Recommend stop continue scale decisions

Recommend stop, continue learning, or scale for aggregate initiatives using beta-binomial posterior rollout economics, independent harm gates, simulation precision, sampling cost, and opportunity decay.

Causal inference & experiment design Optimize CTOs Tech CEOs IT managers Team leads Investors

What it's for

Turns ambiguous pilots and product bets into an executive-ready stop, keep learning, or scale decision that prices both evidence and the opportunity lost while waiting.

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.

Field	Type	Role	Required
initiatives	array of objects (18 fields) ≥ 1 item	Evidence	Yes
max_detail_rows	integer ≥ 1, ≤ 500	Numerical control	Optional
maximum_harm_probability	number ≥ 0.001, ≤ 0.5	Your calibration	Optional
minimum_continue_net_value	number ≥ 0	Your calibration	Optional
scale_probability_threshold	number ≥ 0.5, ≤ 0.999	Your calibration	Optional
seed	integer ≥ 0, ≤ 4294967295	Numerical control	Optional
simulation_confidence_level	number ≥ 0.8, ≤ 0.999	Your calibration	Optional
simulation_draws	integer ≥ 1000, ≤ 200000	Numerical control	Optional
stop_probability_threshold	number ≥ 0.001, ≤ 0.5	Your calibration	Optional

Each `initiatives` record

Field	Type	Required
additional_sample_size	integer (≥ 1, ≤ 100000)	Yes
baseline_success_rate	number (≥ 0, ≤ 1)	Yes
harm_cost_per_event	number (≥ 0)	Yes
harm_prior_alpha	number (> 0)	Yes
harm_prior_beta	number (> 0)	Yes
id	string (non-empty)	Yes
maximum_acceptable_harm_rate	number (≥ 0, ≤ 1)	Yes
observed_harm_trials	integer (≥ 0)	Yes
observed_harms	integer (≥ 0)	Yes
observed_successes	integer (≥ 0)	Yes
observed_trials	integer (≥ 0)	Yes
opportunity_decay_during_learning	number (≥ 0, ≤ 1)	Yes
rollout_cost	number (≥ 0)	Yes
rollout_opportunities	integer (≥ 1)	Yes
sample_cost_per_observation	number (≥ 0)	Yes
success_prior_alpha	number (> 0)	Yes
success_prior_beta	number (> 0)	Yes
value_per_incremental_success	number (≥ 0)	Yes

Example input

{
  "initiatives": [
    {
      "additional_sample_size": 100,
      "baseline_success_rate": 0.5,
      "harm_cost_per_event": 10,
      "harm_prior_alpha": 1,
      "harm_prior_beta": 50,
      "id": "onboarding-pilot",
      "maximum_acceptable_harm_rate": 0.1,
      "observed_harm_trials": 100,
      "observed_harms": 0,
      "observed_successes": 5,
      "observed_trials": 10,
      "opportunity_decay_during_learning": 0.05,
      "rollout_cost": 100,
      "rollout_opportunities": 1000,
      "sample_cost_per_observation": 1,
      "success_prior_alpha": 1,
      "success_prior_beta": 1,
      "value_per_incremental_success": 10
    }
  ],
  "seed": 79,
  "simulation_draws": 5000
}

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.

Example output

{
  "assumptions": [
    "Success and harm observations are prospectively defined Bernoulli outcomes with stable, representative exposure and independent beta priors; trials, failures, zero events, censoring, assignment, interference, and unresolved outcomes are preserved rather than selectively omitted.",
    "Scale value is incremental to the declared baseline and includes rollout cost plus expected harm cost over the same horizon; opportunities remaining after learning explicitly decline, while all omitted delay, capacity, financing, quality, safety, legal, and reversal costs invalidate the economic comparison.",
    "The additional sample is a valid prospective design; future decisions use posterior expected economics plus a one-sided harm bound, and net evidence value is evaluated against the governed immediate policy on identical posterior states rather than against an artificially weak baseline.",
    "Initiatives are aggregate product, policy, project, architecture, or investment cases, never named employees; stop, continue, and scale are review recommendations subject to causal-design, safety, finance, and accountable-owner approval, not automatic deployment or termination actions."
  ],
  "configuration": {
    "current_scale_and_stop_gates_use_simulation_confidence_bounds": true,
    "future_harm_gate_uses_one_sided_beta_normal_approximation": true,
    "maximum_harm_probability": 0.1,
    "minimum_continue_net_value": 0,
    "scale_probability_threshold": 0.9,
    "seed": 79,
    "simulation_confidence_level": 0.95,
    "simulation_draws": 5000,
    "stop_probability_threshold": 0.1
  },
  "decision": "portfolio_requires_value_positive_continued_learning",
  "initiative_diagnostics": [
    {
      "additional_sample_cost": 100,
      "expected_net_scale_value_now": -187.2784,
      "expected_wait_then_decide_value_net_of_sampling": 317.053,
      "harm_probability_interval": [
        0,
        0
      ],
      "immediate_policy_value": 0,
      "initiative_id": "onboarding-pilot",
      "net_value_of_additional_evidence": 317.053,
      "positive_value_probability_interval": [
        0.43501381,
        0.46258619
      ],
      "posterior_harm_rate_mean": 0.0066,
      "posterior_success_rate_mean": 0.5,
      "probability_future_sample_recommends_scale": 0.4506,
      "probability_harm_rate_exceeds_limit": 0,
      "probability_scale_value_is_positive": 0.4488,
      "reason": "additional_evidence_has_positive_net_policy_value",
      "recommendation": "continue",
      "remaining_rollout_opportunities_after_learning": 950
    }

