Standing Rules for the CMSTG Repository

These rules govern all simulation, paper, and commit work. They derive from Phase 1, Phase 2, and Phase 3 experience and are not negotiable without explicit discussion.

1. Project context

This is the CMSTG (Curvature-Memory Scalar-Tensor Gravity) simulation repository, maintained by Christopher R. B. Wilson. CMSTG is a scalar-tensor extension of GR with a non-minimal coupling Λ(Ψ)R and a retarded memory kernel. The Phase 1 canonical action is locked:

Action: S = ∫d⁴x √-g [(1/16πG + Λ(Ψ))R + ½(∇Ψ)² − ½m(Ψ)²Ψ² − V(Ψ)] + S_matter
Λ₀ = 0.003 (locked)
Ψ̄ = 2.62 M_Pl
F_eff = (1 + 2Λ₀Ψ̄²)/2 = 0.521
Phase 1 passes: BAO, CMB TT/EE/TE, RSD, GW speed, Solar System, UV finiteness through two loops
Phase 1 residual: 2.77σ tension with DESI Y1 BAO (structural)

Phase 2 (SIM112–130) exhausted matter-sector extensions. Phase 3 (SIM131–136) exhausted curvature-sourced scalar extensions and produced a structural no-go theorem. Phase 4 (SIM137 onward) is the current programme.

2. Core theoretical discipline

2.1 Foundations before phases

Do not advance to later simulation phases before the earlier phase is closed. If a diagnostic sim (Tier 1) changes what the mechanism sim (Tier 2) should target, update the spec before running.

2.2 Derived vs. phenomenological

Always distinguish:

Derived results: follow from variation of the Lagrangian
Phenomenological parametrizations: functional forms imposed by hand (e.g., f_ψ(a), running Λ₀(a))

Flag phenomenological inputs explicitly in sim output JSON and in any paper text. Do not let phenomenological fits be cited later as derivations.

2.3 Every new Lagrangian term must be checked against

Before any mechanism simulation reports a PASS, verify:

GR recovery: Λ(0) = 0, V(0) = 0, G_eff > 0
Graviton sector: c_T = c (GW170817 bound); m_g = 0 (Ward identity Π_hh(0) = 0)
No tachyon: ω² = k² + m_eff² > 0 for all physical modes
UV behaviour: no new quartic divergences that the memory kernel does not suppress
Causality: retarded Green's function structure preserved

A sim that passes observational criteria but fails any of the above is a FAIL, not a PASS.

2.4 The Sim 8 lesson

V(Ψ) = αΨ² exp(−βΨ²) did not produce monotone decaying halo solutions without a matter source term. The lesson: potential shape alone does not guarantee the desired field profile. Check that the EOM actually admits the solution class you're looking for before scanning parameters.

3. Simulation conventions

3.1 Directory structure

cmstg/
├── RESEARCH_RULES.md            # this file
├── PHASE4_ROADMAP.md            # current roadmap
├── sims/
│   ├── SIM137/
│   │   ├── SPEC.md              # pre-run: action, criteria
│   │   ├── run.py               # executable simulation
│   │   ├── output.json          # numerical results
│   │   ├── RESULT.md            # post-run: verdict, diagnosis
│   │   └── figures/
│   ├── SIM138/
│   └── ...
├── papers/
└── shared/                      # common utilities, solvers

3.2 Output JSON schema

Every sim must produce output.json with at minimum:

{
  "sim_id": "SIM137",
  "timestamp": "ISO8601",
  "action_spec": "Phase 1 canonical" | "<full Lagrangian description>",
  "parameters": { ... },
  "observational_targets": { "dataset": "...", "chi2": ..., "dof": ... },
  "theoretical_checks": {
    "gr_recovery": true,
    "c_T_eq_c": true,
    "no_tachyon": true,
    "ward_identity": true,
    "uv_finite": true
  },
  "verdict": "PASS" | "PARTIAL" | "FAIL",
  "failure_mode": "<structural diagnosis if FAIL or PARTIAL>",
  "derived_vs_phenom": { "<param>": "derived" | "phenomenological" }
}

3.3 Success criteria discipline

Each sim spec states explicit, verifiable success criteria. Loop until all are verified or until a structural obstruction is diagnosed. Do not report PASS on partial evidence.

3.4 Commit after every sim

After each sim completes (PASS, PARTIAL, or FAIL), immediately:

Commit the SIM<N>/ directory with output.json, RESULT.md, figures
Update PHASE4_ROADMAP.md with the verdict and any scope changes
Push to GitHub before starting the next sim

This preserves the paper trail that made Phase 2 and Phase 3 documents strong.

4. Coding style

4.1 Simplicity first

Minimum code that solves the problem
No speculative features, no abstractions for single-use code
Match existing style in shared/ rather than refactoring
If a sim script exceeds 300 lines, factor the common numerical machinery into shared/

4.2 Numerical verification

Every new solver: compare against a known analytic limit
Every new likelihood: verify against the Phase 1 canonical parameters reproduce SIM90/SIM98 χ² within 0.01
Every new ODE integrator: convergence test with step size halving

4.3 External dependencies

Prefer CLASS for Boltzmann work (used in Phase 1)
Prefer clipy for Planck plikHM likelihood (used in SIM97/98)
Use full covariance matrices for BAO (BOSS+eBOSS+Lyα, DESI Y1)
When in doubt about a citation or numerical value, verify with a web search before using it

5. Writing and papers

5.1 When a phase closes, write the conclusions paper

Follow the structure of Papers VI and VII:

Introduction + motivation
Per-sim results with actions, parameters, outcomes
Summary table with verdicts and Δχ²
Structural diagnosis (no-go theorem if applicable)
Directions for the next phase
Canonical parameter set if changed

5.2 No overreach

"CMSTG" is a scalar-tensor extension of GR with a retarded memory kernel. Not "modified gravity," not a TOE candidate in the paper text.
Speculative material (e.g., SM coupling appendix in the master paper) must be explicitly labeled speculative.
The Bayesian evidence is ln B = -0.71 (Inconclusive). Do not claim CMSTG is preferred over ΛCDM; claim it is not excluded.

5.3 Publication strategy

Zenodo with CC BY 4.0 for timestamping and priority protection
Peer review venues: Phys. Rev. D, JCAP for cosmology sims; Class. Quantum Grav. for the canonical action paper

FilesExpand file tree

RESEARCH_RULES.md

Latest commit

History