Skip to content

Nine-node staging-topology create and recovery model ​

This chart models a 3-control + 2-ingress + 4-app Fleet. Those pool counts match protected norn-fleet main at 6335c667693d0333d4239f91edf2efb9772ab63a and the 2026-11-09 18:58 UTC provider inventory. This is topology alignment, not a reproduction of the current protected input parser/workflow or proof of live bootstrap/recovery. The model abstracts inputs, provider calls, and node identities. It focuses on partial create, lost response, exact state/provider readback, and same-intent forward recovery. Persistent staging has no sanctioned full-Fleet destroy path, so this model has no cleanup or retirement action. See the topology source audit.

The aggregate inputsValid gate represents the saved plan and digest, capacity intent and one-use nonce, provider account identity, backend bucket/key/versioning, SSH/admin route and host-key binding, bootstrap secret bundle, and provider permission-scope evidence. The model then represents owner-only bootstrap secret materialization before intent admission. Token-scope evidence is an intended admission requirement; the current preflight only proves account identity and backend access, so this condition is not yet implemented. Generated runtime identities and the per-file provenance, scope, expiry, rotation, and revocation lifecycle are outside the model.

The node count abstracts the fixed 3+2+4 pool distribution and provider IDs. providerMismatch represents provider/state disagreement or an unclassified resource; exact readback clears uncertainty only when IDs and backend state match. It does not model live DigitalOcean APIs, Terraform resource-level dependency order, SSH, Ansible, PKI bytes, secret values, GitHub Actions artifact/environment behavior, or Norn's actual receipt APIs.

This protocol model matches the declared pool counts but is not evidence that the protected workflow implements each transition. Bind the model to the actual input parser and replay each path against controller fixtures before treating it as staging recovery qualification.

TLC result ​

TLC explored 102 states and 210 transitions. All ten configured invariants and the deadlock check passed. Invalid/unadmitted input and unresolved identity mismatch are explicit safe hold states; other holds can read back and recover only when identity matches. The reader-facing chart groups resource-level repetitions into 17 states and 35 arrows, including owner-only secret materialization, partial provider create, mismatch hold, fresh exact readback, same-intent recovery, inventory/PKI, partial configuration, and terminal receipt.

Recreate the model with:

sh
bun "$HOME/.agents/skills/statechart/scripts/statechart.ts" \
  docs/v3/statecharts/fresh-staging-nine-node-recovery/FreshStagingRecovery.tla \
  --config docs/v3/statecharts/fresh-staging-nine-node-recovery/FreshStagingRecovery.cfg \
  --view docs/v3/statecharts/fresh-staging-nine-node-recovery/view.ts \
  --out /tmp/fresh-staging-nine-node-recovery-model

The model was rerun on 2026-11-09 and reproduced 102 states, 210 steps, all ten invariants, and the deadlock check. The current public canary route was read separately at 19:01 UTC; neither result qualifies fresh-Fleet bootstrap or its protected apply/recovery workflow.

The generated chart, TLC report, and tlc.out are retained in this directory.