Attested interpretive units protocol
The protocol governs how Better Robots.txt selects and hashes canonical units.
Selection rule
Only units with interpretive risk should be attested.
A good candidate is a sentence or compact statement that blocks a known misreading:
- policy signal does not prove enforcement;
/checkscore does not prove compliance;- llms.txt guidance does not replace robots.txt;
- first-party sources do not override qualified criticism;
- hash does not prove truth.
Digest rule
Hash the canonical unit envelope, not a weak scalar.
The envelope includes unit ID, claim class, authority scope, canonical source, selector and canonical text.
Audit rule
An audit must fail when:
- a unit lacks canonical text;
- two units share the same digest without justification;
- a unit claims that a hash proves truth;
- a unit converts policy into crawler obedience;
- a unit suppresses qualified external criticism.
The current audit script is scripts/audit-interpretive-authority.mjs.