feat: publish Tend v0.2 Preview 2 source

Export the reviewed allowlisted snapshot from private source commit 8aab3db43f35e6a49aa497f45d73701b13fc9f32 and tree 992132ea4703437dc13ffdbb04a077816c02caf9. This includes routed singleton continuity, deployment evidence, strict schema-2 configuration, restricted transport, and the independently compilable public-tree guard. AI-Assisted: OpenAI Codex Signed-off-by: Cole Speelman <crspeelman@gmail.com>
2026-08-18 06:40:58 -04:00
33 changed files with 1463 additions and 150 deletions
@@ -1 +1 @@
-{"schema_version":1,"source_commit":"4d7094c8b7c61991bfb67b11fc1558724c874eb2","source_tree":"54a2f74804f7acddf3755d7d4da5b97f5fc28381","source_date_epoch":1786920560,"file_count":68}
+{"schema_version":1,"source_commit":"8aab3db43f35e6a49aa497f45d73701b13fc9f32","source_tree":"992132ea4703437dc13ffdbb04a077816c02caf9","source_date_epoch":1787049308,"file_count":76}
@@ -1 +1 @@
-d30559d85177736bed9cbf4e1fa2e0703e3195d42390539efd098f091d64b983  PUBLIC-SNAPSHOT.json
+9c6153ab5b861ec5818591269b2a38a14239f68e89911655aaebec444f54a26d  PUBLIC-SNAPSHOT.json
@@ -17,7 +17,7 @@ shell hooks. Application-specific data activation remains application-specific.

 ```text
 tend check    --config tend.json
-tend package  --config tend.json --version v0.2.0-preview.1 --out dist
+tend package  --config tend.json --version v0.2.0-preview.2 --out dist
 tend push     --target tend-deploy@host --known-hosts FILE --service NAME --artifact FILE --sha256 HEX --approve-sha256 HEX
 tend receive  --policy /etc/tend/receive-policy.json
 tend check-server --policy /etc/tend/receive-policy.json
@@ -38,6 +38,15 @@ through `/run/lock/tend-deploy.lock`. Builds and transfers remain parallel;
 only the short Caddy/service activation phase is host-wide. Tend is still a
 single command, not a daemon.

+Each service also keeps a bounded JSONL deployment-event stream and explicit
+desired, candidate, active, previous, and last-attempt release identities. The
+stream contains only fixed provenance and lifecycle fields; Observatory may
+ingest it later, but an event-write failure never blocks deployment or rollback.
+After Caddy reload, Tend repeatedly probes the configured canonical HTTPS
+origins for the activation window. Blue/green deployments simultaneously keep
+checking the previous slot, restoring the prior handler and inactive-slot state
+if routed traffic or continuity fails.
+
 Dry-run and digest approval are intentional friction. See the
 [schema-2 migration guide](docs/SCHEMA_V2_MIGRATION.md) and the
 [two-service walkthrough](docs/WALKTHROUGH.md).
@@ -48,12 +57,22 @@ dependency-free module graph. Tend supports Linux hosts with systemd and Caddy;
 WSL may be used as a Linux development environment, but native Windows is not a
 supported execution, deployment, or release-gate platform.

-The v0.1 public preview and the restricted v0.2 implementation candidate each
-completed maintenance releases and explicit rollback/reactivation for both
-Gamertan and the Sandwich Hime website using one reviewed candidate. See the dated
+The v0.1 public preview and immutable `v0.2.0-preview.1` each completed
+maintenance releases and explicit rollback/reactivation for both Gamertan and
+the Sandwich Hime website using one reviewed candidate. See the dated
 [dogfood evidence](docs/DOGFOOD_EVIDENCE.md) for exact scope and limitations.
-The v0.2 preview will not be tagged until one identical binary has deployed and
-rolled back both services through the restricted transport.
+
+The additive `v0.2.0-preview.2` candidate keeps that transport and activation
+contract while adding bounded deployment-event JSONL, routed-origin continuity,
+rollback annotations, and the operational findings recorded through real
+dogfooding. Preview 1 remains unchanged. Preview 2 will not be tagged until one
+identical binary has deployed, rolled back, and reactivated Gamertan, the
+Sandwich Hime website, and Gamertan Observatory.
+Operational friction discovered while applying the same contract to new
+services is tracked separately in the
+[dogfood friction ledger](docs/DOGFOOD_FRICTION.md). The ledger preserves the
+fail-closed behavior and records candidate product improvements instead of
+normalizing application-specific deployment workarounds.
 The canonical public repository begins with a sanitized root snapshot rather
 than the private development history.

@@ -19,13 +19,18 @@ The release tag and attached candidate must be built from the final reviewed
 source commit. Documentation-only changes after the recorded campaign require
 one final identical-candidate maintenance pass before tagging.

-`v0.2.0-preview.1` is a separate, additive release line. It requires schema 2,
+`v0.2.0-preview.1` is an immutable, additive release line. It requires schema 2,
 the restricted `push`/`receive` transport, root-owned environment-file
 references, host-wide activation serialization, and HTTPS public-origin smoke.
-It may be tagged only after the exact same v0.2 binary successfully deploys,
-rolls back, and reactivates both Gamertan and the Sandwich Hime website. EQL is
-not part of this generic gate; its SQLite/catalog publication needs a dedicated
-adapter rather than arbitrary hooks.
+Its exact released source remains unchanged.
+
+`v0.2.0-preview.2` adds bounded deployment-event JSONL, routed-origin
+activation continuity, rollback annotations, and the reviewed operational
+friction record. It must preserve every Preview 1 security and release gate.
+The exact same Preview 2 binary must deploy, roll back, and reactivate
+Gamertan, the Sandwich Hime website, and Gamertan Observatory before the tag is
+created. EQL is not part of this generic gate; its SQLite/catalog publication
+needs a dedicated adapter rather than arbitrary hooks.

 `v0.1.0-preview.1` is immutable but withdrawn: its source and module checksums
 are valid, while a fresh `go install` reports the development identity because
@@ -50,12 +50,31 @@ probes it, atomically replaces one imported Caddy handler, validates the full
 Caddy configuration, reloads Caddy, and records the prior active slot.

 Singleton mode starts the new release in a hardened transient systemd unit on a
-separate loopback address, probes it, stops the candidate, changes the current
-release pointer, and restarts the installed singleton unit.
+separate loopback address and probes it. Tend then validates and atomically
+routes the imported Caddy handler to that candidate. While the candidate serves
+the canonical origin, Tend changes the current release pointer, restarts and
+probes the installed fixed-address unit, validates Caddy again, and routes back
+to it. The candidate remains healthy through the activation window and stops
+only after the handoff succeeds.

 After the local post-activation probes, Tend also checks configured HTTPS public
-origins. Any failure before state persistence restores the previously observed
-Caddy bytes and/or release pointers. Rollback is a separate explicit command
+origins throughout a bounded activation window. Blue/green mode also probes the
+previous slot, and singleton mode probes the handoff candidate, for health and
+readiness throughout that window. Any failure before success is recorded
+restores the previously observed Caddy bytes and/or release pointers. State
+records desired, candidate, active, previous, and last-attempt
+release identities, including failed attempts without claiming they became
+active. Rollback is a separate explicit command
 over the recorded state. It rechecks local health/readiness and public reachability,
 but deliberately does not apply a future release's content marker to an older
 release whose routes may differ. Pruning preserves both active and previous releases.
+
+## Evidence boundary
+
+Every attempted activation and explicit rollback emits bounded, versioned JSONL
+events with an operation ID, service, approved artifact digest, source commit,
+release version, phase, slot, elapsed duration, and outcome. Values are
+validated rather than copied from command output. The log contains no
+environment values, arbitrary process output, HTTP bodies, or secret paths.
+Identity, entropy, file, and downstream observability failures are deliberately
+best effort and cannot control Tend's deployment or rollback result.
@@ -1,8 +1,85 @@
 # Preview dogfood evidence

 This is maintainer-run operational evidence, not an independent audit or a
-general reliability claim. It records the acceptance campaign completed on
-August 14, 2026 before Tend's first public preview.
+general reliability claim. It preserves each dated campaign and its limitations
+instead of rewriting earlier observations as though later fixes had already
+existed.
+
+## Final Preview 2 code-candidate campaign — August 18, 2026
+
+The final v0.2 Preview 2 implementation candidate completed two explicit
+rollback-and-reactivation cycles for Gamertan, the Sandwich Hime website, and
+Gamertan Observatory on the production Linux/systemd/Caddy host. The candidate
+routed singleton traffic to the already-proven transient process while the
+fixed-address installed unit restarted, then restored the canonical upstream
+only after loopback and public-origin validation.
+
+### Assessed Tend candidate
+
+- Private implementation source commit:
+  `1fd3b9904c46e817c244196dd5d5a90921ca81a2`.
+- Version: `v0.2.0-preview.2`.
+- Linux/amd64 binary SHA-256:
+  `adb4753d4e865775d50d618c999f10dfac9a0de945ccfd9d0b8f2e80d65f4597`.
+- Gitea release-candidate archive SHA-256:
+  `2bdfee2168cb16883d524cf9703adfa6ed5bc2bf44fbbb896993acc5fe6969ec`.
+- Toolchain: Go 1.26.6, `CGO_ENABLED=0`, `-trimpath`.
+- Trusted verification run 286 and release-candidate run 287 passed. The
+  archive's checksums, SPDX SBOM, embedded version, commit, clean VCS state,
+  target, and Go build information were independently rechecked before host
+  installation.
+
+### Application maintenance artifacts
+
+- Gamertan archive SHA-256:
+  `4700b075640b8b2fb5c17e0e02cf8d96ee67ceee10fe76d108c8b411983a88aa`.
+- Sandwich Hime website archive SHA-256:
+  `46b4da41cf6703fb8818e7d25e3a9c13e57cf700f5608b1888c4adb3352e4d38`.
+- Observatory preview 12 archive SHA-256:
+  `9ef0ddd8ec25d8fb75d6a6887e3ba874df7ebba16d3254f6250fcc646f4fd7f4`.
+
+Every service ended with the intended current release active and the older
+release retained as the explicit rollback target. Gamertan returned to its
+green slot; both singleton services returned to their fixed addresses.
+Candidate ports and the shared lock were free afterward. Caddy and all five
+installed application units were active with zero restarts and no failed
+units.
+
+### Continuity and deployment evidence
+
+A seven-minute workstation probe sampled the Gamertan origin, both Sandwich
+Hime origins, Observatory, and EQL health 1,606 times each throughout the
+campaign. It observed no HTTP failure status. One simultaneous client-side
+disconnect affected all five destinations during a reload. A controlled
+90-second replay therefore observed the same validated no-content-change Caddy
+reload from both the workstation and an independent Linux host. The Linux host
+recorded 450 successful responses and zero failures for every origin; the
+workstation alone repeated one common-mode URL transport error across every
+destination. Caddy retained the same PID with zero restarts and continued
+serving unrelated requests. The common-mode workstation event is recorded as
+an observer-path limitation, not server downtime.
+
+The authoritative deployment-event files finished at 18 Gamertan events, 14
+Sandwich Hime events, and 16 Observatory events. Observatory's agent cursor for
+each stream exactly equalled the corresponding file size, proving complete
+consumption. The agent and applications reported no warning-or-higher journal
+entries during the campaign. Representative public routes returned HTTP 200,
+and the EQL origin remained healthy.
+
+### Findings closed before this campaign
+
+Two earlier pre-activation artifacts exposed a mode-restoration defect under a
+hardened root umask. They failed before route or pointer mutation. Tend now
+reapplies validated archive modes explicitly and tests extraction under umask
+`0077`.
+
+The older singleton strategy briefly exposed an unavailable fixed upstream and
+produced transient Observatory-agent `502`s. The final candidate's routed
+handoff removed that failure in repeated production activation and rollback.
+
+This evidence update changes VCS build metadata but not deployment logic. The
+release policy therefore still requires one last maintenance pass with the
+exact evidence-bearing candidate before the signed public tag is created.

 ## Restricted multi-service campaign — August 16, 2026

@@ -0,0 +1,328 @@
+# Dogfood friction ledger
+
+This document records friction observed through August 18, 2026 while using
+Tend for real maintenance releases. Friction is evidence about the product: it
+should become either a clearer contract, safer automation, or an explicit
+non-goal. It must not become application-specific shell lore.
+
+The findings below are maintainer observations, not implemented promises. Tend
+continues to fail closed while an option is being designed. In particular,
+there is no `--skip-remote` packaging escape hatch, no arbitrary deployment
+hook, and no relaxation of the host-wide activation lock.
+
+## Design standard
+
+A Tend workflow should make the secure path the short, documented path:
+
+- source, artifact, approval, configuration, and active-release identities are
+  explicit and inspectable;
+- credentials are narrowly scoped, briefly available, and absent from
+  artifacts and logs;
+- application validation is declarative, bounded, and cannot become a shell;
+- first installation, maintenance, activation, rollback, and pruning are
+  distinct operations;
+- every mutation has a recorded prior state and a tested restoration path;
+- unrelated services may build and prepare candidates concurrently, while the
+  shared Caddy activation boundary remains serialized.
