diff --git a/Plugin/unraid/Tools/ai_chat_worker.php b/Plugin/unraid/Tools/ai_chat_worker.php
index 2fff51b..a153e56 100644
--- a/Plugin/unraid/Tools/ai_chat_worker.php
+++ b/Plugin/unraid/Tools/ai_chat_worker.php
@@ -129,6 +129,9 @@ if ($explain) {
 }
 
 // What got attached and why, recorded as it happens rather than reconstructed afterwards.
+// Keyed by a stable token with the varying detail — counts, line totals — kept in the value, so a
+// fixture can assert that the log tail was attached without asserting how many lines it had that
+// day. An assertion that breaks whenever a log grows is an assertion nobody keeps.
 $attached = [];
 
 $profile = in_array($profile, ['varaverk', 'chat', 'code', 'troubleshoot'], true) ? $profile : 'varaverk';
@@ -279,7 +282,7 @@ if ($diagnostic) {
     // asked what was wrong, it reported "AI_ENABLED=false" read from the conf *template*
     // while the live value was true. Documentation records defaults; only this block records
     // what is actually set.
-    $attached[] = 'health (' . count(vv_ai_health()) . ' checks)';
+    $attached['health'] = count(vv_ai_health()) . ' checks';
     $diagBlock .= "LIVE SYSTEM STATE (authoritative — measured just now)\n";
     foreach (vv_ai_health() as $c) {
         $mark = ['ok' => 'OK', 'warn' => 'WARNING', 'bad' => 'PROBLEM'][$c['state']] ?? '?';
@@ -290,7 +293,7 @@ if ($diagnostic) {
 
     $logs = vv_ai_recent_logs(40);
     if ($logs) {
-        $attached[] = 'recent warnings (' . count($logs) . ' lines)';
+        $attached['warnings'] = count($logs) . ' lines';
         $diagBlock .= "RECENT WARNINGS AND ERRORS (newest last)\n" . implode("\n", $logs) . "\n\n";
     }
 }
@@ -312,8 +315,7 @@ if ($runTarget !== '' && vv_ai_scope_ok($runTarget)) {
     // next line, and the difference is the whole answer.
     $rec = vv_ai_run_record($runTarget);
     if ($rec['ok']) {
-        $attached[] = 'run record (' . $rec['status']
-                    . ', exit ' . var_export($rec['exit'], true) . ')';
+        $attached['run_record'] = $rec['status'] . ', exit ' . var_export($rec['exit'], true);
         $diagBlock .= 'RUN RECORD for ' . $runTarget . " (authoritative — how the last run ended)\n"
                     . '- status: ' . $rec['status']
                     . ($rec['exit'] !== null ? ' (exit ' . $rec['exit'] . ')' : '') . "\n"
@@ -328,13 +330,13 @@ if ($runTarget !== '' && vv_ai_scope_ok($runTarget)) {
 
     $scopedLog = vv_ai_scoped_log($runTarget, 120);
     if ($scopedLog['ok']) {
-        $attached[] = 'log tail ' . $scopedLog['path'] . ' (' . count($scopedLog['tail'])
-                    . ' of ' . $scopedLog['total'] . ' lines)';
+        $attached['log_tail'] = $scopedLog['path'] . ' (' . count($scopedLog['tail'])
+                              . ' of ' . $scopedLog['total'] . ' lines)';
         $diagBlock .= 'LOG: ' . $scopedLog['path']
                     . ' (' . $scopedLog['total'] . " lines total, newest last)\n"
                     . implode("\n", $scopedLog['tail']) . "\n\n";
     } else {
-        $attached[] = 'log tail MISSING for ' . $runTarget . ' (' . ($scopedLog['error'] ?? '?') . ')';
+        $attached['log_missing'] = $runTarget . ' — ' . ($scopedLog['error'] ?? '?');
         $diagBlock .= "LOG: none found for " . $runTarget . " — it may never have run.\n\n";
     }
 }
