FableBook a live walkthrough

How each decision can be checked

Every recommendation is linked to its source records, model version, alternatives, constraints and expected effect. The completed action and observed result stay on the same record.

The evidence retained for each decision.

Execution

A controlled run uses the same mapping, model, action, optimisation, persistence and API code intended for deployment.

Models

A fixed test compares candidate models with rules-based and statistical baselines before any model can influence a decision.

Source records

Identity, timing, units, authority, freshness and conflicts remain visible as records move between systems.

Actions and value

Every recommendation states the action, expected effect, alternative, outcome window and method used to count value.

Detailed product evidence.

Project Margin Rescue and Range-to-Cash document the recurring decision, required systems, forecasts, constraints, available actions and value method.

Project Margin Rescue Engine

Intervene before forecast drift becomes lost margin, delayed cash or a write-off.

Reference scopeCoupled project portfolio in the supported weekly envelope
Decision pathMapper → model → action/value → optimiser → worker → API
Automation boundaryApproval required; no autonomous financial mutation
Inspect this engine

Range-to-Cash Trading Yield Engine

Choose range actions using their combined effect on cash, margin and service.

Decision grainSKU × branch × customer segment × cutoff
ComparisonCash, contribution, funding and service effects remain separate
Automation boundaryApproval required; no source-system write intent
Inspect this engine
Explore Decision Engines

Inspect one decision record.

Use the reference record below to see how evidence moves from source records to a bounded action and a measured result.

Reference recordProject Margin Rescue
Decision record / 01 of 6Reconstruct the position

The record starts with the systems that describe the same project, joined at the decision cutoff so identity, timing, units and conflicts remain visible.

Evidence in this stage5 retained items
  • ERP and project accounting
  • PSA and timekeeping
  • Billing, AR and WIP
  • Procurement and supplier cost
  • PMO, contract and forecast history

A model must improve the operating decision.

  • Candidate models are tested on records that existed before the outcome and compared with fixed rules-based and statistical baselines.
  • Evaluation covers accuracy, calibration, uncertainty and robustness.
  • Production replays test whether the model changes the selected action or allocation and improves the agreed objective.
  • Permissions, legal requirements and safety constraints remain binding.

Evidence before deployment.

  1. 01
    Controlled product run

    Run the complete decision path against a fixed operating history.

  2. 02
    Client calibration

    Map the client’s records, fit the model system and define where it may be used.

  3. 03
    Production use

    Make the decision with live permissions, constraints, workflow and measurement.

  4. 04
    Financial result

    Compare completed actions and observed outcomes with the agreed baseline and finance records.

Published figures include their limits.

Review the complete decision record.

We will show the evidence available at the cutoff, the forecast, the rejected alternatives and the method for measuring the selected action.

Book a live walkthrough