{
  "asOf": "2026-09-12",
  "version": "Initial planning model v1",
  "currency": "USD",
  "scientistRate": 100,
  "reviewRate": 150,
  "inputRate": 3,
  "outputRate": 15,
  "contingency": 20,
  "marketPool": 3200,
  "programs": [
    {
      "id": "transitions",
      "name": "Repeated anatomical changes",
      "status": "First priority",
      "weeks": "4\u20136",
      "hours": 120,
      "review": 24,
      "inputM": 40,
      "outputM": 8,
      "compute": 180,
      "other": 100,
      "p": 30,
      "prange": [
        15,
        45
      ],
      "commercial": 20,
      "buyers": [
        160,
        320,
        640
      ],
      "price": [
        25000,
        50000,
        100000
      ],
      "product": "Comparative target-evidence service for discovery teams",
      "evidence": "Published organ-loss studies establish a foothold. DNALYZER has not yet produced a new association that replicates across independent transitions.",
      "gate": "A previously unreported association survives ancestry and sequence-quality controls and transfers to an entire evolutionary transition withheld from discovery.",
      "stop": "Stop discovery claims if too few independent transitions are usable or the held-out association fails. Recovering a known association validates the workflow, but does not pass the novelty gate.",
      "why": "Repeated natural changes create useful comparisons, but small event counts, missing sequence and reverse causality make novel transfer uncertain.",
      "next": "Independent replication and targeted functional validation",
      "follow": [
        75000,
        250000
      ],
      "link": "/research/structural-transitions/"
    },
    {
      "id": "regulation",
      "name": "Regulatory activity differences",
      "status": "Priority test",
      "weeks": "5\u20138",
      "hours": 160,
      "review": 32,
      "inputM": 60,
      "outputM": 12,
      "compute": 650,
      "other": 150,
      "p": 25,
      "prange": [
        10,
        40
      ],
      "commercial": 25,
      "buyers": [
        160,
        480,
        960
      ],
      "price": [
        25000,
        75000,
        150000
      ],
      "product": "Regulatory variant and enhancer evaluation for biotech R&D",
      "evidence": "Public enhancer experiments provide measured comparisons. Recovering a published example is a data audit, not a new biological experiment.",
      "gate": "At least 10% lower error than the strongest applicable baseline on held-out loci or studies, with the paired 95% interval excluding no improvement and no construct-family leakage.",
      "stop": "Stop if comparable activity-change labels are insufficient or improvement vanishes when related loci and studies are withheld.",
      "why": "Direct measurements help. Public comparison sets are limited, prior models are strong, and recognizing activity is easier than predicting a change.",
      "next": "Independent assay validation in the intended tissue context",
      "follow": [
        100000,
        400000
      ],
      "link": "/research/structural-transitions/"
    },
    {
      "id": "atlas",
      "name": "Genome\u2013phenome evidence atlas",
      "status": "Foundation",
      "weeks": "4\u20136",
      "hours": 120,
      "review": 40,
      "inputM": 100,
      "outputM": 20,
      "compute": 150,
      "other": 100,
      "p": 45,
      "prange": [
        25,
        65
      ],
      "commercial": 25,
      "buyers": [
        320,
        800,
        1600
      ],
      "price": [
        10000,
        25000,
        50000
      ],
      "product": "Evidence-linked anatomy data subscriptions and research integrations",
      "evidence": "14,329 assembly records, 3,014 species profiles and 87,265 source assertions. One complete genome downloaded and verified; coverage remains partial.",
      "gate": "At least 95% precision on a blinded, stratified assertion audit AND at least 10% lower error on a frozen anatomical task versus coarse-label and ancestry controls.",
      "stop": "Stop scaling ingestion if extra description does not improve the frozen task or source fidelity fails. Accurate extraction alone is a narrower success.",
      "why": "An operating corpus makes a useful resource plausible. Richer text may still add no information beyond ancestry or leak the target trait.",
      "next": "Broader source adjudication, licensing review and buyer validation",
      "follow": [
        50000,
        150000
      ],
      "link": "/research/genome-phenome-atlas/"
    },
    {
      "id": "sequence",
      "name": "Sequence \u2192 knockout survival",
      "status": "Partial signal",
      "weeks": "2\u20134",
      "hours": 60,
      "review": 16,
      "inputM": 20,
      "outputM": 4,
      "compute": 100,
      "other": 50,
      "p": 40,
      "prange": [
        20,
        60
      ],
      "commercial": 15,
      "buyers": [
        160,
        320,
        640
      ],
      "price": [
        15000,
        40000,
        80000
      ],
      "product": "Research-only gene prioritization for preclinical teams",
      "evidence": "Historical protein features reached AUROC 0.667 versus 0.525 for length on 1,639 later-measured mouse genes. This is partial ranking, not percent accuracy.",
      "gate": "At least +0.03 AUROC over the strongest applicable sequence baseline under both temporal and homology-family separation, with a positive paired 95% interval and acceptable calibration.",
      "stop": "Stop product claims if family separation removes the lift or errors remain too large for a defined research use.",
      "why": "A measured temporal signal exists, but the earlier split did not exclude all homologous families. The sequence\u2013essentiality relationship is established prior art.",
      "next": "External-cohort replication and workflow utility study",
      "follow": [
        50000,
        200000
      ],
      "link": "#pilot-evidence"
    },
    {
      "id": "integration",
      "name": "Integrated evidence \u2192 survival",
      "status": "Strongest pilot signal",
      "weeks": "3\u20135",
      "hours": 100,
      "review": 24,
      "inputM": 30,
      "outputM": 6,
      "compute": 150,
      "other": 100,
      "p": 45,
      "prange": [
        25,
        65
      ],
      "commercial": 25,
      "buyers": [
        160,
