RFC-0050 · Design Q&A
Versioned draft snapshot
RFCs, reference catalogs, and evidence reflect Spec 3b2e3e7a114fb826b1445d39a48756b8fff629cd. Counts and pending/status notes belong to that snapshot. Read the dated project update for later progress.
Source repositories are private during staging. Links marked “repository access required” need access to the linked repository.
Q: Where does provenance terminate for provenance statements themselves?
Status: Open
Source: RFC-0050 §§3–4, 10 invariants 1–3 Issue: Every edge-gnomon must carry the same mandatory provenance statements as every other gnomon. Creating one provenance statement therefore creates more statements that also require provenance, with no terminating rule. Recommendation: Define a finite intrinsic provenance envelope or exempt reserved provenance edges from recursive emission while recording their parent transaction and creator in stable metadata.
Q: How can a statement gnomon be the subject of a provenance statement?
Status: Open
Source: RFC-0050 §4; RFC-0036 §§4.1–4.2 Issue: The provenance table makes any created gnomon, including an edge-gnomon, a statement subject. RFC-0036 statement endpoints are Resource IDs, while a graph statement is not a Resource. The physical representation cannot encode the stated relation for statements. Recommendation: Define a first-class gnomon identity that provenance edges can reference, or store statement provenance in an explicitly typed side table or envelope with queryable projection semantics.
Q: How does database enforcement reject direct writes to ordinary resource kinds?
Status: Open
Source: RFC-0050 §§3.1–3.2, 9.2, 10 invariants 7–9
Issue: The shown RLS policy accepts an INSERT with no template marker and no admin session whenever the resource kind is outside graph.* and core.*. This permits exactly the direct application writes the single rule prohibits.
Recommendation: Require a trusted template or privileged-write context for every insert, update, and delete, and test application-role writes to ordinary kinds against the database policy.
Q: Can an application session forge the template marker used by RLS?
Status: Open
Source: RFC-0050 §§3.3, 9.1–9.2 Issue: The database policy trusts presence of a session GUC set by the runtime. The RFC does not establish that a database role used by application code cannot set the same value through SQL, nor that the marker is bound to the actual invocation and transaction. Recommendation: Use a protected database role or security-definer entry point with transaction-local, validated invocation identity rather than trusting a user-settable non-null setting.
Q: How are nested template arguments represented by one shared argument record?
Status: Open
Source: RFC-0050 §§4.2, 6.1
Issue: One core.template-args record is created per top-level invocation, yet nested calls can have different arguments. The text says a nested argHash “overrides” for that level while each template.with edge points to the same record, leaving no place for a per-edge hash or nested argument references.
Recommendation: Give each invocation frame a distinct immutable argument record and link parent frames, or define per-edge typed argument metadata with collision-free identity.
Q: Which audit record stores the full argument values claimed here?
Status: Open
Source: RFC-0050 §4.2; RFC-0047 §3.1 Issue: RFC-0050 says the audit log carries the full pre-redaction argument set, while RFC-0047’s audit schema stores only a post-redaction hash and size. Storing full sensitive arguments would also conflict with RFC-0047’s redaction rules. Recommendation: Align both RFCs on redacted or keyed-hash metadata, and never promise plaintext sensitive arguments in general audit records.
Q: How is the creation invariant applied to legacy gnomons without provenance?
Status: Open
Source: RFC-0050 §§1, 10 invariants 1 and 5, 11.3
Issue: The invariants say every gnomon has provenance and core.provenance is complete, but §11.3 tolerates existing gnomons with no provenance and returns only the gnomon itself. The migration does not define a release threshold or explicit legacy-origin representation.
Recommendation: State an explicit legacy exception in the invariant and projection result, track its count and origin confidence, and define the gate for removing that exception.