Recorded, deterministic proof

Make stale prose fail loudly.

One binding connects a command in source to its README claim. When the source changes alone, the check fails with the exact stale region; repair rewrites that region and nothing else.

Evidence sha256: b762d4623c9637fdee0922acf02db6e82d0fd2851fd18b365d75992c523ba33b

Binding: public-command

Source · command.txtsourcebound check --changed
mismatch
Document · README.mdsourcebound check

The complete loop

One binding. Three observable states.

The source changes in state two. The document is left untouched until the declared repair runs in state three.

01

Current

Source and README agree. The gate has nothing to repair.

sourcebound check exit 0

[current] public-command: README.md
02

Drift caught

The source changed; the README did not. The gate names the stale binding.

sourcebound check exit 1

[drift] public-command: README.md
--- README.md
+++ README.md (derived)
@@ -7,5 +7,5 @@
 ## Command
 
 <!-- sourcebound:begin public-command -->
-sourcebound check
+sourcebound check --changed
 <!-- sourcebound:end public-command -->
03

Repaired and verified

The declared region is regenerated, then the same check passes.

sourcebound drive exit 0

[repaired] public-command: README.md
--- README.md
+++ README.md (derived)
@@ -7,5 +7,5 @@
 ## Command
 
 <!-- sourcebound:begin public-command -->
-sourcebound check
+sourcebound check --changed
 <!-- sourcebound:end public-command -->
drive: repaired 1 document(s); implemented policy checks passed

sourcebound check exit 0

[current] public-command: README.md

Why this demonstration exists

See the failure before adding the gate.

Maintainers deciding whether source-bound documentation is worth adding to a repository.

What is running

  • No account, credentials, backend, or database.
  • The outputs below come from a generated temporary repository.

Evidence boundary

What this proves.

Try the same loop in a repository.