LatticeLock
Methodology

How LatticeLock tests a website

Every result comes from three layers of evidence, from a quick capability check to cryptographic proof. This page explains each test, how sites are graded, how the India study is run, and what a scan cannot see.

3layers of evidence per site
~15short handshakes per scan, nothing else
100%of ready sites proven with a real ML-KEM exchange
4,652Indian sites measured in November 2026

Part 1: What happens when one site is scanned

A scan opens about 15 short connections to the site. Each is an ordinary encrypted handshake, the same opening exchange a browser makes before loading a page. No pages, logins or forms are touched.

Layer 1. Support: can the server do post-quantum?

We test seven key-exchange methods one at a time: the hybrid X25519MLKEM768 (used by Chrome, Firefox and Cloudflare), two other standard hybrids, two pure ML-KEM methods, the obsolete pre-standard Kyber draft, and classical x25519 as a baseline. For each, we offer the server only that method with no key data. Under TLS 1.3 rules a server that supports it must reply asking for a key of that type; one that does not, refuses.

Layer 2. Proof: does a real post-quantum exchange work?

For every server that passes Layer 1 we perform a real exchange:

  1. We generate a genuine ML-KEM-768 key (NIST FIPS 203) plus a classical X25519 key and send both.
  2. The server returns its half: an ML-KEM ciphertext and its own X25519 key.
  3. We combine both into the shared secret, ML-KEM first, as RFC 10024 specifies, and derive the handshake keys with the TLS 1.3 key schedule.
  4. We decrypt the server's first encrypted message.

The decryption uses authenticated encryption (AES-GCM), which only succeeds if both sides calculated exactly the same secret. A server cannot pass by merely claiming support. Our test suite confirms the check fails when the secret is deliberately corrupted.

Layer 3. Real visitors: do Chrome users actually get it?

We offer exactly what Chrome offers, post-quantum and classical together, and record which one the server picks. A server that can do post-quantum but chooses the classical option leaves its real visitors unprotected, even though Layers 1 and 2 pass.

General security health

We also record whether TLS 1.3 (required for post-quantum) is supported, whether obsolete TLS 1.0 and 1.1 are still enabled, and whether the certificate is trusted, strong enough and not about to expire.

Platform detection

One normal web request lets us read the response headers, which usually reveal what serves the site (Cloudflare, Akamai, AWS, Google Cloud, Azure, nginx, Apache, IIS and others). That decides which step-by-step fix plan the site owner sees.

Part 2: Grading

CheckPoints
Post-quantum proven with a real handshake (Layer 2)30
Real visitors get it (Layer 3)10
TLS 1.3 supported20
TLS 1.0 and 1.1 switched off10
Certificate trusted10
Certificate key strong enough10
Certificate valid for 14 or more days10

A is 90 or more, B 75, C 60, D 40, F below. Two rules are stricter than the points: a server that claims post-quantum support but fails the real handshake is treated as not protected, and a server that gives real visitors the classical option cannot receive an A.

Part 3: The India study

  1. How the list is built: hand-checked major Indian brands (many big Indian companies use .com addresses), plus Indian domains from the research-grade Tranco ranking of popular websites, taken in popularity order. Each site is placed in one of ten sectors: by its official domain type (.gov.in and .nic.in are government; .ac.in, .edu.in and .res.in are education), by the hand-checked list, or by keywords in its name (for example "bank", "hospital", "news"). Sites that match no sector are grouped as Business & other. Each sector has a fixed share of the list, so no single sector dominates.
  2. Which sites count where: every site that answers counts in the national figure. The sector figures use only sites whose sector is certain (hand-checked brands, and official .gov.in, .nic.in and .ac.in domains), because name keywords alone sometimes misfile a site, for example a job portal whose name contains "times". All other sites are shown together in one row, "Other sites (sector not confirmed)", so the sector table adds up to the national total.
  3. Unconfirmed results: a site that advertises post-quantum support but fails the real-handshake proof is counted as not ready, and we do not name or describe it until the cause has been re-tested.
  4. A fixed list: the same list is re-scanned each month, so changes over time reflect real changes, not a different sample.
  5. Domain fallback: if a bare domain does not work, we try its www address.
  6. Exclusions: sites that do not exist, refuse connections or never complete a secure handshake are skipped and left out of every figure, not counted as failing (348 in November 2026).
  7. CDN classification: for each ready site we record whether a known CDN answered on its behalf. Some CDNs hide themselves, so "through a known CDN" is a minimum figure.
  8. Publication: only sector totals are published, never a named list of organisations. Any site owner can check their own result with the free scanner.
  9. Benchmark: global figures come from F5 Labs, 2026 State of PQC on the Web (top 1M sites).
  10. Monthly repeat to build the trend line.

Part 4: How the scanner itself is checked

Part 5: Limitations

  1. Only the front door is visible. Behind a CDN, the connection from the CDN to the organisation's own server cannot be seen by any outside scan.
  2. One vantage point. Scans run from one location in India; large sites sometimes behave differently by region.
  3. A snapshot. A site can change the day after a scan, which is why we re-measure monthly.
  4. The list shapes the result. These are major sites chosen by sector, not a random sample of all Indian websites.
  5. Platform detection is a best estimate from what a site's headers reveal.
  6. Certificates are not post-quantum anywhere yet. Public certificate authorities do not issue ML-DSA certificates, so this is reported for information but not penalised. Authentication cannot be attacked retroactively, unlike key exchange.

Check your own site   Read the India report →