Research & evidence

Measured progress. Clear decisions.

Phaneon is building a controlled path from uncertain demand to reviewable planning choices. This page separates implemented software, measured public results and the customer evidence that still has to be earned.

Current product evidenceGuided evaluation platform implementedPolaris 3.8 · Evaluation and stabilizationRigel 0.7 · Bounded, no-write evaluationEvidence reviewed August 2026

Evidence at a glance

A clear boundary between capability and proof.

The platform is available for guided demonstrations and narrowly scoped evaluation. Production authorization and customer outcomes are separate milestones, not implied by software checks.

Working now

One reviewable planning workflow.

Polaris 3.8 connects data context, forecast review and scenarios. Rigel 0.7evaluates inventory and Production P2production-flow choices without writing to operational systems.

Measured already

Focused checks passed; a challenger was rejected.

13/13 focused Aurora adapter checks pass. In a matched 480-series public comparison, Aurora did not earn promotion and the existing route was retained.

To be proven with design partners

Operational forecast quality and business value.

The native system route still needs eligible operational demand data. Customer value, integration readiness, security approval and production authorization require separate customer-specific evidence.

What works now

From forecast review to bounded planning experiments.

Polaris contains the governed forecasting workspace and Aurora adapter while integrated stabilization continues. Rigel tests replenishment and production-flow choices before operational commitment. Native Aurora quality on eligible operational data and customer value remain evaluation questions, not current claims.

Evaluation evidenceCurrent implementation · Focused contracts · No production authorization
Polaris workspace3.8

The current public line is in evaluation and stabilization, with guided demonstrations available.

Aurora adapter checks13/13

Focused integration and fallback checks passed against the exact adapter source. This is mechanism evidence, not forecast-quality proof.

Rigel boundary0.7

Inventory and production-flow simulation under the Production P2 operating contract, bounded and no-write.

Matched public panel480

Series compared on the same public data. Aurora did not earn promotion; the legacy route was retained.

01

Forecast review workspaceA governed adapter, guarded fallback, explanations and scenarios in one planner view. Native system performance still needs eligible operational data.

02

Inventory policy studioCompare recommended orders, expected service and modeled cost under matched assumptions, without system write-back.

03

Production-flow workspaceUnder the Production P2 contract, model shifts, resources, queues, routings and lots; review modeled throughput, WIP and bottlenecks.

Aurora integration evidence

The adapter is integrated—and promotion still has to be earned.

The strongest current Aurora evidence is mechanism evidence: the adapter is connected to Polaris and focused contracts pass. In the matched public comparison, native Aurora did not earn promotion; retaining the legacy route shows the protection gate working.

Implemented

A governed adapter inside the Polaris evaluation line.

Declared calendar meaning, deterministic execution, caching and per-series routing are connected inside Polaris 3.8 while integrated stabilization continues.

Focused verification

13 of 13 adapter checks passed.

The current-source checks cover integration contracts and the guarded behavior expected when inputs or outputs are unsafe.

Measured honestly

The matched public result was not promoted.

The retained route performed better on the locked 480-series comparison. The integrated system route still needs eligible operational demand data before a native performance claim.

Matched public evaluation

Aurora did not earn promotion on the locked 480-series panel.

Native Aurora and the retained Polaris route were evaluated on the same locked M4 monthly cohort. The legacy route produced the better promotion evidence, so the gate retained it. This is useful negative evidence—not customer proof and not production authorization.

Evaluation series480Same locked cohort for both routes
Evaluation observations2,880Six observations per series
Aurora matched WIS469.5Lower is better
Retained route WIS410.5Lower score retained
WIS (lower is better)480 matched series
Polaris legacy route410.5
Aurora v2 matched469.5

Did not earn promotion; the Polaris legacy route was retained. The lower legacy score is the stronger result on this named public cohort.

Measured result and next improvement

The negative result is retained rather than reframed as an Aurora win. The next meaningful test is the integrated system route on eligible operational demand data, using a frozen baseline and predeclared promotion measures. Until then, no native superiority or calibrated-performance claim is made.

Scientific methodology

Evidence is designed into the workflow.

The objective is not to reward a sophisticated model. It is to improve a planning decision while protecting the current method whenever new evidence is not strong enough.

01

Define the planning question

Name the decision, dataset, horizon, cadence and operational use before evaluating a model.

02

Freeze the comparison

Fix the data split, software subject, metrics and practical baseline before examining the result.

03

Test through time

Use time-ordered holdouts or rolling origins so future observations never leak into training.

04

Measure several failure modes

Track point error, bias, interval quality and decision-relevant effects instead of one composite score.

05

Retain the safer method

A challenger that fails the agreed gate does not replace the current method or simple baseline.

06

Carry evidence forward

Bind forecasts, scenarios and simulation results to versioned lineage so a planner can review what changed.

Business interpretation

What a design-partner evaluation can test.

Better evidence matters only if it helps a team see risk earlier, compare realistic options and preserve a clear reason for the final decision.

Focus attention

Review where uncertainty and consequence meet.

Test whether exceptions and explanations direct planners to the decisions that deserve expert attention.

Compare options

Test inventory and flow choices under matched assumptions.

Evaluate whether Rigel makes service, cost, capacity and congestion trade-offs clearer before operational commitment.

Protect continuity

Keep the current method until a challenger earns trust.

Baseline protection and no-write evaluation allow learning without forcing an early operational change.

How to read the evidence

Four evidence levels, kept separate.

01

Implemented capability

Current-source checks show that named forecasting, governance and simulation contracts behave as specified.

02

Measured public behavior

Public-panel and synthetic studies show behavior under named, reproducible conditions—including negative results.

03

Customer value

Inventory, service, planning-time and economic value must be measured on representative conditions with an agreed baseline.

04

Operational authorization

Production deployment and system write-back require separate customer-specific security, integration and operating approval.

Design-partner pathway

Measure one planning question end to end.

A controlled evaluation connects product capability to a real decision without requiring production write-back. The scope stays narrow enough to learn quickly and stop cleanly if the value is not material.

Request a planning review
  1. 01

    Choose one product family, planning horizon and consequential decision.

  2. 02

    Freeze the current method, dataset boundary and success measures.

  3. 03

    Compare Polaris and Rigel evidence without writing to operational systems.

  4. 04

    Review the result together and proceed only if the measured value is material.

Public evaluation source

M4 Monthly Dataset, Zenodo DOI 10.5281/zenodo.4656480, licensed CC BY 4.0. The matched public comparison is not manufacturing or customer evidence and does not authorize product activation.