@@ -348,7 +350,7 @@ if ($runTarget !== '' && vv_ai_scope_ok($runTarget)) {
 if ($can('incidents') && $scope !== '') {
     $past = vv_ai_incidents_for($scope, 4);
     if ($past) {
-        $attached[] = 'operator incidents (' . count($past) . ')';
+        $attached['incidents'] = (string)count($past);
         $diagBlock .= "PREVIOUSLY ON THIS, WRITTEN BY THE OPERATOR AFTER IT WAS RESOLVED\n"
                     . "Confirmed outcomes, not guesses. If the current symptom matches one of "
                     . "these, say so and lead with it. If it clearly does not, ignore them "
@@ -372,7 +374,7 @@ if ($can('conf_lookup')) {
         }
     }
     if ($seen) {
-        $attached[] = 'conf keys resolved (' . count($seen) . ')';
+        $attached['conf_keys'] = (string)count($seen);
         $diagBlock .= "WHERE THESE SETTINGS ACTUALLY LIVE (looked up just now, authoritative)\n"
                     . implode("\n", $seen) . "\n"
                     . "If this is a different file from the one they have open, say so plainly.\n\n";
@@ -653,7 +655,7 @@ if ($explain) {
     if (!$attached) {
         echo '  (nothing', $sources ? " — no evidence beyond the passages)\n" : ")\n";
     }
-    foreach ($attached as $a) echo '  ', $a, "\n";
+    foreach ($attached as $k => $detail) printf("  %-12s %s\n", $k, $detail);
     echo "\n";
 
     echo 'RETRIEVED  ', count($sources), " passages\n";
diff --git a/Plugin/unraid/Tools/ai_explain_check.sh b/Plugin/unraid/Tools/ai_explain_check.sh
new file mode 100755
index 0000000..01554a6
--- /dev/null
+++ b/Plugin/unraid/Tools/ai_explain_check.sh
@@ -0,0 +1,158 @@
+#!/bin/bash
+# ═══════════════════════════════════════════════════════════════════════════════════════════════
+# PURPOSE
+#   Runs every case in ai_explain_fixtures.txt through the worker's --explain mode and checks the
+#   routing it reports against what the fixture says it should be. Catches a guard that has been
+#   undone by a later guard, which is the failure this subsystem keeps producing.
+#
+# OPERATIONAL MODEL
+#   Never calls the model. --explain stops on the line where deterministic assembly ends, so a
+#   full pass costs about a second per case and returns the same answer every time. There is no
+#   Ollama dependency, no token spend, and no flaky wording to chase.
+#
+#   Not scheduled and deliberately not in any orchestrator. This is a development check — it runs
+#   when the routing changes, not every night. Nothing on the running system depends on it.
+#
+# RUNTIME MODES
+#   ai_explain_check.sh            check every fixture
+#   ai_explain_check.sh --verbose  print the full explain report for each case
+#   ai_explain_check.sh <pattern>  only cases whose question matches the pattern
+#
+# OPERATIONAL SAFEGUARDS
+#   Asserts routing, never wording.
+#       Which capabilities a profile holds and which evidence was attached are decided before the
+#       model is asked anything. Asserting on generated prose would fail for reasons that tell
+#       nobody anything and the check would be ignored within a fortnight.
+#
+#   A malformed assertion fails loudly rather than passing quietly.
+#       An unrecognised key is an error, not a skip. A typo in an assertion that silently passes
+#       is worse than no assertion, because the line still reads as covered.
+#
+#   Exits non-zero on any failure, so it can gate a commit.
