Research note / Cycle 013
A reply to this question
Authenticated toy responses reject altered or differently bound replies; verifier reset and joint rollback still defeat the model’s freshness assumptions.
Give the authority an actual reply
Lumen needs a receiver to recover an existing decision after losing its reply, without consuming another challenge or trusting an old snapshot. The current-epoch study supplied an authority record but did not authenticate a response. Haddad and Chen now implement a local toy exchange: an authority reply carries a realm, session nonce, request, payload and epoch, together with an HMAC-SHA256 tag over their canonical JSON representation.
The program checks the tag and the active realm/nonce/request/payload before applying the existing bound-recovery and epoch rules. This is executed software, but its key and nonce strings are fixed public fixture data. There is no network authority, unpredictable challenge generator, protected key store or durable service. The exercise tests the implemented decision path, not deployment security.
Authentic does not mean applicable
A correctly authenticated reply for the active values recovers the existing epoch-8 decision and closes that local verification session. A valid tag on a different nonce, request, payload or realm still holds. Changing the epoch after signing fails authentication. A response signed with another test key also fails, and absence of a response holds.
Matching issuer and witness snapshots at epoch 7 hold against the authenticated epoch-8 response. An epoch-7 reply against epoch-8 records also holds; the verifier does not decide which account should win. Withdrawn authority remains withdrawn. A second use of a closed session holds, and none of these recovery reads consumes a new challenge.
The verifier also needs a history
Resetting the verifier and reusing its old nonce reopens the same active values. The recorded authentic response passes again. A nonce label does not prove uniqueness after state loss. This explicit failure remains in the retained output rather than being treated as a fresh successful exchange.
Rolling the records and responding authority together to epoch 7 still produces exactly the same verifier inputs as a valid epoch-7 history. A key holder can authenticate false or stale content. The next task is therefore durable nonce non-reuse and actual epoch/key custody outside common recovery snapshots, with defined behavior when those records cannot be recovered. No lunar timing, live response latency or protected monotonic advancement has been measured.
Results at a glance
| Fixture history | Decision |
|---|---|
| current | recover_bound_current_model_decision |
| old_nonce | hold_response_binding |
| wrong_request | hold_response_binding |
| wrong_payload | hold_response_binding |
| wrong_realm | hold_response_binding |
| tampered_epoch | hold_authentication |
| wrong_key | hold_authentication |
| missing_response | hold_no_response |
| stale_records | hold_epoch_not_current |
| authority_behind | hold_epoch_not_current |
| all_domains_rollback | recover_bound_current_model_decision |
| valid_epoch_7_history | recover_bound_current_model_decision |
| withdrawn_authority | hold_authority |
| closed_session | hold_session_closed |
| reset_reuses_nonce | recover_bound_current_model_decision |