        480,
        960
      ],
      "price": [
        25000,
        60000,
        120000
      ],
      "product": "Evidence integration and prioritization for discovery teams",
      "evidence": "Combined biological features reached AUROC 0.841\u20130.851 versus 0.744\u20130.771 for cell-fitness-only on fixed gene splits. These are a different cohort from the sequence pilot.",
      "gate": "At least +0.03 AUROC over the strongest single-source and published baseline on a genuinely external cohort, with improved calibration and a positive paired 95% interval.",
      "stop": "Stop if date, source or gene-family leakage explains the lift, or the external cohort fails. An internal refit audit is not external biological replication.",
      "why": "The measured integration gain is encouraging. Transfer to independent measurements and superiority to existing tools remain untested.",
      "next": "Prospective research utility and independent cohort validation",
      "follow": [
        75000,
        250000
      ],
      "link": "#pilot-evidence"
    },
    {
      "id": "timing",
      "name": "Developmental failure timing",
      "status": "Rework before scaling",
      "weeks": "3\u20135",
      "hours": 100,
      "review": 32,
      "inputM": 40,
      "outputM": 8,
      "compute": 250,
      "other": 100,
      "p": 15,
      "prange": [
        5,
        30
      ],
      "commercial": 10,
      "buyers": [
        80,
        160,
        320
      ],
      "price": [
        15000,
        35000,
        75000
      ],
      "product": "Developmental-stage prioritization for preclinical research",
      "evidence": "Accuracy was 54.2% versus 49.8% for length, but only 4 of 155 middle-stage cases were correctly classified. The cohort includes already-lethal mouse genes only.",
      "gate": "Middle-stage recall \u226530%, balanced accuracy \u226550%, and lower log loss than the strongest baseline on untouched domain-family groups, with a positive paired 95% interval for log-loss improvement.",
      "stop": "Stop if better average accuracy continues to hide failure of the middle class. Do not extrapolate to arbitrary genes or human patients.",
      "why": "The existing failure is substantial. Better context could help, but the available labels and class imbalance limit confidence.",
      "next": "Independent timing cohort and use-case validation",
      "follow": [
        75000,
        250000
      ],
      "link": "#pilot-evidence"
    },
    {
      "id": "anatomy",
      "name": "Broad anatomy prediction",
      "status": "Paused after negative result",
      "weeks": "2\u20133",
      "hours": 40,
      "review": 16,
      "inputM": 10,
      "outputM": 2,
      "compute": 80,
      "other": 50,
      "p": 10,
      "prange": [
        3,
        20
      ],
      "commercial": 10,
      "buyers": [
        80,
        160,
        320
      ],
      "price": [
        10000,
        25000,
        50000
      ],
      "product": "Comparative phenotype analysis for R&D groups",
      "evidence": "An initial 47-mammal pilot did not reliably beat both relatedness baselines. In a stronger 32-mammal test, adding gene profiles made error 0.43% worse than body mass plus relatedness.",
      "gate": "A changed representation reduces error by \u226510% beyond body mass plus relatedness on untouched families, with a positive paired 95% interval; original species are not a fresh confirmation set.",
      "stop": "Do not spend this budget to rerun the same features. Reopen only with a materially different representation and a usable new holdout.",
      "why": "Two failed baseline comparisons lower the prior. Larger token budgets alone do not fix the information gap.",
      "next": "Independent family replication, only after a positive gate",
      "follow": [
        50000,
        150000
      ],
      "link": "/research/anatomy-controls/"
    },
    {
      "id": "llm",
      "name": "Does LLM assistance add value?",
      "status": "Method check",
      "weeks": "2\u20133",
      "hours": 50,
      "review": 20,
      "inputM": 30,
      "outputM": 6,
      "compute": 80,
      "other": 50,
      "p": 30,
      "prange": [
        15,
        50
      ],
      "commercial": 20,
      "buyers": [
        320,
        800,
        1600
      ],
      "price": [
        5000,
        20000,
        40000
      ],
      "product": "Audited research-curation workflow for biotech teams",
      "evidence": "LLMs helped write the existing workflows. No controlled experiment has measured their incremental scientific value or total billed token cost.",
      "gate": "At least 20% lower fully costed expense per correct sourced assertion versus a non-LLM workflow, at \u226595% precision on a blinded audit, with a positive paired 95% interval for savings.",
      "stop": "Stop if savings disappear after expert correction, or if unsupported assertions increase. An LLM explanation is not an experimental observation.",
      "why": "Extraction and coding are plausible uses. Expert correction, source access and verification can consume the apparent savings.",
      "next": "Independent workflow replication and customer pilot",
      "follow": [
        25000,
        100000
      ],
      "link": "#methodology"
    }
  ],
  "forecastType": "Uncalibrated analyst judgment; conditional on budget, usable data and staffing",
  "commercialEvent": "Three distinct organizations each paying at least USD 25000 for a research pilot by month 36, conditional on passing the test and funded follow-on work",
  "marketType": "Annual US research-tool TAM scenario; overlapping buyers, do not sum",
  "budgetRangeMultipliers": [
    0.65,
    1.6
  ],
  "historicalSpend": "Not metered in this model",
  "sources": [
    {
      "url": "https://ncses.nsf.gov/pubs/nsf25354/assets/data-tables/tables/nsf25354-tab019.pdf",
      "basis": "2023 biotechnology R&D company reference pool; buyer relevance and prices are assumptions"
    },
    {
      "url": "https://www.bls.gov/ooh/life-physical-and-social-science/medical-scientists.htm",
      "basis": "May 2025 R&D-industry median salary; burden and reviewer rates are assumptions"
    },
    {
      "url": "https://modal.com/pricing",
      "basis": "Resource-price context checked 2026-09-12; per-path compute budgets are allowances"
    }
  ]
}