Analyzer

Paste one prompt and one response. Get a composite score across eleven dimensions, a bucket, and a single coaching tip — anchored to the dimension most likely to move the next interaction.

Phase 3.14 · worker live · v32 narrative Pre-pilot, design-intent. Endpoint api.tvcrpro.com/score · scoring runs server-side under the patent-pending TVCR formulation.

Try the analyzer

Prompt paste the human-authored prompt

Response paste the AI-generated response

Context optional · one line

The three fields below are optional. They stay in your browser, are never sent to the scoring engine, and disappear when you close this tab. They only adjust the wording of the framing paragraph below the score — your TVCR score is unchanged by them.

Stakes optional · context only · does not change the score

Deliverable type optional · what tangible thing did this interaction help produce?

Estimated outcome optional · context only · stays in your browser

Scoring runs on the TVCRpro Light Worker. Prompt and response are processed for scoring and discarded; the optional Stakes, Deliverable type, and Estimated outcome fields stay in your browser and never reach the worker. See /security.

What the analyzer does

You paste one interaction

One prompt. One response. Optional context line. No login, no team setup, no history.

It scores eleven dimensions

Five PQS dimensions on the prompt. Six BVS dimensions on the response. Token counts read from the model output.

You get one coaching tip

Anchored to the lowest-scoring dimension that has the most leverage. One sentence. No essay.

Sample report shape

The card below illustrates the structure the analyzer returns when you click Compute. Numbers shown here are illustrative; live results render in the card directly under the form.

The 78 below is not a grade in a vacuum. It is the mileage for this interaction — the value extracted, normalized by the tokens burned. Read against your own previous scores for the same kind of work, it tells you whether this interaction got better or worse mileage than the last one.

Analyzer · sample report illustrative

Prompt

I am the CEO of Northwind Logistics, a 24-person B2B SaaS company. We are designing a new internal analytics service. Workload is 80% structured reads with 6–12 table joins, plus event ingestion bursts up to 4,000 events/sec. We have zero production database experience and SOC 2 is in scope within 12 months. Recommend Postgres or MongoDB. Output format: 1) recommendation, 2) three-line rationale, 3) two top risks, 4) two-week proof-of-concept plan, 5) one decision-stopper that would change the answer.

Response

Recommendation: PostgreSQL. … (full response, 968 tokens, redacted in this sample)

Composite

78 Competent PQS 82 · BVS 74

PQS — Prompt Quality 82

SPSpecificity86
CCContextual Completeness78
SFStrategic Framing84
CDConstraint Definition80
IEIteration Efficiency82

BVS — Business Value 74

DADecision Advancement80
IQInformation Quality72
AGAction Generation68
EGEfficiency Gain76
OVOrganizational Value70
RUReusability78

Coaching tip · anchored to AG (lowest BVS)

Add an explicit owner and due-date format to the action list. Specify the verb form expected for each task — “publish,” “draft,” “schedule” — so the response converts cleanly into work items.

What the analyzer is not

Not a vendor-review or third-party-risk tool. It scores one interaction; it does not assess the model itself.
Not a productivity dashboard. It returns one report at a time, not a stream of metrics.
Not an AI training pipeline. Submitted text is processed for scoring and discarded per the published retention policy.
Not a leaderboard. Comparisons across people or teams are not surfaced from the public analyzer.

Status

The analyzer is live in Phase 3.14. Scoring runs through the TVCRpro Light Worker at api.tvcrpro.com. Full dimensional definitions live on /methodology; eleven real captured pairs live on /comparisons showing the same scoring shape applied to actual model responses.

Pilot interest: write to pilot@tvcrpro.com. NDA-first.