Platform
Integrations
All use cases
IndustriesPricingProductField Notes
Sign inGet started
Otiox Command · Accuracy & Calibration

An ontology that predicts must also be held to account.

Get startedView platform
Every prediction Otiox makes becomes a fact in its own model — frozen as it was made, bound to the record it concerns, and later joined to what actually happened. Detector by detector, you see whether 'seventy percent' truly meant seventy percent, and where confidence has yet to be earned.
app.otiox.comFig. 00
Accuracy & calibration
Every prediction is frozen as it was made and checked against what then happened.
43 predictions with a known outcomeBrier score 0.2564last 180 daysModerate
Stockout riskNot enough data
Brier
0.1433
Came true
83 %
Said / happened
90 % / 83 %
6 decided · 5 happened
Invoice payment riskNot enough data
Brier
0.0400
Came true
100 %
Said / happened
80 % / 100 %
9 decided · 9 happened
Late productionNot enough data
Brier
0.1810
Came true
71 %
Said / happened
76 % / 71 %
14 decided · 10 happened
Lot expiringNot enough data
Brier
0.7275
Came true
0 %
Said / happened
85 % / 0 %
6 decided · 0 happened
01

Confidence, measured.

Most systems describe their forecasts as intelligent. Otiox shows its record instead: every call it made, what happened next, and the score that follows.
Prediction ledger
Written once, never edited. The score is computed from these rows only.
Frozen atRecordSaidOutcome
02 Sep, 08:14MO-PRED-000182 %Happened
05 Sep, 08:11RM-STEEL-0464 %Averted
09 Sep, 08:15INV-2026-018880 %Happened
14 Sep, 08:10LOT-2026-09785 %Did not
21 Sep, 08:12SO-PRED-000257 %Still open
01

Frozen predictions

The likelihood, the record and the moment are written once and never edited. No quiet revision after the fact, no benefit of hindsight.

A record that cannot be rewritten

Accuracy & calibration
Every prediction is frozen as it was made and checked against what then happened.
43 predictions with a known outcomeBrier score 0.2564last 180 daysModerate
Stockout riskNot enough data
Brier
0.1433
Came true
83 %
Said / happened
90 % / 83 %
6 decided · 5 happened
Invoice payment riskNot enough data
Brier
0.0400
Came true
100 %
Said / happened
80 % / 100 %
9 decided · 9 happened
Late productionNot enough data
Brier
0.1810
Came true
71 %
Said / happened
76 % / 71 %
14 decided · 10 happened
Lot expiringNot enough data
Brier
0.7275
Came true
0 %
Said / happened
85 % / 0 %
6 decided · 0 happened
02

A Brier score, explained

One measure of overall accuracy, in plain terms: zero is perfect, a quarter is a coin toss. Shown for the engine as a whole and for each detector.

Reliability, understood at a glance

Accuracy & calibration
Every prediction is frozen as it was made and checked against what then happened.
43 predictions with a known outcomeBrier score 0.2564last 180 daysModerate
Stockout riskNot enough data
Brier
0.1433
Came true
83 %
Said / happened
90 % / 83 %
6 decided · 5 happened
Invoice payment riskNot enough data
Brier
0.0400
Came true
100 %
Said / happened
80 % / 100 %
9 decided · 9 happened
Late productionNot enough data
Brier
0.1810
Came true
71 %
Said / happened
76 % / 71 %
14 decided · 10 happened
Lot expiringNot enough data
Brier
0.7275
Came true
0 %
Said / happened
85 % / 0 %
6 decided · 0 happened
03

Calibration per detector

A reliability chart for every detector compares what it said with how often it came true. Over- and under-confidence become visible immediately.

Trust placed where it is deserved

Command · accuracy
Was it right?
Predictions frozen and scored
MeasureValueMeaning
Known outcomesFortyLast six months
Brier scoreNear a quarterCoin-toss level
Calibrated detectorsNone yetNeed more data
Confidence shown as is
Score

Check every prediction against what happened.

Trust earned, not assumed.

04

An honest 'not enough data'

A detector whose outcomes are all of one kind cannot yet be calibrated — and Otiox says so, rather than inventing a confidence it does not have.

No false precision

Interventions
What was done about a risk, and what happened next, compared with the same kind of finding when nothing was done.
+ Log intervention
By actionModerate
Release the reservation
2 times · 2 likely helped
Likely helped in 100 % of decided cases
Post the completion
2 times · 2 likely helped
Likely helped in 100 % of decided cases
Reconcile lots with stock
1 time · 1 likely helped
Likely helped in 100 % of decided cases
Move PO arrival date
1 time · 1 likely helped
Likely helped in 100 % of decided cases
Swap the lot
1 time · 1 likely helped
Likely helped in 100 % of decided cases
Escalate PO approval
1 time · 1 likely helped
Likely helped in 100 % of decided cases
"Likely helped" means the risk did not happen after the action; it does not prove the action caused it.
05

Outcomes, including the averted

Each prediction resolves as happened, did not happen, averted by an action, or still open. A risk you prevented is not held against the engine.