+#
+# DEPENDS ON
+#   Plugin/unraid/Tools/ai_chat_worker.php   --explain mode
+#   Plugin/unraid/Tools/ai_explain_fixtures.txt
+# ═══════════════════════════════════════════════════════════════════════════════════════════════
+
+set -uo pipefail
+
+HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+WORKER="$HERE/ai_chat_worker.php"
+FIXTURES="$HERE/ai_explain_fixtures.txt"
+
+VERBOSE=false
+FILTER=""
+for a in "$@"; do
+    case "$a" in
+        --verbose) VERBOSE=true ;;
+        *)         FILTER="$a" ;;
+    esac
+done
+
+[[ -f "$WORKER"   ]] || { echo "missing worker: $WORKER"; exit 2; }
+[[ -f "$FIXTURES" ]] || { echo "missing fixtures: $FIXTURES"; exit 2; }
+
+PASS=0; FAIL=0; SKIP=0
+FAILED_LINES=()
+
+trim() { local s="$1"; s="${s#"${s%%[![:space:]]*}"}"; s="${s%"${s##*[![:space:]]}"}"; printf '%s' "$s"; }
+
+# One field out of the explain report. Everything it reads is a fixed label printed by --explain.
+field() {
+    local report="$1" label="$2"
+    printf '%s' "$report" | grep -m1 -E "^ *$label " | sed -E "s/^ *$label +//" | sed -E 's/ +$//'
+}
+
+lineno=0
+while IFS= read -r raw || [[ -n "$raw" ]]; do
+    lineno=$((lineno + 1))
+    line="$(trim "$raw")"
+    [[ -z "$line" || "$line" == \#* ]] && continue
+
+    IFS='|' read -r q prof scope kind expect <<< "$line"
+    q="$(trim "$q")"; prof="$(trim "$prof")"; scope="$(trim "$scope")"
+    kind="$(trim "$kind")"; expect="$(trim "$expect")"
+    [[ -z "$q" ]] && continue
+
+    if [[ -n "$FILTER" && "$q" != *"$FILTER"* ]]; then SKIP=$((SKIP + 1)); continue; fi
+
+    report="$(php "$WORKER" --explain "$q" "$prof" "$scope" "$kind" 2>&1)"
+    if [[ $? -ne 0 ]]; then
+        echo "✗ line $lineno: explain failed — $q"
+        printf '%s\n' "$report" | head -3 | sed 's/^/    /'
+        FAIL=$((FAIL + 1)); FAILED_LINES+=("$lineno"); continue
+    fi
+
+    # PROFILE prints "asked -> used (escalated)" on a handoff and just the profile otherwise.
+    got_profile="$(printf '%s' "$report" | grep -m1 '^PROFILE' | sed -E 's/^PROFILE +//')"
+    got_profile="${got_profile##*-> }"
+    got_profile="$(printf '%s' "$got_profile" | sed -E 's/ *\(.*\)$//')"
+    got_caps="$(printf '%s' "$report" | grep -m1 '^CAPS' | sed -E 's/^CAPS +//')"
+    got_target="$(field "$report" 'named target')"
+    got_run="$(field "$report" 'run outcome')"
+    got_diag="$(field "$report" 'diagnostic')"
+    got_attached="$(printf '%s' "$report" | sed -n '/^ATTACHED/,/^$/p' | tail -n +2 \
+                    | awk 'NF {print $1}' | tr '\n' ',' )"
+
+    problems=()
+    for assert in $expect; do
+        key="${assert%%=*}"; want="${assert#*=}"
+        case "$key" in
+            profile)
+                [[ "$got_profile" == "$want" ]] || problems+=("profile: want $want, got $got_profile") ;;
+            target)
+                if [[ "$want" == "none" ]]; then
+                    [[ "$got_target" == "(none resolved)" ]] || problems+=("target: want none, got $got_target")
+                else
+                    [[ "$got_target" == "$want" ]] || problems+=("target: want $want, got $got_target")
+                fi ;;
+            run)
+                exp=$([[ "$want" == "yes" ]] && echo YES || echo no)
+                [[ "$got_run" == "$exp" ]] || problems+=("run outcome: want $exp, got $got_run") ;;
+            diag)
+                exp=$([[ "$want" == "yes" ]] && echo YES || echo no)
+                [[ "$got_diag" == "$exp" ]] || problems+=("diagnostic: want $exp, got $got_diag") ;;
+            caps)
+                if [[ "$want" == "none" ]]; then
