Audit forecast ensemble lineage integrity
Audit whether a claimed forecast ensemble is a complete, independently sealed and point-in-time evidence set rather than duplicated consensus, then score only mature uncontaminated outcomes.
What it's for
Makes plan-vs-reality, probabilistic estimation and board-confidence claims auditable by proving that apparent consensus existed before the answer and was not merely the same upstream opinion repeated.
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 |
|---|---|---|---|
| as_of_ms | number ≥ 0 | Your calibration | Yes |
| eligible_assignments | array of objects (7 fields) | Evidence | Yes |
| max_detail_rows | integer ≥ 1, ≤ 500 | Numerical control | Optional |
| maximum_expected_calibration_error | number ≥ 0, ≤ 1 | Your calibration | Optional |
| maximum_missing_assignment_fraction | number ≥ 0, ≤ 1 | Your calibration | Optional |
| maximum_post_cutoff_fraction | number ≥ 0, ≤ 1 | Your calibration | Optional |
| minimum_mature_questions | integer ≥ 1, ≤ 100000 | Your calibration | Optional |
| questions | array of objects (7 fields) | Evidence | Yes |
| resolved_outcomes | array of objects (6 fields) ≥ 0 items | Evidence | Yes |
| sealed_forecasts | array of objects (9 fields) ≥ 0 items | Evidence | Yes |
Each sealed_forecasts
record
| Field | Type | Required |
|---|---|---|
| evidence_verified | boolean | Yes |
| forecast_at_ms | number (≥ 0) | Yes |
| id | string (non-empty) | Yes |
| immutable_hash | string (non-empty) | Yes |
| information_cutoff_ms | number (≥ 0) | Yes |
| observed_source_refs | array of string | Yes |
| probability | number (≥ 0, ≤ 1) | Yes |
| question_id | string (non-empty) | Yes |
| source_ref | string (non-empty) | Yes |
{
"as_of_ms": 400,
"eligible_assignments": [
{
"eligible_at_ms": 1,
"evidence_verified": true,
"id": "sealed-assignment-0-expert-a",
"independence_required": true,
"question_id": "sealed-question-0",
"source_ref": "expert-a",
"source_type_id": "human"
},
{
"eligible_at_ms": 1,
"evidence_verified": true,
"id": "sealed-assignment-0-expert-b",
"independence_required": true,
"question_id": "sealed-question-0",
"source_ref": "expert-b",
"source_type_id": "model"
},
{
"eligible_at_ms": 11,
"evidence_verified": true,
"id": "sealed-assignment-1-expert-a",
"independence_required": true,
"question_id": "sealed-question-1",
"source_ref": "expert-a",
"source_type_id": "human"
},
{
"eligible_at_ms": 11,
"evidence_verified": true,
"id": "sealed-assignment-1-expert-b",
"independence_required": true,
"question_id": "sealed-question-1",
"source_ref": "expert-b",
"source_type_id": "model"
},
{
"eligible_at_ms": 21,
"evidence_verified": true,
"id": "sealed-assignment-2-expert-a",
"independence_required": true,Truncated for display — the full payload is 1722 lines.
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": [
"The registered question, resolution rule, eligible source perimeter and sealed cutoff were frozen before forecasts were submitted.",
"Observed-source declarations are complete; exact duplicate probabilities are a herding diagnostic, not proof of copying or misconduct."
],
"counts": {
"contaminated_questions_excluded": 0,
"eligible_assignments": 60,
"mature_scored_questions": 30,
"questions": 30,
"sealed_forecasts": 60,
"source_types": 2
},
"decision": "ensemble_evidence_integrity_passed",
"failed_gates": [],
"integrity": {
"duplicate_assignments": 0,
"exact_duplicate_probability_fraction": 1,
"future_information_forecasts": 0,
"missing_assignment_fraction": 0,
"post_cutoff_fraction": 0,
"sealed_independence_breaches": 0,
"unexpected_forecasts": 0
},
"limitations": [
"Passing proves lineage and retrospective calibration only; it does not prove source independence, causal forecasting skill, intent, or that the ensemble is useful for a particular decision.",
"Intervention-contaminated outcomes are excluded from natural-history scoring rather than silently counted as forecast errors."
