Reconcile hierarchical delivery forecasts mint

Reconcile independently produced portfolio, product, team, repository, or workstream forecasts into one additive hierarchy using shrinkage MinT: learn the cross-level residual covariance on training forecasts, prove coherence, gate accuracy on later untouched periods, and return coherent current forecasts with uncertainty intervals.

Forecasting & survival Audit CTOs Tech CEOs IT managers Team leads Investors

What it's for

Gives executives, product owners, and investors one coherent answer when portfolio, product, and delivery forecasts disagree—while testing whether reconciliation improves later forecasts instead of merely forcing totals to match.

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
actual_coherence_tolerance	number ≥ 0, ≤ 0.1	Your calibration	Optional
confidence_level	number ≥ 0.8, ≤ 0.99	Your calibration	Optional
covariance_shrinkage	number ≥ 0, ≤ 1	Your calibration	Optional
current_base_forecasts	object	Evidence	Yes
enforce_nonnegative	boolean	Your calibration	Optional
hierarchy_nodes	array of objects (2 fields) ≥ 2 items	Evidence	Yes
historical_forecasts	array of objects (5 fields) ≥ 16 items	Evidence	Yes
max_detail_rows	integer ≥ 1, ≤ 500	Numerical control	Optional
maximum_test_rmse_ratio	number ≥ 0.5, ≤ 5	Your calibration	Optional

Each `historical_forecasts` record

Field	Type	Required
actuals	object	Yes
base_forecasts	object	Yes
id	string (non-empty)	Yes
period	integer	Yes
split	one of "train", "test"	Yes

Example input

{
  "current_base_forecasts": {
    "portfolio": 47,
    "product-a": 30,
    "product-b": 18
  },
  "hierarchy_nodes": [
    {
      "id": "portfolio",
      "parent_id": null
    },
    {
      "id": "product-a",
      "parent_id": "portfolio"
    },
    {
      "id": "product-b",
      "parent_id": "portfolio"
    }
  ],
  "historical_forecasts": [
    {
      "actuals": {
        "portfolio": 32,
        "product-a": 20,
        "product-b": 12
      },
      "base_forecasts": {
        "portfolio": 32,
        "product-a": 20.4,
        "product-b": 12
      },
      "id": "batch2-hierarchy-0",
      "period": 0,
      "split": "train"
    },
    {
      "actuals": {
        "portfolio": 32.900000000000006,
        "product-a": 20.6,
        "product-b": 12.3
      },
      "base_forecasts": {
        "portfolio": 34.582941969615796,

Truncated for display — the full payload is 263 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.

Example output

{
  "decision": "hierarchical_forecast_reconciliation_supported",
  "executive_summary": {
    "all_reported_forecasts_are_additively_coherent": true,
    "root_base_forecast": 47,
    "root_id": "portfolio",
    "root_reconciled_forecast": 47.9549,
    "test_rmse_ratio": 0.3193
  },
  "forecasts": [
    {
      "base_forecast": 47,
      "interval_lower": 47.3134,
      "interval_upper": 48.5963,
      "is_leaf": false,
      "node_id": "portfolio",
      "reconciled_forecast": 47.9549
    },
    {
      "base_forecast": 30,
      "interval_lower": 29.463,
      "interval_upper": 30.464,
      "is_leaf": true,
      "node_id": "product-a",
      "reconciled_forecast": 29.9635
    },
    {
      "base_forecast": 18,
      "interval_lower": 17.6214,
      "interval_upper": 18.3614,
      "is_leaf": true,
      "node_id": "product-b",
      "reconciled_forecast": 17.9914
    }
  ],
  "guardrails": [
    "MinT coherence improves arithmetic consistency, not the information content of weak base forecasts.",
    "The untouched test gate must be repeated after hierarchy, metric, cadence, or forecast-model changes.",
    "Intervals condition on the supplied hierarchy and residual history; they do not include unknown scope changes.",
    "This aggregate hierarchy must not be converted into named-person performance rankings."
  ],
  "hierarchy": {
    "covariance_shrinkage": 0.2,
    "leaf_count": 2,

Truncated for display — the full payload is 66 lines.

