#!/usr/bin/env python3 """Rank capability gaps as research projects (item #9 follow-on). Reads models/capability_gap_roadmap.jsonl (the (gap_type, theme) groups built by build_gap_roadmap.py) and merges groups belonging to the same capability theme into one research project: one concrete deliverable that would close the theme's tool + likelihood + derivation gaps together. Each project is scored by: impact = sum over distinct blocked models of adjusted_priority_score x severity multiplier (a model counts once, at its worst-severity gap in the theme) impact_sf = same, counting only gap entries that block structure-formation observables cost = mean effort of the member gap entries, on the scale small=1, medium=3, large=10, major_program=30 (unknown=3) — rough person-months of the typical closing work in the theme (the median is degenerate: almost every theme's median entry is 'medium') value_ratio = impact / cost ("science unblocked per unit effort") Emits models/research_projects_ranked.jsonl and the human-readable report ../../research_project_ranking.md (project root), with three orderings: by impact, by value ratio, and by structure-formation impact. """ import json import statistics from collections import defaultdict from pathlib import Path REGISTRY = Path(__file__).resolve().parent MODELS = REGISTRY / "models" IN_JSONL = MODELS / "capability_gap_roadmap.jsonl" OUT_JSONL = MODELS / "research_projects_ranked.jsonl" OUT_MD = REGISTRY.parent.parent / "research_project_ranking.md" SEV_MULT = {"blocks_all_observables": 1.0, "blocks_key_observables": 0.8, "degrades_precision": 0.4, "minor": 0.1, "unknown": 0.3} EFFORT_COST = {"small": 1, "medium": 3, "large": 10, "major_program": 30, "unknown": 3} # theme -> (project title, deliverable statement) PROJECTS = { "Lyman-α forest modeling & likelihood": ( "A reusable Lyman-α forest likelihood for non-CDM cosmologies", "A public likelihood package that takes an arbitrary suppressed or " "oscillatory linear power spectrum (WDM-like, interacting, mixed, " "fuzzy) and returns constraints from eBOSS/DESI and high-resolution " "flux-power data, with an emulator-based nonlinear mapping so users " "never rerun hydrodynamical simulations."), "Boltzmann-code extension (CLASS/CAMB/ETHOS)": ( "A maintained general non-CDM Boltzmann-code framework", "A supported CLASS/ETHOS-style branch implementing interacting, " "decaying, warm, mixed and self-interacting species behind one " "interface, with standardized transfer-function output that " "downstream structure-formation pipelines can consume."), "Relic-abundance solver extension (micrOMEGAs/MadDM/DarkSUSY)": ( "A community model-implementation library for relic-abundance solvers", "A curated, validated library of FeynRules/CalcHEP model files plus " "solver extensions (freeze-in, co-annihilation, resonances, " "non-standard histories) so registry models get relic densities from " "micrOMEGAs/MadDM instead of bespoke by-hand estimates."), "Indirect-detection likelihood (γ-ray / cosmic-ray / ν)": ( "A unified indirect-detection likelihood suite", "One package wrapping Fermi dwarf-spheroidal, Galactic-center, " "AMS-02 antiproton/positron and neutrino-telescope likelihoods with " "user-supplied annihilation/decay spectra, replacing per-paper " "digitized limits."), "Direct-detection recast & likelihood": ( "A direct-detection recast framework beyond the standard WIMP", "Recast machinery for XENONnT/LZ/CRESST-class results covering " "sub-GeV kinematics, non-standard form factors, inelastic and " "boosted scenarios, exposing proper likelihoods rather than " "published 90% curves."), "Dwarf-galaxy dynamics & density profiles": ( "A dwarf-galaxy structure likelihood", "A likelihood connecting model predictions for inner density " "profiles, dynamical