Ontology Navigator takes an AI investment decision from the first question to a record two named people sign. People score, value is always a range, and doing nothing is always on the table.
Likely 0.9m
Two named people sign the record.The software never signs.
And 5,565 jobs and outcomes the Navigator reads, each one a row, never invented.
Active rows, counted on 29 October 2026.
EARNED, one question per turn, then one report.
Each takes ratings from people. None supplies a rating of its own.
The pitch with the best demo moves first, its value is quoted as a single number, and nobody writes down what would show it failed. Ontology Navigator slows that moment down on purpose.
“The medical writing assistant will save 1.2m a year. Approve it this quarter.”
“Value 0.4m to 1.6m, likely 0.9m. Compared with doing nothing and doing it internally. Investigate first: one trial, one measure, a stop rule.”
The owner chooses the result. Two named people sign. The dissent stays on the record.
Six questions, one letter each, across the Navigator's five stages. You bring the ratings and the judgment. The plugin asks the right question at each step. The connector holds the ontology, runs the calculators on your ratings and keeps the brief.
Thirteen facilitated skills for Claude. One question per turn, then one report, from framing the decision to reviewing the portfolio.
A paid service, open by invitation during the beta. It holds the ontology, runs the calculators on your ratings and keeps the brief between sessions.
Each skill asks one question per turn, then writes one report.
Installing the plugin does not give access to the Connector.
Each step leaves something your team can challenge: the value range, the private ratings, the ranked options against doing nothing, the readiness gaps, the trial and the signed record.
Likely 0.9m
Basis stated by the team: hours saved on first drafts, at the team’s own rate. A single figure is refused.
Ranked by opportunity from importance and satisfaction, rated by three people alone.
| Criterion | Weight | Writing assistant | Reuse engine | Doing it internally | Doing nothing |
|---|---|---|---|---|---|
| Value to the outcome | 40 | 3 | 3 | 2 | 0 |
| Feasibility | 25 | 3 | 2 | 3 | 4 |
| Risk | 20 | 2 | 3 | 3 | 4 |
| Time to value | 15 | 3 | 2 | 1 | 0 |
| Score out of 100 | 70.0 | 65.0 | 57.5 | 45.0 |
Criteria scored 0 to 4 by the team. The calculator adds up their ratings; it does not choose the result.
Illustrative example. The decision, people, ratings and figures are fictional; the layout shows the kind of output each step produces. Every calculator takes ratings from people.
42,726 active entities and 42,036 relations across 13 pillars. The Navigator reads a structured map of the work: value drivers, the outcomes behind them, and the measures, capabilities and processes that connect them. Where the map has no row, the answer says so.
Counted on 29 October 2026: active entities across 13 pillars, shared codes counted once.
The plugin carries 87 test prompts, including ones it must refuse. Their results are not published yet. When they are, they appear here with their method and their limits.
Each scorer rates alone before discussion; an AI name as a scorer is refused.
A wide spread between scorers is flagged, never averaged away.
Low, likely and high, with its basis; a single number is refused.
Doing nothing and doing it internally sit on every comparison.
Proceed, Investigate first, or Handle it internally; the calculator does not choose.
The record is prepared for two people; the software never marks it signed.
Rules enforced in the connector, not in a prompt. Not a test result.
Thirteen EARNED skills.
Bring one AI decision with an owner and more than one option, and frame it.
For the ontology, the calculators and the brief between sessions.
The Navigator prepares the record. The people who own the decision make it and sign it.