+
+## Observed findings
+
+| ID | Status | Finding | Current safe behavior | Candidate direction |
+| --- | --- | --- | --- | --- |
+| F-01 | Workflow mitigation | CI packaging needs proof that the exact commit was pushed. | Packaging verifies the configured remote and fails if Git cannot authenticate. The trusted workflow retains only its job-scoped read-only checkout credential. | Make the proof credential or an authenticated source attestation an explicit Tend input and evidence boundary. |
+| F-02 | Open | Application-owned unit and configuration changes are outside the binary activation transaction. | Operators stage, validate, back up, and restore those files separately. | Add an allowlisted configuration transaction or record and validate exact configuration digests. |
+| F-03 | Open | A singleton candidate may not reproduce the installed unit's arguments and application configuration. | It receives the production environment file and an isolated listen override; activation still fails closed. | Add a bounded application preflight and explicit, validated candidate invocation. |
+| F-04 | Partially resolved | Artifact production, review, approval, transfer, and activation require several identity checks. | v0.2 standardizes the evidence bundle and restricted transfer; the operator still repeats the approved digest intentionally. | Add offline inspection and approval ergonomics without combining build authority and production authority. |
+| F-05 | Partially resolved | Tend maintenance assumes an existing adopted service and current release. | `check-server` validates a prepared host, while first installation remains a separately reviewed operator procedure. | Define an explicit `install` or `adopt` transaction rather than silently treating bootstrap as maintenance. |
+| F-06 | Partially resolved | Binary rollback can be unsafe when service configuration has changed incompatibly. | v0.2 records desired, candidate, active, previous, and last-attempt releases, but external configuration compatibility is operator-owned. | Bind non-secret configuration identities and preflight results to release state. |
+| F-07 | Open | `GOPROXY=off` does not prove that every module metadata lookup is available locally. | Packaging stops before artifact creation when Go cannot resolve the complete pinned module graph. | Separate checksum-verified dependency resolution from a network-disabled, cache-completeness-checked build stage. |
+| F-08 | Open | Packaging an otherwise clean pushed commit from a linked Git worktree fails Go's required VCS-status stamp. | Tend stops before artifact creation; package the same exact commit from a clean standalone clone. | Detect linked worktrees during `check`/`package`, explain the supported source shape, and assess a provenance-preserving worktree build that does not weaken `-buildvcs=true`. |
+| F-09 | Resolved in Preview 2 | Restarting a fixed-address singleton briefly exposed Caddy to an unavailable upstream. | Tend now routes the imported handler to the candidate, restarts and proves the installed unit behind that handoff, then routes back while the candidate remains healthy through the activation window. | Preserve the production regression campaign and failure-injection matrix. |
+| F-10 | Resolved in Preview 2 | A hardened root umask could make an extracted application binary executable only by root. | Tend reapplies every validated archive mode explicitly; extraction is tested under umask `0077`. | Preserve the mode and non-root candidate tests. |
+| F-11 | Evidence practice | A single external observer can report common-mode client or network errors as apparent multi-service downtime. | Production campaigns retain server state and use an independent observer before classifying a continuity failure. | Standardize multi-vantage continuity evidence without making an observability dependency part of deployment authority. |
+
+## F-01: pushed-commit proof in CI
+
+### Observation
+
+During an Observatory release-candidate run, verification intentionally checked
+out source without persisted credentials. Tend later attempted to prove that
+`HEAD` existed on the configured private Gitea branch and Git could not
+authenticate. Packaging stopped before producing an artifact. This was the
+correct fail-closed outcome, but the credential lifecycle was not obvious from
+the workflow contract.
+
+The trusted workflow uses a job-scoped read-only checkout credential so Tend
+can perform the independent remote proof. It is not a long-lived repository or
+deployment credential, and it is removed by checkout cleanup at the end of the
+job. Production credentials remain unavailable to the build job.
+
+### Options to assess
+
+1. Support an explicit read-only Git credential file or credential helper for
+   package-time proof. Never accept a token in arguments, configuration,
+   manifests, artifacts, or logs.
+2. Accept an authenticated CI source-attestation document binding repository,
+   branch, commit, tree, workflow identity, and event identity. Define which
+   CI issuers are trusted and preserve the attestation with release evidence.
+3. Accept a signed, pre-verified source bundle whose identity and policy can be
+   checked without network access.
+
+Do not add a generic skip flag. An unavailable proof must remain a packaging
+failure unless an equally strong proof mode was selected explicitly.
+
+## F-02: application configuration is not binary activation
+
+### Observation
+
+Tend can transact an immutable binary, release pointers, a systemd restart, a
+validated Caddy handler, health checks, and rollback state. It does not
+currently transact application configuration, systemd unit changes, credential
+bindings, or server-local secret files. A first deployment that changes these
+files therefore needs a separate backup, validation, installation, and
+restoration procedure.
+
+This boundary is safe but easy to overlook: restoring the previous binary does
+not restore an incompatible unit or application configuration.
+
+### Options to assess
+
+1. Add strict managed-file entries with exact source and destination paths,
+   content digest, owner, group, mode, and validation type. Permit only regular
+   files beneath configured roots; reject symlinks and unknown destinations.
+2. Add a separate `tend configure` transaction that backs up, atomically
+   replaces, validates, and restores supported systemd/Caddy/application files.
+3. Keep configuration externally managed, but require Tend to record desired
+   and active configuration digests and prove candidate/rollback compatibility.
+
+None of these options should permit arbitrary shell commands. Secret values
+remain referenced server-side and must never enter release artifacts or state.
+
+## F-03: candidate fidelity
+
+### Observation
+
+A hardened singleton candidate receives the configured production environment
+file and Tend's isolated listen-address override. It does not currently model
+the installed systemd unit's complete argument vector. An application whose
+configuration path is supplied by unit arguments may therefore start a
+candidate with defaults instead of the intended production configuration.
+That preserves live-state isolation, but it may prove only executable startup.
+A missing credential, incompatible configuration field, filesystem permission,
+or data migration requirement can then surface at activation time.
+
+### Options to assess
+
+- Define a fixed application preflight command or protocol that validates the
+  production configuration, credentials, permissions, and data compatibility
+  without binding the live port or mutating live state.
+- Add a strict candidate argument vector to schema 2. Validate each argument,
+  reject secret values and paths outside the application contract, and pass it
+  directly to systemd without a shell.
+- Permit an allowlisted `check` argument vector, never a shell string, with
+  bounded time/output and an explicitly non-mutating application contract.
+- Record which checks ran against the binary alone and which ran against the
+  actual production configuration so the evidence cannot overstate coverage.
+
+## F-04: artifact review and approval ergonomics
+
+### Observation
+
+The secure flow deliberately separates build, review, digest approval,
+transfer, candidate validation, and activation. v0.2 now produces a consistent
+archive, manifest, SBOM, and checksum set, then transfers one exact artifact
+through its restricted receiver. In practice, operators still need a
+predictable way to discover that CI artifact, inspect it, approve exactly one
+digest, and retain the review evidence. Without a first-class review path,
+correct manual steps are easy to reconstruct differently for each application.
+
+### Options to assess
+
+- Add `tend inspect` for offline, non-mutating verification and a concise human
+  and JSON summary of source, build, dependency, and archive identities.
+- Add an approval record that binds the artifact digest, service, target,
+  approver, and expiry without containing a credential.
+- Standardize one Gitea artifact layout and documented download-to-activation
+  workflow. Keep production credentials unavailable to verification jobs.
+
+Approval must remain explicit; better ergonomics must not turn a successful
+build into an automatic production mutation.
+
+## F-05: first installation and adoption
+
+### Observation
+
+The maintenance workflow expects an installed service, valid configuration,
+and a current release pointer. `check-server` validates the restricted receiver
+policy and prepared service boundary, but it does not create them. Observatory
+bootstrap therefore required operator-managed service account, directories,
+credentials, unit, configuration, and an initial current release before Tend
+could own later maintenance safely.
+
+### Options to assess
+
+- `tend install`: a deliberately broader, separately approved transaction with
+  a strict schema and complete rollback of every supported created object.
+- `tend adopt`: validate an existing service and release, copy or identify its
+  immutable artifact, establish state, and refuse ambiguous ownership.
+- Keep bootstrap out of Tend, but ship a versioned acceptance checklist and a
+  machine-readable `check-server` result that maintenance can require.
+
+Installation and adoption must not be inferred from a missing state file.
+
+## F-06: release and configuration identity
+
+### Observation
+
+Content-addressed releases and v0.2 state make desired, candidate, active,
+previous, and last-attempt binary identities clear. A live service is still a
+combination of its binary, application configuration, credential bindings,
+unit, routing fragment, and sometimes data schema. Tend cannot yet fully
+explain that combined identity or determine whether a retained binary is
+compatible with the current external configuration.
+
+### Options to assess
+
+- Extend the existing desired, candidate, active, previous, and last-attempt
+  release state with non-secret configuration and unit digests.
+- Require rollback compatibility declarations or read-only application
+  preflight before changing traffic.
+- Expose the identities and last validation results through
+  `tend status --json` for deployment evidence and future Observatory
+  ingestion.
+
+Configuration records contain digests and approved metadata only—not secret
+contents.
+
+## F-07: offline module-cache completeness
+
+### Observation
+
+An exact Observatory package attempt used `GOPROXY=off` and a previously used
+module cache. Source archives for the application dependencies were present,
+but Go still needed several module metadata records while Tend enumerated the
+complete build graph. Go refused the lookup and Tend stopped before building
+or writing an artifact. Re-enabling the checksum-verified public proxy supplied
+the missing metadata; the resulting package was byte-identical to the trusted
+CI candidate.
+
+This is safe failure, but `GOPROXY=off` alone is not evidence of a hermetic
+build. A cache can be partially populated even when ordinary builds happen to
+succeed.
+
+### Options to assess
+
+- Add a resolver stage that runs with the pinned toolchain, proxy, and checksum
+  database, emits the complete module inventory, and materializes a bounded,
+  read-only cache for the builder.
+- Run Tend's package stage with networking disabled and require every module,
+  checksum, source archive, and metadata record to come from that reviewed
+  cache.
+- Add a non-mutating cache-completeness check that reports missing module
+  identities before the expensive double build.
+- Consider a strictly verified vendored-source mode where repository size and
+  update review are acceptable; do not silently change dependency modes.
+
+Do not treat `GONOSUMDB`, `GOPRIVATE`, or a proxy bypass as an offline-build
+solution. They alter verification or routing policy rather than proving cache
+completeness.
+
+## F-08: linked-worktree VCS stamping
+
+### Observation
+
+An exact clean, pushed Observatory main commit passed verification and public
+snapshot isolation from a detached linked Git worktree. Tend accepted its
+source and remote provenance, then Go 1.26.6 stopped both `tend package` and an
+equivalent direct build at the required `-buildvcs=true` step with `error
+obtaining VCS status: exit status 128`. Ordinary Git status and commit queries
+from the same worktree succeeded.
+
+A fresh standalone SSH clone of the identical commit packaged successfully
+twice. Both archives were byte-identical and carried the expected commit, Go
+version, VCS settings, checksums, manifest, and SBOM. This isolates the failure
+to the linked-worktree build shape rather than the application source or module
+graph.
+
+The failure is safe: Tend did not create a partial artifact and must not switch
+to `-buildvcs=false`, because the package gate independently verifies the
+embedded VCS revision and clean state. Until the interaction is resolved, use
+a clean standalone clone for release packaging.
+
+### Options to assess
+
+- Detect a `.git` indirection file during `tend check` and fail before the
+  expensive double build with a precise standalone-clone instruction.
+- Reproduce the interaction in a package integration test against the minimum
+  supported Go release and determine whether it is a Go toolchain limitation
+  or an invocation/environment defect.
+- If linked worktrees can be supported, require the resulting build record to
+  carry the exact expected `vcs.revision` and `vcs.modified=false`; do not
+  synthesize those settings or disable VCS stamping.
+- Consider an explicit, signed source-bundle input as part of the broader
+  source-attestation design. It must remain at least as strong as current
+  pushed-commit proof.
+
+## F-09: singleton traffic continuity
+
+### Observation
+
+During the August 18 Observatory maintenance exercise, the transient candidate
+passed health, readiness, and content checks. Tend then stopped that candidate,
+changed the singleton pointer, and restarted the installed fixed-address unit.
+Caddy still targeted the fixed address during that restart, so the Observatory
+agent observed a small number of transient HTTP `502` responses. Its durable
+spool retried successfully and no accepted telemetry was lost, but the routed
+origin was not continuously available.
+
+The corrected activation contract treats the candidate as a traffic handoff,
+not merely a preflight process. Tend validates and routes the imported Caddy
+handler to the candidate before changing the release pointer. It restarts and
+probes the fixed-address unit without public traffic, routes back only after
+that unit passes, and keeps the candidate healthy throughout the bounded
+activation window. Any failure restores the former pointer and exact handler
+bytes. If restoration itself cannot be completed, Tend leaves the proven
+candidate routed and running for explicit operator recovery instead of causing
+a known outage.
+
+Regression tests cover successful handoff, restart failure, public-origin
+failure during the activation window, pointer and handler restoration, Caddy
+validation/reload boundaries, and candidate cleanup. Final production evidence
+must still repeat deploy, rollback, and reactivation with the exact release
+binary before this finding is treated as released.
+
+## F-10: archive modes under a hardened umask
+
+### Observation
+
+The first two August 18 maintenance candidates stopped before activation. The
+release archives correctly recorded executable mode `0755`, but extraction by
+root under umask `0077` left the installed application binary mode `0700`.
+The unprivileged transient candidate could not execute it. Tend emitted failed
+candidate evidence, retained the prior release and route, and did not expose
+the failed binary to traffic.
+
+Extraction now reapplies the already validated archive mode after file content
+is closed. A Linux regression test sets umask `0077`, extracts the release, and
+requires the installed binary to remain executable by the service identity.
+The successful maintenance campaign used that corrected Tend binary.
+
+## Prioritization
+
+The recommended implementation order is:
+
+1. final production proof of the singleton traffic handoff;
+2. a checksum-verified resolver and network-disabled package contract;
+3. an early linked-worktree diagnostic and a provenance-preserving support
+   decision;
+4. `tend inspect` and a standardized CI artifact/approval contract;
+5. exact release plus configuration identity in status and state;
+6. a bounded application preflight contract;
+7. restricted transfer and receive with host policy;
+8. explicit adoption for existing services;
+9. managed configuration only after its restoration and failure-injection
+   model is as strong as binary activation.
+
+Friction entries should be updated with the implementing version, tests, and
+dogfood evidence when resolved. Resolved entries remain in this ledger so the
+reason for the security boundary is not lost.
@@ -14,12 +14,19 @@ been exercised.
 5. Set every service's `deployment.lock_file` to
   `/run/lock/tend-deploy.lock`.
 6. Add one or more query-free HTTPS `deployment.public_smoke` checks.
-7. Update installed systemd units to read the same environment file as the
+7. Add a per-service `deployment.event_log` below its release root and a
+   bounded `deployment.activation_window_seconds` value. Keep the log
+   root-owned and grant collectors read access explicitly.
+8. For singleton services, add the full Caddy configuration, imported handler,
+   and one-upstream handler-template paths. The template must be reviewed and
+   contain exactly one `{{UPSTREAM}}` marker so Tend can keep traffic on the
+   candidate while the fixed-address unit restarts.