heating and star-cluster survival in dwarfs to " "stellar-kinematics datasets, with marginalization over baryonic " "feedback."), "Milky Way satellite / subhalo-count likelihood": ( "A subhalo-abundance-to-satellite-counts pipeline", "A pipeline from a suppressed subhalo mass function to the observed " "Milky Way satellite luminosity function (DES/Rubin selection " "functions included), packaged as a likelihood any transfer " "function can be pushed through."), "Axion / fuzzy-DM field solver": ( "A public wave-dark-matter solver suite", "Maintained Schrödinger–Poisson / axion-field solvers with soliton, " "minicluster and oscillon modules, calibrated scaling relations, and " "hooks to structure-formation observables."), "N-body / hydrodynamic simulation campaign": ( "A targeted non-CDM simulation campaign with public halo products", "A coordinated N-body/hydro campaign spanning the registry's " "recurring non-CDM physics (velocity-dependent SIDM, decays, " "late-forming DM), releasing halo catalogs and calibrations others " "can build likelihoods on."), "Sterile-neutrino production solver": ( "An extended sterile-neutrino production code", "A quantum-kinetic production solver going beyond " "Dodelson–Widrow/Shi–Fuller: new mediators, self-interactions, and " "non-standard expansion histories, emitting phase-space " "distributions ready for Boltzmann codes."), "PBH formation & abundance pipeline": ( "An end-to-end primordial-black-hole pipeline", "One pipeline from a primordial P(k) (or collapse mechanism) to an " "extended PBH mass function and current constraints (microlensing, " "CMB accretion, GW rates), replacing monochromatic by-hand " "estimates."), "PTA / stochastic GW background likelihood": ( "A modern PTA stochastic-background likelihood for dark sectors", "NANOGrav-15yr/EPTA-class likelihoods packaged for dark-sector " "sources (strings, phase transitions, superradiance) so cards stop " "relying on pre-2017 exclusions."), "GW detector forecasts & likelihoods (LIGO/LISA/ET)": ( "Interferometer forecasts and likelihoods for DM-sector signals", "Shared LIGO/LISA/ET sensitivity and event-rate machinery for " "dark-sector mergers, backgrounds and continuous waves."), "CMB likelihood & ΔN_eff machinery": ( "A CMB energy-injection and ΔN_eff constraint toolkit", "Standardized Planck(+ACT/SPT) likelihood wrappers for energy " "injection, annihilation/decay, and ΔN_eff from dark radiation, " "consumable without a full cosmology-pipeline setup."), "BBN code extension": ( "A BBN code extension for dark-sector energy injection", "PRIMAT/AlterBBN-class extensions handling hadronic/electromagnetic " "injection, non-standard expansion and MeV-scale sectors, with a " "packaged likelihood."), "Collider / beam-dump recast": ( "A collider and beam-dump recast library for DM benchmarks", "Recast implementations (LHC, Belle II, fixed-target/beam-dump) of " "the registry's recurring signatures, with acceptances and " "likelihoods, in a CheckMATE/Contur-style framework."), "Dark-photon / light-mediator constraint recast": ( "A light-mediator constraint compilation with proper likelihoods", "A maintained, recastable compilation of dark-photon/fifth-force/" "light-scalar constraints (beyond static exclusion plots) mapped to " "arbitrary coupling structures."), "Self-interacting DM halo modeling": ( "A velocity-dependent SIDM halo-modeling kit", "Calibrated halo-level models (gravothermal evolution, core " "formation/collapse) for velocity-dependent cross sections, plus " "the likelihood connecting sigma(v)/m to rotation curves, clusters " "and dwarfs."), "Stellar-stream perturbation likelihood": ( "A stellar-stream perturbation likelihood", "A likelihood from subhalo population predictions to observed " "stream density/track