],
"method": "complete_sealed_forecast_lineage_calibration_audit_v1",
"passed": true,
"question_diagnostics": [
{
"complete_sealed_forecasts": 2,
"context_id": "delivery",
"eligible_sources": 2,
"ensemble_probability": 0.15,
"exact_duplicate_forecasts": 2,
"mature_uncontaminated_outcome": true,
"question_id": "sealed-question-0"
},
{
"complete_sealed_forecasts": 2,
"context_id": "delivery",
"eligible_sources": 2,Truncated for display — the full payload is 315 lines.
How it works
Forecasting & survival — Estimate when something completes or fails, with censoring and unresolved work handled honestly rather than dropped.
- 1 Freeze each question, resolution rule, base rate, eligible source perimeter and sealed cutoff before any submitted forecast is visible.
- 2 Reconcile exactly one immutable forecast for every eligible question-source pair and reject future information, post-cutoff submissions, undeclared sources and independence breaches.
- 3 Exclude intervention-contaminated or immature outcomes, reconstruct Brier/log loss/calibration, and report exact-probability herding as a diagnostic rather than an accusation.
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
- Training examples precede their outcomes, censoring and unresolved work are represented, and deployment populations remain comparable to validation cohorts.
- Question registration, eligibility, source exposure declarations, immutable hashes, resolution timestamps and contamination flags are complete and cannot be edited retrospectively.
- A predictive interval or risk estimate is not a deadline promise, causal explanation, or individual-performance judgment.
- Agreement is not independent confirmation unless lineage supports it, and duplicate probabilities or failed integrity never prove copying, deception or individual underperformance.
Minimum evidence
- questions: required and organization-defined
- eligible_assignments: required and organization-defined
- sealed_forecasts: at least 0 rows/items
- resolved_outcomes: at least 0 rows/items
- as_of_ms: required and organization-defined
How to validate it
Use chronological train/calibration/test windows, compare proper scores and decision value with a simple baseline, and recalibrate only from outcomes resolved after prediction time.
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
- one complete question-source spine joining every eligible assignment to exactly one pre-cutoff forecast, its immutable hash and exposure declaration, plus at most one later mature uncontaminated outcome
- question/context taxonomy, base-rate epoch, eligible source perimeter, independence definition, sealed cutoff, outcome resolution and maturity, intervention contamination, evidence verification, completeness/calibration gates and correction ownership
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": "audit whether a claimed forecast ensemble" }
→ finds "audit_forecast_ensemble_lineage_integrity"
gitrevio_capability_describe
{ "capability_id": "audit_forecast_ensemble_lineage_integrity" }
→ returns the input schema and agent guidance shown on this page
gitrevio_capability_run
{ "capability_id": "audit_forecast_ensemble_lineage_integrity", "arguments": { ... } }
→ returns the result shown aboveWorks in Claude Desktop, Claude Code, Cursor, Cline, Continue.dev, Goose and Aider. See the MCP server.
Related tools
Forecast dependency adjusted consensus
Combine independently sealed human and model forecasts while learning context base rates and source reliability on earlier questions, discounting empirical and declared information dependence, and abstaining unless later questions beat both base rates and naive consensus.
Optimize forecast elicitation portfolio
Choose which independent human or model forecasts to obtain next by learning chronologically validated contextual directional skill, simulating conservative entropy reduction, pricing decision relevance and removing duplicated information under budget and source-capacity constraints.
Audit analytical specification multiverse
Audit whether an analytical conclusion survives a prespecified multiverse of admissible windows, cohorts, metrics, and models using aligned bootstrap draws, a weighted specification curve, practical-effect support gates, and descriptive choice-influence diagnostics.
Audit analytics challenger independence
Audit whether an analytical challenger supplies genuinely independent error information: use paired temporal moving-block bootstrap bounds on error correlation, incumbent-failure catch rate and common-mode joint failure, with simultaneous Bonferroni control across every screened challenger and explicit evidence gates.
Audit analytics function calibration readiness
Gate analytical functions on paired out-of-time decision loss against a frozen baseline using temporal moving-block bootstrap, autocorrelation- and weight-adjusted effective sample size, evidence coverage, lower confidence bounds, improvement probability and recent degradation rather than declaring a model calibrated from training fit.
Audit analytics transportability
Audit whether locally calibrated analytical functions retain decision-loss improvement across target-similar operating environments using similarity-weighted random-effects meta-analysis, between-environment variance, I-squared, sign consistency and a conservative target prediction interval.