+9. Update installed systemd units to read the same environment file as the
   transient candidate.
-8. Install a root-owned `0600` receive policy mapping each service name to its
+10. Install a root-owned `0600` receive policy mapping each service name to its
   exact configuration and artifact-size ceiling.
-9. Run `tend check-server` as root before accepting a transfer.
-10. Validate, activate, rollback, and reactivate one service at a time. Confirm
+11. Run `tend check-server` as root before accepting a transfer.
+12. Validate, activate, rollback, and reactivate one service at a time. Confirm
    unrelated services never restart.

 Tend does not discover `.env`, infer old values, rewrite a production file, or
@@ -7,11 +7,17 @@
 - Release roots, state, pointers, and Caddy files reject symlink substitution
  at their checked boundaries.
 - Configuration is strict JSON and is never interpolated into a shell command.
- Candidate health is established before traffic or the singleton current
+- Candidate health is established before traffic or any current release
  pointer changes. New deployments also satisfy configured content smoke
  checks; rollback uses health and readiness because future-release content
  markers are not valid requirements for an older retained release.
 - Caddy configuration validates before reload.
+- Canonical routed origins and the previous blue/green slot or singleton
+  handoff candidate remain under probe for the configured activation window; a
+  failure restores the old handler and release pointer.
+- Desired, candidate, active, previous, and failed-attempt identities remain
+  distinct in state. Bounded deployment events contain no arbitrary command
+  output or environment values and cannot block deployment.
 - An activation failure restores the previously observed state.
 - Active and previous releases survive pruning.
 - The restricted receiver accepts one versioned bounded stream, one allowlisted
@@ -9,10 +9,14 @@ binary at `/usr/local/bin/tend`; it does not need Git or Go.
 1. Create `/etc/tend/services`, `/etc/tend/environment`, and
   `/var/lib/tend/incoming`. The incoming and environment directories are
   root-owned mode `0700`.
-2. Install one schema-2 file per service and one root-owned mode-`0600`
+2. Install one schema-2 file per service, one per-service deployment-event log,
+   and one root-owned mode-`0600`
   environment file per service. A singleton's shared environment file must
   not define its configured listen key; its installed unit owns the live
-   address and Tend overrides only the transient candidate.
+   address and Tend overrides only the transient candidate. Give each
+   singleton an imported Caddy handler and a root-owned handler template with
+   exactly one `{{UPSTREAM}}` marker; this is the bounded traffic handoff while
+   its fixed-address unit restarts.
 3. Install the receive policy, forced `authorized_keys` entry, and exact sudoers
   rule from `examples/server/` after replacing every placeholder. The sudoers
   fragment preserves only `SSH_ORIGINAL_COMMAND`; the root receiver requires
@@ -46,9 +50,18 @@ services share Caddy. Tend does not stop the other application.

 ## Failure and recovery exercises

- Change a candidate marker: activation must fail before state is stored.
+- Change a candidate marker: activation must fail, preserve the active release,
+  and record the failed attempt without calling the candidate active.
 - Make a Caddy template invalid: validation must fail and restore prior bytes.
- Make the public marker unavailable: Tend must restore the former slot/pointer.
+- Make the public marker fail after an initially successful request: the
+  activation window must catch the transient routed failure and restore the
+  former slot/pointer.
+- For a singleton, verify repeated canonical-origin requests remain successful
+  while Tend routes to the candidate, restarts the fixed-address unit, and
+  returns traffic to it. Inject failure at both Caddy reloads and require the
+  prior handler and pointer to be restored.
+- Stop the previous blue/green slot during the activation window: Tend must
+  restore the old Caddy handler instead of accepting reduced continuity.
 - Run `tend rollback --activate` for one service and verify the other service's
  units, pointers, and public origin did not change.
 - Interrupt a transfer: no release becomes active and the incomplete incoming
@@ -18,13 +18,24 @@ the isolated candidate address. Secret values never enter `tend.json`.
 The singleton example follows the same split: its shared root-only environment
 file omits the configured listen key, the installed unit owns the live address,
 and Tend supplies only the transient candidate address. This prevents a shared
-environment file from overriding the isolated candidate port.
+environment file from overriding the isolated candidate port. Its imported
+Caddy handler is also managed from one reviewed template. Tend temporarily
+routes the canonical origin to the proven candidate while the fixed-address
+unit restarts, then returns traffic to that unit only after it passes its local
+checks.

 Production configuration belongs outside the source checkout, owned by root,
 and not group- or world-writable. The Caddy handler template is an entire
 imported handler fragment; the enclosing site, matchers, and routing precedence
 remain operator-owned.

+`event_log` is a per-service, root-owned JSONL evidence stream below that
+service's release root. Grant an Observatory agent read access explicitly; do
+not make the release root broadly readable. `activation_window_seconds` keeps
+canonical routed probes active after Caddy reload and keeps the previous
+blue/green slot—or the singleton handoff candidate—under health/readiness
+observation until the activation is recorded.
+
 `server/` demonstrates the schema-2 receive policy, forced OpenSSH command,
 restricted sudo entry, two independent service configurations, and secret-file
 placement. The values are placeholders, not an installation script.
@@ -7,9 +7,11 @@
    "root": "/opt/example-site",
    "lock_file": "/run/lock/tend-deploy.lock",
    "state_file": "/opt/example-site/tend-state.json",
+    "event_log": "/opt/example-site/deployment-events.jsonl",
    "health_path": "/healthz",
    "readiness_path": "/readyz",
    "candidate_timeout_seconds": 30,
+    "activation_window_seconds": 10,
    "smoke": [{ "path": "/", "contains": "Example site" }],
    "public_smoke": [{ "url": "https://example.test/", "contains": "Example site" }],
    "blue_green": {
@@ -0,0 +1,2 @@
+# Managed by Tend. The enclosing site and route matchers remain operator-owned.
+reverse_proxy {{UPSTREAM}}
@@ -0,0 +1,2 @@
+# Managed by Tend. The enclosing site and route matchers remain operator-owned.
+reverse_proxy {{UPSTREAM}}
@@ -7,9 +7,11 @@
    "root": "/opt/docs-site",
    "lock_file": "/run/lock/tend-deploy.lock",
    "state_file": "/opt/docs-site/tend-state.json",
+    "event_log": "/opt/docs-site/deployment-events.jsonl",
    "health_path": "/healthz",
    "readiness_path": "/readyz",
    "candidate_timeout_seconds": 30,
+    "activation_window_seconds": 10,
    "smoke": [{ "path": "/", "contains": "Documentation" }],
    "public_smoke": [{ "url": "https://docs.example.test/", "contains": "Documentation" }],
    "singleton": {
@@ -18,7 +20,10 @@
      "candidate_address": "127.0.0.1:18102",
      "listen_env": "DOCS_LISTEN",
      "current_link": "/opt/docs-site/current",
-      "previous_link": "/opt/docs-site/previous"
+      "previous_link": "/opt/docs-site/previous",
+      "caddy_config": "/etc/caddy/Caddyfile",
+      "caddy_handler": "/etc/caddy/docs-site-handler.caddy",
+      "caddy_handler_template": "/etc/tend/caddy/docs-site.template"
    }
  }
 }
@@ -7,9 +7,11 @@
    "root": "/opt/example-site",
    "lock_file": "/run/lock/tend-deploy.lock",
    "state_file": "/opt/example-site/tend-state.json",
+    "event_log": "/opt/example-site/deployment-events.jsonl",
    "health_path": "/healthz",
    "readiness_path": "/readyz",
    "candidate_timeout_seconds": 30,
+    "activation_window_seconds": 10,
    "smoke": [{ "path": "/", "contains": "Example site" }],
    "public_smoke": [{ "url": "https://example.test/", "contains": "Example site" }],
    "singleton": {
@@ -18,7 +20,10 @@
      "candidate_address": "127.0.0.1:18092",
      "listen_env": "EXAMPLE_LISTEN",
      "current_link": "/opt/example-site/current",
-      "previous_link": "/opt/example-site/previous"
+      "previous_link": "/opt/example-site/previous",
+      "caddy_config": "/etc/caddy/Caddyfile",
+      "caddy_handler": "/etc/caddy/example-site-handler.caddy",
+      "caddy_handler_template": "/etc/tend/caddy/example-site.template"
    }
  }
 }
@@ -0,0 +1,2 @@
+# Managed by Tend. The enclosing site and route matchers remain operator-owned.
+reverse_proxy {{UPSTREAM}}
@@ -7,9 +7,11 @@
    "root": "/opt/example-site",
    "lock_file": "/run/lock/tend-deploy.lock",
    "state_file": "/opt/example-site/tend-state.json",
+    "event_log": "/opt/example-site/deployment-events.jsonl",
    "health_path": "/healthz",
    "readiness_path": "/readyz",
    "candidate_timeout_seconds": 30,
+    "activation_window_seconds": 10,
    "smoke": [{ "path": "/", "contains": "Example site" }],
    "public_smoke": [{ "url": "https://example.test/", "contains": "Example site" }],
    "singleton": {
@@ -18,7 +20,10 @@
      "candidate_address": "127.0.0.1:18092",
      "listen_env": "EXAMPLE_LISTEN",
      "current_link": "/opt/example-site/current",
-      "previous_link": "/opt/example-site/previous"
+      "previous_link": "/opt/example-site/previous",
+      "caddy_config": "/etc/caddy/Caddyfile",
+      "caddy_handler": "/etc/caddy/example-site-handler.caddy",
+      "caddy_handler_template": "/etc/tend/caddy/example-site.template"
    }
  }
 }
@@ -56,9 +56,11 @@ type Deployment struct {
 	Root                 string        `json:"root"`
 	LockFile             string        `json:"lock_file"`
 	StateFile            string        `json:"state_file"`
+	EventLog             string        `json:"event_log"`
 	HealthPath           string        `json:"health_path"`
 	ReadinessPath        string        `json:"readiness_path"`
 	CandidateTimeoutSecs int           `json:"candidate_timeout_seconds"`
+	ActivationWindowSecs int           `json:"activation_window_seconds"`
 	Smoke                []Smoke       `json:"smoke"`
 	PublicSmoke          []PublicSmoke `json:"public_smoke"`
 	BlueGreen            *BlueGreen    `json:"blue_green,omitempty"`
@@ -91,12 +93,15 @@ type Slot struct {
 }

 type Singleton struct {
-	Unit             string `json:"unit"`
-	Address          string `json:"address"`
-	CandidateAddress string `json:"candidate_address"`
-	ListenEnv        string `json:"listen_env"`
-	CurrentLink      string `json:"current_link"`
-	PreviousLink     string `json:"previous_link"`
+	Unit                 string `json:"unit"`
+	Address              string `json:"address"`
+	CandidateAddress     string `json:"candidate_address"`
+	ListenEnv            string `json:"listen_env"`
+	CurrentLink          string `json:"current_link"`
+	PreviousLink         string `json:"previous_link"`
+	CaddyConfig          string `json:"caddy_config"`
+	CaddyHandler         string `json:"caddy_handler"`
+	CaddyHandlerTemplate string `json:"caddy_handler_template"`
 }

 func Load(path string) (Config, error) {
@@ -187,12 +192,21 @@ func (c Config) Validate() error {
 	if filepath.Clean(d.StateFile) == filepath.Clean(d.Root) || !within(d.Root, d.StateFile) {
 		return errors.New("deployment.state_file must be below deployment.root")
 	}
+	if err := safeAbsolute("deployment.event_log", d.EventLog); err != nil {
+		return err
+	}
+	if filepath.Clean(d.EventLog) == filepath.Clean(d.Root) || !within(d.Root, d.EventLog) || filepath.Clean(d.EventLog) == filepath.Clean(d.StateFile) {
+		return errors.New("deployment.event_log must be a distinct file below deployment.root")
+	}
 	if !safeHTTPPath(d.HealthPath) || !safeHTTPPath(d.ReadinessPath) {
 		return errors.New("health and readiness paths must be absolute HTTP paths")
 	}
 	if d.CandidateTimeoutSecs < 2 || d.CandidateTimeoutSecs > 300 {
 		return errors.New("candidate_timeout_seconds must be between 2 and 300")
 	}
