tokyo Type 1 passed, or Avida replacement. |
 
 ### Phases A–L
 
-Phases A–G are the public `0.3.0b3` substrate. Phases H–L are on `main` only until a later release.
+Phases A–G remain the earlier `0.3.0b3` substrate. Phases H–L + HE01 SCHEMA v7 + discovery ClaimGate pipeline ship in public `0.3.0b4`.
 
 | Phase | What landed | Claim ceiling |
 |---|---|---|
@@ -133,22 +141,22 @@ North star: eventually produce honest collective-work / intelligence-*pathway* o
 
 ## Installation
 
-The published wheel is `codontrace==0.3.0b3` (Phases A–G). This branch reports development identity `0.3.0b4.dev0` (H–L plus hygiene / hard experiment 01). Install from source if you need the post–Phase G pathway code.
+The published wheel is `codontrace==0.3.0b4` (Phases H–L + HE01 SCHEMA v7 + discovery ClaimGate pipeline). `main` is `0.3.0b4.dev1` and is not a PyPI recut. Older `0.3.0b3` remains the A–G substrate tip.
 
 Python `3.11–3.14`. CI smokes `ubuntu-latest`, `windows-latest`, and `macos-latest` on that range.
 
-### From PyPI (`0.3.0b3`)
+### From PyPI (`0.3.0b4`)
 
 ```bash
-pip install codontrace==0.3.0b3
+pip install codontrace==0.3.0b4
 ```
 
 Optional research extras:
 
 ```bash
-pip install "codontrace[research]==0.3.0b3"
-pip install "codontrace[causal]==0.3.0b3"
-pip install "codontrace[qd]==0.3.0b3"
+pip install "codontrace[research]==0.3.0b4"
+pip install "codontrace[causal]==0.3.0b4"
+pip install "codontrace[qd]==0.3.0b4"
 ```
 
