Add an energy-audited SED hydrogen hypothesis tournament

.github/pull_request_template.md

@@ -4,7 +4,7 @@
 
 ## Verification
 
-- [ ] `pytest`: 112 passed, 2 xfailed, **0 skipped** (paste the summary line)
+- [ ] `pytest`: 134 passed, 2 xfailed, **0 skipped** (paste the summary line)
 - [ ] `pytest -m rust` run if the integrator/backends changed
 - [ ] No certificate rule weakened, no check silenced, no `FAIL` converted to a skip
 - [ ] Which oracles/certificates cover this change:

.github/workflows/test.yml

@@ -28,6 +28,7 @@ jobs:
           python examples/demo_vacuum_covariance.py
           python examples/demo_rectification.py
           python examples/demo_hydrogen_coulomb.py
+          python examples/demo_hypothesis_tournament.py
       - name: certificate bundles round-trip, and tampering is rejected
         run: |
           python - <<'PY'
@@ -61,7 +62,7 @@ jobs:
       - run: pip install .
       - run: |
           cd /tmp
-          python -c "import blueberry_circus as bc; print(bc.__version__); from blueberry_circus import oracles, symplectic, certify"
+          python -c "import blueberry_circus as bc; assert bc.__version__ == '0.3.0'; from blueberry_circus import oracles, symplectic, certify, tournament"
       # A directory named nanarch_certify on sys.path must not shadow the
       # vendored, SHA-pinned certificate layer.
       - run: |

.gitignore

@@ -5,6 +5,9 @@ __pycache__/
 build/
 dist/
 examples/out/
+tournament-*.json
+tournament-*.npz
+tournament-summary.json
 rust/**/target/
 src/*.egg-info/
 target/

CITATION.cff

@@ -11,7 +11,9 @@ abstract: >-
   provides an isotropic 3-D random-phase ZPF mode sampler, a non-runaway
   Landau-Lifshitz radiation-reaction integrator, a Strawberry-Fields-style
   Program/Engine API, physical Bohr normalization, independently integrated
-  near-ionization rectification, and re-checkable PASS/FAIL/NULL certificates
+  near-ionization rectification, a complete point-charge drift surface, and an
+  energy-audited four-arm hypothesis tournament with reproducible finite-mode
+  confidence intervals. Re-checkable PASS/FAIL/NULL certificates remain
   anchored to the exactly-solvable oscillator oracle <x^2> = hbar/(2 m omega0).
 type: software
 authors:
@@ -24,7 +26,9 @@ keywords:
   - "hydrogen ground state"
   - "radiation reaction"
   - "certified computation"
-version: "0.2.0"
+  - "energy audit"
+  - "hypothesis tournament"
+version: "0.3.0"
 date-released: "2026-08-13"
 license: "Apache-2.0"
 repository-code: "https://github.com/NanarchTech/BlueberryCircus"

CONTRIBUTING.md

@@ -22,7 +22,7 @@ guarantee. This is a research prototype maintained alongside other work.
 ```bash
 python3 -m venv .venv && source .venv/bin/activate
 pip install -e ".[dev]"
-pytest                      # core suite: must be 112 passed, 2 xfailed, 0 skipped
+pytest                      # core suite: must be 134 passed, 2 xfailed, 0 skipped
 sh scripts/build_rust.sh && pytest -m rust   # if your change touches the integrator
 pip install ".[jax]" && pytest -m jax        # if your change touches backends
 ```

README.md

@@ -6,7 +6,7 @@
 
 [![test](https://github.com/NanarchTech/BlueberryCircus/actions/workflows/test.yml/badge.svg)](https://github.com/NanarchTech/BlueberryCircus/actions/workflows/test.yml)
 ![status](https://img.shields.io/badge/status-simulation--first%20prototype-1f6feb)
-![tests](https://img.shields.io/badge/tests-112%20core%20passed%20·%202%20xfail%20·%200%20skip-2ea043)
+![tests](https://img.shields.io/badge/tests-134%20core%20passed%20·%202%20xfail%20·%200%20skip-2ea043)
 ![python](https://img.shields.io/badge/python-3.10%2B-1f6feb)
 ![backends](https://img.shields.io/badge/backends-numpy%20·%20rust%20·%20jax-d29922)
 ![deps](https://img.shields.io/badge/runtime%20deps-numpy%20only-2ea043)
@@ -18,6 +18,13 @@
 
 BlueberryCircus is a Python laboratory for testing stochastic electrodynamics, not a proof that classical vacuum noise reproduces hydrogen. Its linear oscillator remains an exact benchmark, while its Coulomb sector evaluates three narrower claims: Puthoff’s circular-orbit absorption and radiation balance, Nieuwenhuizen’s near-ionization rectification threshold \(L_c=16/(5\pi\sqrt3)\), and Setterfield’s proposed static vacuum co-scaling, which leaves dimensionless hydrogen dynamics unchanged after time rescaling. The v0.1.0 escape run is retained only as an accelerated numerical stress test because its damping ratio is about 13,000 times the physical Bohr-unit value and it unbinds in less than one tenth of an orbit, so it cannot stand as a reproduction of physical hydrogen self-ionization; certificates verify numerical claims under explicit assumptions, not the truth of SED itself.
 
