eval: guard authenticated live-daemon registry recovery

CATALOG.md

@@ -1,7 +1,7 @@
 <!-- Generated by bin/generate-catalog.sh from cell KDL and evidence/*.tsv. Do not edit. -->
 # Current eval catalog and evidence
 
-This is the sole current corpus surface: **58 included cells** (18 model-free,
+This is the sole current corpus surface: **59 included cells** (19 model-free,
 40 model-backed) and **4 retired exclusions**. Inventory fields are derived from the active
 KDL and executable judge scripts. Latest accepted PASS evidence remains distinct from the append-only run
 history: a failed last run is visible without being advertised as accepted evidence, and cells with no
@@ -50,6 +50,7 @@ harness-native loader plus canonical hook file.
 | `presence-ding-matrix` | model-free | — | 0 | none | `60s` | 6 | — | **NO STRUCTURED RUN** |
 | `pty-attach-machine-stream` | model-free | — | 0 | none | `90s` | 5 | — | **NO STRUCTURED RUN** |
 | `pty-attach-only` | model-free | — | 0 | none | `90s` | 7 | — | **NO STRUCTURED RUN** |
+| `pty-live-recovery` | model-free | — | 0 | none | `120s` | 10 | — | **NO STRUCTURED RUN** |
 | `pty-send-peek` | model-free | — | 0 | none | `120s` | 5 | **PASS** 2026-07-28T19:53:59Z, 13/13, source [`b7da882`](https://github.com/compoundingtech/evals/commit/b7da8826b503e0408e309af8702fb684d28e3a19), [receipt](evidence/stable-main-phase4-runs-20260728.json) | **PASS** 2026-07-28T19:53:59Z, 13/13, source [`b7da882`](https://github.com/compoundingtech/evals/commit/b7da8826b503e0408e309af8702fb684d28e3a19), [receipt](evidence/stable-main-phase4-runs-20260728.json) |
 | `reconcile-retire-keep` | model-free | — | 0 | none | `90s` | 6 | — | **NO STRUCTURED RUN** |
 | `render-target-safety` | model-free | — | 0 | none | `90s` | 5 | — | **NO STRUCTURED RUN** |
@@ -103,6 +104,7 @@ while the last-run column makes a recorded failure distinct from a cell with no
 - `host-lock-health-negatives` / `cell` — deterministic native doctor negative probe; no harness seat
 - `pty-attach-machine-stream` / `cell` — deterministic installed PTY attach composition probe; no harness seat
 - `pty-attach-only` / `cell` — deterministic PTY dead-attach policy probe; no harness seat
+- `pty-live-recovery` / `cell` — deterministic PTY live-registry recovery probe; no harness seat
 - `pty-send-peek` / `cell` — deterministic PTY transport probe; no harness seat
 - `presence-ding-matrix` / `cell` — deterministic native presence and DING probe; no harness seat
 - `reconcile-retire-keep` / `cell` — deterministic native reconcile lifecycle probe; no harness seat
@@ -114,7 +116,7 @@ while the last-run column makes a recorded failure distinct from a cell with no
 - `targeted-reconcile-isolation` / `cell` — deterministic selected reconcile isolation probe; no bus-connected agent
 - `two-networks-coexist` / `cell` — deterministic catalog-isolation probe; no harness seat
 
-These **19** rows are generated from structured exclusions. A cell-level row is required for
+These **20** rows are generated from structured exclusions. A cell-level row is required for
 every derived model-free cell. The one-shot docs cold reader is a model-backed grader, but not a bus-connected
 agent: it has no st2 identity, DING sidecar, or lifecycle hooks.
 

cells/pty-live-recovery/README.md

@@ -0,0 +1,44 @@
+# pty-live-recovery
+
+Held-out, model-free black-box coverage for the selected live-daemon recovery
+protocol from [PTY PR #128](https://github.com/compoundingtech/pty/pull/128),
+squash-merged at exact source
+`9eb958c5aae026d5c05690ab72b528662c55708d`.
+
+Set `EVALS_PTY_PR128_ROOT` to a clean checkout of that merged source after
+`npm ci && npm run build`. The cell rejects any other Git head, package lock,
+or built CLI/server artifacts. Its synthetic PTY roots and providers live only
+inside the temporary eval catalog; it launches no model or provider harness.
+
+## What it proves
+
+- `pty recover <name> --snapshot <metadata.json>` is the only accepted surface.
+- A private root advertises the complete protocol, secret, process-start,
+  launch-identity, root/recovery-directory identity, and metadata-revision
+  capability.
+- Recovery preserves the daemon PID, generation, process-start token, launch
+  identity, existing client, and provider while rotating the secret and signed
+  metadata revision.
+- A fresh client reaches the republished pathname, and every synthetic provider
+  still has exactly one launch and one `session_start`.
+- A rotated snapshot cannot replay; a pre-tag snapshot cannot roll back current
+  metadata; and a secret-tampered snapshot cannot publish a registry. Each
+  refusal leaves the original daemon alive and a current valid snapshot usable.
+- Group/world-accessible root or `.recovery` modes fail before request/result,
+  lock, socket, PID, or metadata publication.
+- A daemon started without a recovery capability is refused without signal,
+  restart, relaunch, or a recoverability claim.
+- Cleanup binds every captured PID to its process-start token, uses bounded
+  TERM then KILL only for those exact identities, fails if any survives, and
+  exercises the KILL fallback with a TERM-resistant synthetic sentinel.
+
+The experiment is exact-source merged evidence, not a published PTY release artifact.
+It deliberately replaces the incompatible PR #127 oracle rather than merely
+retargeting its metadata.
+
+## Run it
+
+```sh
+EVALS_PTY_PR128_ROOT=/path/to/clean/pty-pr128 \
+  st2 eval ./cells/pty-live-recovery/
+```

cells/pty-live-recovery/fixture/live.sh

@@ -0,0 +1,9 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+marker="${1:?marker path required}"
+printf 'run\n' >>"$marker"
+printf 'LIVE-RECOVERY-READY\n'
+while IFS= read -r line; do
+  printf 'LIVE-RECOVERY-ACK:%s\n' "$line"
+done