TVCRpro

Tokens are the fuel of AI. TVCR is the mileage.

Benchmarked against your own trajectory, not against other users or models.

TVCR schematic gauge A schematic half-circle gauge with a single needle. Labeled TOKENS IN on the left and VALUE OUT on the right. Beneath the arc, three component labels read PQS, BVS, and TVCR — the three elements that combine into the Token Value Conversion Ratio. TOKENS IN VALUE OUT PQS BVS TVCR

Every human-to-AI interaction burns tokens to produce business value. TVCRpro measures the ratio. A single interaction is scored across eleven dimensions — five for the prompt, six for the response — and normalized for what it cost. The score is calibrated on representative interactions, the evaluator is version-locked per scoring epoch, and the result is one number that holds whether the interaction is an executive picking a vendor, an account manager closing a customer, or a foreman aligning a crew on safety, quality, and a shared goal — what changes is which dimensions carry weight, not the math.

Pre-pilot, design-intent.

How it works → Try the analyzer →

Vendor-reviewer checklist

The four questions a vendor reviewer asks first, answered above the fold.

What it is

A measurement architecture that scores a single human-to-AI interaction across eleven dimensions and combines them into a composite normalized by token consumption.

Read the methodology →

How it is calibrated

Rubrics are documented for all eleven dimensions. The evaluator is version-locked per scoring epoch and recalibrated against expert-judged seed sets when a vendor model changes.

Calibration approach →

Security posture

  • Email and identity in Switzerland (Proton AG).
  • Workspace data in Canada (Microsoft 365 Canada tenant).
  • Deletion-on-withdrawal via Power Automate workflow.

Full security brief →

What it’s not

  • Not a vendor-review or third-party-risk tool.
  • Not a productivity suite.
  • Not an AI training pipeline.
  • Not a public ranking of users, teams, or models.

Full limitations →

Eleven dimensions, two suites

Five dimensions for the human input. Six dimensions for the AI-generated output. Combined into TVCR.

PQS — five trainable prompt-quality dimensions

  1. Specificity SP
  2. Contextual Completeness CC
  3. Strategic Framing SF
  4. Constraint Definition CD
  5. Iteration Efficiency IE

BVS — six outcome-anchored business-value dimensions

  1. Decision Advancement DA
  2. Information Quality IQ
  3. Action Generation AG
  4. Efficiency Gain EG
  5. Organizational Value OV
  6. Reusability RU
TVCR = f(PQS, BVS, tokens)

Exact weighting and token-normalization constants are retained as trade secret and protected by the patent’s claims. Production rubric anchors are not disclosed in public materials.

Read the full methodology →

How scores look in practice

Eleven side-by-side examples of scored interactions, drawn from synthetic validation. Three previewed below.

Reproducible

Document drafting — vague vs. specific

Same task, two prompts. The specific prompt scores higher in PQS Specificity and Constraint Definition and produces fewer follow-up turns.

Realistic

Decision-making meeting — focused vs. drifting

Same agenda, two transcripts. The focused meeting scores higher in Decision Advancement and lower in token consumption per decision reached.

Library

Eight more, ready to read

Code generation, customer communication, data analysis, research synthesis, strategic planning, risk assessment, problem-solving, stakeholder alignment.

See all 11 comparisons →

Scoring is the mechanism. Coaching is the consequence.

A score on its own is just a number. The point of scoring is the coaching tip that comes with it — the single, specific, dimension-anchored thing the next interaction should do differently. Over ninety days of repeated scored interactions, that coaching loop produces a measurable trajectory in interaction quality.

The 90-day trajectory framework is published as a concept on the methodology page. The math, the projection coefficients, and the habit-recommendation mappings are not.

Read about the trajectory framework →