 ### From source (`main`, may be ahead of PyPI)
@@ -163,7 +171,7 @@ python -m pip install -e ".[dev,research,causal,qd]"
 python -c "import codontrace; print(codontrace.__version__)"
 ```
 
-A PyPI install of the current tip should print `0.3.0b3`. This checkout prints `0.3.0b4.dev0`. Do not treat the version tuple as a phase fence.
+A PyPI install of the current tip prints `0.3.0b4`. An editable install from `main` prints `0.3.0b4.dev1`. Do not treat the version tuple as a phase fence.
 
 ---
 
@@ -271,12 +279,12 @@ Full policy: [`CLAIMS.md`](CLAIMS.md).
 Library-complete beta is not “done.” The next work is **not** more empty Phase
 letters.
 
-1. **Hygiene** — keep `main` free of probe junk and claim-language drift.
-2. **Widen CI** — more of the existing suite on hosted runners, still honest.
-3. **`HARD_EXPERIMENT_01`** — one hard division-of-labor / ablation / isolation experiment, not another scaffold. Wave 2 adds a standalone ClaimGate auditor around that evidence; it does not raise the ceiling.
-4. **Candidate** — `collective_intelligence_candidate` only if the full honest flags, including replay, are actually earned.
-5. **Goldsby-grade** — literature-scale specialist / messaging work; `is_goldsby_2012_pnas_experiment` stays false until that experiment exists.
-6. **OEE later** — Tokyo Type 1 *pass* and open-ended intelligence remain blocked until evidence exists.
+1. **Do not recut PyPI `0.3.0b4`.** The next public wheel is `0.3.0b5` only after CI is green on a release identity.
+2. **Keep `main` on `0.3.0b4.dev1`** until that cut. Mark stale GitHub handoff releases as pre-release so Latest stays on [`v0.3.0b4`](https://github.com/Parvaz-Jamei/codontrace-genesis/releases/tag/v0.3.0b4).
+3. **HE03 research campaign** — run only against the locked prereg; do not fabricate `results_v1`.
+4. **Lint/type inventory** — `lint-type` remains non-blocking until the ruff/mypy backlog is reduced in its own PR.
+5. **Candidate claims** — `collective_intelligence_candidate` only if the full honest flags, including replay, are actually earned.
+6. **OEE later** — Tokyo Type 1 *pass* and open-ended intelligence remain blocked.
 
 ---
 
@@ -310,6 +318,7 @@ boundaries.
 | Document | Purpose |
 |---|---|
 | [`CLAIMS.md`](CLAIMS.md) | Allowed, candidate, and blocked claims |
+| [`docs/protocol/CLAIM_LADDER_PROTOCOL_v0.1.md`](docs/protocol/CLAIM_LADDER_PROTOCOL_v0.1.md) | Claim-ladder protocol v0.1 (design + grade ALife claims; Wave 3) |
 | [`docs/WHY_NOT_INTELLIGENCE_YET.md`](docs/WHY_NOT_INTELLIGENCE_YET.md) | Literature-vs-reality barrier map; not close to AGI |
 | [`docs/PHASE_H_CI_AI_PATH.md`](docs/PHASE_H_CI_AI_PATH.md) | Phase H: RAG + ablation / effect-size harnesses |
 | [`docs/PHASE_I_CI_EVIDENCE.md`](docs/PHASE_I_CI_EVIDENCE.md) | Phase I: heldout / evolved DoL / MLS / export-of-fitness |
@@ -325,7 +334,7 @@ boundaries.
 | [`docs/rag/README.md`](docs/rag/README.md) | Literature RAG corpus (measurement design, not intelligence evidence) |
 | [`REPRODUCIBILITY.md`](REPRODUCIBILITY.md) | Install, validation tiers, artifact preservation |
 | [`BENCHMARKS.md`](BENCHMARKS.md) | Benchmark protocols and claim boundaries |
-| [`RELEASE_EVIDENCE.md`](RELEASE_EVIDENCE.md) | Release evidence for the public `0.3.0b3` wheel |
+| [`RELEASE_EVIDENCE.md`](RELEASE_EVIDENCE.md) | Release evidence pack (see file for which public wheel it covers) |
 
 Phase literature checklists: [`PHASE_D_LITERATURE.md`](docs/PHASE_D_LITERATURE.md),
 [`PHASE_E_LITERATURE.md`](docs/PHASE_E_LITERATURE.md),
@@ -359,7 +368,7 @@ versioned software release.
 @software{codontrace_genesis_2026,
   title = {CodonTrace Genesis},
   author = {Jamei, Parvaz},
-  version = {0.3.0b3},
+  version = {0.3.0b4},
   doi = {10.5281/zenodo.20337435},
   url = {https://github.com/Parvaz-Jamei/codontrace-genesis}
 }

{codontrace-0.3.0b4.dev0 → codontrace-0.3.0b5}/README.md

@@ -35,6 +35,8 @@ CodonTrace Genesis is **not** currently presented as:
 - proof of open-ended intelligence, or a Tokyo Type 1 *pass* (Channon 2024)
 - a biological evolution simulator or wet-lab chemistry engine
 - a replacement for Avida, MABE, DEAP, QDax, pyribs, or similar tools
+- a physical-robot research platform (ESP32 Moj-ه is an engineering stub / `SimEsp32Bridge` only)
+- an Avida/MABE literature-compatible campaign runner (ClaimGate adapters are skeletons until audited published `.dat`/CSV exist)
 
 The project is ambitious. Claims must pass evidence gates. See [`CLAIMS.md`](CLAIMS.md) and [`docs/WHY_NOT_INTELLIGENCE_YET.md`](docs/WHY_NOT_INTELLIGENCE_YET.md).
 
@@ -45,16 +47,20 @@ The project is ambitious. Claims must pass evidence gates. See [`CLAIMS.md`](CLA
 | Field | Current status |
 |---|---|
 | Package | `codontrace` |
-| Public PyPI tip | `0.3.0b3` — Phase A–G (life-loop + materials). Confirm on [PyPI](https://pypi.org/project/codontrace/). |
-| GitHub `main` / this PR | Development identity `0.3.0b4.dev0` (Phases H–L plus hygiene / hard experiment 01). |
+| Public PyPI tip | `0.3.0b4` — Phases H–L + HE01 SCHEMA v7 + discovery ClaimGate pipeline. Confirm on [PyPI](https://pypi.org/project/codontrace/). |
+| GitHub `main` | Development identity `0.3.0b4.dev1` (ahead of the published cut). |
 | Python | `3.11–3.14` |
 | DOI | `10.5281/zenodo.20337435` |
 | License | `AGPL-3.0-or-later` |
+| Official GitHub release | [`v0.3.0b4`](https://github.com/Parvaz-Jamei/codontrace-genesis/releases/tag/v0.3.0b4) (set older handoff tags to pre-release so Latest stays here) |
+| HE01 | SCHEMA v7 locked; ceiling at most `intervention_supported` when the full rule holds |
+| HE02 | Research null after analysis v1b; ceiling stays `runtime_observation` |
+| HE03 | Code + prereg present; research `results_v1` is absent on purpose |
 | Claim ceiling | Research software and evidence infrastructure. Not a proof of AGI, consciousness, collective intelligence, Tokyo Type 1 passed, or Avida replacement. |
 
 ### Phases A–L
 
-Phases A–G are the public `0.3.0b3` substrate. Phases H–L are on `main` only until a later release.
+Phases A–G remain the earlier `0.3.0b3` substrate. Phases H–L + HE01 SCHEMA v7 + discovery ClaimGate pipeline ship in public `0.3.0b4`.
 
 | Phase | What landed | Claim ceiling |
 |---|---|---|
@@ -81,22 +87,22 @@ North star: eventually produce honest collective-work / intelligence-*pathway* o
 
 ## Installation
 
-The published wheel is `codontrace==0.3.0b3` (Phases A–G). This branch reports development identity `0.3.0b4.dev0` (H–L plus hygiene / hard experiment 01). Install from source if you need the post–Phase G pathway code.
+The published wheel is `codontrace==0.3.0b4` (Phases H–L + HE01 SCHEMA v7 + discovery ClaimGate pipeline). `main` is `0.3.0b4.dev1` and is not a PyPI recut. Older `0.3.0b3` remains the A–G substrate tip.
 
 Python `3.11–3.14`. CI smokes `ubuntu-latest`, `windows-latest`, and `macos-latest` on that range.
 
-### From PyPI (`0.3.0b3`)
+### From PyPI (`0.3.0b4`)
 