Prevention counted, not hidden

Accuracy & calibration
Every prediction is frozen as it was made and checked against what then happened.
43 predictions with a known outcomeBrier score 0.2564last 180 daysModerate
Stockout riskNot enough data
Brier
0.1433
Came true
83 %
Said / happened
90 % / 83 %
6 decided · 5 happened
Invoice payment riskNot enough data
Brier
0.0400
Came true
100 %
Said / happened
80 % / 100 %
9 decided · 9 happened
Late productionNot enough data
Brier
0.1810
Came true
71 %
Said / happened
76 % / 71 %
14 decided · 10 happened
Lot expiringNot enough data
Brier
0.7275
Came true
0 %
Said / happened
85 % / 0 %
6 decided · 0 happened
06

Readiness, visible

How many predictions have a known outcome, over which window, and how many detectors have enough evidence to stand on their own.

An engine seen maturing over time

02

Every score traces back to its evidence.

Each figure on the page is computed from a single append-only ledger of predictions. Open any score and you arrive at the exact predictions and outcomes that produced it.
  1. 01
    The moment each prediction was made
  2. 02
    The record it concerned, with its full history
  3. 03
    What the engine said, and what then happened
  4. 04
    Averted outcomes linked to the intervention behind them
app.otiox.com
Prediction ledger
Written once, never edited. The score is computed from these rows only.
Frozen atRecordSaidOutcome
02 Sep, 08:14MO-PRED-000182 %Happened
05 Sep, 08:11RM-STEEL-0464 %Averted
09 Sep, 08:15INV-2026-018880 %Happened
14 Sep, 08:10LOT-2026-09785 %Did not
21 Sep, 08:12SO-PRED-000257 %Still open
03

How a prediction earns its score

Five steps — automatic, and entirely on the record.
  1. 01

    Predict

    A detector raises a risk with a likelihood

  2. 02

    Freeze

    The claim is written once and cannot be edited

  3. 03

    Observe

    The record is followed until the outcome is known

  4. 04

    Resolve

    Happened, did not, averted, or still open

  5. 05

    Score

    Brier score and calibration update per detector

04

Knowing which forecasts deserve your weight.

Calibration turns 'the system says eighty percent' into something a business can plan around. When a detector's eighty percent comes true roughly eighty percent of the time, buffers, promises and purchases can rest on it.
  1. 01
    Brier score and calibration error per detector
  2. 02
    How often each detector came true, and what it said
  3. 03
    Decided, happened, did-not, averted and open counts
  4. 04
    Rolling windows of 30, 90 or 180 days
app.otiox.com
Command · accuracy
Was it right?
Predictions frozen and scored
MeasureValueMeaning
Known outcomesFortyLast six months
Brier scoreNear a quarterCoin-toss level
Calibrated detectorsNone yetNeed more data
Confidence shown as is
Score

Check every prediction against what happened.

Trust earned, not assumed.

05

The idea, applied.

Scenarios from the demo workspace, walked through step by step — on the same data you receive when you sign up.
01

Forecast accuracy

Every prediction frozen and checked against what happened.
02

Decisions with evidence

Who decided, on what evidence, and what followed.
03

Lead-time drift

The supplier grew slower. The reorder point did not notice.
06

One ontology, seven ways of seeing it.

Every Command view reads the same things, the same relationships and the same history. An answer found in one becomes evidence in all the others.
01

Predictive Engine

Risks, forecasts and what-if scenarios, explained
02

Investigations & Causal Map

Why things happened, with evidence for and against
03

Decisions & Interventions

Who decided, on what evidence, and whether it worked
04

Explorer & Dependency Map

Everything a record is related to, and what is at risk
05

Time Machine

Any record or stock exactly as it was on a date
06

Rules & Controls

Company rules monitored, with attested control runs
07

Questions

01What is a Brier score?
A measure of how close probabilistic predictions were to what happened. Zero is perfect; 0.25 is what you would score by always saying fifty percent. Lower is better.
02Can anyone edit a prediction after the fact?
No. Predictions are frozen when they are made. Outcomes are added later; the original claim never changes.
03Why do some detectors show 'Not enough data'?
Calibration requires a mix of outcomes. If every case so far has gone the same way, a score would be meaningless, so Otiox says so instead of displaying false confidence.
04Does preventing a risk count against the engine?
No. When an intervention is recorded against a risk and the risk does not occur, the outcome is marked as averted and kept apart from 'did not happen'.
Otiox

Hold every forecast to account.

Open the demo workspace and review 43 real predictions, the outcomes that followed, and the scores they earned.

Get startedLearn more