diff --git a/composer.json b/composer.json
index 6cb96acd..3d185aea 100644
--- a/composer.json
+++ b/composer.json
@@ -18,11 +18,6 @@
     "bin": [
         "hydra-gates/bin/hydra-gates"
     ],
-    "autoload": {
-        "psr-4": {
-            "OCA\\OpenRegister\\Contract\\": "hydra-gates/contracts/"
-        }
-    },
     "require": {
         "php": ">=8.1"
     },
diff --git a/docs/hydra/operations/app-health.md b/docs/hydra/operations/app-health.md
index 850cffd3..6975abca 100644
--- a/docs/hydra/operations/app-health.md
+++ b/docs/hydra/operations/app-health.md
@@ -287,10 +287,19 @@ day. Neither number was a decision; both were the absence of one.
 
 The cost was not theoretical. hydra-gates v1.8.0 shipped an
 `ObjectServiceInterface` **without `patchObject()`**, and because the package
-claims `OCA\OpenRegister\Contract\` in its composer autoload — a *longer*
-psr-4 prefix than openregister's own — a stale copy in **any** app's vendor
-directory defined that contract for the whole instance. Measured:
-softwarecatalog's vendor was supplying openregister's interface. Separately,
+*used to claim* `OCA\OpenRegister\Contract\` in its composer autoload — a
+*longer* psr-4 prefix than openregister's own — a stale copy in **any** app's
+vendor directory defined that contract for the whole instance. Measured:
+softwarecatalog's vendor was supplying openregister's interface.
+
+> ✅ **That mechanism is gone.** The prefix was removed from the package
+> (ConductionNL/.github#531); the contracts still ship, but a consumer now opts
+> in from its own test bootstrap behind `interface_exists()`, which is
+> order-independent. Four apps needed the opt-in — buildiq, decidiq, filinq and
+> stackiq. The others either ship their own contract stubs (dossiq, pipelinq,
+> shillinq), own the real one (openregister), or never reference it.
+
+Separately,
 nc-vue's AI-companion singleton landed in 2.7.0, so every app below it rendered
 a **second** companion hex beside hermiq's on every page. Both fixes had existed
 upstream for weeks.
diff --git a/hydra-gates/README.md b/hydra-gates/README.md
index 5f346332..b2aa053b 100644
--- a/hydra-gates/README.md
+++ b/hydra-gates/README.md
@@ -146,6 +146,52 @@ gap, not a pass — see *Reading a green*.
 
 ---
 
+## The shipped OpenRegister contracts are opt-in, not autoloaded
+
+`hydra-gates/contracts/` ships `ObjectServiceInterface` and
+`ObjectEntityInterface` so a leaf app can typehint OpenRegister's data-access
+surface under PHPUnit without the OpenRegister app installed.
+
+They are **not** autoloaded. This package used to declare
+
+```json
+"autoload": { "psr-4": { "OCA\\OpenRegister\\Contract\\": "hydra-gates/contracts/" } }
+```
+
+and that was a bug, not a convenience. The prefix is *longer* than
+openregister's own `OCA\OpenRegister\` → `lib/`, PSR-4 is longest-prefix-wins,
+and the package installs into every consumer's `vendor/` — so whichever app's
+autoloader registered first defined OpenRegister's contract **for the whole
+process**. Measured on a real instance: softwarecatalog's vendored copy was
+supplying openregister's own interface, and a v1.8.0 copy without
+`patchObject()` broke callers of the real one (ConductionNL/.github#531).
+
+A consumer that needs these interfaces requires them from its own test
+bootstrap, behind a guard:
+
+```php
+foreach (['ObjectEntityInterface', 'ObjectServiceInterface'] as $contract) {
+	if (interface_exists('\\OCA\\OpenRegister\\Contract\\' . $contract) === false) {
+		$shipped = __DIR__ . '/../vendor/conduction/hydra-gates/hydra-gates/contracts/' . $contract . '.php';
+		if (file_exists($shipped) === true) {
+			require_once $shipped;
+		}
+	}
+}
+```
+
+Put it **before** anything that implements the interface — a stub entity that
+`implements \OCA\OpenRegister\Contract\ObjectEntityInterface` fatals inside the
+bootstrap otherwise, which is a dead run rather than a failed test. And put it
+in the bootstrap `phpunit.xml` actually loads: several apps ship two or three.
+
+`interface_exists()` rather than a fallback autoloader, because
+`spl_autoload_register` appends relative to *registration order*, and
+registration order across independently loaded apps is exactly the thing nobody
+controls. Asking whether the interface is resolvable is order-independent.
+
+---
+
 ## What it needs at runtime
 
 `bash`, `git`, `python3` (about twenty gates are Python helpers) and `node`