Verification at the fault lines

Catches AI hallucinations in WordPress code

Every AI coding tool hallucinates function names, deprecated APIs, and wrong parameters. Every detection tool uses another AI to check. VFault checks against the actual source code.

$ vfault check "Use wp_is_rest_request() to check REST calls"

NOT FOUND — HALLUCINATION CAUGHT
  wp_is_rest_request
  Did you mean: wp_is_serving_rest_request

$ vfault check "Get user data with get_currentuserinfo()"

DEPRECATED in 4.5.0
  Use wp_get_current_user() instead

$ vfault check "Use wp_enqueue_script() and add_action()"

VERIFIED
  wp_enqueue_script (since 2.1.0)
  add_action (since 1.2.0)
<1ms
Lookup speed
0%
False positives
100%
Hallucination catch rate
25,398
Verified triples
Every competitor uses AI to check AI
VFault checks against the actual WordPress source code. No AI in the verification layer. The function either exists or it doesn't.
VFault Every competitor
MethodSource code lookupLLM-as-judge
Speed<1ms152ms+
False positives0%5–10%
Provides correctionsYes + closest matchNo — score only
Can verifier hallucinate?ImpossibleYes
CostFree CLIPaid
Five verification layers
Every claim is checked against 25,398 triples parsed directly from the WordPress source code.

Hallucinated functions

Catches non-existent function names and suggests the correct one. wp_register_scripts → wp_register_script

Deprecated functions

Flags deprecated functions with the version they were removed and the replacement to use instead.

Outdated parameters

Compares stated parameter signatures against the current source. Catches changes like $in_footer → $args in WP 6.3.

Fabricated hooks

Verifies action and filter hooks against 2,346 hooks extracted from WordPress core. Non-existent hooks are flagged.

Clean code confirmed

Correct functions are verified with the version they were introduced, source file, and full parameter signature.

{}

Class methods

Recognises wpdb, WP_REST_Request, and WP_Query methods without false positives on object-oriented code.

Tested against three frontier models
50+ WordPress questions across 10 categories. Standard and hard-mode testing.
Grok (xAI) — 50 standard questions
94 claims, 0 hallucinations
All verified. No false positives.
Gemini (Google) — 50 standard questions
93 claims, 0 hallucinations
All verified. No false positives.
Claude (Anthropic) — 8 hard questions
61 claims, 4 hallucinations
Gate caught all 4 with correction suggestions.
Deliberately fabricated input
100% catch rate
Every fake function name caught. Corrections provided for all.
Free and open source
VFault is free to download and use. Hosted API, VS Code extension, and team plans coming soon.
Paid tiers — coming soon
In development
 
  • Hosted API (no setup needed)
  • VS Code extension
  • Additional language shards
  • CI/CD GitHub Action
  • Team dashboard
  • Custom codebase shards
Follow on GitHub

Stop trusting AI output blindly

VFault checks against the source. Free, open source, ready to use.