+Version 0.3.0 adds an energy-audited hypothesis tournament. It computes the
+complete point-charge `D(E,L)` surface before testing a driven Setterfield map,
+a reciprocal finite shell, an inverse-square control, and a conservative
+multipole-storage surrogate. Its finite-mode statistics validate the
+perturbative response kernel, not a nonlinear long-time trajectory. See the
+[tournament specification](docs/tournament.md).
+
 Artifact label: simulation-repo
 
 ## Features
@@ -30,6 +37,10 @@ Artifact label: simulation-repo
 - Analytic oracles: Bohr radius, ground-state energy, angular momentum, oscillator variance
 - Physical Bohr units and an independently integrated near-ionization threshold
 - Static Setterfield co-scaling map with a trajectory-conjugacy regression test
+- Complete point-charge drift surface with perihelion-adapted quadrature
+- Seven-channel energy ledgers and a closed five-state classification vocabulary
+- Four preregistered hypothesis arms with exact zero-parameter recovery controls
+- Chunkable 32-seed/2,048-mode research command with deterministic JSON/NPZ output
 - Phase-space covariance with symplectic readout
 - Seeded ensemble runs for power balance
 - Cole–Zou moving spectral window
@@ -92,6 +103,9 @@ print(o.hydrogen_ground_state_energy(bc.SI) / bc.E_CHARGE)  # -13.605693 (eV)
 | **Rectification threshold** | $L_c=$ **0.5880841551** | independent improper quadrature vs $16/(5\pi\sqrt3)$ | `residual_le_tol` |
 | Critical perihelion | $r_p=$ **0.172921 $a_0$** | $L_c^2/2$ near-ionization asymptote | — |
 | Setterfield static co-scaling | trajectory conjugacy **<10⁻⁹** | $x_U(Ut)=x_1(t)$ and mapped velocities | regression |
+| Point-charge drift surface | **18 preregistered cells** | full Eq. (2.34), both endpoint limits | regression |
+| Inverse-square defining maximum | $H_{\max}=$ **7.327** at $\mu\approx0.590$ | printed kernel gives $d_c\approx-53.69$, not prose −35.8 | quadrature |
+| Multipole closed Hamiltonian | relative total-energy error **<10⁻⁶** | no ZPF or damping | conservation |
 | **Orbit conservation** (no radiation) | $\Delta E/E \sim 10^{-14}$ | exact | `residual_le_tol` |
 | **numpy ↔ Rust agreement** | spring case **bit-identical**, orbit 7×10⁻¹⁴ | — | — |
 | **Independent recheck** | Rust re-derives the verdicts | rejects tampered bundles | separate stack |
@@ -122,7 +136,6 @@ Operation (| apply)  ─▶  Program.compile()  ─▶  Backend  ─▶  Result
 | | check | passes if | tier |
 |---|---|---|---|
 | **O0** | field statistics | discrete → continuum, rel-err < 5% | A |
-| **O1** | Puthoff circular-orbit power balance | $P_{\rm abs}=P_{\rm rad}$ and $m\omega_0r_0^2=\hbar$ | A |
 | **O2** | **spring variance, the gate** | $\langle x^2\rangle=\hbar/2m\omega_0$ to ~1% | A |
 | **O3** | hydrogen radial density | → $4r^2e^{-2r}$ (CPU-day ensembles) | B · `xfail` |
 | **O4** | phase-space conjecture | N–L §3 (NULL where the dynamics don't reach) | B |
@@ -135,6 +148,9 @@ O2 remains the normalization gate: if the simulated field does not give $\hbar/2
 - The published long-duration 3-D studies report self-ionization, including later relativistic and renormalized-noise attempts. BlueberryCircus does not claim to reproduce those physical timescales with its accelerated fixture.
 - Nieuwenhuizen rectification is a conditional drift in energy space. It does not establish equilibrium vacuum-energy extraction, usable net work, or evasion of detailed balance; broken symmetry alone is insufficient.
 - Setterfield co-scaling is represented as a speculative static hypothesis. Its invariants and trajectory conjugacy show that the static profile is dynamically inert after time reparameterization.
+- A time-dependent Setterfield profile is an externally driven variable-mass Hamiltonian; any apparent suppression with nonzero parameter work is classified `ACTIVE_CONTROL`.
+- The finite-shell and multipole arms are response/surrogate models. They are not validated electron or proton structure models. The stochastic tournament is perturbative and does not replace a physical-timescale nonlinear ensemble.
+- Direct quadrature of Nieuwenhuizen's printed inverse-square kernel gives $d_c\approx-53.69$, while the paper's prose quotes −35.8 from its endpoint. Both are retained explicitly; neither establishes a physical inverse-square force.
 - Matching the quantum 1s density remains a CPU-day frontier, marked strict-`xfail` rather than faked.
 - **The orbit is chaotic.** The code is deterministic and byte-reproducible on a fixed machine, but quantities like `r_max` depend on floating-point summation order and shift across machines. The certified quantities, meaning the conservation laws and the tolerance-gated checks, are stable. Raw chaotic outputs are not, and shouldn't be quoted as if they were.
 - Non-relativistic, with dipole and point-charge approximations and a finite, band-limited background field. Each run is faithful only out to bounded times.
@@ -151,25 +167,28 @@ Framework and code © Joe Pecoraro / Nanarch Technologies, Inc., Apache-2.0. The
 - B. Setterfield, *ZPE and Atomic Constants’ Behavior*, [behaviorzpe3.html](https://www.barrysetterfield.org/behaviorzpe3.html) (speculative scaling proposal, tested here as a hypothesis only).
 - T. M. Nieuwenhuizen, *Stochastic Electrodynamics: Renormalized Noise in the Hydrogen Ground-State Problem*, **Front. Phys. 8, 335 (2020)**, [doi:10.3389/fphy.2020.00335](https://doi.org/10.3389/fphy.2020.00335).
 - G. Moddel & O. Dmitriyeva, *Extraction of Zero-Point Energy from the Vacuum*, [arXiv:0910.5893](https://arxiv.org/abs/0910.5893) (equilibrium/detailed-balance assessment).
+- J. A. E. Rodríguez, extended-charge motivation, [arXiv:1201.6168](https://arxiv.org/abs/1201.6168) (the implemented multipole oscillator is a clearly labeled surrogate).
 