 ```bash
-pip install codontrace==0.3.0b3
+pip install codontrace==0.3.0b4
 ```
 
 Optional research extras:
 
 ```bash
-pip install "codontrace[research]==0.3.0b3"
-pip install "codontrace[causal]==0.3.0b3"
-pip install "codontrace[qd]==0.3.0b3"
+pip install "codontrace[research]==0.3.0b4"
+pip install "codontrace[causal]==0.3.0b4"
+pip install "codontrace[qd]==0.3.0b4"
 ```
 
 ### From source (`main`, may be ahead of PyPI)
@@ -111,7 +117,7 @@ python -m pip install -e ".[dev,research,causal,qd]"
 python -c "import codontrace; print(codontrace.__version__)"
 ```
 
-A PyPI install of the current tip should print `0.3.0b3`. This checkout prints `0.3.0b4.dev0`. Do not treat the version tuple as a phase fence.
+A PyPI install of the current tip prints `0.3.0b4`. An editable install from `main` prints `0.3.0b4.dev1`. Do not treat the version tuple as a phase fence.
 
 ---
 
@@ -219,12 +225,12 @@ Full policy: [`CLAIMS.md`](CLAIMS.md).
 Library-complete beta is not “done.” The next work is **not** more empty Phase
 letters.
 
-1. **Hygiene** — keep `main` free of probe junk and claim-language drift.
-2. **Widen CI** — more of the existing suite on hosted runners, still honest.
-3. **`HARD_EXPERIMENT_01`** — one hard division-of-labor / ablation / isolation experiment, not another scaffold. Wave 2 adds a standalone ClaimGate auditor around that evidence; it does not raise the ceiling.
-4. **Candidate** — `collective_intelligence_candidate` only if the full honest flags, including replay, are actually earned.
-5. **Goldsby-grade** — literature-scale specialist / messaging work; `is_goldsby_2012_pnas_experiment` stays false until that experiment exists.
-6. **OEE later** — Tokyo Type 1 *pass* and open-ended intelligence remain blocked until evidence exists.
+1. **Do not recut PyPI `0.3.0b4`.** The next public wheel is `0.3.0b5` only after CI is green on a release identity.
+2. **Keep `main` on `0.3.0b4.dev1`** until that cut. Mark stale GitHub handoff releases as pre-release so Latest stays on [`v0.3.0b4`](https://github.com/Parvaz-Jamei/codontrace-genesis/releases/tag/v0.3.0b4).
+3. **HE03 research campaign** — run only against the locked prereg; do not fabricate `results_v1`.
+4. **Lint/type inventory** — `lint-type` remains non-blocking until the ruff/mypy backlog is reduced in its own PR.
+5. **Candidate claims** — `collective_intelligence_candidate` only if the full honest flags, including replay, are actually earned.
+6. **OEE later** — Tokyo Type 1 *pass* and open-ended intelligence remain blocked.
 