+	if d.ActivationWindowSecs < 1 || d.ActivationWindowSecs > 120 {
+		return errors.New("activation_window_seconds must be between 1 and 120")
+	}
 	if len(d.Smoke) == 0 || len(d.Smoke) > 32 {
 		return errors.New("deployment.smoke must contain 1 to 32 checks")
 	}
@@ -298,6 +312,14 @@ func validateSingleton(root string, s Singleton) error {
 	if s.CurrentLink == s.PreviousLink {
 		return errors.New("current and previous links must differ")
 	}
+	for label, path := range map[string]string{"caddy_config": s.CaddyConfig, "caddy_handler": s.CaddyHandler, "caddy_handler_template": s.CaddyHandlerTemplate} {
+		if err := safeAbsolute("deployment.singleton."+label, path); err != nil {
+			return err
+		}
+	}
+	if filepath.Clean(s.CaddyHandler) == filepath.Clean(s.CaddyHandlerTemplate) {
+		return errors.New("singleton Caddy handler and template must be different files")
+	}
 	return nil
 }

@@ -14,8 +14,8 @@ func validConfig() Config {
 		Build:         Build{Package: "./cmd/site", Binary: "example-site", Branch: "main"},
 		Deployment: Deployment{
 			Strategy: "blue_green", Root: "/opt/example-site", LockFile: SharedLockFile,
-			StateFile: "/opt/example-site/state.json", HealthPath: "/healthz", ReadinessPath: "/readyz",
-			CandidateTimeoutSecs: 30, Smoke: []Smoke{{Path: "/", Contains: "Example"}},
+			StateFile: "/opt/example-site/state.json", EventLog: "/opt/example-site/deployment-events.jsonl", HealthPath: "/healthz", ReadinessPath: "/readyz",
+			CandidateTimeoutSecs: 30, ActivationWindowSecs: 10, Smoke: []Smoke{{Path: "/", Contains: "Example"}},
 			PublicSmoke: []PublicSmoke{{URL: "https://example.test/", Contains: "Example"}},
 			BlueGreen: &BlueGreen{
 				CaddyConfig: "/etc/caddy/Caddyfile", CaddyHandler: "/etc/caddy/example.caddy",
@@ -34,6 +34,24 @@ func TestValidateAcceptsBlueGreen(t *testing.T) {
 	}
 }

+func TestValidateSingletonRequiresDistinctCaddyHandoffFiles(t *testing.T) {
+	cfg := validConfig()
+	cfg.Deployment.Strategy = "singleton_candidate"
+	cfg.Deployment.BlueGreen = nil
+	cfg.Deployment.Singleton = &Singleton{
+		Unit: "example-site.service", Address: "127.0.0.1:8092", CandidateAddress: "127.0.0.1:18092", ListenEnv: "EXAMPLE_LISTEN",
+		CurrentLink: "/opt/example-site/current", PreviousLink: "/opt/example-site/previous", CaddyConfig: "/etc/caddy/Caddyfile",
+		CaddyHandler: "/etc/caddy/example-site.caddy", CaddyHandlerTemplate: "/etc/tend/caddy/example-site.template",
+	}
+	if err := cfg.Validate(); err != nil {
+		t.Fatal(err)
+	}
+	cfg.Deployment.Singleton.CaddyHandlerTemplate = cfg.Deployment.Singleton.CaddyHandler
+	if err := cfg.Validate(); err == nil {
+		t.Fatal("expected shared handler/template path to be rejected")
+	}
+}
+
 func TestValidateRejectsHostileValues(t *testing.T) {
 	tests := map[string]func(*Config){
 		"unknown strategy":    func(c *Config) { c.Deployment.Strategy = "shell" },
@@ -14,6 +14,7 @@ import (
 	"time"

 	"gamertan.com/tend/internal/config"
+	"gamertan.com/tend/internal/eventlog"
 	"gamertan.com/tend/internal/state"
 )

@@ -29,6 +30,7 @@ type Report struct {
 	Release         string `json:"release,omitempty"`
 	ActiveRelease   string `json:"active_release,omitempty"`
 	PreviousRelease string `json:"previous_release,omitempty"`
+	EventWarnings   int    `json:"event_warnings,omitempty"`
 }
 type Status struct {
 	State            *state.Record   `json:"state,omitempty"`
@@ -37,14 +39,18 @@ type Status struct {
 }

 type Manager struct {
-	Operator Operator
-	Now      func() time.Time
-	Prepare  func(config.Config, string, string, string) (string, error)
-	Inspect  func(config.Config, string, string, string) error
+	Operator     Operator
+	Now          func() time.Time
+	Prepare      func(config.Config, string, string, string) (string, error)
+	Inspect      func(config.Config, string, string, string) error
+	ReadIdentity func(string) (releaseIdentity, error)
+	OperationID  func() (string, error)
+	AppendEvent  func(string, eventlog.Event) error
+	Sleep        func(context.Context, time.Duration) error
 }

 func NewManager(operator Operator) Manager {
-	return Manager{Operator: operator, Now: time.Now, Prepare: prepareRelease, Inspect: inspectArtifact}
+	return Manager{Operator: operator, Now: time.Now, Prepare: prepareRelease, Inspect: inspectArtifact, ReadIdentity: readReleaseIdentity, OperationID: eventlog.OperationID, AppendEvent: eventlog.Append, Sleep: sleepContext}
 }

 func (m Manager) Deploy(ctx context.Context, cfg config.Config, request Request) (Report, error) {
@@ -66,10 +72,44 @@ func (m Manager) Deploy(ctx context.Context, cfg config.Config, request Request)
 	if err != nil {
 		return Report{}, err
 	}
+	started := m.Now()
+	eventWarnings := 0
+	identity, identityErr := m.ReadIdentity(release)
+	if identityErr != nil {
+		eventWarnings++
+	}
+	operationID := ""
+	if m.OperationID != nil {
+		operationID, err = m.OperationID()
+		if err != nil {
+			eventWarnings++
+			operationID = ""
+		}
+	}
+	emit := func(phase, slot, outcome string) {
+		if m.AppendEvent == nil || identityErr != nil || operationID == "" {
+			return
+		}
+		event := eventlog.Event{Version: eventlog.Version, OperationID: operationID, Service: cfg.Service.Name, ArtifactDigest: request.ApprovedSHA256, Commit: identity.Commit, ReleaseVersion: identity.Version, Phase: phase, Slot: slot, DurationMillis: max(0, m.Now().Sub(started).Milliseconds()), Outcome: outcome, ObservedAt: m.Now().UTC().Format(time.RFC3339Nano)}
+		if eventErr := m.AppendEvent(cfg.Deployment.EventLog, event); eventErr != nil {
+			eventWarnings++
+		}
+	}
 	record, err := loadOrBootstrap(cfg, m.Now())
 	if err != nil {
 		return Report{}, err
 	}
+	attemptAt := m.Now().UTC().Format(time.RFC3339)
+	record.DesiredRelease = release
+	record.CandidateRelease = release
+	record.LastAttemptRelease = release
+	record.LastAttemptOutcome = "running"
+	record.LastAttemptAt = attemptAt
+	record.UpdatedAt = attemptAt
+	if err := state.Store(cfg.Deployment.StateFile, cfg.Deployment.Root, record); err != nil {
+		return Report{}, err
+	}
+	emit("candidate", inactiveSlot(cfg, record), "running")
 	switch cfg.Deployment.Strategy {
 	case "blue_green":
 		err = m.deployBlueGreen(ctx, cfg, record, release)
@@ -79,13 +119,64 @@ func (m Manager) Deploy(ctx context.Context, cfg config.Config, request Request)
 		err = errors.New("unsupported strategy")
 	}
 	if err != nil {
-		return Report{}, err
+		failed := record
+		failed.CandidateRelease = ""
+		failed.LastAttemptOutcome = "failed"
+		failed.UpdatedAt = m.Now().UTC().Format(time.RFC3339)
+		_ = state.Store(cfg.Deployment.StateFile, cfg.Deployment.Root, failed)
+		emit("activation", inactiveSlot(cfg, record), "failed")
+		return Report{EventWarnings: eventWarnings}, err
 	}
+	emit("activation", inactiveSlot(cfg, record), "succeeded")
 	updated, err := state.Load(cfg.Deployment.StateFile, cfg.Deployment.Root, cfg.Deployment.Strategy)
 	if err != nil {
 		return Report{}, err
 	}
-	return Report{Validated: true, Mutation: "activated", Release: release, ActiveRelease: updated.ActiveRelease, PreviousRelease: updated.PreviousRelease}, nil
+	return Report{Validated: true, Mutation: "activated", Release: release, ActiveRelease: updated.ActiveRelease, PreviousRelease: updated.PreviousRelease, EventWarnings: eventWarnings}, nil
+}
+
+type releaseIdentity struct {
+	Version string `json:"version"`
+	Commit  string `json:"commit"`
+}
+
+func readReleaseIdentity(release string) (releaseIdentity, error) {
+	b, err := os.ReadFile(filepath.Join(release, "RELEASE.json"))
+	if err != nil {
+		return releaseIdentity{}, fmt.Errorf("read installed release identity: %w", err)
+	}
+	if len(b) > 1<<20 {
+		return releaseIdentity{}, errors.New("installed release identity is too large")
+	}
+	var identity releaseIdentity
+	if err := json.Unmarshal(b, &identity); err != nil {
+		return releaseIdentity{}, errors.New("decode installed release identity")
+	}
+	if identity.Version == "" || identity.Commit == "" {
+		return releaseIdentity{}, errors.New("installed release identity is incomplete")
+	}
+	return identity, nil
+}
+
+func inactiveSlot(cfg config.Config, record state.Record) string {
+	if cfg.Deployment.Strategy == "singleton_candidate" {
+		return "singleton"
+	}
+	if record.ActiveSlot == "blue" {
+		return "green"
+	}
+	return "blue"
+}
+
+func sleepContext(ctx context.Context, duration time.Duration) error {
+	timer := time.NewTimer(duration)
+	defer timer.Stop()
+	select {
+	case <-ctx.Done():
+		return ctx.Err()
+	case <-timer.C:
+		return nil
+	}
 }

 func loadOrBootstrap(cfg config.Config, now time.Time) (state.Record, error) {
@@ -103,13 +194,13 @@ func loadOrBootstrap(cfg config.Config, now time.Time) (state.Record, error) {
 		if err != nil {
 			return state.Record{}, fmt.Errorf("bootstrap active slot: %w", err)
 		}
-		return state.Record{SchemaVersion: 1, Strategy: cfg.Deployment.Strategy, ActiveSlot: slot, ActiveRelease: release, UpdatedAt: now.UTC().Format(time.RFC3339)}, nil
+		return state.Record{SchemaVersion: 1, Strategy: cfg.Deployment.Strategy, DesiredRelease: release, ActiveSlot: slot, ActiveRelease: release, UpdatedAt: now.UTC().Format(time.RFC3339)}, nil
 	case "singleton_candidate":
 		release, err := resolveReleaseLink(cfg.Deployment.Root, cfg.Deployment.Singleton.CurrentLink)
 		if err != nil {
 			return state.Record{}, fmt.Errorf("bootstrap singleton: %w", err)
 		}
-		return state.Record{SchemaVersion: 1, Strategy: cfg.Deployment.Strategy, ActiveSlot: "singleton", ActiveRelease: release, UpdatedAt: now.UTC().Format(time.RFC3339)}, nil
+		return state.Record{SchemaVersion: 1, Strategy: cfg.Deployment.Strategy, DesiredRelease: release, ActiveSlot: "singleton", ActiveRelease: release, UpdatedAt: now.UTC().Format(time.RFC3339)}, nil
 	}
 	return state.Record{}, errors.New("unsupported strategy")
 }
@@ -180,7 +271,11 @@ func (m Manager) deployBlueGreen(ctx context.Context, cfg config.Config, record
 	if err = m.probePublic(ctx, cfg, true); err != nil {
 		return fmt.Errorf("public-origin smoke failed: %w", err)
 	}
-	next := state.Record{SchemaVersion: 1, Strategy: cfg.Deployment.Strategy, ActiveSlot: inactive, ActiveRelease: release, PreviousSlot: record.ActiveSlot, PreviousRelease: record.ActiveRelease, UpdatedAt: m.Now().UTC().Format(time.RFC3339)}
+	previous := slotConfig(bg, record.ActiveSlot)
+	if err = m.continuityWindow(ctx, cfg, previous.Address, true); err != nil {
+		return fmt.Errorf("activation continuity failed: %w", err)
+	}
+	next := state.Record{SchemaVersion: 1, Strategy: cfg.Deployment.Strategy, DesiredRelease: release, ActiveSlot: inactive, ActiveRelease: release, PreviousSlot: record.ActiveSlot, PreviousRelease: record.ActiveRelease, LastAttemptRelease: release, LastAttemptOutcome: "succeeded", LastAttemptAt: record.LastAttemptAt, UpdatedAt: m.Now().UTC().Format(time.RFC3339)}
 	if err = state.Store(cfg.Deployment.StateFile, cfg.Deployment.Root, next); err != nil {
 		return err
 	}
@@ -188,62 +283,10 @@ func (m Manager) deployBlueGreen(ctx context.Context, cfg config.Config, record
 }

 func (m Manager) deploySingleton(ctx context.Context, cfg config.Config, record state.Record, release string) (err error) {
-	single := *cfg.Deployment.Singleton
-	candidateUnit := cfg.Service.Name + "-tend-candidate.service"
-	env := map[string]string{single.ListenEnv: single.CandidateAddress}
-	binary := filepath.Join(release, cfg.Build.Binary)
-	if err = m.Operator.StartCandidate(ctx, candidateUnit, binary, cfg.Service.EnvironmentFile, env); err != nil {
+	if err = m.activateSingletonRelease(ctx, cfg, release, true); err != nil {
 		return err
 	}
-	defer func() {
-		stopCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
-		defer cancel()
-		_ = m.Operator.Stop(stopCtx, candidateUnit)
-	}()
-	if err = m.probeAll(ctx, cfg, single.CandidateAddress); err != nil {
-		return fmt.Errorf("candidate failed: %w", err)
-	}
-	oldCurrent, err := resolveReleaseLink(cfg.Deployment.Root, single.CurrentLink)
-	if err != nil {
-		return err
-	}
-	oldPrevious, previousErr := resolveReleaseLink(cfg.Deployment.Root, single.PreviousLink)
-	currentChanged := false
-	previousChanged := false
-	defer func() {
-		if err == nil {
-			return
-		}
-		if currentChanged {
-			_ = replaceSymlink(single.CurrentLink, oldCurrent)
-			_ = m.Operator.Restart(ctx, single.Unit)