perturbations (GD-1, Pal 5, Rubin-era " "samples), usable by any model that predicts a subhalo mass " "function."), "Strong-lensing substructure likelihood": ( "A strong-lensing substructure likelihood", "Flux-ratio and gravitational-imaging likelihoods for suppressed or " "enhanced substructure, packaged for arbitrary transfer functions."), "21-cm signal modeling & likelihood": ( "A 21-cm signal modeling and likelihood package for non-CDM", "21cmFAST-class modeling plus EDGES/HERA/SKA likelihoods for " "models altering the timing of structure formation or injecting " "energy."), "Nonlinear-structure emulator": ( "Nonlinear-structure emulators for non-CDM cosmologies", "Emulators mapping modified linear spectra to nonlinear power, " "halo mass functions and profiles, trained on the simulation " "campaign above."), "Phase-transition & defect-network computation": ( "A dark-sector phase-transition and defect computation toolkit", "CosmoTransitions-class tooling extended to the registry's dark " "sectors (nucleation, GW spectra, string/wall networks) with " "validated defaults."), "CMB spectral distortions": ( "A spectral-distortion prediction and constraint module", "mu/y-distortion computation for dark-sector energy release with " "FIRAS(+PIXIE-forecast) likelihoods."), "Isocurvature machinery": ( "An isocurvature constraint pipeline", "From a model's isocurvature production mechanism to Planck " "isocurvature likelihoods, for axion-like and multi-field " "scenarios."), "Compact-object capture & heating": ( "A compact-object capture and heating calculator", "A public calculator for DM capture, thermalization and heating in " "neutron stars/white dwarfs, with observational comparison to " "cooling data."), } def main() -> None: groups = [json.loads(l) for l in IN_JSONL.open()] by_theme = defaultdict(list) for g in groups: by_theme[g["theme"]].append(g) projects = [] skipped_unclassified = 0 for theme, grps in by_theme.items(): if theme not in PROJECTS: skipped_unclassified += sum(g["n_gaps"] for g in grps) continue title, deliverable = PROJECTS[theme] members = [m for g in grps for m in g["members"]] gap_types = sorted({g["gap_type"] for g in grps}) # one contribution per model: worst-severity gap wins best = {} best_sf = {} for m in members: mult = SEV_MULT.get(m["severity"], 0.3) contrib = m["score"] * mult mid = m["model_id"] if contrib > best.get(mid, (-1, None))[0]: best[mid] = (contrib, m) if m["blocks_structure_formation"] and \ contrib > best_sf.get(mid, (-1, None))[0]: best_sf[mid] = (contrib, m) impact = sum(c for c, _ in best.values()) impact_sf = sum(c for c, _ in best_sf.values()) costs = [EFFORT_COST.get(m["estimated_effort"], 3) for m in members] cost = round(statistics.mean(costs), 1) eff = defaultdict(int) sev = defaultdict(int) for m in members: eff[m["estimated_effort"]] += 1 sev[m["severity"]] += 1 top_models = sorted((mm for _, mm in best.values()), key=lambda mm: -mm["score"]) projects.append({ "project": title, "theme": theme, "deliverable": deliverable, "gap_types": gap_types, "impact": round(impact, 1), "impact_structure_formation": round(impact_sf, 1), "cost": cost, "value_ratio": round(impact / cost, 1), "n_models": len(best), "n_models_structure_formation": len(best_sf), "n_gaps": len(members), "effort_counts": dict(eff), "severity_counts": dict(sev), "top_models": [{ "model_id": mm["model_id"], "model_name": mm["model_name"], "arxiv": mm["arxiv"], "score": mm["score"], "severity": mm["severity"], "description": mm["description"], "proposed_resolution": mm["proposed_resolution"], } for mm in top_models[:10]], }) projects.sort(key=lambda p: -p["impact"]) for i, p in enumerate(projects, 1): p["rank_impact"] = i for i, p in enumerate(sorted(projects, key=lambda p: -p["value_ratio"]), 