diff --git a/docs/schema-proposal.md b/docs/schema-proposal.md
index 6ca70c3..cab4a49 100644
--- a/docs/schema-proposal.md
+++ b/docs/schema-proposal.md
@@ -82,6 +82,58 @@ enrichment:
   submitted upstream
   ([LiturgicalCalendarAPI PR #718](https://github.com/Liturgical-Calendar/LiturgicalCalendarAPI/pull/718)).
 
+## Identifier durability
+
+Three cases are already on this repository's record. None is a mistake of execution; each
+is what a name-derived identifier does when the name, the type, or the source index moves.
+
+- **The strip rule is already broken at scale.** The scheme requires generic type words to
+  be stripped, because "the *type* is an attribute, not part of the identity" — yet 58
+  seeded IDs carry the Italian type word (`circ:it-arcidiocesi-di-acerenza`,
+  `circ:it-arcidiocesi-di-agrigento`, `circ:it-arcidiocesi-di-amalfi-cava-de-tirreni`, and
+  55 more), and 31 ordinariates carry theirs in ten languages:
+  `circ:au-military-ordinariate-of-australia`, `circ:de-deutsches-militarordinariat`,
+  `circ:it-ordinariato-militare-per-l-italia`, `circ:es-arzobispado-castrense-de-espana`,
+  `circ:pl-ordynariat-polowy-wojska-polskiego`, `circ:sk-vojensky-ordinariat-slovenska`,
+  `circ:hu-magyarorszagi-katonai-ordinariatus`, `circ:lt-lietuvos-kariuomenes-ordinariatas`,
+  `circ:ba-vojni-ordinarijat-u-bosne-i-hercegovine`, `circ:br-ordinariado-militar-do-brasil`,
+  and the rest. Those two cohorts are 89 of the seed's 2,935 entries (two further entries,
+  `circ:fr-diocese-aux-armees-francaises` and `circ:us-archdiocese-for-the-military-services`,
+  carry a type word outside them). Rule 1 holds that type is not identity; 89 identifiers
+  record the opposite, and correcting them would rename 89 live IDs.
+- **A live ID has already been renamed.** `circ:it-opus-dei` became `circ:int-opus-dei`
+  (commit `5067fb8`) once the country segment proved wrong for a supranational personal
+  prelature. The correction is right; what had to change to make it is the identifier that
+  was supposed to be stable. "Seed and its limits" above records the same case.
+- **Two sees are told apart by an ordinal.** `circ:cn-xinjiang-1` and `circ:cn-xinjiang-2`
+  "carry ordinal qualifiers pending proper disambiguation" — an arbitrary fact about
+  arrival order in the source index, not about either circumscription, and one that will be
+  wrong for at least one of them the moment the disambiguation lands.
+
+Each case dissolves when the canonical identifier is machine-readable and the
+human-readable layer is guaranteed beside it — both, not one at the cost of the other:
+
+```text
+id:      R7kQp2mXf4LdTbz9Ns3Hc1        # canonical, machine-readable, minted once
+                                       # (illustrative value: shape only, not a minted ID)
+alias:   circ:it-arcidiocesi-di-acerenza   # permanent, resolvable, never reused
+labels:  "Arcidiocesi di Acerenza"@it · "Archidioecesis Acheruntina"@la ·
+         "Archdiocese of Acerenza"@en
+type:    archdiocese                   # an attribute, exactly as rule 1 already asks
+```
+
+Under that shape the strip rule needs no enforcement, because no type word is load-bearing;
+the Opus Dei correction changes `nation` and adds a label, leaving the alias
+`circ:it-opus-dei` resolvable forever alongside `circ:int-opus-dei`; and the two Xinjiang
+sees each hold a distinct canonical ID from the moment they are seeded, with the pending
+disambiguation arriving as labels rather than as a renumbering.
+
+The general argument — why canonical identifiers should be machine-readable, what that
+costs, and how the human-readable layer is guaranteed rather than left optional — is set
+out once in *Identifier Durability: Machine-Readable Canonical IRIs* (CDCF
+`foundation-docs`, `research/identifier-durability-opaque-canonical-iris.md`) and is not
+restated here.
+
 ## Open questions for the committee
 
 1. The namespace prefix (`circ:`) and its coordination with the other registries'
@@ -94,3 +146,8 @@ enrichment:
 5. Whether Ordinariates for Eastern-rite faithful without their own hierarchy, and
    pastoral structures like apostolic exarchates in the diaspora, need a dedicated
    type taxonomy beyond the Annuario's.
+6. Whether the canonical ID of a circumscription should be machine-readable and minted
+   once, with every slug the scheme above produces (`circ:us-boston`,
+   `circ:it-arcidiocesi-di-acerenza`, `circ:it-opus-dei`) kept as a permanent resolvable
+   alias and every see name carried as a multilingual label — keeping this scheme intact
+   as the human-readable layer rather than replacing it (see "Identifier durability").