Forecast intervention effect half life

What it's for

Turns one-off transformation claims into an executive durability forecast: leaders can see when an observed process, platform, reliability, or AI-assistant improvement is likely to fade and schedule evidence review before value silently disappears.

What you give it

Inputs split into evidence read from your connected systems, calibration your team owns, and numerical controls that affect precision but never the result's meaning.

{
  "current_interventions": [
    {
      "elapsed_periods": 0,
      "id": "review-policy",
      "initial_effect": 1,
      "reinforcement_cost": 500,
      "value_per_effect_unit_per_period": 1000
    }
  ],
  "direction": "higher_is_better",
  "half_life_grid_points": 80,
  "historical_effects": [
    {
      "cohort_id": "cohort-0",
      "effect_estimate": 0.9,
      "elapsed_periods": 0,
      "id": "effect-0-0",
      "standard_error": 0.05
    },
    {
      "cohort_id": "cohort-0",
      "effect_estimate": 0.7727171322029717,
      "elapsed_periods": 1,
      "id": "effect-0-1",
      "standard_error": 0.05
    },
    {
      "cohort_id": "cohort-0",
      "effect_estimate": 0.6656854249492381,
      "elapsed_periods": 2,
      "id": "effect-0-2",
      "standard_error": 0.05
    },
    {
      "cohort_id": "cohort-0",
      "effect_estimate": 0.5756828460010884,
      "elapsed_periods": 3,
      "id": "effect-0-3",
      "standard_error": 0.05
    },
    {
      "cohort_id": "cohort-0",
      "effect_estimate": 0.5,

What you get back

This is the actual output of running the example above — computed by the same function the platform calls, not an illustration.

{
  "assumptions": [
    "Within the governed epoch, each cohort's benefit is a persistent floor plus a cohort-specific exponentially decaying component with a shared half-life.",
    "Historical effect estimates are comparable, directionally aligned, and accompanied by valid standard errors from identified or appropriately adjusted designs.",
    "Current initial effects are on the same scale as history; value conversion is owner-governed and does not turn association into causal ROI."
  ],
  "decision": "reinforcement_review_likely_within_horizon",
  "executive_summary": {
    "expected_aggregate_value_at_horizon": 256.5455,
    "half_life_p10": 3.3897,
    "half_life_p50": 4.0034,
    "half_life_p90": 4.7282,
    "persistent_effect_posterior_mean": 0.0831,
    "probability_any_below_practical_effect_now": 0,
    "probability_any_below_practical_effect_within_horizon": 1
  },
  "intervention_forecast": [
    {
      "effect_at_horizon_interval": [
        0.2098,
        0.3003
      ],
      "expected_effect_at_horizon": 0.2565,
      "intervention_id": "review-policy",
      "median_first_below_period": 7,
      "probability_below_practical_effect_within_horizon": 1,
      "reinforcement_cost": 500
    }
  ],
  "limitations": [
    "The half-life is a local empirical durability parameter, not a biological constant or a benchmark transferable across companies.",
    "A reinforcement review flag does not prove that reinforcement will restore the effect; that action needs its own effect and cost evidence.",
    "High posterior boundary mass means the supplied half-life grid or decay family is inadequate and the decision should be treated as unstable."
  ],
  "method": "profiled_bayesian_exponential_effect_half_life_v1",
  "model_diagnostics": {
    "posterior_boundary_mass": 0,
    "selected_grid_half_life": 4.0034,
    "selected_noise_scale": 1,
    "standardized_weighted_rmse": 0.0008
  },
  "period_forecast": [
    {
      "aggregate_effect_p10": 1,

How it works

Sequential Bayesian & bandits — Learn while deciding — update beliefs as evidence arrives and choose where the next unit of effort is worth spending.

1. 1 Directionally align comparable longitudinal effect estimates and their standard errors; require at least five cohorts with four or more observations beginning at elapsed period zero.
2. 2 For every log-spaced half-life candidate, profile a weighted model with one shared persistent effect and one cohort-specific decaying amplitude, inflate uncertainty when residual dispersion exceeds reported standard errors, and convert relative likelihood into posterior grid weight.
3. 3 Draw half-life and persistent-effect uncertainty, decay each current intervention from its locally supplied initial effect and age, and aggregate effect and governed value conversion across the forecast horizon.
4. 4 Report posterior half-life, grid-boundary mass, model misfit, period distributions, and first-below-threshold probabilities; flag a reinforcement review rather than claiming reinforcement itself will work.

Before you trust it

Every tool in the catalog ships with the conditions under which its answer is meaningful — and the conditions under which it should abstain instead of guessing.

Calibrating it to your org

Calibration workflow

1. 1 Define the management decision, target outcome, aggregate unit, privacy boundary, cadence, and prediction/intervention horizon for this organization.
2. 2 Build a tenant-scoped historical cohort using only information available before each prediction or decision; preserve zero periods, censoring, assignment probabilities, and unresolved outcomes when the method requires them.
3. 3 Estimate statistical parameters on training history, but obtain costs, utilities, risk tolerance, practical-effect thresholds, capacity, and policy constraints from accountable decision owners.
4. 4 Validate on later time windows or held-out aggregate units at the deployment grain, against a simple baseline and the function-specific validation strategy.
5. 5 Deploy only if the returned decision clears evidence, overlap, calibration, robustness, and guardrail checks; warning, unsupported, schema-gap, and fallback decisions are abstentions.
6. 6 Monitor realized outcomes, data drift, coverage, and decision regret; recalibrate at a governed cadence or after a detected regime/definition change, never merely because a stakeholder dislikes the result.

Call it from your AI

You don't wire up 388 tools in your MCP client. The GitRevio MCP server exposes 18 tools, three of which let an agent search the catalog, read a tool's schema, and run it — so the assistant finds this one on its own.

gitrevio_capabilities_search
  { "q": "learn how quickly a governed interventions" }
  → finds "forecast_intervention_effect_half_life"

gitrevio_capability_describe
  { "capability_id": "forecast_intervention_effect_half_life" }
  → returns the input schema and agent guidance shown on this page

gitrevio_capability_run
  { "capability_id": "forecast_intervention_effect_half_life", "arguments": { ... } }
  → returns the result shown above

Works in Claude Desktop, Claude Code, Cursor, Cline, Continue.dev, Goose and Aider. See the MCP server.

Related tools

See every tool in Measurement integrity →