+                    [[ "$got_caps" == "(none"* ]] || problems+=("caps: want none, got $got_caps")
+                else
+                    [[ "$got_caps" == *"$want"* ]] || problems+=("caps: want $want in [$got_caps]")
+                fi ;;
+            has)
+                IFS=',' read -ra want_keys <<< "$want"
+                for w in "${want_keys[@]}"; do
+                    [[ ",$got_attached" == *",$w,"* ]] || problems+=("missing attachment: $w")
+                done ;;
+            hasnt)
+                IFS=',' read -ra bad_keys <<< "$want"
+                for b in "${bad_keys[@]}"; do
+                    [[ ",$got_attached" == *",$b,"* ]] && problems+=("attached but must not be: $b")
+                done ;;
+            *)
+                problems+=("unknown assertion '$key' — typo, or a key this checker does not know") ;;
+        esac
+    done
+
+    if [[ ${#problems[@]} -eq 0 ]]; then
+        PASS=$((PASS + 1))
+        printf '\033[32m✓\033[0m %-6s %s\n' "$prof" "$q"
+    else
+        FAIL=$((FAIL + 1)); FAILED_LINES+=("$lineno")
+        printf '\033[31m✗\033[0m %-6s %s\n' "$prof" "$q"
+        for p in "${problems[@]}"; do echo "      $p"; done
+    fi
+
+    [[ "$VERBOSE" == true ]] && printf '%s\n\n' "$report" | sed 's/^/    /'
+done < "$FIXTURES"
+
+echo
+echo "─────────────────────────────────────────────"
+printf 'passed %d   failed %d' "$PASS" "$FAIL"
+[[ $SKIP -gt 0 ]] && printf '   skipped %d' "$SKIP"
+echo
+if [[ $FAIL -gt 0 ]]; then
+    echo "failing fixture lines: ${FAILED_LINES[*]}"
+    exit 1
+fi
+exit 0
diff --git a/Plugin/unraid/Tools/ai_explain_fixtures.txt b/Plugin/unraid/Tools/ai_explain_fixtures.txt
new file mode 100644
index 0000000..d22bd83
--- /dev/null
+++ b/Plugin/unraid/Tools/ai_explain_fixtures.txt
@@ -0,0 +1,66 @@
+# ═══════════════════════════════════════════════════════════════════════════════════════════════
+# Expected routing for the AI assistant, one case per line, checked by ai_explain_check.sh.
+#
+# Every line here is a bug that was found by reading an answer and noticing it was wrong. That is
+# the expensive way to find them and it does not scale: the operator is the test suite, and the
+# operator is busy. Written down, each one costs a second to re-check forever.
+#
+# These assert ROUTING, never wording. What a profile is allowed, which script a question names,
+# which gates fire and what evidence gets attached are all decided before the model is involved,
+# so they are identical every run. The prose is not and is deliberately not asserted — a test that
+# depends on how the model phrases something fails for reasons nobody wants to read about.
+#
+# FORMAT
+#   question | profile | scope | kind | assertions
+#
+#   Blank fields are allowed. Assertions are space-separated:
+#     profile=X      the profile that ends up answering, after any handoff
+#     target=X       resolved run target, or 'none'
+#     run=yes|no     the run-outcome gate
+#     diag=yes|no    the diagnostic gate
+#     caps=none      the profile holds no capabilities at all
+#     has=a,b        every one of these must be attached
+#     hasnt=a,b      none of these may be attached
+#
+#   Attachment keys: health warnings run_record log_tail log_missing incidents conf_keys
+# ═══════════════════════════════════════════════════════════════════════════════════════════════
+
+# ── Run-outcome questions must arrive with the run, not with directions to the log panel ───────
+# The phrasing that started it. "went last" matched; "the run went" did not, one word order apart.
