Publication is a decision followed by a separate observation

D/I: release_cycle.py is an in-memory, trusted-input model. Its roles are fixture labels. It does not authenticate humans or agents, cryptographically attest decisions, enforce process isolation, serve a website, detect private content or create legal permission. Required-check booleans model externally provided results; the script does not perform those checks itself.

Let the candidate binding be

$$b=H(r,s,H(x),e,p,c,u),$$

where $r$ is a release identifier, $s$ a source revision, $x$ the bytes, $e$ the evidence-set digest, $p$ the policy version, $c$ a content class and $u$ the destination. Canonical JSON fixes the encoding in this model. The delegation has a separate digest binding publisher, scope, destination, policy and validity interval. The verifier’s decision names both digests and its expiry.

T (model invariant): publication can occur through the model’s publish method only if its current candidate/delegation pair matches a recorded decision, the publisher matches the configured fixture role, required scope and policy match, and neither decision nor delegation has expired or been revoked. This follows from the guards preceding the only insertion into deliveries. It assumes trusted unmodified runtime state and use of the public methods; direct mutation of Python dictionaries defeats the boundary and is outside the claim. Hash collision resistance is assumed when a digest is used as a binding.

Publication first produces delivered_unobserved. A separate observer reports missing, binding_mismatch, mismatch or matches about retrieved fixture bytes. A matching historical receipt remains a valid description of that observation even after the authority expires. Currentness is computed separately and names the last observation rather than asserting continuous monitoring.

The demo walks through failed delivery, a missing observation, successful retry, idempotent repeat, observation, correction requiring a new decision, and later revocation. The tests mutate each bound component, expand a delegation, reuse expired authority, roll time backward, omit checks, substitute a verifier and tamper with the delivered bytes.

python3 papers/07-release-packet/models/release_cycle.py
python3 -m unittest discover -s papers/07-release-packet/models -p 'test_release_cycle.py' -v

This canary closes an explanatory gap in the prior dossier: it demonstrates a complete synthetic sequence and several failures. It does not close the operational gap between the toy model and the hosted Commons or public website. Real deployment needs authenticated decisions, independent credentials and observation, immutable artifacts, transactional delivery, rights handling and an accountable operator.

Red-team repairs additionally bind amendment successors, reject identifier cycles/rebinding, and prevent rejected future-dated calls from advancing the toy clock.