AssayStart the demo

Brand content, tested before it ships.

Assay checks captions, emails, product pages and creatives against a brand's approved rules, facts and palette. Every finding quotes its evidence, cites its source and comes with a fix that keeps your message.

Start the demo, no sign-upCheck your own draft
Palette · VIS-01ΔE 18Off palette; the limit is 10
Headline contrast · A11Y-011.7:1Fails AA; needs 3:1
Streaks of gold rubbed onto a black touchstone.
Gold rubbed on a touchstone. The colour of the streak told an assayer how pure the metal was.

Try it on a caption

The rule checks run as you type, in your browser, against Velvra's published rulebook and Claim Ledger. No account, and nothing is sent anywhere.

Or try

Velvra rulebook v3.2 · Instagram

4 findings

OMG (finding 3, medium severity: Never hype) our new sunblock (finding 4, low severity: Use the product name) is clinically proven to erase dark spots in just 7 days (finding 1, high severity: Claim contradicts an approved fact)! The best serum in India (finding 2, high severity: Unsupported superlative). Shop now via the link in bio
  1. 1Claim contradicts an approved fact HighClaim · CLAIM-01 · F-112 · fix: “shown in our 4-week study to help skin look more even (82% saw a visibly more even skin tone)*”
  2. 2Unsupported superlative HighClaim · CLAIM-03 · fix: “Lightweight protection for every day”
  3. 3Never hype MediumTone · VOICE-02 · fix: “remove”
  4. 4Use the product name LowTerminology · TERM-04 · fix: “Daily Shield SPF 50”

How it works

  1. Check
  2. Fix
  3. Approve
  4. Stamp

Check: Rule checks run in the browser and Gemini judges tone. Each finding quotes the draft and cites the brand book page or the approved fact.

Fix: Each fix rewrites only its own words, uses the fact's approved wording and footnote, and is checked against the creator's key message.

Approve: Claims go to Compliance, then the draft to a Reviewer. A creator can't wave away a high-severity finding alone.

Stamp: Approval writes a Hallmark: the content's SHA-256, the rulebook version and every decision, chained to the one before.

Monsoon launch: Instagram post

Velvra rulebook v3.2 · Women 22–35, metro India

4 findings, all fixedv2 · intent keptIn review · Compliance clearedApproved · Hallmark #1

Our new Daily Shield SPF 50 is shown in our 4-week study to help skin look more even (82% saw a visibly more even skin tone)!* Lightweight protection for every day. Shop now via the link in bio.*In-house study, 40 participants, 2025.

The sample Velvra creative: a Daily Shield tube on Rose Clay, after the fixes.
Headline contrast
1.7:15.7:1
Palette distance
ΔE 18ΔE 2
  1. 1Contradicts F-112: the study ran 4 weeks, not 7 daysCLAIM-01 · Brand Book p.18 §6.1HighFixed
  2. 2A superlative with no fact behind itCLAIM-03 · Brand Book p.19 §6.3HighFixed
  3. 3Never hypeVOICE-02 · Brand Book p.6 §2.2MediumFixed
  4. 4Say Daily Shield SPF 50, not sunblockTERM-04 · Brand Book p.12 §4.2LowFixed

Meera · Compliance · cleared CLAIM-01 and CLAIM-03

Arjun · Reviewer · approved v2

sha256 …

Beyond a style checker

Cited findings

An AI finding must quote real text and cite a real rule, or the evidence verifier drops it and says how many it dropped.

1ClaimHighRule check + Claim Ledger

Claim contradicts an approved fact

clinically proven to erase dark spots in just 7 days

“7 days” doesn't match F-112, which says 4 weeks. “Clinically proven” is not approved wording for an in-house study.

CLAIM-01 · Velvra Brand Book v3 p.18 §6.1 · F-112 · owner R&D · valid to 31 Mar 2027

Fixes that keep the message

Fixes replace only their own words, use the fact's approved wording, and are checked against the creator's Intent Card.

Suggested minimal edit

our new sunblock Daily Shield SPF 50 is clinically proven to erase dark spots in just 7 days shown in our 4-week study to help skin look more even

Adds footnote: *In-house study, 40 participants, 2025.

Intent kept: key message intact and strengthened

Guideline Health

Dismissals and contests show which rules people find unclear, with the real examples and a drafted clarification.

Unclear score, seeded demo history

  • MSG-0146
  • VOICE-0226
  • TERM-040
  • CLAIM-010
  • A11Y-020

MSG-01: 2 dismissed and 12 contested in 14 findings.

Hallmarks

Each approval records the content's hash, the rulebook version, every decision and the AI used, chained to the last.

Hallmark · verified

Rulebook v3.2 · Compliance and Reviewer approvals

sha256 …

  • CLAIM-01 · fixed, then approved by Compliance
  • CLAIM-03 · fixed, then approved by Compliance
  • VOICE-02 · dismissed: “launch-day post, tone agreed”

Six kinds of issue, each with its source:

hype in the tone (VOICE-02), banned words like “whitening” (TERM-05), instant-results promises (MSG-02), a discount the Claim Ledger says is wrong (F-107), a pink ΔE 18 off the palette (VIS-01), and a headline at 1.7:1 contrast (A11Y-01).

Four men in an assay office stand around a very large balance, with bars of gold on one pan.
Weighing gold in an assay office in New York, from a magazine of 1917.

Test it, then mark it

An assay office tests a metal for purity, and only then strikes a hallmark into it. Assay does the same for brand content: it tests a draft against the rules, facts and palette the brand has approved, then writes a Hallmark that anyone can check later.

Two hands hold a steel punch on a small piece of silver and strike it with a hammer.
A steel punch and a hammer, striking a mark into silver.

Made in Mumbai

CodeStar 30, Round 3

Source on GitHub

Built to be trusted.

It works without AI: claims, terms, disclaimers, contrast and palette are checked in code, and Gemini judges only tone and messaging. Drafts stay in your browser; only redacted paragraphs reach the AI route, and Confidential mode sends nothing. People decide: nothing is applied automatically.

Built by Shlok Sudhakar Shinde, team 24K Magic, at SPIT Mumbai. The About page lists every service, library and licence, and the free-tier data terms.