Truncated for display — the full payload is 56 lines.

How it works

Causal inference & experiment design — Separate what a change caused from what merely moved alongside it.

1 Freeze success and harm outcome definitions, priors, complete observed trials including failures and zero events, baseline rate, rollout opportunities, incremental value, harm cost, rollout cost, and the prospective additional-sample design.
2 Simulate posterior scale economics and harm exceedance, requiring confidence bounds—not point estimates—to clear scale or stop gates; evaluate a future sample by updating beta posteriors and applying a one-sided harm bound.
3 Compare wait-then-decide value net of sample cost and lost opportunities with the governed immediate policy, then return scale only with precision-supported value and safety, continue only when learning repays cost and decay, otherwise stop with the binding reason.

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.

Assumptions & guardrails

Assignment or adoption timing, interference policy, overlap, attrition, and outcome availability match the estimand encoded by the method.
Trials are representative Bernoulli exposures with stable success and harm processes, valid assignment and outcome resolution; scale economics are incremental and complete, while the future sample informs the same target population before remaining opportunities expire.
A causal label is warranted only when design and diagnostic requirements pass; otherwise treat the result as descriptive or abstain.
A posterior gate cannot validate causal design or guarantee safety, and stop does not mean failure or blame; initiatives are aggregate cases, never employees, and every recommendation remains an accountable stage-gate review.

Minimum evidence

initiatives: at least 1 rows/items

How to validate it

Preserve the assignment/adoption design and validate overlap, pre-period or placebo diagnostics, attrition, interference policy, and cluster-level uncertainty; never tune on the estimated effect.

Calibrating it to your org

Same for everyone

The mathematical kernel, validation rules, method version, and JSON output semantics are organization-independent; no tenant-trained coefficients or company benchmark is embedded in the function.

Specific to you

tenant-calibrated beta priors, representative resolved trial cohort, incremental rollout value net of harm and implementation, opportunity decay while learning, and effective sample semantics after clustering, censoring and interference
initiative and target-population perimeter, causal design, success/harm definitions, prior provenance, baseline, value and cost horizon, harm limit, scale/stop confidence gates, additional sample, learning cost and opportunity decay

Calibration workflow

1 Define the management decision, target outcome, aggregate unit, privacy boundary, cadence, and prediction/intervention horizon for this organization.
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 Estimate statistical parameters on training history, but obtain costs, utilities, risk tolerance, practical-effect thresholds, capacity, and policy constraints from accountable decision owners.
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 Deploy only if the returned decision clears evidence, overlap, calibration, robustness, and guardrail checks; warning, unsupported, schema-gap, and fallback decisions are abstentions.
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": "recommend stop continue learning or scale" }
  → finds "recommend_stop_continue_scale_decisions"

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

gitrevio_capability_run
  { "capability_id": "recommend_stop_continue_scale_decisions", "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

Calculate value of delay to decide

Calculate a period-by-period value-of-delay curve that separates prospectively available information from waiting cost and changing action economics under coherent scenarios.

Decision analysis

Optimize learning vs earning allocation

Solve the exact finite-horizon Beta–Bernoulli bandit for allocating scarce units between a known earning baseline and uncertain actions that earn and update their posterior.

Sequential Bayesian & bandits

Audit commercial technical commitment integrity

Audit signed commercial promises against explicitly allocated technical scope, dependency order, funded capacity, acceptance criteria and evidence; expose orphan scope, double allocation, cycles, late plans and maximum contractual penalty without interpreting legal rights from engineering activity.

Constrained optimization

Audit decision execution fidelity

Audit whether approved decisions actually became verified implementation at the promised aggregate-unit and component grain, with whole-unit bootstrap uncertainty and simultaneous gates for fidelity, overdue scope, unverifiable evidence, exceptions and critical gaps.

Statistical audit & measurement

Audit KPI threshold bunching

Detect a post-target excess concentration immediately above a governed KPI threshold: compare within-unit pre/post local mass and above-versus-below mirror asymmetry, bootstrap whole units, report density bins and a smoothed log-density jump, and explicitly refuse to equate bunching with individual gaming or intent.