 The certificate layer is vendored inside the package (`blueberry_circus/_vendor/nanarch_certify`, a near-verbatim mirror of Nanarch's canonical copy), so `import blueberry_circus` works from a fresh checkout. See [`PROVENANCE.md`](PROVENANCE.md) and [`NOTICE`](NOTICE).
 
 ## Run it
 
 ```bash
-pytest                                     # core suite: 112 passed, 2 xfailed, 0 skipped
+pytest                                     # core suite: 134 passed, 2 xfailed, 0 skipped
 sh scripts/build_rust.sh && pytest -m rust # optional: Rust backend cross-language tests
 pip install ".[jax]" && pytest -m jax      # optional: JAX backend tests
 pytest -m verify                           # optional: needs BLUEBERRY_VERIFY_BIN
 python examples/demo_sho_ground_state.py   # certified spring ground state
 python examples/demo_vacuum_covariance.py  # full vacuum covariance certificate
 python examples/demo_rectification.py      # certified Puthoff + O5 threshold
 python examples/demo_hydrogen_coulomb.py   # accelerated Coulomb stress fixture
+python examples/demo_hypothesis_tournament.py # reduced schema/ledger smoke
+blueberry-tournament --profile preregistered --arm all # manifest only
 ```
 
 <div align="center">
 
-**[docs/STATUS.md](docs/STATUS.md)** · **[docs/theory.md](docs/theory.md)** · **[docs/comparison.md](docs/comparison.md)**
+**[docs/STATUS.md](docs/STATUS.md)** · **[docs/theory.md](docs/theory.md)** · **[docs/tournament.md](docs/tournament.md)** · **[docs/comparison.md](docs/comparison.md)**
 
 *Nanarch Technologies — Photonic & Quantum Intelligence Systems*
 

REPRODUCIBILITY.md

@@ -20,6 +20,8 @@ python3 examples/demo_sho_ground_state.py     # -> out/sho_ground_state_certific
 python3 examples/demo_vacuum_covariance.py    # -> out/vacuum_covariance_certificate.json
 python3 examples/demo_rectification.py         # -> out/rectification_certificate.json
 python3 examples/demo_hydrogen_coulomb.py     # -> out/hydrogen_certificate.json, out/hydrogen_radial.npz
+python3 examples/demo_hypothesis_tournament.py # -> out/hypothesis_tournament_smoke.json
+blueberry-tournament --profile preregistered --arm all # manifest only; no computation
 ```
 
 Tests set `pythonpath = ["src"]` via `pyproject.toml`; `conftest.py` and the
@@ -29,7 +31,7 @@ without installation.
 ## Verified result (2026-08-13, macOS workstation: CPython 3.14.2)
 