+lets check the daily orch log and see how the run went | varaverk | Scheduler | | target=Orchestrators/daily_sync_maintenance run=yes has=run_record,log_tail
+lets look at daily orch log and see how it went last run | varaverk | Scheduler | | run=yes has=run_record,log_tail
+how did the daily orch go | varaverk | | | target=Orchestrators/daily_sync_maintenance run=yes has=run_record,log_tail
+how did the weekly orch go last run | varaverk | | | target=Orchestrators/weekly_sync_maintenance run=yes has=run_record
+did the watchdog orchestrator run | varaverk | | | target=Orchestrators/watchdog_orchestrator run=yes
+give me a rundown of the daily orch | varaverk | | | run=yes has=run_record
+how long did the daily orch take | varaverk | | | run=yes has=run_record
+
+# Works from the AI tab, which sends no scope at all — resolution is from the question, not the page.
+how did the daily orch go last run | varaverk | | | target=Orchestrators/daily_sync_maintenance has=run_record
+
+# ── Definitional questions must NOT be answered with last night's log ──────────────────────────
+what does the daily orchestrator do | varaverk | | | target=Orchestrators/daily_sync_maintenance run=no hasnt=run_record,log_tail
+how does the daily orch work | varaverk | | | run=no hasnt=run_record,log_tail
+what is the daily orchestrator | varaverk | | | run=no hasnt=run_record
+
+# ── Ambiguity resolves to nothing rather than guessing a script ────────────────────────────────
+# A scored match here attaches the wrong log and answers confidently about a run nobody asked
+# about, which is indistinguishable from a right answer unless you already knew.
+how did sync go | varaverk | | | target=none run=no
+what does RSYNC_ENABLED do | varaverk | | | target=none run=no has=conf_keys
+
+# ── General Chat holds nothing, and hands Varaverk questions up rather than deferring ──────────
+how was your day | chat | | | profile=chat caps=none hasnt=health,log_tail,incidents,conf_keys
+what does arr_sync.sh do | chat | | | profile=varaverk
+is RSYNC_ENABLED on right now | chat | | | profile=varaverk has=conf_keys
+how did the daily orch go | chat | | | profile=varaverk run=yes has=run_record
+
+# Chat opened against a script must not be handed the operator's own incident notes about it.
+how was your day | chat | Orchestrators/daily_sync_maintenance | | profile=chat hasnt=incidents
+
+# ── Troubleshoot gets the log for whatever is open, plus the right to file a bug ───────────────
+why did the weekly orch fail | troubleshoot | Orchestrators/weekly_sync_maintenance | | diag=yes has=log_tail,run_record
+what is going on here | troubleshoot | Orchestrators/daily_sync_maintenance | | diag=yes has=log_tail
+
+# ── The code profile answers from the model alone: no passages, no live state ──────────────────
+write me a script that copies a folder | code | | | hasnt=health,log_tail,incidents,conf_keys
diff --git a/Plugin/unraid/include/ai.php b/Plugin/unraid/include/ai.php
index fb5d801..db045fa 100644
--- a/Plugin/unraid/include/ai.php
+++ b/Plugin/unraid/include/ai.php
@@ -131,7 +131,13 @@ function vv_ai_profile_can(string $profile, string $cap): bool {
 // a class of question that is neither answered nor deferred.
 function vv_ai_chat_needs_varaverk(string $question): bool {
     return vv_ai_mentions_varaverk($question)
-        || (bool)preg_match('/\b[\w.-]+\.sh\b|\b[A-Z][A-Z0-9]*(_[A-Z0-9]+)+\b/', $question);
+        || (bool)preg_match('/\b[\w.-]+\.sh\b|\b[A-Z][A-Z0-9]*(_[A-Z0-9]+)+\b/', $question)
+        // Naming something that has a log is as Varaverk-specific as naming a script file, and
+        // this is how an operator actually refers to them: "the daily orch", not
+        // "daily_sync_maintenance.sh". Without this, "how did the daily orch go" asked in General
+        // Chat was refused correctly and then went nowhere — the refusal is right, but the
+        // question was answerable one profile up and the handoff could not see that.
+        || vv_ai_resolve_run_target($question) !== '';
 }
 
 function vv_ai_config(): array {