gnomonlearn
Docs/RFC-0045/Design Q&A

RFC-0045 · 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.

← Read the authored RFC

Q: How is a remote bundle authenticated beyond its self-declared hash?

Status: Open

Source: RFC-0045 §§2–3.3, 4.2 Issue: The runtime verifies downloaded bytes against a hash embedded in the same bundle. A substituted bundle can carry its own matching hash, so this does not establish that it is the bundle selected by the deployer. Recommendation: Pin the expected digest outside the bundle in deployment configuration or a signed install plan, verify it before loading, and define signature and trust-root behavior where required.

Q: Which listener serves health endpoints when no HTTP serve is declared?

Status: Open

Source: RFC-0045 §§4.7, 5.2, 6.2, 7 Issue: /healthz and /readyz always bind GNOMON_HOST:GNOMON_HTTP_PORT, while HTTP serve declarations may bind that same port or none at all. The RFC does not define listener sharing, route precedence, or whether a separate management listener exists. Recommendation: Define a management listener or explicit sharing rule, including conflict detection, exposure policy, and behavior before protocol binding completes.

Q: When can shutdown mark an idempotency key incomplete after cancellation?

Status: Open

Source: RFC-0045 §8.3; RFC-0043 §§6.2, 9.2 Issue: A request can commit graph writes and then be cancelled before its response or separately persisted idempotency result completes. Marking the key incomplete permits a duplicate mutation on retry. Recommendation: Make idempotency completion atomic with graph commit and derive shutdown retry behavior from durable transaction state, including the committed-but-unanswered case.

Q: How is the final shutdown audit emitted after backend connections close?

Status: Open

Source: RFC-0045 §8.2 phases S6–S7; RFC-0047 Issue: The shutdown sequence closes backend connections before the mandatory final audit event. If audit storage uses that backend, the event cannot be durably emitted in the stated order. Recommendation: Emit and flush the shutdown audit before closing its transport, with a bounded failure policy, or require an independent audit sink that remains live through S7.

Q: What guarantee can a runtime provide on out-of-memory termination?

Status: Open

Source: RFC-0045 §§9.3, 10 invariant 8 Issue: The RFC requires a clean exit code and no state corruption on out-of-memory, but an operating-system OOM kill is uncatchable and does not allow the runtime to choose exit code 1 or run cleanup. Recommendation: Require transactional crash consistency and recovery after abrupt termination; define exit-code behavior only for catchable in-process allocation failures.