feat(scrape): recover JSON summary from host compose run log

Reuse shared recover helper before deleting the temp compose log when
DCE_RUN_SUMMARY_FILE is missing after a successful host scrape.

docs/plans/2026-06-04-072-feat-host-runner-summary-recovery-plan.md

@@ -0,0 +1,64 @@
+---
+title: "feat: Host runner recover JSON summary from compose log"
+type: feat
+status: complete
+date: 2026-06-04
+origin: /lfg — plan 071 deferred host-runner recovery when stdout is not teed to a persistent file
+---
+
+# feat: Host runner recover JSON summary from compose log
+
+## Summary
+
+`run-discord-scrape-host.sh` already tees compose output to a temporary run log. Before deleting that log on success, recover `DCE_RUN_SUMMARY_FILE` from the last `DCE_JSON_SUMMARY:` line when the file is missing or empty.
+
+## Problem Frame
+
+Operator validation (plan 071) recovers summaries from its teed log. Direct host scrapes (`run-discord-scrape-host.sh scrape`) capture compose stdout in a temp file but discard it after success. When the container logs `DCE_JSON_SUMMARY` but cannot write the mapped file, operators lose machine-readable totals unless they manually grep the scrollback.
+
+## Requirements
+
+| ID | Requirement |
+|----|-------------|
+| R1 | Shared `recover_json_summary_if_missing(run_log, dest_file)` skips when dest exists and is non-empty |
+| R2 | `run_subcommand_with_retry` calls recovery on successful scrape/preflight runs before deleting the temp log |
+| R3 | Recovery runs only when `DCE_RUN_SUMMARY_FILE` is set and file is missing or zero-length |
+| R4 | Success prints `JSON summary recovered from run log:` to stderr |
+| R5 | `run-operator-validation.sh` uses the shared helper instead of inline extract |
+| R6 | Host smoke covers recovery from a synthetic run log |
+| R7 | `DCE_MIN_FREE_MB=0 ./scripts/run-all-smokes.sh` → 22/22 |
+
+## Implementation Units
+
+### U1. Shared recovery helper
+
+**Files:** `scripts/lib/scrape-summary-json.sh`, `scripts/tests/scrape-summary-json-smoke.sh`
+
+**Approach:** Add `recover_json_summary_if_missing`; extend smoke with dest-already-exists skip case.
+
+### U2. Host runner wiring
+
+**Files:** `scripts/run-discord-scrape-host.sh`, `scripts/tests/run-discord-scrape-host-smoke.sh`
+
+**Approach:** Source lib in host runner; call recovery before `rm -f "$output_file"` on both success paths in `run_subcommand_with_retry`.
+
+### U3. Validation refactor + docs
+
+**Files:** `scripts/run-operator-validation.sh`, `docs/recurring-scrape-merge-readiness.md`
+
+**Approach:** Replace inline extract block with shared helper; add Plan 072 stamp.
+
+## Verification
+
+```bash
+DCE_MIN_FREE_MB=0 ./scripts/run-all-smokes.sh
+```
+
+## Scope Boundaries
+
+### Deferred
+
+- Live KotOR catch-up on host
+- Auto-enable JSON summary on bare `host.sh scrape` without env vars
+- Operator-proof JSON summary parity
+- Merging multiple per-target summaries into one JSON artifact

docs/recurring-scrape-merge-readiness.md

@@ -172,6 +172,8 @@ DCE_MIN_FREE_MB=0 ./scripts/run-operator-validation.sh \
 
 **Plan 071 (2026-06-04):** When summary file write fails, operator validation recovers JSON from the last `DCE_JSON_SUMMARY:` line in the teed log.
 
+**Plan 072 (2026-06-04):** Host runner recovers JSON summary from the captured compose run log before deleting the temp file.
+
 **Disk:** ~65 GiB free on `/home` (2026-05-30); large channel merges still need headroom.
 
 ## CI note (fork PRs)

scripts/lib/scrape-summary-json.sh

@@ -24,3 +24,12 @@ extract_json_summary_from_log() {
   mkdir -p "$(dirname "$dest_file")"
   jq . <<<"$json_payload" >"$dest_file"
 }
+
+recover_json_summary_if_missing() {
+  local run_log=$1
+  local dest_file=$2
+
+  [[ -n "$run_log" && -n "$dest_file" ]] || return 1
+  [[ -s "$dest_file" ]] && return 1
+  extract_json_summary_from_log "$run_log" "$dest_file"
+}

scripts/run-discord-scrape-host.sh

@@ -511,6 +511,18 @@ try_interactive_reauth() {
   "$reauth_script"
 }
 
+recover_scrape_summary_from_run_log() {
+  local output_file=$1
+  local dest_file=${DCE_RUN_SUMMARY_FILE:-}
+
+  [[ -n "$dest_file" ]] || return 0
+  # shellcheck source=lib/scrape-summary-json.sh
+  source "$SCRIPT_DIR/lib/scrape-summary-json.sh"
+  if recover_json_summary_if_missing "$output_file" "$dest_file"; then
+    printf 'JSON summary recovered from run log: %s\n' "$dest_file" >&2
+  fi
+}
+
 run_subcommand_with_retry() {
   local subcommand=$1
   shift
@@ -522,6 +534,7 @@ run_subcommand_with_retry() {
 
   compose_run_args run_args "$subcommand" "$@"
   if "${run_args[@]}" 2>&1 | tee "$output_file"; then
+    recover_scrape_summary_from_run_log "$output_file"
     rm -f "$output_file"
     return 0
   fi
@@ -545,6 +558,7 @@ run_subcommand_with_retry() {
   compose_run_args run_args "$subcommand" "$@"
 
