Energy-Flow Cosmology (EFC) — Likelihood Ledger

Pitch Validation Likelihood Evaluation Models White Paper Roadmap Gaps External Predictions Atlas Changelog

Morten Magnusson · Symbiose Research, Sandnes, Norway · ORCID: 0009-0002-4860-5095 · April 2026 · CC-BY-4.0

Status: This ledger closes the Dataset → Likelihood gap identified in the 2026-04-23 system audit. Every empirical test in the Validation Ledger must declare here which likelihood, which covariance, which priors, which code. Tests without a Likelihood Ledger row are not eligible for inclusion in the Evaluation Ledger global score. Scope today: schema frozen, worked example included, bulk population pending pipeline declaration by maintainer.

1. Purpose

The Validation Ledger records what was tested and how it scored. The Likelihood Ledger records how the score was computed. Without this layer, two identically-named tests in different pipelines (e.g. direct chi² vs full Cobaya posterior) can produce different verdicts — the system becomes non-reproducible.

Every row here binds a test_id from the Validation Ledger to a fully declared pipeline: likelihood type, covariance source, parameter priors, code, entry point, inputs, outputs, and execution status.

2. Schema (summary)

Field	Meaning	Required
`test_id`	Foreign key into `validation-ledger/data/tests.json`	yes
`likelihood_type`	`gaussian` · `gaussian_cov` · `binned_chi2` · `mcmc_posterior` · `analytic` · `external`	yes
`covariance`	Path or DOI of covariance matrix; `diagonal` if independent errors	yes
`parameter_priors`	dict of parameter → {type, range / mean ± sigma / ref}	yes
`code`	`cobaya` · `mgcamb` · `cosmoSIS` · `direct_python` · `external`	yes
`code_version`	Git SHA / package version / external tag	recommended
`entry_point`	Repo-relative script or external URL	yes when code is not external
`inputs`	Dataset DOIs / repo refs	yes
`outputs`	Subset of: `chi2`, `best_fit`, `posterior_summary`, `evidence`, `log_likelihood`, `predictive_samples`	yes
`status`	declared runnable executed frozen	yes

Full JSON Schema: docs/likelihood-ledger/schema.json.

3. Pipeline registry (initial scan)

Status as of ledger creation. Entries marked GAP have no pipeline declared and block their Validation Ledger counterparts from the Evaluation Ledger.

Category	Likelihood type	Code	Entry point	Status
BAO (DESI, eBOSS)	`gaussian_cov`	`direct_python`	`reproduce_bao.py`	runnable
CMB sanity	`gaussian`	`direct_python`	`reproduce_cmb_sanity.py`	runnable
SPARC rotation curves	`binned_chi2`	`direct_python`	`reproduce_sparc.py`	runnable
EFC integration (background)	`analytic`	`direct_python`	`reproduce_efc.py`	runnable
fσ₈ / S₈ (growth)	`gaussian_cov`	TBD (Cobaya likely)	TBD	GAP
Cosmic shear (DES, KiDS, Euclid)	`mcmc_posterior`	`cobaya` + `mgcamb`	TBD	GAP
ISW cross-correlation	`gaussian`	`direct_python`	TBD	GAP
GW propagation (KC6)	`gaussian`	`direct_python`	TBD	GAP
Full CMB likelihood (plik_lite TTTEEE)	`mcmc_posterior`	`cobaya`	TBD	GAP

4. Worked example (template)

Template entry: fσ₈ growth test declared

{
  "test_id": "s8_growth_fsigma8_2026",
  "likelihood_type": "gaussian_cov",
  "covariance": "data/cov/fsigma8_compilation_2026.json",
  "parameter_priors": {
    "Omega_m": {"type": "uniform", "range": [0.20, 0.40]},
    "sigma8":  {"type": "uniform", "range": [0.70, 0.90]},
    "mu_efc":  {"type": "uniform", "range": [0.80, 1.20]}
  },
  "code": "direct_python",
  "code_version": "efc@<git-sha>",
  "entry_point": "pipelines/fsigma8_runner.py",
  "inputs": [
    "doi:10.xxxx/desi-y3-fsigma8",
    "doi:10.xxxx/des-y6-3x2pt"
  ],
  "outputs": ["chi2", "best_fit", "posterior_summary"],
  "status": "declared"
}

Once a run produces numbers, status advances to executed and the result is recorded in the Validation Ledger. If the declaration is then frozen for publication, it advances to frozen.

5. Linking to the rest of the stack

Ledger	Role	Direction
Validation Ledger	Defines `test_id` vocabulary	Likelihood Ledger references it (FK)
Evaluation Ledger	Global score function, falsification verdict	Consumes Likelihood Ledger rows with `status ∈ {executed, frozen}`
Model Comparison	EFC vs ΛCDM vs MG on identical likelihood	Consumes Likelihood Ledger row via `likelihood_id`
Gap Analysis	Tracks missing pipelines	Every GAP row here should appear as a gap row there
Stage-IV Roadmap	Maps survey releases to tests	Each roadmap test should eventually gain a Likelihood Ledger row

6. Sync hook (proposed, not yet written)

scripts/maintenance/efc_likelihood_sync.py mirrors the pattern of efc_ledger_impact_sync.py:

Reads every test_id from validation-ledger/data/tests.json.
For each vl_category ∈ {physics_test, phenomenological} without a likelihood row, emits a warning.
Validates that inputs DOIs exist in validation-ledger/data/evidence-register.json.
Idempotent — re-runs are no-ops.

7. Open questions

Should likelihood entries live as separate JSON files per test, or as array entries inside data/likelihoods.json? Proposal: single-array file, mirroring tests.json.
How are external likelihood codes (Cobaya, MGCAMB) version-pinned? Proposal: Git SHA in code_version; conda env export as supplementary artefact.
Should parameter_priors inline ranges, or reference validation-ledger/data/parameters.json? Proposal: reference by default, allow inline only when a test uses a non-standard prior with documented rationale.

Bottom line

What this ledger fixes	Makes every empirical score reproducible by declaring its pipeline explicitly. No test enters the global score without a row here.
Hard dependency	Validation Ledger — provides the `test_id` vocabulary every row references.
Hard consumers	Evaluation Ledger and Model Comparison — neither can fire without executed likelihood rows.
Current gaps	fσ₈ / S₈, cosmic shear, ISW, GW propagation, full CMB likelihood. All marked GAP until a pipeline is wired.
Status	Schema frozen; pipeline registry populated with all current `reproduce_*.py` scripts; bulk population of new pipelines follows maintainer review.