Skip to content
The trust layer between spec and software

The PR passed.
The requirement didn't.

Norma verifies that software matches the specification before it ships.

AI agents write the code. Norma compares your requirements to the implementation, finds the drift, and generates the fix.

Works with: Claude Code · Cursor · Windsurf · GitHub Copilot · Bolt

Spec Fidelity model: claude
0
/ 100
auth-api ⚠ 1 contradiction
01

Every check was green. The product was still wrong.

Requirement
Limit withdrawals to $500 per day.
Implementation
const dailyLimit = 5000;
Existing tools
Unit tests pass
Linter passes
Pull request approved
Norma
Contradiction detected
41 Spec Fidelity

Nothing was broken. The code was clean. It was off by a factor of ten on the one thing that mattered, and the only tool that read the requirement caught it.

Read the spec. Find the drift. Generate the fix.

Norma doesn't stop at detection.

Requirement missing Norma explains why Norma suggests the changes Norma generates a repair prompt AI fixes the implementation
02

See Norma catch it.

norma · analyze
cancel appointments withdrawal limit password hashing
edit either box, then run
Illustrative. The full tool judges with a model and reads your whole codebase.
03

One more step in the loop. The one that was missing.

Spec AI builds Norma verifies Ship
01 EXTRACT

Requirements

The spec becomes atomic, testable claims with stable ids.

02 INDEX

Implementation

The code is indexed and the relevant files retrieved per claim.

03 JUDGE

Verdicts

Each claim is met, partial, missing, or contradicted.

04 SCORE

Fidelity

Weighted by priority into one reproducible number.

met · fullpartial · halfmissing · zerocontradicted · zero, plus a flag
04

Why existing tools miss this

ToolWhat it checks
TestsBehavior
LintersStyle
Code reviewCode quality
Security scannersVulnerabilities
NormaRequirement fidelity

Every other tool measures your code against an external standard. Norma is the only one that measures it against your spec. That gap is the whole reason it exists.

05

Generation is solved. Verification is the missing half.

Generation
  • Claude Code
  • Cursor
  • Windsurf
  • Bolt
Verification + Repair
  • Norma

The new software stack writes code faster than anyone can review it. Norma is the layer that checks the output against the intent and generates the fix when they diverge, so the speed does not cost you correctness.

06

Built for teams shipping with AI

FOUNDERS

Founders

Verify outsourced and AI-generated work before it reaches customers.

PRODUCT

Product Managers

Catch requirement drift before release, not in the bug report.

ENGINEERING

Engineering Managers

Add a spec-fidelity gate to CI and stop low-fidelity PRs at the door.

DEVELOPERS

Developers

Validate your implementation against intent without writing the eval.

A missed requirement costs
rework QA cycles delayed launches customer trust token spend
Norma catches drift before a human wastes time on it.
07

Introducing the Spec Fidelity Score

A single number for how faithfully an implementation matches its specification. Read it like Lighthouse reads performance. Spec Fidelity is the category, and Norma is the tool that measures it.

95+
Ship
Faithful to the spec. Merge with confidence.
80 to 94
Review
Mostly there. A human should look at the gaps before release.
below 80
Investigate
Material drift, or a contradiction. Do not ship on trust.

Every Spec Fidelity Score should stamp the model and prompt version, so it is reproducible and contestable. The report has an open schema, which is how the number becomes a shared standard rather than one vendor's metric.

08

One command. Or one workflow.

terminal# install once, then run on any repo with a spec
$ npm i -g @norma-sh/norma
$ export ANTHROPIC_API_KEY=sk-...
$ norma check
# no install needed: npx @norma-sh/norma check

  Spec Fidelity Score  55/100
  drift located, per requirement
.github/workflows/norma.ymlon: pull_request
steps:
  - uses: norma-sh/norma@v1
    with:
      threshold: "80"
      anthropic-api-key: ${{ secrets... }}
▲ scores every PR, comments the result, fails below threshold
09

Designed for AI-native development

Norma sits on top of whatever wrote your code. It only needs a spec and an implementation, so it works with the tools your team already uses.

Claude CodeCursorWindsurfGitHub CopilotBoltLovableDevin

and any spec format: Spec Kit, AGENTS.md, or a plain PRD.

Did the build keep its promises?

AI can write the code.
Who verifies the intent?