 ---
 
@@ -258,6 +264,7 @@ boundaries.
 | Document | Purpose |
 |---|---|
 | [`CLAIMS.md`](CLAIMS.md) | Allowed, candidate, and blocked claims |
+| [`docs/protocol/CLAIM_LADDER_PROTOCOL_v0.1.md`](docs/protocol/CLAIM_LADDER_PROTOCOL_v0.1.md) | Claim-ladder protocol v0.1 (design + grade ALife claims; Wave 3) |
 | [`docs/WHY_NOT_INTELLIGENCE_YET.md`](docs/WHY_NOT_INTELLIGENCE_YET.md) | Literature-vs-reality barrier map; not close to AGI |
 | [`docs/PHASE_H_CI_AI_PATH.md`](docs/PHASE_H_CI_AI_PATH.md) | Phase H: RAG + ablation / effect-size harnesses |
 | [`docs/PHASE_I_CI_EVIDENCE.md`](docs/PHASE_I_CI_EVIDENCE.md) | Phase I: heldout / evolved DoL / MLS / export-of-fitness |
@@ -273,7 +280,7 @@ boundaries.
 | [`docs/rag/README.md`](docs/rag/README.md) | Literature RAG corpus (measurement design, not intelligence evidence) |
 | [`REPRODUCIBILITY.md`](REPRODUCIBILITY.md) | Install, validation tiers, artifact preservation |
 | [`BENCHMARKS.md`](BENCHMARKS.md) | Benchmark protocols and claim boundaries |
-| [`RELEASE_EVIDENCE.md`](RELEASE_EVIDENCE.md) | Release evidence for the public `0.3.0b3` wheel |
+| [`RELEASE_EVIDENCE.md`](RELEASE_EVIDENCE.md) | Release evidence pack (see file for which public wheel it covers) |
 
 Phase literature checklists: [`PHASE_D_LITERATURE.md`](docs/PHASE_D_LITERATURE.md),
 [`PHASE_E_LITERATURE.md`](docs/PHASE_E_LITERATURE.md),
@@ -307,7 +314,7 @@ versioned software release.
 @software{codontrace_genesis_2026,
   title = {CodonTrace Genesis},
   author = {Jamei, Parvaz},
-  version = {0.3.0b3},
+  version = {0.3.0b4},
   doi = {10.5281/zenodo.20337435},
   url = {https://github.com/Parvaz-Jamei/codontrace-genesis}
 }

codontrace-0.3.0b5/docs/AGENT_COMPLETION_STATUS.md

@@ -0,0 +1,27 @@
+# Agent wave completion status (2026-09-17)
+
+Product: CodonTrace Genesis `0.3.0b4`. Not a PyPI bump.
+
+## This continuation
+
+Official live entry point after the swallowed-`dz` bug:
+
+```python
+from codontrace.genesis.he02_contrasts import run_hard_experiment_02_v1b
+```
+
+Frozen-row analysis:
+
+```bash
+python -m codontrace.genesis.he02_contrasts
+```
+
+`run_hard_experiment_02` on `main` still contains the two-argument call.
+`run_hard_experiment_02_v1b` rescores that campaign before return.
+ClaimGate is not raised.
+
+## Still not shipped
+
+- GitHub Release tag / PyPI cut
+- HE03 research `results_v1`
+- lint/mypy backlog

{codontrace-0.3.0b4.dev0 → codontrace-0.3.0b5}/docs/HARD_EXPERIMENT_01.md

@@ -13,10 +13,13 @@ is frozen in an earlier commit. Campaign payloads record `prereg_digest`.
 
 **Claim ceiling:** `runtime_observation` unless the treatment assay passes,
 the preregistered decision rule holds, **and** ClaimGate allows existing
-`intervention_supported`. Smoke (`n=12`) is `exploratory_only` and stays
-`runtime_observation`. If the treatment arm has mean adoptions ≈ 0 or mean
-extinction ≈ 1, `assay_failed` is recorded and the ceiling stays
-`runtime_observation` even when paired CIs are numerically defined.
+`intervention_supported` (SCHEMA v7 research artifact may earn that ceiling;
+LOCK text alone never auto-grants). Smoke (`n=12`) is `exploratory_only` and
+stays `runtime_observation`. If the treatment arm has mean adoptions ≈ 0 or
+mean extinction ≈ 1, or treatment seed variance is zero, `assay_failed` is
+recorded and the ceiling stays `runtime_observation` even when paired CIs are
+numerically defined. Confirmatory negative control is `capsules_content_null`
+(shuffled is sensitivity-only).
 