How it works

Forecasting & survival — Estimate when something completes or fails, with censoring and unresolved work handled honestly rather than dropped.

1 Freeze one additive metric, hierarchy version, forecast issue time, horizon, and chronological train/test split; reject cycles, multiple roots, incomplete node vectors, and realized actuals that do not sum under the declared hierarchy.
2 Estimate the full base-forecast residual covariance on training periods only, shrink it toward its diagonal for stability, and form the MinT projection through the hierarchy summing matrix and generalized inverse.
3 Apply the frozen projection to untouched test forecasts, compare all-node RMSE with the unreconciled baseline, and require every reconciled vector to satisfy the additive identities before deployment.
4 Project the current base forecasts, optionally enforce nonnegative leaf values transparently, propagate residual covariance into intervals, and abstain if the later-period RMSE ratio exceeds the governed tolerance.

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.
The target measure is genuinely additive across every declared parent/child edge at one unit, cadence, horizon, cutoff, and metric version.
Historical base forecasts are point-in-time snapshots made before outcomes, and residual dependence is sufficiently stable across the represented deployment epoch.
Untouched test periods represent current hierarchy behavior; covariance shrinkage and nonnegative projection are prespecified rather than tuned on current results.
A predictive interval or risk estimate is not a deadline promise, causal explanation, or individual-performance judgment.
Arithmetic coherence does not make weak base forecasts accurate, causal, strategically aligned, or complete.
A passing aggregate test applies only to the represented hierarchy, metric, horizon, and forecast producers; it is not a universal MinT endorsement.
Hierarchy adjustments are model outputs, not evidence that a team or person overstated or understated work.

Minimum evidence

hierarchy_nodes: at least 2 rows/items
historical_forecasts: at least 16 rows/items
current_base_forecasts: 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

complete point-in-time base forecast vectors for every node and forecast vintage, split chronologically into training and untouched test periods
one additive summing matrix plus training-only cross-node forecast residual covariance
hierarchy and metric versions, forecast origin and horizon, outcome finalization, split, actual-coherence tolerance, covariance shrinkage, test RMSE tolerance, interval level, and nonnegative policy

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": "reconcile independently produced portfolio product team" }
  → finds "reconcile_hierarchical_delivery_forecasts_mint"

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

gitrevio_capability_run
  { "capability_id": "reconcile_hierarchical_delivery_forecasts_mint", "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

Detect operational critical slowing down

Detect early-warning patterns associated with an aggregate system losing resilience before a possible regime transition: locally detrend rolling windows, track rising lag-one autocorrelation, variance, and spectral reddening, compare endpoint shifts with a frozen reference regime, and control multiplicity under a circular moving-block bootstrap.

Decision analysis

Forecast organizational change load capacity

Forecast whether the organization's planned portfolio of migrations, launches, reorganizations, policy changes, and platform transitions exceeds aggregate operating capacity: select a saturating distributed-lag change-load model on pretest history, beat an autoregressive baseline on later periods, then simulate peak strain and limit-breach probability.

Forecasting & survival

Aggregate risk register copula

Aggregate risk-register occurrence and lognormal severity marginals through a validated Gaussian copula into expected loss, VaR, CVaR, dependence amplification, and tail shares.

Statistical audit & measurement

Allocate restless bandit interventions

Allocate scarce recurring interventions across evolving Markov units with Whittle indices, explicit indexability checks, and paired policy simulation.

Sequential Bayesian & bandits

Attribute commercial dependency tail loss

Calculate expected loss, VaR and CVaR for commercial value concentrated in shared technical components, then allocate every modeled tail-loss dollar exactly once across components with normalized negative-log survival hazard rather than overlapping leave-one-out sensitivities.