   if "${run_args[@]}" 2>&1 | tee "$output_file"; then
+    recover_scrape_summary_from_run_log "$output_file"
     rm -f "$output_file"
     return 0
   fi

scripts/run-operator-validation.sh

@@ -327,12 +327,10 @@ main() {
   local pipeline_status=${PIPESTATUS[0]}
 
   if (( export_json_summary )) && [[ -n "${DCE_RUN_SUMMARY_FILE:-}" ]]; then
-    if [[ ! -s "${DCE_RUN_SUMMARY_FILE}" ]]; then
+    # shellcheck source=lib/scrape-summary-json.sh
-      # shellcheck source=lib/scrape-summary-json.sh
+    source "$SCRIPT_DIR/lib/scrape-summary-json.sh"
-      source "$SCRIPT_DIR/lib/scrape-summary-json.sh"
+    if recover_json_summary_if_missing "$LOG_FILE" "$DCE_RUN_SUMMARY_FILE"; then
-      if extract_json_summary_from_log "$LOG_FILE" "$DCE_RUN_SUMMARY_FILE"; then
+      printf 'JSON summary recovered from log: %s\n' "$DCE_RUN_SUMMARY_FILE"
-        printf 'JSON summary recovered from log: %s\n' "$DCE_RUN_SUMMARY_FILE"
-      fi
     fi
   fi
 

scripts/tests/run-discord-scrape-host-smoke.sh

@@ -83,6 +83,12 @@ if [[ "$mode" == "streaming" ]]; then
   exit 0
 fi
 
+if [[ "$mode" == "json-summary-log-only" ]]; then
+  printf '[2026-06-04T12:00:00Z] DCE_JSON_SUMMARY: {"version":1,"totals":{"merged":42,"unchanged":0,"created":0,"skipped":0,"skipped_oom":0,"messages_appended":7}}\n' >&2
+  printf 'run succeeded\n'
+  exit 0
+fi
+
 printf 'run succeeded\n'
 EOF
 chmod +x "$FAKE_DOCKER"
@@ -331,4 +337,34 @@ grep -q 'env:DCE_RUN_SUMMARY_FILE=/logs/host-smoke-summary.json' "$COMPOSE_SUMMA
   exit 1
 }
 
+HOST_RECOVER_SUMMARY="$TMP_DIR/host-run-recovered.summary.json"
+HOST_RECOVER_STDERR="$TMP_DIR/host-recover-stderr.txt"
+rm -f "$HOST_RECOVER_SUMMARY"
+printf '0' >"$CALL_COUNT"
+env -u DISCORD_TOKEN \
+  DCE_SKIP_SCRAPE_LOCK=1 \
+  DCE_REPO_ROOT="$REPO_ROOT" \
+  DCE_DOCKER_BIN="$FAKE_DOCKER" \
+  DCE_ENV_FILE="$ENV_FILE" \
+  DCE_COMPOSE_FILE="$COMPOSE_FILE" \
+  DCE_RUN_SUMMARY_FILE="$HOST_RECOVER_SUMMARY" \
+  FAKE_DOCKER_CALL_COUNT="$CALL_COUNT" \
+  FAKE_DOCKER_TOKEN_FILE="$TOKEN_FILE" \
+  FAKE_DOCKER_MODE=json-summary-log-only \
+  "$REPO_ROOT/scripts/run-discord-scrape-host.sh" scrape --target demo \
+  >/dev/null 2>"$HOST_RECOVER_STDERR"
+[[ -s "$HOST_RECOVER_SUMMARY" ]] || {
+  echo "expected host runner to recover summary from compose run log" >&2
+  exit 1
+}
+jq -e '.totals.merged == 42 and .totals.messages_appended == 7' "$HOST_RECOVER_SUMMARY" >/dev/null || {
+  echo "recovered host summary JSON content mismatch" >&2
+  exit 1
+}
+grep -q 'JSON summary recovered from run log:' "$HOST_RECOVER_STDERR" || {
+  echo "expected recovery notice on stderr" >&2
+  cat "$HOST_RECOVER_STDERR" >&2
+  exit 1
+}
+
 echo "run-discord-scrape-host smoke test passed"

scripts/tests/scrape-summary-json-smoke.sh

@@ -43,6 +43,23 @@ jq -e '.totals.merged == 9' "$OUT_FILE" >/dev/null || {
   exit 1
 }
 
+EXISTING="$TMP_DIR/existing.summary.json"
+printf '{"version":1,"totals":{"merged":1}}\n' >"$EXISTING"
+if recover_json_summary_if_missing "$LOG_FILE" "$EXISTING" 2>/dev/null; then
+  printf 'ERROR: recover should skip when dest already non-empty\n' >&2
+  exit 1
+fi
+
+RECOVER_OUT="$TMP_DIR/recover-via-helper.summary.json"
+recover_json_summary_if_missing "$LOG_FILE" "$RECOVER_OUT" || {
+  printf 'ERROR: recover_json_summary_if_missing failed\n' >&2
+  exit 1
+}
+jq -e '.totals.merged == 9' "$RECOVER_OUT" >/dev/null || {
+  printf 'ERROR: recover helper wrote wrong content\n' >&2
+  exit 1
+}
+
 if extract_json_summary_from_log "$TMP_DIR/missing.log" "$OUT_FILE" 2>/dev/null; then
   printf 'ERROR: extract should fail on missing log\n' >&2
   exit 1