Skip to content

The Bootstrap Create Carries Its Case and Nested Objects Inline

Context and Problem Statement

A participant seeds its replica of a case from the bootstrap Create(VulnerabilityCase). The protocol let that message carry the case, and each object nested in it, inline or by reference. A separate protocol rule says a receiver fetches an object it lacks and needs from the actor that holds it.

Read together, the two rules ask a receiver to fetch part of the bootstrap from someone. The only candidate source is the CASE_MANAGER the message introduces, and the receiver does not trust it until it accepts this very message. The reference implementation already refused a bootstrap that named a participant only by URI, and its specs called that refusal a shortcut of the prototype (#5373).

Should the bootstrap be allowed to carry a reference, so that a receiver must fetch it, or must it carry everything inline?

Decision Drivers

  • A receiver must be able to seed a complete replica from a message it can check.
  • A receiver must not fetch from a source before it trusts that source.
  • The protocol should not require a behavior its reference implementation does not have.
  • Other implementations share the wire, so the rule must say what a receiver may refuse.

Considered Options

  • A. The bootstrap carries its case and every nested object inline, and a receiver does not fetch any of them.
  • B. The bootstrap may name an object by reference, and the receiver fetches it from the CASE_MANAGER the message names.
  • C. The bootstrap may name an object by reference, and the receiver fetches it from the sender.

Decision Outcome

Chosen option: "A", because it is the only option in which a receiver never has to trust a source before the message that establishes trust.

The invariant is:

  • The bootstrap Create(VulnerabilityCase) carries the complete case and its participant records inline, the objects its receiver cannot already hold. A report the receiver offered may be named by reference, because the receiver holds it (Amendment 2026-11-02).
  • A receiver does not fetch an object the protocol requires inline. A bootstrap whose participant record is missing or named only by reference is a protocol error, and the receiver refuses it.

The CASE_MANAGER's own Create(VulnerabilityCase) to the Case Owner, which the Case Owner seeds from before it trusts the CASE_MANAGER, carries its participant records inline for the same reason. Every other message keeps the general rule: an object may travel inline or by reference, and a receiver fetches what it lacks and needs.

Consequences

  • Good, because the trust root no longer depends on a fetch from an actor the receiver does not yet trust.
  • Good, because the protocol now states the refusal the reference implementation already performs, so the refusal is no longer a prototype shortcut.
  • Good, because the receiver's checks run on one message, with nothing left to fetch.
  • Bad, because a large case makes a large bootstrap message.
  • Neutral, because the receiver's refusal covers participant records, which its later checks read; a report the receiver already holds, such as the one it submitted, it reads from its own store when the sender names it by reference.
  • Neutral, because a Create whose own object is a bare URI cannot be recognised as a bootstrap until its type is known, and a receiver that holds no case has no host to fetch it from, so it is not processed.

Validation

  • A test sends a bootstrap whose participant is a bare URI and finds it refused before anything is stored.
  • A test checks that the bootstrap a sender builds carries the case and its participants as full objects.
  • A test gives a sender a case naming a report its own store cannot read and finds the bootstrap is sent, the report named by its id.
  • A test gives a sender a case naming a participant its own store cannot read and finds the bootstrap is not sent.

Pros and Cons of the Options

A. Everything inline

  • Good, because the receiver needs no second source.
  • Good, because it matches what the reference implementation sends and accepts.
  • Bad, because the message grows with the case.

B. Fetch from the CASE_MANAGER

  • Good, because the bootstrap message stays small.
  • Bad, because the receiver must ask the CASE_MANAGER before it trusts it, and a forged bootstrap could point it at a forged one.
  • Bad, because the CASE_MANAGER's read serves only participants of the case, and the receiver is not one until it accepts the bootstrap.

C. Fetch from the sender

  • Good, because the sender is the actor the reporter chose.
  • Bad, because the sender may have delegated the case to another CASE_MANAGER and need not hold every object.
  • Bad, because it adds a fetch exchange that exists only for the bootstrap.

Amendment — 2026-11-02

The invariant is narrowed to the objects the bootstrap's receiver cannot already hold; the chosen option, inline and never fetched, is unchanged. The maintainer decided that the protocol requires an object inline only when its receiver cannot already hold it (#5691). The reporter offered the report the bootstrap names, so the bootstrap may name that report by reference; the case and its participant records stay inline. The reference implementation still carries each report its sender holds inline, but that is a choice of the prototype, not a requirement of this decision.

The first invariant read:

The bootstrap Create(VulnerabilityCase) carries the complete case and every domain object nested in it inline.

The neutral consequence on reports read:

a report the receiver already holds, such as the one it submitted, it can still read from its own store if a non-conforming sender names it by reference.

A sender that names the report by reference is no longer non-conforming.

The second validation item read:

A test checks that the bootstrap a sender builds carries the case, its participants and its reports as full objects.

The title is kept so that links to this record stay valid; "nested objects" in it now means the participant records.

More Information

Generated spec requirements: case-bootstrap-trust.yaml CBT-01-002, CBT-01-007; actor-knowledge-model.yaml AKM-05-007; case-proposal.yaml CP-09-004.