 **Blocked:** `intelligence`, `collective_intelligence`, `agi`,
 `tokyo_type1_passed`, `avida_replacement`, and related ClaimGate aliases.
@@ -474,6 +477,47 @@ Boot–yoked limitation under Amd 05). Pilot trail:
 | `capsules_shuffled` | 30 | 42.314 | 8.715 | 542.53 |
 | `oracle_capsule` | 30 | 78.397 | 0.252 | 542.53 |
 
+
+### Results (research v7) — SCHEMA v7 / LOCK
+
+Single lock digest: [`HARD_EXPERIMENT_01_PREREG_AMENDMENT_LOCK.md`](HARD_EXPERIMENT_01_PREREG_AMENDMENT_LOCK.md)
+(`prereg_amendment_lock_digest`
+`818b4efe364d5248fc0f1bd52e28e0e860579f4934a6e047e69730171e39159c`).
+Artifact: [`hard_experiment_01/results_v7.json`](hard_experiment_01/results_v7.json)
+(SCHEMA `hard_experiment_01_v7`). Campaign digest
+`28f812c5aa1e84bad5b8ab61162c29d2d8db80f6e3ccb8e3bbb2861c95f390b9`.
+Scale: 30 seeds (`11`…`40`), 40 ticks, population 16. Primary outcome:
+`receiver_mean_terminal_runtime_atp`. Food layout:
+`every_cell_seed_amounts_v7` (coverage 1.0; per-cell amounts × `{0.75,1.0,1.25}`
+from seed RNG). Role layout: seed Fisher–Yates (unchanged). Replay matched:
+**True**. `assay_failed`: **False**. Decision rule: **PASS**. Sensitivity:
+`shuffled_better_than_capsules_off` (expected peer-rotation; non-blocking).
+**Confirmatory null:** `capsules_content_null` mean = `capsules_off` mean = 28
+(null − off = 0; lower 95% BCa ≤ 0). ClaimGate ceiling:
+**`intervention_supported`** (earned by full decision rule + gate flags; LOCK
+alone does not auto-grant). Pilot trail: `hard_experiment_01/pilot_v7.json`
+(seeds 1000–1009 cleared: assay PASS, content_null PASS, treatment sd > 0).
+
+| Arm | n | mean | sd | adoption attempts |
+|---|---:|---:|---:|---:|
+| `source_bias_on` | 30 | 58.669 | 13.254 | (see JSON) |
+| `source_bias_off` | 30 | 42.180 | 15.205 | (see JSON) |
+| `capsules_off` | 30 | 28.000 | 0.000 | 0.0 |
+| `capsules_content_null` | 30 | 28.000 | 0.000 | (channel active; WAIT token) |
+| `capsules_shuffled` | 30 | 41.518 | 8.219 | (sensitivity) |
+| `oracle_capsule` | 30 | 78.376 | 0.284 | (positive control) |
+
+Primary contrasts (Holm): on vs off mean Δ ≈ 16.49, dz ≈ 2.37, CI95 excludes 0;
+on vs content_null mean Δ ≈ 30.67, dz ≈ 2.31, CI95 excludes 0.
+
+**Reading (honest).** Seed-contingent food amounts + role permute give
+treatment across-seed sd > 0 (not 30 bitwise replays). Confirmatory
+`capsules_content_null` severs informational usefulness (equals channel-off;
+does not “win”). Legacy shuffled still beats off by construction — sensitivity
+only. Ceiling at most `intervention_supported`; not collective intelligence /
+AGI / Avida replacement. `capsules_off` / content_null sd = 0 remains the
+structural WAIT basal plateau (Amd 03 honesty), not a silent assay fail.
+
 ## Wave 1d″ — evidence honesty (no new claim)
 
 See [`hard_experiment_01/WAVE_1D_DOUBLE_PRIME_HONESTY.md`](hard_experiment_01/WAVE_1D_DOUBLE_PRIME_HONESTY.md)