{
  "study_id": "pm25-health-v2",
  "protocol_sha256": "78b4a0c505bc948ca0b6e19ec0624a603e160850a745491dbc8368e5ac4c78fb",
  "inputs_sha256": {
    "baseline": "9b8f14099d32fa86d01c4257f81d56518293fb73df9cddc3523be82d3c35564a",
    "track_a": "f3c255cf0866b1fc75c574842feb8e6538e4a9c28d739945dfa8507f572c6d46",
    "track_b_validation": "8836d2854fc5faac0bdebc45e15c8e7dd335ea6dcf993a55cdeee630fb718099"
  },
  "baseline": {
    "role": "Frozen reproducible reference; not preferred causal model",
    "primary_estimate": {
      "estimate": 0.0471,
      "cluster_se": 0.1349,
      "test_statistic": 0.349,
      "p_value": 0.7271,
      "ci95_low": -0.2174,
      "ci95_high": 0.3116
    },
    "claim": "The fully adjusted conventional panel did not distinguish its PM2.5 coefficient from zero."
  },
  "track_a": {
    "status": "implemented on the existing country panel",
    "primary": {
      "id": "A1",
      "label": "Primary: two-way fixed effects with lagged confounders",
      "claim": "association",
      "formula": "life ~ pm10 + log_gdp_lag1 + urban10_lag1 + C(iso3) + C(year)",
      "n": 4220,
      "countries": 186,
      "estimate": 0.400655,
      "standard_error": 0.127886,
      "ci95_low": 0.149998,
      "ci95_high": 0.651311,
      "p_value": 0.002013,
      "uncertainty": "Country-clustered sandwich covariance"
    },
    "country_trend": {
      "id": "A2",
      "label": "Primary plus country-specific linear trends",
      "claim": "association",
      "formula": "life ~ pm10 + lag1(log_gdp) + lag1(urban) + country FE + year FE + country-specific linear trends (one redundant slope omitted)",
      "n": 4220,
      "countries": 186,
      "estimate": -0.037905,
      "standard_error": 0.073927,
      "ci95_low": -0.182802,
      "ci95_high": 0.106992,
      "p_value": 0.60875,
      "uncertainty": "Country-clustered sandwich covariance"
    },
    "future_exposure_placebo": {
      "id": "A8",
      "label": "Future-exposure placebo",
      "claim": "association",
      "formula": "life ~ pm10_lead1 + log_gdp_lag1 + urban10_lag1 + C(iso3) + C(year)",
      "n": 4034,
      "countries": 186,
      "estimate": 0.363643,
      "standard_error": 0.12393,
      "ci95_low": 0.12074,
      "ci95_high": 0.606546,
      "p_value": 0.003767,
      "uncertainty": "Country-clustered sandwich covariance"
    },
    "decision": {
      "primary_model": "A1",
      "stable_direction_across_core_models": false,
      "all_core_intervals_exclude_zero": false,
      "future_exposure_placebo_pass": false,
      "evidence_grade": "A1-inconclusive-association",
      "conclusion": "Track A does not identify a stable independent association: direction and/or interval conclusions change across trend, change and exposure-window specifications. This is evidence of design sensitivity, not evidence that PM2.5 is harmless.",
      "causal_claim": false
    }
  },
  "track_b": {
    "status": "Estimator and falsification workflow validated on a deterministic synthetic fixture; empirical PM policy input not yet registered",
    "validation_decision": {
      "pretrend_pass": true,
      "exposure_first_stage_pass": true,
      "negative_control_pass": true,
      "identification_checks_pass": true,
      "conclusion": "The deterministic validation fixture recovers the simulated exposure reduction and health improvement while leaving the negative-control outcome near zero. This validates the workflow, not a real PM2.5 causal claim."
    },
    "empirical_causal_estimate": null
  },
  "combined_decision": {
    "substantive_conclusion": "The current country-level panel does not identify the direction or magnitude of a causal PM2.5 effect on period life expectancy.",
    "what_is_supported": "The large negative cross-country association is strongly confounded by national development conditions, and longitudinal estimates are sensitive to trend control, change specification, exposure window and temporal placebo tests.",
    "what_is_not_supported": "Neither no-effect language nor a numerical causal life-expectancy effect is supported by the current data.",
    "methodological_conclusion": "A reusable exposure-health template should use Track A to expose associational sensitivity and Track B to gate causal language on an explicit assignment mechanism and falsification checks.",
    "next_empirical_gate": "Register a real policy or natural-experiment panel and an annual age-standardized cardiorespiratory outcome before producing a Track B PM estimate."
  }
}