-		}
-		if previousChanged {
-			if previousErr == nil {
-				_ = replaceSymlink(single.PreviousLink, oldPrevious)
-			} else {
-				_ = removeSymlink(single.PreviousLink)
-			}
-		}
-	}()
-	if err = replaceSymlink(single.PreviousLink, oldCurrent); err != nil {
-		return err
-	}
-	previousChanged = true
-	if err = replaceSymlink(single.CurrentLink, release); err != nil {
-		return err
-	}
-	currentChanged = true
-	if err = m.Operator.Restart(ctx, single.Unit); err != nil {
-		return err
-	}
-	if err = m.probeAll(ctx, cfg, single.Address); err != nil {
-		return fmt.Errorf("post-activation smoke failed: %w", err)
-	}
-	if err = m.probePublic(ctx, cfg, true); err != nil {
-		return fmt.Errorf("public-origin smoke failed: %w", err)
-	}
-	next := state.Record{SchemaVersion: 1, Strategy: cfg.Deployment.Strategy, ActiveSlot: "singleton", ActiveRelease: release, PreviousSlot: "singleton", PreviousRelease: record.ActiveRelease, UpdatedAt: m.Now().UTC().Format(time.RFC3339)}
+	next := state.Record{SchemaVersion: 1, Strategy: cfg.Deployment.Strategy, DesiredRelease: release, ActiveSlot: "singleton", ActiveRelease: release, PreviousSlot: "singleton", PreviousRelease: record.ActiveRelease, LastAttemptRelease: release, LastAttemptOutcome: "succeeded", LastAttemptAt: record.LastAttemptAt, UpdatedAt: m.Now().UTC().Format(time.RFC3339)}
 	if err = state.Store(cfg.Deployment.StateFile, cfg.Deployment.Root, next); err != nil {
 		return err
 	}
@@ -263,6 +306,21 @@ func (m Manager) Rollback(ctx context.Context, cfg config.Config) (state.Record,
 	if record.PreviousRelease == "" {
 		return state.Record{}, errors.New("no previous release is recorded")
 	}
+	started := m.Now()
+	identity, identityErr := m.ReadIdentity(record.PreviousRelease)
+	digest, digestErr := releaseDigest(record.PreviousRelease)
+	operationID := ""
+	if m.OperationID != nil {
+		operationID, _ = m.OperationID()
+	}
+	emit := func(outcome string) {
+		if m.AppendEvent == nil || identityErr != nil || digestErr != nil || operationID == "" {
+			return
+		}
+		event := eventlog.Event{Version: eventlog.Version, OperationID: operationID, Service: cfg.Service.Name, ArtifactDigest: digest, Commit: identity.Commit, ReleaseVersion: identity.Version, Phase: "rollback", Slot: record.PreviousSlot, DurationMillis: max(0, m.Now().Sub(started).Milliseconds()), Outcome: outcome, ObservedAt: m.Now().UTC().Format(time.RFC3339Nano)}
+		_ = m.AppendEvent(cfg.Deployment.EventLog, event)
+	}
+	emit("running")
 	switch cfg.Deployment.Strategy {
 	case "blue_green":
 		err = m.rollbackBlueGreen(ctx, cfg, record)
@@ -272,10 +330,29 @@ func (m Manager) Rollback(ctx context.Context, cfg config.Config) (state.Record,
 		err = errors.New("unsupported strategy")
 	}
 	if err != nil {
+		emit("failed")
 		return state.Record{}, err
 	}
+	emit("succeeded")
 	return state.Load(cfg.Deployment.StateFile, cfg.Deployment.Root, cfg.Deployment.Strategy)
 }
+
+func releaseDigest(release string) (string, error) {
+	name := filepath.Base(release)
+	if !strings.HasPrefix(name, "sha256-") {
+		return "", errors.New("release is not content addressed")
+	}
+	digest := strings.TrimPrefix(name, "sha256-")
+	if len(digest) != 64 {
+		return "", errors.New("release digest is invalid")
+	}
+	for _, character := range digest {
+		if !strings.ContainsRune("0123456789abcdef", character) {
+			return "", errors.New("release digest is invalid")
+		}
+	}
+	return digest, nil
+}
 func (m Manager) rollbackBlueGreen(ctx context.Context, cfg config.Config, record state.Record) (err error) {
 	bg := *cfg.Deployment.BlueGreen
 	slot := slotConfig(bg, record.PreviousSlot)
@@ -321,36 +398,148 @@ func (m Manager) rollbackBlueGreen(ctx context.Context, cfg config.Config, recor
 	if err = m.probePublic(ctx, cfg, false); err != nil {
 		return err
 	}
-	next := state.Record{SchemaVersion: 1, Strategy: record.Strategy, ActiveSlot: record.PreviousSlot, ActiveRelease: record.PreviousRelease, PreviousSlot: record.ActiveSlot, PreviousRelease: record.ActiveRelease, UpdatedAt: m.Now().UTC().Format(time.RFC3339)}
+	next := state.Record{SchemaVersion: 1, Strategy: record.Strategy, DesiredRelease: record.PreviousRelease, ActiveSlot: record.PreviousSlot, ActiveRelease: record.PreviousRelease, PreviousSlot: record.ActiveSlot, PreviousRelease: record.ActiveRelease, LastAttemptRelease: record.PreviousRelease, LastAttemptOutcome: "rolled_back", LastAttemptAt: m.Now().UTC().Format(time.RFC3339), UpdatedAt: m.Now().UTC().Format(time.RFC3339)}
 	return state.Store(cfg.Deployment.StateFile, cfg.Deployment.Root, next)
 }
 func (m Manager) rollbackSingleton(ctx context.Context, cfg config.Config, record state.Record) (err error) {
+	if err = m.activateSingletonRelease(ctx, cfg, record.PreviousRelease, false); err != nil {
+		return err
+	}
+	next := state.Record{SchemaVersion: 1, Strategy: record.Strategy, DesiredRelease: record.PreviousRelease, ActiveSlot: "singleton", ActiveRelease: record.PreviousRelease, PreviousSlot: "singleton", PreviousRelease: record.ActiveRelease, LastAttemptRelease: record.PreviousRelease, LastAttemptOutcome: "rolled_back", LastAttemptAt: m.Now().UTC().Format(time.RFC3339), UpdatedAt: m.Now().UTC().Format(time.RFC3339)}
+	return state.Store(cfg.Deployment.StateFile, cfg.Deployment.Root, next)
+}
+
+// activateSingletonRelease keeps public traffic on a proven process while the
+// installed fixed-address unit changes release. The transient candidate first
+// receives traffic, remains healthy through the handoff, and is stopped only
+// after Caddy points back to the verified installed unit.
+func (m Manager) activateSingletonRelease(ctx context.Context, cfg config.Config, release string, checkMarkers bool) (err error) {
 	single := *cfg.Deployment.Singleton
+	candidateUnit := cfg.Service.Name + "-tend-candidate.service"
+	env := map[string]string{single.ListenEnv: single.CandidateAddress}
+	binary := filepath.Join(release, cfg.Build.Binary)
+	if err = m.Operator.StartCandidate(ctx, candidateUnit, binary, cfg.Service.EnvironmentFile, env); err != nil {
+		return err
+	}
+	stopCandidate := true
+	defer func() {
+		if !stopCandidate {
+			return
+		}
+		stopCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+		defer cancel()
+		_ = m.Operator.Stop(stopCtx, candidateUnit)
+	}()
+	probeLocal := m.probeHealthReadiness
+	if checkMarkers {
+		probeLocal = m.probeAll
+	}
+	if err = probeLocal(ctx, cfg, single.CandidateAddress); err != nil {
+		return fmt.Errorf("candidate failed: %w", err)
+	}
+
+	oldHandler, err := os.ReadFile(single.CaddyHandler)
+	if err != nil {
+		return fmt.Errorf("read current Caddy handler: %w", err)
+	}
 	oldCurrent, err := resolveReleaseLink(cfg.Deployment.Root, single.CurrentLink)
 	if err != nil {
 		return err
 	}
-	if err = replaceSymlink(single.CurrentLink, record.PreviousRelease); err != nil {
-		return err
-	}
+	oldPrevious, previousErr := resolveReleaseLink(cfg.Deployment.Root, single.PreviousLink)
+	handlerChanged := false
+	currentChanged := false
+	previousChanged := false
 	defer func() {
-		if err != nil {
-			_ = replaceSymlink(single.CurrentLink, oldCurrent)
-			_ = m.Operator.Restart(ctx, single.Unit)
+		if err == nil {
+			return
+		}
+		recoveryErr := error(nil)
+		if currentChanged {
+			if restoreErr := replaceSymlink(single.CurrentLink, oldCurrent); restoreErr != nil {
+				recoveryErr = errors.Join(recoveryErr, restoreErr)
+			} else if restoreErr = m.Operator.Restart(ctx, single.Unit); restoreErr != nil {
+				recoveryErr = errors.Join(recoveryErr, restoreErr)
+			} else if restoreErr = m.probeHealthReadiness(ctx, cfg, single.Address); restoreErr != nil {
+				recoveryErr = errors.Join(recoveryErr, restoreErr)
+			}
+		}
+		if previousChanged {
+			var restoreErr error
+			if previousErr == nil {
+				restoreErr = replaceSymlink(single.PreviousLink, oldPrevious)
+			} else {
+				restoreErr = removeSymlink(single.PreviousLink)
+			}
+			recoveryErr = errors.Join(recoveryErr, restoreErr)
+		}
+		if handlerChanged && recoveryErr == nil {
+			if restoreErr := atomicWrite(single.CaddyHandler, oldHandler, 0o644); restoreErr != nil {
+				recoveryErr = errors.Join(recoveryErr, restoreErr)
+			} else if restoreErr = m.Operator.ValidateCaddy(ctx, single.CaddyConfig); restoreErr != nil {
+				recoveryErr = errors.Join(recoveryErr, restoreErr)
+			} else if restoreErr = m.Operator.ReloadCaddy(ctx); restoreErr != nil {
+				recoveryErr = errors.Join(recoveryErr, restoreErr)
+			}
+		}
+		if recoveryErr != nil && handlerChanged {
+			stopCandidate = false
+			err = errors.Join(err, fmt.Errorf("singleton recovery incomplete; candidate remains routed for operator recovery: %w", recoveryErr))
 		}
 	}()
+
+	candidateHandler, err := renderHandler(single.CaddyHandlerTemplate, single.CandidateAddress)
+	if err != nil {
+		return err
+	}
+	if err = atomicWrite(single.CaddyHandler, candidateHandler, 0o644); err != nil {
+		return err
+	}
+	handlerChanged = true
+	if err = m.Operator.ValidateCaddy(ctx, single.CaddyConfig); err != nil {
+		return fmt.Errorf("candidate Caddy validation failed: %w", err)
+	}
+	if err = m.Operator.ReloadCaddy(ctx); err != nil {
+		return fmt.Errorf("candidate Caddy reload failed: %w", err)
+	}
+	if err = m.probePublic(ctx, cfg, checkMarkers); err != nil {
+		return fmt.Errorf("candidate public-origin smoke failed: %w", err)
+	}
+
+	if err = replaceSymlink(single.PreviousLink, oldCurrent); err != nil {
+		return err
+	}
+	previousChanged = true
+	if err = replaceSymlink(single.CurrentLink, release); err != nil {
+		return err
+	}
+	currentChanged = true
 	if err = m.Operator.Restart(ctx, single.Unit); err != nil {
 		return err
 	}
-	if err = m.probeHealthReadiness(ctx, cfg, single.Address); err != nil {
+	if err = probeLocal(ctx, cfg, single.Address); err != nil {
+		return fmt.Errorf("post-activation smoke failed: %w", err)
+	}
+	installedHandler, err := renderHandler(single.CaddyHandlerTemplate, single.Address)
+	if err != nil {
 		return err
 	}
-	if err = m.probePublic(ctx, cfg, false); err != nil {
+	if err = atomicWrite(single.CaddyHandler, installedHandler, 0o644); err != nil {
 		return err
 	}
-	_ = replaceSymlink(single.PreviousLink, record.ActiveRelease)
-	next := state.Record{SchemaVersion: 1, Strategy: record.Strategy, ActiveSlot: "singleton", ActiveRelease: record.PreviousRelease, PreviousSlot: "singleton", PreviousRelease: record.ActiveRelease, UpdatedAt: m.Now().UTC().Format(time.RFC3339)}
-	return state.Store(cfg.Deployment.StateFile, cfg.Deployment.Root, next)
+	if err = m.Operator.ValidateCaddy(ctx, single.CaddyConfig); err != nil {
+		return fmt.Errorf("installed Caddy validation failed: %w", err)
+	}
+	if err = m.Operator.ReloadCaddy(ctx); err != nil {
+		return fmt.Errorf("installed Caddy reload failed: %w", err)
+	}
+	if err = m.probePublic(ctx, cfg, checkMarkers); err != nil {
+		return fmt.Errorf("public-origin smoke failed: %w", err)
+	}
+	if err = m.continuityWindow(ctx, cfg, single.CandidateAddress, checkMarkers); err != nil {
+		return fmt.Errorf("activation continuity failed: %w", err)
+	}
+	return nil
 }

 func (m Manager) Status(ctx context.Context, cfg config.Config) (Status, error) {
@@ -463,6 +652,33 @@ func (m Manager) probePublic(ctx context.Context, cfg config.Config, checkMarker
 	return nil
 }

+func (m Manager) continuityWindow(ctx context.Context, cfg config.Config, previousAddress string, checkMarkers bool) error {
+	steps := cfg.Deployment.ActivationWindowSecs * 4
+	if steps < 1 {
+		steps = 1
+	}
+	for step := 0; step < steps; step++ {
+		if err := m.probePublic(ctx, cfg, checkMarkers); err != nil {
+			return err
+		}
+		if previousAddress != "" {
+			if err := m.probeHealthReadiness(ctx, cfg, previousAddress); err != nil {
+				return fmt.Errorf("previous slot lost continuity: %w", err)
+			}
+		}
+		if step+1 < steps {
+			sleep := m.Sleep
+			if sleep == nil {
+				sleep = sleepContext
+			}
+			if err := sleep(ctx, 250*time.Millisecond); err != nil {
+				return err
+			}
+		}
+	}
+	return nil
+}
+
 func (m Manager) probe(ctx context.Context, cfg config.Config, address string, checks []config.Smoke) error {
 	timeout := time.Duration(cfg.Deployment.CandidateTimeoutSecs) * time.Second
 	for _, check := range checks {
@@ -5,6 +5,7 @@ package deploy
 import (
 	"context"
 	"errors"