1): p["rank_value"] = i for i, p in enumerate(sorted(projects, key=lambda p: -p["impact_structure_formation"]), 1): p["rank_structure_formation"] = i with OUT_JSONL.open("w") as f: for p in projects: f.write(json.dumps(p, ensure_ascii=False) + "\n") print(f"wrote {OUT_JSONL}: {len(projects)} projects " f"({skipped_unclassified} unclassified gaps excluded)") # ---- markdown report -------------------------------------------------- def table(rows, key, sf=False): lines = ["| # | project | impact | SF impact | models | cost | " "value ratio |", "|---|---------|-------:|----------:|-------:|-----:|" "------------:|"] for p in rows: lines.append( f"| {p[key]} | {p['project']} | {p['impact']:,.0f} | " f"{p['impact_structure_formation']:,.0f} | {p['n_models']} | " f"{p['cost']:g} | {p['value_ratio']:,.0f} |") return "\n".join(lines) md = ["# Research-project ranking from the capability-gap registry", "", "Each project is one capability theme from the gap roadmap, " "phrased as a concrete deliverable that would close its tool, " "likelihood and derivation gaps together. Scores:", "", "- **impact** — summed `adjusted_priority_score` of the distinct " "models the project unblocks, each weighted by how badly it is " "blocked (blocks_all_observables 1.0, blocks_key 0.8, " "degrades_precision 0.4, minor 0.1, unknown 0.3).", "- **SF impact** — the same sum counting only gap entries that " "block structure-formation observables.", "- **cost** — mean effort of the member gap entries " "(small=1, medium=3, large=10, major_program=30; rough " "person-months of the typical closing work in the theme).", "- **value ratio** — impact / cost: science unblocked per unit " "effort.", "", "Machine-readable version with full member lists: " "`registry/models/research_projects_ranked.jsonl`; underlying " "groups: `registry/models/capability_gap_roadmap.jsonl`.", "", "## Ranking by impact", "", table(projects, "rank_impact"), "", "## Ranking by value ratio (impact per unit effort)", "", table(sorted(projects, key=lambda p: -p["value_ratio"]), "rank_value"), "", "## Ranking by structure-formation impact", "", table(sorted(projects, key=lambda p: -p["impact_structure_formation"]), "rank_structure_formation"), "", "## Project descriptions (impact order)", ""] for p in projects: md += [f"### {p['rank_impact']}. {p['project']}", "", f"**Deliverable.** {p['deliverable']}", "", f"Impact {p['impact']:,.0f} (structure formation " f"{p['impact_structure_formation']:,.0f}) · unblocks " f"{p['n_models']} models ({p['n_models_structure_formation']} " f"via structure-formation observables) · {p['n_gaps']} gap " f"entries ({', '.join(p['gap_types'])}) · typical effort " f"{max(p['effort_counts'], key=p['effort_counts'].get)}.", "", "Highest-priority blocked models:", ""] for mm in p["top_models"][:3]: md.append(f"- **{mm['model_name']}** " f"([arXiv:{mm['arxiv']}](https://arxiv.org/abs/" f"{mm['arxiv']}), score {mm['score']:.0f}, " f"{mm['severity']}): {mm['description']}") md.append("") OUT_MD.write_text("\n".join(md)) print(f"wrote {OUT_MD}") print("\nTop 10 by impact:") for p in projects[:10]: print(f" {p['rank_impact']:2d}. {p['project']} — impact " f"{p['impact']:,.0f}, SF {p['impact_structure_formation']:,.0f}," f" models {p['n_models']}, cost {p['cost']:g}, value " f"{p['value_ratio']:,.0f}") print("Top 5 by value ratio:") for p in sorted(projects, key=lambda p: -p["value_ratio"])[:5]: print(f" {p['rank_value']:2d}. {p['project']} — value " f"{p['value_ratio']:,.0f} (impact {p['impact']:,.0f}, cost " f"{p['cost']:g})") print("Top 5 by SF impact:") for p in sorted(projects, key=lambda p: -p["impact_structure_formation"])[:5]: print(f" {p['rank_structure_formation']:2d}. {p['project']} — SF " f"{p['impact_structure_formation']:,.0f}") if __name__ == "__main__": main()