Skip to content
MetaMinds

Tier 1 · instant download

Hallucination evaluation starter kit

Measure how often the system is wrong, on a set you keep and can rerun.

Buy it now · $490

What you get

  • Golden-dataset structure
  • Groundedness rubric
  • CI wiring guide, so the number is produced on every change rather than once

$490

One payment. USD.

Your receipt, invoice and download link go here. You can still change it at the payment step, and the file follows whichever address pays.

It decides the tax treatment printed on your invoice. Nothing else.

Buying as

$490 once. You are buying from NexManas Tech Pvt Ltd, trading as MetaMinds, which issues your invoice. Payment is taken on Razorpay’s own page, so your card details never reach this site.

All products and prices
Who this is for

Three ways to tell whether this applies to you.

If none of these describes your situation, this is the wrong product, and the catalog has nine others.

  • Your system answers questions and being confidently wrong is a real cost
  • You need a number for how often it is wrong, on a set you keep and can rerun
  • You are tired of judging quality by reading a handful of outputs
What is in it

The file, before you buy it

Read from the artifact itself rather than described, so what is printed here and what arrives cannot drift apart.

Every column, as it ships

  1. 01item_id
  2. 02question
  3. 03expected_behaviour
  4. 04gold_passage_ids
  5. 05gold_answer_summary
  6. 06category
  7. 07difficulty
  8. 08source_of_question
  9. 09added_date
  10. 10last_reviewed
  11. 11holdout
  12. 12run_label
  13. 13assigned_label
  14. 14reviewer_note

hallucination-eval-starter-kit.html

7 sections, in the same download.

  • Why a correct answer can still be a failure
  • The five-label groundedness rubric
  • The golden dataset
  • Who applies the labels
  • Wiring it into CI
  • What MetaMinds runs
  • What the number still does not tell you
14 columns, with 3 worked rows and 15 blank rows ready to fill, every one named above.
What it resolves

What makes a golden set last

The columns split three ways, and the split is the method. Most home-made evaluation sets have only the first group.

14 columns that split three ways: 4 you write once, 7 that keep the set honest over time, 3 that one run fills in.

A set with only the first group is the one every team builds and then stops trusting. holdout and last_reviewed are what stop it being quietly tuned against, and source_of_question is what lets somebody else judge whether it was ever representative.

4Written onceThe item itself, and what a grounded answer would cite.
  • question
  • expected_behaviour
  • gold_passage_ids
  • gold_answer_summary
7Kept honestProvenance, difficulty, and whether an item is held out of tuning.
  • item_id
  • category
  • difficulty
  • source_of_question
  • added_date
  • last_reviewed
  • holdout
3Per runWhat a single evaluation pass records against each item.
  • run_label
  • assigned_label
  • reviewer_note

SourceColumn names read from deliverables/T1-ASR-02/golden-dataset-template.csv at build time.

Before you buy

Delivery and license

Stated here rather than in the terms, because these are the questions asked before a purchase and not after it.

Delivery
Immediate download.
License
Single organization, internal use, perpetual. Adapt it for your own systems freely. Do not resell it or publish it as your own.

HTML and CSV · 14-column golden-dataset template, five-label rubric, CI wiring guide · Version 1.0 · 21 September 2026

Related

What usually goes with this

Each of these does something this one does not, and the line under it says what. They are separate items, not an upsell path.

  • RAG architecture decision guide

    Groundedness is mostly a retrieval property, so most of what moves the number is in the architecture decisions rather than the prompt.

    $290

  • Prompt injection test suite

    This measures whether answers are grounded. The injection suite measures whether they can be steered.

    $390

Questions

Before you ask

These are the same strings this page puts in its structured data, so a person and an answer engine read identical text.

Can I buy this today?

Yes. It is $490, one payment, in USD, and the file downloads as soon as the payment clears.

What if I want the work done rather than the file?

The fixed-scope engagements on the services page cover the same ground with a person doing the work, at a published price and a fixed duration. Those are available now: a vendor AI assessment at $2,500 over three days is the smallest, an EU AI Act readiness assessment is $4,900 over one week, and an ISO/IEC 42001 gap assessment is $8,900 over two to three weeks.

What exactly is Hallucination evaluation starter kit?

Measure how often the system is wrong, on a set you keep and can rerun. It is a Tier 1 instant product from MetaMinds, listed at $490 in USD as a single payment. Tier 1 is the entry rung of a three-tier catalog: instant products from $99 to $890, fixed-scope engagements from $2,500 to $8,900, and scoped builds from $18,000.

What happens after I pay?

The download link and your invoice are emailed to you immediately, and the link stays good for 30 days and up to 20 downloads. Because a file you have already downloaded cannot be returned, the checkout asks you to confirm before paying that downloading ends your 14-day right to withdraw. That is Article 16(m) of the Consumer Rights Directive, and it is why the box is there and why it starts unticked.

It downloads the moment you pay.

No call, no scoping, no waiting. If you would rather a person did this work, the engagements on the services page cover the same ground at a fixed price and a fixed duration.

Buy Hallucination evaluation starter kit