+	"fmt"
 	"os"
 	"path/filepath"
 	"strings"
@@ -12,18 +13,23 @@ import (
 	"time"

 	"gamertan.com/tend/internal/config"
+	"gamertan.com/tend/internal/eventlog"
 	"gamertan.com/tend/internal/state"
 )

 type fakeOperator struct {
 	failReload              bool
+	failReloadAt            int
+	reloads                 int
 	failRestartUnit         string
+	failRestartOnce         bool
 	rejectMarkers           bool
 	active                  map[string]bool
 	starts, stops, restarts []string
 	probes                  []string
 	publicProbes            []string
 	failPublic              bool
+	failPublicAfter         int
 	candidateEnvironment    map[string]string
 	candidateFile           string
 }
@@ -31,6 +37,9 @@ type fakeOperator struct {
 func (f *fakeOperator) Restart(_ context.Context, unit string) error {
 	f.restarts = append(f.restarts, unit)
 	if unit == f.failRestartUnit {
+		if f.failRestartOnce {
+			f.failRestartUnit = ""
+		}
 		return errors.New("injected restart failure")
 	}
 	f.active[unit] = true
@@ -59,7 +68,7 @@ func (f *fakeOperator) StartCandidate(_ context.Context, unit, binary, environme
 }
 func (f *fakeOperator) ProbeURL(_ context.Context, value, contains string) error {
 	f.publicProbes = append(f.publicProbes, value)
-	if f.failPublic {
+	if f.failPublic || (f.failPublicAfter > 0 && len(f.publicProbes) >= f.failPublicAfter) {
 		return errors.New("injected public smoke failure")
 	}
 	if f.rejectMarkers && contains != "" {
@@ -92,13 +101,44 @@ func TestPublicSmokeFailureRestoresBlueGreenHandlerAndSlot(t *testing.T) {
 	if err != nil || target != old {
 		t.Fatalf("green=%q err=%v", target, err)
 	}
-	if _, err = os.Stat(cfg.Deployment.StateFile); !os.IsNotExist(err) {
-		t.Fatal("failed public smoke wrote state")
+	record, err := state.Load(cfg.Deployment.StateFile, cfg.Deployment.Root, cfg.Deployment.Strategy)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if record.ActiveRelease != old || record.LastAttemptOutcome != "failed" || record.CandidateRelease != "" || record.LastAttemptRelease != fresh {
+		t.Fatalf("failed attempt state=%+v", record)
+	}
+}
+
+func TestContinuityFailureRestoresBlueGreenRoute(t *testing.T) {
+	cfg, old, fresh := baseConfig(t, "blue_green")
+	handler := filepath.Join(cfg.Deployment.Root, "handler.caddy")
+	template := filepath.Join(cfg.Deployment.Root, "handler.template")
+	original := []byte("reverse_proxy 127.0.0.1:8090\n")
+	_ = os.WriteFile(handler, original, 0o644)
+	_ = os.WriteFile(template, []byte("reverse_proxy {{UPSTREAM}}\n"), 0o644)
+	blue := filepath.Join(cfg.Deployment.Root, "slots", "blue")
+	green := filepath.Join(cfg.Deployment.Root, "slots", "green")
+	_ = replaceSymlink(blue, old)
+	_ = replaceSymlink(green, old)
+	cfg.Deployment.BlueGreen = &config.BlueGreen{CaddyConfig: filepath.Join(cfg.Deployment.Root, "Caddyfile"), CaddyHandler: handler, CaddyHandlerTemplate: template, BootstrapActive: "blue", Blue: config.Slot{Unit: "example-blue.service", Address: "127.0.0.1:8090", Link: blue}, Green: config.Slot{Unit: "example-green.service", Address: "127.0.0.1:8091", Link: green}}
+	operator := &fakeOperator{active: map[string]bool{}, failPublicAfter: 3}
+	if _, err := manager(operator, fresh).Deploy(context.Background(), cfg, Request{Activate: true, ApprovedSHA256: strings.Repeat("a", 64)}); err == nil || !strings.Contains(err.Error(), "continuity") {
+		t.Fatalf("expected continuity failure, got %v", err)
+	}
+	body, _ := os.ReadFile(handler)
+	if string(body) != string(original) {
+		t.Fatalf("handler not restored: %q", body)
+	}
+	target, err := resolveReleaseLink(cfg.Deployment.Root, green)
+	if err != nil || target != old {
+		t.Fatalf("green=%q err=%v", target, err)
 	}
 }
 func (f *fakeOperator) ValidateCaddy(context.Context, string) error { return nil }
 func (f *fakeOperator) ReloadCaddy(context.Context) error {
-	if f.failReload {
+	f.reloads++
+	if f.failReload || (f.failReloadAt > 0 && f.reloads == f.failReloadAt) {
 		return errors.New("injected reload failure")
 	}
 	return nil
@@ -117,7 +157,7 @@ func baseConfig(t *testing.T, strategy string) (config.Config, string, string) {
 	if err := os.MkdirAll(filepath.Join(root, "releases"), 0o755); err != nil {
 		t.Fatal(err)
 	}
-	old := filepath.Join(root, "releases", "legacy-old")
+	old := filepath.Join(root, "releases", "sha256-"+strings.Repeat("c", 64))
 	fresh := filepath.Join(root, "releases", "sha256-"+strings.Repeat("a", 64))
 	for _, dir := range []string{old, fresh} {
 		if err := os.Mkdir(dir, 0o755); err != nil {
@@ -127,11 +167,34 @@ func baseConfig(t *testing.T, strategy string) (config.Config, string, string) {
 			t.Fatal(err)
 		}
 	}
-	cfg := config.Config{SchemaVersion: 2, Service: config.Service{Name: "example-site", AllowedHost: "example.test", EnvironmentFile: "/etc/tend/environment/example-site.env"}, Build: config.Build{Package: "./cmd/site", Binary: "app", Branch: "main"}, Deployment: config.Deployment{Strategy: strategy, Root: root, LockFile: filepath.Join(root, "deploy.lock"), StateFile: filepath.Join(root, "state.json"), HealthPath: "/healthz", ReadinessPath: "/readyz", CandidateTimeoutSecs: 2, Smoke: []config.Smoke{{Path: "/", Contains: "Example"}}, PublicSmoke: []config.PublicSmoke{{URL: "https://example.test/", Contains: "Example"}}}}
+	cfg := config.Config{SchemaVersion: 2, Service: config.Service{Name: "example-site", AllowedHost: "example.test", EnvironmentFile: "/etc/tend/environment/example-site.env"}, Build: config.Build{Package: "./cmd/site", Binary: "app", Branch: "main"}, Deployment: config.Deployment{Strategy: strategy, Root: root, LockFile: filepath.Join(root, "deploy.lock"), StateFile: filepath.Join(root, "state.json"), EventLog: filepath.Join(root, "deployment-events.jsonl"), HealthPath: "/healthz", ReadinessPath: "/readyz", CandidateTimeoutSecs: 2, ActivationWindowSecs: 1, Smoke: []config.Smoke{{Path: "/", Contains: "Example"}}, PublicSmoke: []config.PublicSmoke{{URL: "https://example.test/", Contains: "Example"}}}}
 	return cfg, old, fresh
 }
 func manager(operator Operator, fresh string) Manager {
-	return Manager{Operator: operator, Now: func() time.Time { return time.Unix(100, 0).UTC() }, Prepare: func(config.Config, string, string, string) (string, error) { return fresh, nil }, Inspect: func(config.Config, string, string, string) error { return nil }}
+	return Manager{Operator: operator, Now: func() time.Time { return time.Unix(100, 0).UTC() }, Prepare: func(config.Config, string, string, string) (string, error) { return fresh, nil }, Inspect: func(config.Config, string, string, string) error { return nil }, ReadIdentity: func(string) (releaseIdentity, error) {
+		return releaseIdentity{Version: "v0.2.0-preview.1", Commit: strings.Repeat("b", 40)}, nil
+	}, OperationID: func() (string, error) { return strings.Repeat("d", 32), nil }, Sleep: func(context.Context, time.Duration) error { return nil }}
+}
+
+func singletonSettings(t *testing.T, cfg config.Config, currentRelease string) (*config.Singleton, string) {
+	t.Helper()
+	handler := filepath.Join(cfg.Deployment.Root, "handler.caddy")
+	template := filepath.Join(cfg.Deployment.Root, "handler.template")
+	if err := os.WriteFile(handler, []byte("reverse_proxy 127.0.0.1:8092\n"), 0o640); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(template, []byte("reverse_proxy {{UPSTREAM}}\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	current := filepath.Join(cfg.Deployment.Root, "current")
+	if err := replaceSymlink(current, currentRelease); err != nil {
+		t.Fatal(err)
+	}
+	return &config.Singleton{
+		Unit: "example-site.service", Address: "127.0.0.1:8092", CandidateAddress: "127.0.0.1:18092", ListenEnv: "EXAMPLE_LISTEN",
+		CurrentLink: current, PreviousLink: filepath.Join(cfg.Deployment.Root, "previous"), CaddyConfig: filepath.Join(cfg.Deployment.Root, "Caddyfile"),
+		CaddyHandler: handler, CaddyHandlerTemplate: template,
+	}, handler
 }

 func TestBlueGreenActivationAndRollback(t *testing.T) {
@@ -158,13 +221,28 @@ func TestBlueGreenActivationAndRollback(t *testing.T) {
 	}
 	operator := &fakeOperator{active: map[string]bool{"example-blue.service": true, "example-green.service": true}}
 	m := manager(operator, fresh)
-	report, err := m.Deploy(context.Background(), cfg, Request{Activate: true})
+	var events []eventlog.Event
+	m.AppendEvent = func(_ string, event eventlog.Event) error {
+		events = append(events, event)
+		return nil
+	}
+	report, err := m.Deploy(context.Background(), cfg, Request{Activate: true, ApprovedSHA256: strings.Repeat("a", 64)})
 	if err != nil {
 		t.Fatal(err)
 	}
 	if report.ActiveRelease != fresh || report.PreviousRelease != old {
 		t.Fatalf("report=%+v", report)
 	}
+	deployed, err := state.Load(cfg.Deployment.StateFile, cfg.Deployment.Root, cfg.Deployment.Strategy)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if deployed.DesiredRelease != fresh || deployed.CandidateRelease != "" || deployed.LastAttemptRelease != fresh || deployed.LastAttemptOutcome != "succeeded" {
+		t.Fatalf("deployment identity state=%+v", deployed)
+	}
+	if len(events) != 2 || events[0].Phase != "candidate" || events[0].Outcome != "running" || events[1].Phase != "activation" || events[1].Outcome != "succeeded" || events[0].OperationID != events[1].OperationID {
+		t.Fatalf("events=%+v", events)
+	}
 	if info, err := os.Stat(handler); err != nil || info.Mode().Perm() != 0o640 {
 		t.Fatalf("handler mode=%v err=%v", info.Mode().Perm(), err)
 	}
@@ -180,6 +258,9 @@ func TestBlueGreenActivationAndRollback(t *testing.T) {
 	if record.ActiveRelease != old || record.PreviousRelease != fresh {
 		t.Fatalf("rollback=%+v", record)
 	}
+	if len(events) != 4 || events[2].Phase != "rollback" || events[2].Outcome != "running" || events[3].Phase != "rollback" || events[3].Outcome != "succeeded" || events[2].OperationID != events[3].OperationID || events[2].ArtifactDigest != strings.Repeat("c", 64) {
+		t.Fatalf("rollback events=%+v", events)
+	}
 	if len(operator.probes) != 4 {
 		t.Fatalf("rollback probes=%#v", operator.probes)
 	}
@@ -190,6 +271,58 @@ func TestBlueGreenActivationAndRollback(t *testing.T) {
 	}
 }

+func TestDeploymentEvidenceCanNeverBlockActivationOrRollback(t *testing.T) {
+	cfg, old, fresh := baseConfig(t, "singleton_candidate")
+	cfg.Deployment.Singleton, _ = singletonSettings(t, cfg, old)
+	m := manager(&fakeOperator{active: map[string]bool{}}, fresh)
+	appendCalls := 0
+	m.AppendEvent = func(string, eventlog.Event) error {
+		appendCalls++
+		return errors.New("injected event failure")
+	}
+	report, err := m.Deploy(context.Background(), cfg, Request{Activate: true, ApprovedSHA256: strings.Repeat("a", 64)})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if report.EventWarnings != 2 || report.ActiveRelease != fresh {
+		t.Fatalf("report=%+v", report)
+	}
+	if _, err := m.Rollback(context.Background(), cfg); err != nil {
+		t.Fatal(err)
+	}
+	record, err := state.Load(cfg.Deployment.StateFile, cfg.Deployment.Root, cfg.Deployment.Strategy)
+	if err != nil || record.ActiveRelease != old || appendCalls != 4 {
+		t.Fatalf("record=%+v appends=%d err=%v", record, appendCalls, err)
+	}
+}
+
+func TestDeploymentEvidenceIdentityAndEntropyAreBestEffort(t *testing.T) {
+	for _, test := range []struct {
+		name   string
+		damage func(*Manager)
+	}{
+		{"identity", func(manager *Manager) {
+			manager.ReadIdentity = func(string) (releaseIdentity, error) {
+				return releaseIdentity{}, errors.New("injected identity failure")
+			}
+		}},
+		{"entropy", func(manager *Manager) {
+			manager.OperationID = func() (string, error) { return "", errors.New("injected entropy failure") }
+		}},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			cfg, old, fresh := baseConfig(t, "singleton_candidate")
+			cfg.Deployment.Singleton, _ = singletonSettings(t, cfg, old)
+			manager := manager(&fakeOperator{active: map[string]bool{}}, fresh)
+			test.damage(&manager)
+			report, err := manager.Deploy(context.Background(), cfg, Request{Activate: true, ApprovedSHA256: strings.Repeat("a", 64)})
+			if err != nil || report.EventWarnings != 1 || report.ActiveRelease != fresh {
+				t.Fatalf("report=%+v err=%v", report, err)
+			}
+		})
+	}
+}
+
 func TestBlueGreenCaddyFailureRestoresHandlerAndSlot(t *testing.T) {
 	cfg, old, fresh := baseConfig(t, "blue_green")