 ```
-pytest             112 passed, 2 xfailed, 0 skipped    (core, default)
+pytest             134 passed, 2 xfailed, 0 skipped    (core, default)
 pytest -m rust       4 passed                          (Rust cdylib built)
 pytest -m jax        4 passed                          (JAX installed)
 pytest -m verify     3 skipped                         (no BLUEBERRY_VERIFY_BIN on this run)
@@ -53,6 +55,9 @@ pytest -m verify     3 skipped                         (no BLUEBERRY_VERIFY_BIN
 | Puthoff circular `P_abs` vs `P_rad` | **<1×10⁻¹²** | analytic balance |
 | Nieuwenhuizen improper quadrature vs `16/(5π√3)` | **9.5×10⁻¹³ absolute** | Eq. (2.30) vs Eq. (2.31) |
 | Setterfield `U=1` vs `U=4` mapped trajectory | **<1×10⁻⁹** | static conjugacy |
+| full point-charge surface | **18 finite cells** plus both analytic endpoints | Nieuwenhuizen Eq. (2.34) |
+| inverse-square printed-kernel maximum | **`Hmax≈7.327`, `dc≈-53.69`** | transformed quadrature, orders 48/64 |
+| tournament random-phase convergence | **32 stored seeds; 2,048→4,096 modes** | normalized change `<10%` or `NULL` |
 
 ## Caveats (house honesty policy)
 
@@ -66,3 +71,11 @@ pytest -m verify     3 skipped                         (no BLUEBERRY_VERIFY_BIN
   ensemble averaging over seeds and longer integration. The *linear
   ground state itself* is nonetheless certified exactly via the quadrature oracle
   (C2), which needs no time integration.
+- Tournament JSON is deterministic and authoritative: it stores the complete
+  config, seeds, resolutions, confidence intervals, and ledgers. Optional NPZ
+  arrays are redundant. A full run must be requested explicitly with
+  `blueberry-tournament --profile preregistered ... --execute`; ordinary CI
+  executes only the reduced schema/ledger smoke.
+- The tournament's stochastic layer samples the perturbative quadratic response
+  kernel. It is not a long-time nonlinear trajectory and cannot be cited as a
+  reproduction of physical hydrogen self-ionization or stabilization.

docs/STATUS.md

@@ -1,14 +1,14 @@
 # BlueberryCircus verification status
 
-v0.2.0. Verified on a macOS workstation (CPython 3.14.2; exact dependency and
+v0.3.0. Verified on a macOS workstation (CPython 3.14.2; exact dependency and
 tool versions are printed by the release gate) on 2026-08-13. Pass counts are
 environment-specific; reproduce locally before citing them.
 
 ## Suite
 
 | Command | Result | Requires |
 |---|---|---|
-| `pytest` (default = core) | **112 passed · 2 xfailed (strict) · 0 skipped** | numpy only |
+| `pytest` (default = core) | **134 passed · 2 xfailed (strict) · 0 skipped** | numpy only |
 | `pytest -m rust` | 4 passed | `sh scripts/build_rust.sh` |
 | `pytest -m jax` | 4 passed | `pip install ".[jax]"` |
 | `pytest -m verify` | 3 passed with a verifier binary; skips without | `BLUEBERRY_VERIFY_BIN` |
@@ -51,6 +51,29 @@ inertness, not the empirical scaling proposal.
 physical Bohr value and sustained positive energy begins at `0.06956` orbit.
 It is retained only as an accelerated numerical stress test.
 
+## Hypothesis tournament (`tournament.py`)
+
+The baseline report evaluates Nieuwenhuizen's complete finite-energy
+point-charge surface over 18 preregistered `(E,L)` cells. It recovers the
+near-ionization sign change, agrees with the PR1 asymptote within one percent at
+the independently evaluated `E=-0.001` probe, and recovers the exact circular
+endpoint. The default stochastic
+gate executes 32 stored seeds at 2,048 and 4,096 modes, halving the nominal
+timestep and returning `NULL` when normalized drift changes by ten percent or
+more.
+
+All four zero-parameter arms recover the identical baseline cell. The driven
+Setterfield Hamiltonian records `partial H/partial t` as external work; the
+finite shell applies one reciprocal form factor to both powers; the
+inverse-square arm integrates the printed source kernel; and the
+Rodríguez-inspired multipole surrogate conserves its closed total Hamiltonian.
+None can emit `STABLE_GROUND_STATE`.
+
+The inverse-square audit is a new negative reproducibility result. Orders 48
+and 64 give `H_max≈7.327` near `mu≈0.590`, hence the defining
+`d_c=-H_max²≈-53.69`. The paper's prose `-35.8` squares only its quoted
+`H(0)=5.99`, despite stating an all-`mu` criterion. Reports retain both values.
+
 **Cross-language enclosure:** the Rust `cdylib` backend is **bit-identical** to
 numpy on the SHO case (max|Δx| = 0.0) and agrees to ~7e-14 on the Kepler orbit;
 the JAX backend agrees to ≤1e-6.