 	handler := filepath.Join(cfg.Deployment.Root, "handler.caddy")
@@ -215,37 +348,47 @@ func TestBlueGreenCaddyFailureRestoresHandlerAndSlot(t *testing.T) {
 	if err != nil || target != old {
 		t.Fatalf("green=%q err=%v", target, err)
 	}
-	if _, err := os.Stat(cfg.Deployment.StateFile); !os.IsNotExist(err) {
-		t.Fatal("failed activation wrote state")
+	failed, err := state.Load(cfg.Deployment.StateFile, cfg.Deployment.Root, cfg.Deployment.Strategy)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if failed.ActiveRelease != old || failed.LastAttemptOutcome != "failed" || failed.CandidateRelease != "" {
+		t.Fatalf("failed state=%+v", failed)
 	}
 }

 func TestSingletonRestartFailureRestoresPointers(t *testing.T) {
 	cfg, old, fresh := baseConfig(t, "singleton_candidate")
-	previous := filepath.Join(cfg.Deployment.Root, "previous")
-	current := filepath.Join(cfg.Deployment.Root, "current")
-	_ = replaceSymlink(current, old)
-	cfg.Deployment.Singleton = &config.Singleton{Unit: "example-site.service", Address: "127.0.0.1:8092", CandidateAddress: "127.0.0.1:18092", ListenEnv: "EXAMPLE_LISTEN", CurrentLink: current, PreviousLink: previous}
-	operator := &fakeOperator{active: map[string]bool{}, failRestartUnit: "example-site.service"}
+	settings, handler := singletonSettings(t, cfg, old)
+	cfg.Deployment.Singleton = settings
+	operator := &fakeOperator{active: map[string]bool{}, failRestartUnit: "example-site.service", failRestartOnce: true}
 	m := manager(operator, fresh)
 	if _, err := m.Deploy(context.Background(), cfg, Request{Activate: true}); err == nil {
 		t.Fatal("expected failure")
 	}
-	target, err := resolveReleaseLink(cfg.Deployment.Root, current)
+	target, err := resolveReleaseLink(cfg.Deployment.Root, cfg.Deployment.Singleton.CurrentLink)
 	if err != nil || target != old {
 		t.Fatalf("current=%q err=%v", target, err)
 	}
-	if _, err := os.Lstat(previous); !os.IsNotExist(err) {
+	if _, err := os.Lstat(cfg.Deployment.Singleton.PreviousLink); !os.IsNotExist(err) {
 		t.Fatal("previous pointer was not restored")
 	}
+	body, err := os.ReadFile(handler)
+	if err != nil || string(body) != "reverse_proxy 127.0.0.1:8092\n" {
+		t.Fatalf("handler=%q err=%v", body, err)
+	}
+	if info, err := os.Stat(handler); err != nil || info.Mode().Perm() != 0o640 {
+		t.Fatalf("handler mode=%v err=%v", info.Mode().Perm(), err)
+	}
+	if operator.active["example-site-tend-candidate.service"] {
+		t.Fatal("candidate was not stopped after successful restoration")
+	}
 }

-func TestStatePersistsOnlyAfterSuccessfulActivation(t *testing.T) {
+func TestStateRecordsSuccessfulActivation(t *testing.T) {
 	cfg, old, fresh := baseConfig(t, "singleton_candidate")
-	current := filepath.Join(cfg.Deployment.Root, "current")
-	previous := filepath.Join(cfg.Deployment.Root, "previous")
-	_ = replaceSymlink(current, old)
-	cfg.Deployment.Singleton = &config.Singleton{Unit: "example-site.service", Address: "127.0.0.1:8092", CandidateAddress: "127.0.0.1:18092", ListenEnv: "EXAMPLE_LISTEN", CurrentLink: current, PreviousLink: previous}
+	settings, handler := singletonSettings(t, cfg, old)
+	cfg.Deployment.Singleton = settings
 	operator := &fakeOperator{active: map[string]bool{}}
 	m := manager(operator, fresh)
 	if _, err := m.Deploy(context.Background(), cfg, Request{Activate: true}); err != nil {
@@ -261,4 +404,72 @@ func TestStatePersistsOnlyAfterSuccessfulActivation(t *testing.T) {
 	if operator.candidateFile != cfg.Service.EnvironmentFile || len(operator.candidateEnvironment) != 1 || operator.candidateEnvironment["EXAMPLE_LISTEN"] != "127.0.0.1:18092" {
 		t.Fatalf("candidate file=%q environment=%#v", operator.candidateFile, operator.candidateEnvironment)
 	}
+	body, err := os.ReadFile(handler)
+	if err != nil || string(body) != "reverse_proxy 127.0.0.1:8092\n" {
+		t.Fatalf("handler=%q err=%v", body, err)
+	}
+	if operator.reloads != 2 || operator.active["example-site-tend-candidate.service"] {
+		t.Fatalf("reloads=%d active=%#v", operator.reloads, operator.active)
+	}
+}
+
+func TestSingletonContinuityFailureRestoresHandlerPointersAndService(t *testing.T) {
+	cfg, old, fresh := baseConfig(t, "singleton_candidate")
+	settings, handler := singletonSettings(t, cfg, old)
+	cfg.Deployment.Singleton = settings
+	operator := &fakeOperator{active: map[string]bool{"example-site.service": true}, failPublicAfter: 4}
+	m := manager(operator, fresh)
+	if _, err := m.Deploy(context.Background(), cfg, Request{Activate: true}); err == nil || !strings.Contains(err.Error(), "continuity") {
+		t.Fatalf("expected continuity failure, got %v", err)
+	}
+	current, err := resolveReleaseLink(cfg.Deployment.Root, cfg.Deployment.Singleton.CurrentLink)
+	if err != nil || current != old {
+		t.Fatalf("current=%q err=%v", current, err)
+	}
+	body, err := os.ReadFile(handler)
+	if err != nil || string(body) != "reverse_proxy 127.0.0.1:8092\n" {
+		t.Fatalf("handler=%q err=%v", body, err)
+	}
+	if operator.active["example-site-tend-candidate.service"] {
+		t.Fatal("candidate was not stopped after continuity restoration")
+	}
+}
+
+func TestSingletonCaddyReloadFailuresRestorePriorRoute(t *testing.T) {
+	for _, reload := range []int{1, 2} {
+		t.Run(fmt.Sprintf("reload-%d", reload), func(t *testing.T) {
+			cfg, old, fresh := baseConfig(t, "singleton_candidate")
+			settings, handler := singletonSettings(t, cfg, old)
+			cfg.Deployment.Singleton = settings
+			operator := &fakeOperator{active: map[string]bool{"example-site.service": true}, failReloadAt: reload}
+			if _, err := manager(operator, fresh).Deploy(context.Background(), cfg, Request{Activate: true}); err == nil || !strings.Contains(err.Error(), "Caddy reload failed") {
+				t.Fatalf("expected Caddy reload failure, got %v", err)
+			}
+			current, err := resolveReleaseLink(cfg.Deployment.Root, settings.CurrentLink)
+			if err != nil || current != old {
+				t.Fatalf("current=%q err=%v", current, err)
+			}
+			body, err := os.ReadFile(handler)
+			if err != nil || string(body) != "reverse_proxy 127.0.0.1:8092\n" {
+				t.Fatalf("handler=%q err=%v", body, err)
+			}
+			if operator.active["example-site-tend-candidate.service"] {
+				t.Fatal("candidate was not stopped after route restoration")
+			}
+		})
+	}
+}
+
+func TestSingletonIncompleteRecoveryKeepsProvenCandidateRunning(t *testing.T) {
+	cfg, old, fresh := baseConfig(t, "singleton_candidate")
+	settings, _ := singletonSettings(t, cfg, old)
+	cfg.Deployment.Singleton = settings
+	operator := &fakeOperator{active: map[string]bool{"example-site.service": true}, failReload: true}
+	_, err := manager(operator, fresh).Deploy(context.Background(), cfg, Request{Activate: true})
+	if err == nil || !strings.Contains(err.Error(), "candidate remains routed for operator recovery") {
+		t.Fatalf("expected explicit incomplete recovery, got %v", err)
+	}
+	if !operator.active["example-site-tend-candidate.service"] {
+		t.Fatal("proven candidate was stopped despite incomplete route restoration")
+	}
 }
@@ -196,6 +196,15 @@ func extractArtifact(artifact, stage string) error {
 			_ = out.Close()
 			return err
 		}
+		// OpenFile modes are filtered through the caller's umask. Tend is
+		// commonly invoked by a root account with umask 0077, while release
+		// binaries must remain executable by their dedicated service users.
+		// Reapply the validated, name-derived mode explicitly before the file
+		// becomes part of an immutable release.
+		if err := out.Chmod(mode); err != nil {
+			_ = out.Close()
+			return err
+		}
 		if err := out.Sync(); err != nil {
 			_ = out.Close()
 			return err
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+
+//go:build linux
+
+package deploy
+
+import (
+	"archive/tar"
+	"compress/gzip"
+	"os"
+	"path/filepath"
+	"syscall"
+	"testing"
+)
+
+func TestExtractArtifactAppliesReleaseModesUnderRestrictiveUmask(t *testing.T) {
+	dir := t.TempDir()
+	artifact := filepath.Join(dir, "release.tar.gz")
+	file, err := os.OpenFile(artifact, os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600)
+	if err != nil {
+		t.Fatal(err)
+	}
+	gz := gzip.NewWriter(file)
+	tw := tar.NewWriter(gz)
+	entries := map[string][]byte{
+		"bundle/app":            []byte("executable"),
+		"bundle/BUILDINFO.json": []byte("{}"),
+		"bundle/RELEASE.json":   []byte("{}"),
+		"bundle/SBOM.spdx.json": []byte("{}"),
+		"bundle/SHA256SUMS":     []byte("checksums"),
+	}
+	for name, body := range entries {
+		if err := tw.WriteHeader(&tar.Header{Name: name, Typeflag: tar.TypeReg, Mode: 0o600, Size: int64(len(body))}); err != nil {
+			t.Fatal(err)
+		}
+		if _, err := tw.Write(body); err != nil {
+			t.Fatal(err)
+		}
+	}
+	if err := tw.Close(); err != nil {
+		t.Fatal(err)
+	}
+	if err := gz.Close(); err != nil {
+		t.Fatal(err)
+	}
+	if err := file.Close(); err != nil {
+		t.Fatal(err)
+	}
+
+	oldUmask := syscall.Umask(0o077)
+	t.Cleanup(func() { syscall.Umask(oldUmask) })
+	stage := filepath.Join(dir, "stage")
+	if err := os.Mkdir(stage, 0o700); err != nil {
+		t.Fatal(err)
+	}
+	if err := extractArtifact(artifact, stage); err != nil {
+		t.Fatal(err)
+	}
+
+	for name, want := range map[string]os.FileMode{
+		"app":            0o755,
+		"BUILDINFO.json": 0o644,
+		"RELEASE.json":   0o644,
+		"SBOM.spdx.json": 0o644,
+		"SHA256SUMS":     0o644,
+	} {
+		info, err := os.Stat(filepath.Join(stage, name))
+		if err != nil {
+			t.Fatal(err)
+		}
+		if got := info.Mode().Perm(); got != want {
+			t.Fatalf("%s mode=%#o want=%#o", name, got, want)
+		}
+	}
+}
@@ -0,0 +1,115 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+
+package eventlog
+
+import (
+	"crypto/rand"
+	"encoding/hex"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"regexp"
+	"strings"
+	"syscall"
+	"time"
+)
+
+const Version = 1
+
+var safeValue = regexp.MustCompile(`^[A-Za-z0-9._:/@+-]{1,256}$`)
+var hexDigest = regexp.MustCompile(`^[0-9a-f]{64}$`)
+var gitCommit = regexp.MustCompile(`^[0-9a-f]{40}$`)
+var operationID = regexp.MustCompile(`^[0-9a-f]{32}$`)
+
+type Event struct {
+	Version        int    `json:"version"`
+	OperationID    string `json:"operation_id"`
+	Service        string `json:"service"`
+	ArtifactDigest string `json:"artifact_digest"`
+	Commit         string `json:"commit"`
+	ReleaseVersion string `json:"release_version"`
+	Phase          string `json:"phase"`
+	Slot           string `json:"slot,omitempty"`
+	DurationMillis int64  `json:"duration_ms"`
+	Outcome        string `json:"outcome"`
+	ObservedAt     string `json:"observed_at"`
+}
+
+func OperationID() (string, error) {
+	b := make([]byte, 16)
+	if _, err := rand.Read(b); err != nil {
+		return "", errors.New("cryptographic randomness unavailable")
+	}
+	return hex.EncodeToString(b), nil
+}
+
+func Append(path string, event Event) error {
+	if err := event.validate(); err != nil {
+		return err
+	}
+	if !filepath.IsAbs(path) || filepath.Clean(path) != path {
+		return errors.New("event log path must be absolute and clean")
+	}
+	if err := os.MkdirAll(filepath.Dir(path), 0o750); err != nil {
+		return fmt.Errorf("create event directory: %w", err)
+	}
+	if info, err := os.Lstat(path); err == nil {
+		if !info.Mode().IsRegular() || info.Mode()&os.ModeSymlink != 0 || info.Mode().Perm()&0o022 != 0 {
+			return errors.New("event log must be a non-writable regular non-symlink file")
+		}
+	} else if !errors.Is(err, os.ErrNotExist) {
+		return fmt.Errorf("inspect event log: %w", err)
+	}
+	b, err := json.Marshal(event)
+	if err != nil {
+		return fmt.Errorf("encode deployment event: %w", err)
+	}
+	if len(b) > 4096 {
+		return errors.New("deployment event exceeds bound")
+	}
+	b = append(b, '\n')
+	fd, err := syscall.Open(path, syscall.O_WRONLY|syscall.O_APPEND|syscall.O_CREAT|syscall.O_CLOEXEC|syscall.O_NOFOLLOW, 0o640)
+	if err != nil {
+		return fmt.Errorf("open event log: %w", err)
+	}
+	file := os.NewFile(uintptr(fd), path)
+	if file == nil {
+		_ = syscall.Close(fd)
+		return errors.New("open event log file")
+	}
+	defer file.Close()
+	n, err := file.Write(b)
+	if err != nil || n != len(b) {
+		return errors.New("write complete deployment event")
+	}
+	if err := file.Sync(); err != nil {
+		return fmt.Errorf("sync deployment event: %w", err)
+	}
+	return nil
+}
+
+func (e Event) validate() error {
+	if e.Version != Version || !operationID.MatchString(e.OperationID) {
+		return errors.New("deployment event identity is invalid")
+	}
+	if !hexDigest.MatchString(e.ArtifactDigest) || !gitCommit.MatchString(e.Commit) {
+		return errors.New("deployment event provenance is invalid")
+	}
+	for label, value := range map[string]string{"service": e.Service, "artifact_digest": e.ArtifactDigest, "commit": e.Commit, "release_version": e.ReleaseVersion, "phase": e.Phase, "outcome": e.Outcome} {
+		if !safeValue.MatchString(value) || strings.ContainsRune(value, '\x00') {
+			return fmt.Errorf("deployment event %s is invalid", label)
+		}
+	}
+	if e.Slot != "" && !safeValue.MatchString(e.Slot) {
+		return errors.New("deployment event slot is invalid")
+	}
+	if e.DurationMillis < 0 {
+		return errors.New("deployment event duration is invalid")
+	}
+	if _, err := time.Parse(time.RFC3339Nano, e.ObservedAt); err != nil {
+		return errors.New("deployment event timestamp is invalid")
+	}
+	return nil
+}
@@ -0,0 +1,59 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+
+package eventlog
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+)
+
+func TestAppendBoundedEvent(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "events.jsonl")
+	id, err := OperationID()
+	if err != nil {
+		t.Fatal(err)
+	}
+	event := Event{Version: 1, OperationID: id, Service: "site", ArtifactDigest: strings.Repeat("a", 64), Commit: strings.Repeat("b", 40), ReleaseVersion: "v0.2.0-preview.1", Phase: "activation", Slot: "green", Outcome: "succeeded", ObservedAt: time.Now().UTC().Format(time.RFC3339Nano)}
+	if err := Append(path, event); err != nil {
+		t.Fatal(err)
+	}
+	b, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var decoded Event
+	if err := json.Unmarshal(b, &decoded); err != nil {
+		t.Fatal(err)
+	}
+	if decoded.OperationID != id || strings.Contains(string(b), "secret") {
+		t.Fatalf("unexpected event: %s", b)
+	}
+	if info, _ := os.Stat(path); info.Mode().Perm() != 0o640 {
+		t.Fatalf("mode=%04o", info.Mode().Perm())
+	}
+}
+
+func TestAppendRejectsUnboundedValuesAndSymlink(t *testing.T) {
+	id, _ := OperationID()
+	event := Event{Version: 1, OperationID: id, Service: "site\nsecret", ArtifactDigest: "digest", Commit: "commit", ReleaseVersion: "version", Phase: "activation", Outcome: "failed", ObservedAt: time.Now().UTC().Format(time.RFC3339Nano)}
+	if err := Append(filepath.Join(t.TempDir(), "events.jsonl"), event); err == nil {
+		t.Fatal("expected unsafe value rejection")
+	}
+	dir := t.TempDir()
+	target := filepath.Join(dir, "target")
+	if err := os.WriteFile(target, nil, 0o640); err != nil {
+		t.Fatal(err)
+	}
+	link := filepath.Join(dir, "events.jsonl")
+	if err := os.Symlink(target, link); err != nil {
+		t.Skip(err)
+	}
+	event.Service = "site"
+	if err := Append(link, event); err == nil {
+		t.Fatal("expected symlink rejection")
+	}
+}
@@ -17,13 +17,18 @@ import (
 const SchemaVersion = 1

 type Record struct {
-	SchemaVersion   int    `json:"schema_version"`
-	Strategy        string `json:"strategy"`
-	ActiveSlot      string `json:"active_slot"`
-	ActiveRelease   string `json:"active_release"`
-	PreviousSlot    string `json:"previous_slot,omitempty"`
-	PreviousRelease string `json:"previous_release,omitempty"`
-	UpdatedAt       string `json:"updated_at"`
+	SchemaVersion      int    `json:"schema_version"`
+	Strategy           string `json:"strategy"`
+	DesiredRelease     string `json:"desired_release,omitempty"`
+	CandidateRelease   string `json:"candidate_release,omitempty"`
+	ActiveSlot         string `json:"active_slot"`
+	ActiveRelease      string `json:"active_release"`
+	PreviousSlot       string `json:"previous_slot,omitempty"`
+	PreviousRelease    string `json:"previous_release,omitempty"`
+	LastAttemptRelease string `json:"last_attempt_release,omitempty"`
+	LastAttemptOutcome string `json:"last_attempt_outcome,omitempty"`
+	LastAttemptAt      string `json:"last_attempt_at,omitempty"`
+	UpdatedAt          string `json:"updated_at"`
 }

 func Load(path, root, strategy string) (Record, error) {
@@ -68,6 +73,26 @@ func (r Record) Validate(root, strategy string) error {
 			return fmt.Errorf("previous release: %w", err)
 		}
 	}
+	for label, release := range map[string]string{"desired release": r.DesiredRelease, "candidate release": r.CandidateRelease, "last attempt release": r.LastAttemptRelease} {
+		if release != "" {
+			if err := releaseBelow(root, release); err != nil {
+				return fmt.Errorf("%s: %w", label, err)
+			}
+		}
+	}
+	if r.LastAttemptOutcome != "" {
+		switch r.LastAttemptOutcome {
+		case "running", "succeeded", "failed", "rolled_back":
+		default:
+			return errors.New("last attempt outcome is invalid")
+		}
+		if _, err := time.Parse(time.RFC3339, r.LastAttemptAt); err != nil {
+			return errors.New("last attempt timestamp is invalid")
+		}
+	}
+	if r.LastAttemptOutcome == "running" && r.CandidateRelease == "" {
+		return errors.New("running attempt requires a candidate release")
+	}
 	if strategy == "blue_green" && r.PreviousRelease != "" && r.PreviousSlot == r.ActiveSlot {
 		return errors.New("previous slot must differ from active slot")
 	}
@@ -16,7 +16,8 @@ func TestStoreLoadRoundTripAndRejectSymlink(t *testing.T) {
 		t.Fatal(err)
 	}
 	path := filepath.Join(root, "state.json")
-	record := Record{SchemaVersion: 1, Strategy: "singleton_candidate", ActiveSlot: "singleton", ActiveRelease: release, UpdatedAt: time.Unix(1, 0).UTC().Format(time.RFC3339)}
+	at := time.Unix(1, 0).UTC().Format(time.RFC3339)
+	record := Record{SchemaVersion: 1, Strategy: "singleton_candidate", DesiredRelease: release, ActiveSlot: "singleton", ActiveRelease: release, LastAttemptRelease: release, LastAttemptOutcome: "succeeded", LastAttemptAt: at, UpdatedAt: at}
 	if err := Store(path, root, record); err != nil {
 		t.Fatal(err)
 	}
@@ -24,8 +25,8 @@ func TestStoreLoadRoundTripAndRejectSymlink(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	if loaded.ActiveRelease != release {
-		t.Fatalf("release=%q", loaded.ActiveRelease)
+	if loaded.ActiveRelease != release || loaded.DesiredRelease != release || loaded.LastAttemptOutcome != "succeeded" {
+		t.Fatalf("state=%+v", loaded)
 	}
 	if err := os.Remove(path); err != nil {
 		t.Fatal(err)
@@ -37,6 +38,15 @@ func TestStoreLoadRoundTripAndRejectSymlink(t *testing.T) {
 		t.Fatal("expected symlink refusal")
 	}
 }
+
+func TestRecordRequiresCandidateForRunningAttempt(t *testing.T) {
+	root := filepath.Join(t.TempDir(), "service")
+	release := filepath.Join(root, "releases", "sha256-aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa")
+	record := Record{SchemaVersion: 1, Strategy: "singleton_candidate", DesiredRelease: release, ActiveSlot: "singleton", ActiveRelease: release, LastAttemptRelease: release, LastAttemptOutcome: "running", LastAttemptAt: time.Unix(1, 0).UTC().Format(time.RFC3339), UpdatedAt: time.Unix(1, 0).UTC().Format(time.RFC3339)}
+	if err := record.Validate(root, "singleton_candidate"); err == nil {
+		t.Fatal("expected missing candidate rejection")
+	}
+}
 func TestRecordRejectsReleaseOutsideRoot(t *testing.T) {
 	record := Record{SchemaVersion: 1, Strategy: "singleton_candidate", ActiveSlot: "singleton", ActiveRelease: "/tmp/other/release", UpdatedAt: time.Unix(1, 0).UTC().Format(time.RFC3339)}
 	if err := record.Validate("/opt/example", "singleton_candidate"); err == nil {
@@ -14,9 +14,11 @@
    "root": "/opt/tend-release-test",
    "lock_file": "/run/lock/tend-deploy.lock",
    "state_file": "/opt/tend-release-test/tend-state.json",
+    "event_log": "/opt/tend-release-test/deployment-events.jsonl",
    "health_path": "/healthz",
    "readiness_path": "/readyz",
    "candidate_timeout_seconds": 5,
+    "activation_window_seconds": 5,
    "smoke": [{ "path": "/", "contains": "Tend" }],
    "public_smoke": [{ "url": "https://example.test/", "contains": "Tend" }],
    "singleton": {
@@ -25,7 +27,10 @@
      "candidate_address": "127.0.0.1:19091",
      "listen_env": "TEND_RELEASE_TEST_LISTEN",
      "current_link": "/opt/tend-release-test/current",
-      "previous_link": "/opt/tend-release-test/previous"
+      "previous_link": "/opt/tend-release-test/previous",
+      "caddy_config": "/etc/caddy/Caddyfile",
+      "caddy_handler": "/etc/caddy/tend-release-test-handler.caddy",
+      "caddy_handler_template": "/etc/tend/caddy/tend-release-test.template"
    }
  }
 }
@@ -0,0 +1,30 @@
+#!/usr/bin/env bash
+# SPDX-License-Identifier: AGPL-3.0-only
+set -euo pipefail
+root=$(CDPATH= cd -- "$(dirname -- "$0")/.." && pwd)
+allow=$root/scripts/public-snapshot.allow
+LC_ALL=C sort -c "$allow"
+[[ $(LC_ALL=C sort "$allow" | uniq -d | wc -l) -eq 0 ]]
+mapfile -t files <"$allow"
+[[ ${#files[@]} -gt 0 ]]
+for file in "${files[@]}"; do
+	[[ -n $file && $file != /* && $file != *..* && $file != .gitea/* && $file != .github/* ]]
+	git -C "$root" cat-file -e "HEAD:$file"
+done
+work=$(mktemp -d)
+trap 'rm -rf -- "$work"' EXIT
+mkdir -m 0700 "$work/tree"
+git -C "$root" archive HEAD -- "${files[@]}" | tar -xf - -C "$work/tree"
+test ! -e "$work/tree/.git"
+test ! -e "$work/tree/.gitea"
+test ! -e "$work/tree/.github"
+private_pattern='/home/[[:alnum:]_.-]+/|BEGIN (RSA|OPENSSH|EC) PRIVATE KEY|gitea[-_]api[[:alnum:]_.-]*token'
+if (cd "$work/tree" && rg -n --hidden --glob '!scripts/export-public.sh' --glob '!scripts/check-public-tree.sh' "$private_pattern" .); then
+	echo "private material found in public tree" >&2
+	exit 1
+fi
+(cd "$work/tree" && ./scripts/check-licenses.sh)
+(cd "$work/tree" && GOWORK=off go test -count=1 ./...)
+(cd "$work/tree" && GOWORK=off go vet ./...)
+(cd "$work/tree" && GOWORK=off CGO_ENABLED=0 go build -buildvcs=false -mod=readonly -trimpath -o "$work/tend" ./cmd/tend)
+echo "public tree compiles independently"
@@ -44,7 +44,8 @@ done < <(find "$stage" -type f -print | LC_ALL=C sort)
 epoch=$(git -C "$root" show -s --format=%ct "$commit")
 printf '{"schema_version":1,"source_commit":"%s","source_tree":"%s","source_date_epoch":%s,"file_count":%d}\n' "$commit" "$tree" "$epoch" "${#files[@]}" >"$stage/PUBLIC-SNAPSHOT.json"
 (cd "$stage" && sha256sum PUBLIC-SNAPSHOT.json >PUBLIC-SNAPSHOT.sha256)
-if (cd "$stage" && rg -n --hidden --glob '!.git/**' --glob '!PUBLIC-SNAPSHOT.json' --glob '!scripts/export-public.sh' '/home/cole|BEGIN (RSA|OPENSSH|EC) PRIVATE KEY|gitea-api\.token' .); then
+private_pattern='/home/[[:alnum:]_.-]+/|BEGIN (RSA|OPENSSH|EC) PRIVATE KEY|gitea[-_]api[[:alnum:]_.-]*token'
+if (cd "$stage" && rg -n --hidden --glob '!.git/**' --glob '!PUBLIC-SNAPSHOT.json' --glob '!scripts/export-public.sh' "$private_pattern" .); then
 	echo "private material found" >&2
 	exit 1
 fi
@@ -8,6 +8,7 @@ SECURITY.md
 cmd/tend/main.go
 docs/ARCHITECTURE.md
 docs/DOGFOOD_EVIDENCE.md
+docs/DOGFOOD_FRICTION.md
 docs/PUBLIC_SNAPSHOT.md
 docs/SCHEMA_V2_MIGRATION.md
 docs/THREAT_MODEL.md
@@ -19,6 +20,8 @@ examples/blue-green/example-site@.service
 examples/blue-green/tend.json
 examples/local/.env.example
 examples/server/authorized_keys.example
+examples/server/caddy/docs-site.template
+examples/server/caddy/example-site.template
 examples/server/environment/docs-site.env.example
 examples/server/environment/example-site.env.example
 examples/server/example-singleton.service
@@ -28,6 +31,7 @@ examples/server/services/example-site.json
 examples/server/slots/example-site-blue.env
 examples/server/slots/example-site-green.env
 examples/server/tend-receive.sudoers
+examples/singleton/caddy-handler.template
 examples/singleton/tend.json
 go.mod
 internal/config/config.go
@@ -42,7 +46,10 @@ internal/deploy/operator_test.go
 internal/deploy/ownership_linux.go
 internal/deploy/ownership_other.go
 internal/deploy/release.go
+internal/deploy/release_mode_linux_test.go
 internal/deploy/release_test.go
+internal/eventlog/eventlog.go
+internal/eventlog/eventlog_test.go
 internal/packager/packager.go
 internal/packager/packager_test.go
 internal/process/run.go
@@ -62,6 +69,7 @@ internal/version/version.go
 internal/version/version_test.go
 release/tend.json
 scripts/check-licenses.sh
+scripts/check-public-tree.sh
 scripts/export-public.sh
 scripts/public-snapshot.allow
 scripts/test-public-snapshot.sh
@@ -4,6 +4,7 @@ set -euo pipefail
 root=$(CDPATH= cd -- "$(dirname -- "$0")/.." && pwd)
 cd "$root"
 ./scripts/check-licenses.sh
+./scripts/check-public-tree.sh
 go test -count=1 ./...
 go test -race -count=1 ./...
 go vet ./...