[E13–S9 15/15] Add a bounded daemon-first MCP server

.gitattributes

@@ -1,8 +1,8 @@
 # The compatibility bundle is a byte-for-byte oracle, not a text file.
 document/internal/compattest/testdata/document-compat-v1.json -text
-# The evidence golden bundle is a byte-for-byte oracle, not a text file.
+# Document goldens are byte-for-byte protocol, rendering, and vector oracles.
 document/testdata/normalized-evidence-v1.golden.json -text
-# The rendition golden bundle is a byte-for-byte oracle, not a text file.
-document/testdata/rendition-v1.golden.md -text
-# The processing profile golden bundle is a byte-for-byte oracle, not a text file.
 document/testdata/profile-v1.golden.json -text
+document/testdata/rendition-v1.golden.md -text
+document/testdata/source-metadata-v1.golden.json -text
+document/testdata/vector-set-v1.golden.bin -text

.roborev.toml

@@ -1,3 +1,9 @@
+agent = "pi"
+model = "zai/glm-5.3"
+review_agent = "pi"
+review_model = "zai/glm-5.3"
+review_reasoning = "high"
+
 review_guidelines = """
 docbank is a single-user personal document archive: a local daemon owns
 the vault (SQLite + content-addressed blobs) and the CLI, agents, and a
@@ -138,3 +144,7 @@ Do NOT flag issues that only apply to public-facing, multi-tenant, or
 internet-exposed services. Focus on bugs, logic errors, data corruption
 risks, and code quality issues.
 """
+
+# Cadence: reviews are batched, not per-commit. No post-commit hook is
+# installed; every 5 commits on a working branch, run a range review:
+#   roborev review <last-reviewed-sha>..HEAD

Makefile

@@ -93,7 +93,10 @@ install-hooks:
 docs-install:
 	cd docs && uv sync --frozen
 
-docs-build:
+bridge-contract:
+	go test -tags fts5 ./document/bridge -run '^TestBridgeContractNormativeDocuments'
+
+docs-build: bridge-contract
 	cd docs && ./zensical-docs.sh build
 
 docs-serve:

cmd/docbank/backfill_retry.go

@@ -0,0 +1,56 @@
+package main
+
+import "time"
+
+const maxBackfillRetryDelay = 10 * time.Minute
+
+type backfillRetryState struct {
+	failures  uint
+	notBefore time.Time
+}
+
+type backfillRetrySet map[string]backfillRetryState
+
+func newBackfillRetrySet() backfillRetrySet {
+	return make(backfillRetrySet)
+}
+
+func (retries backfillRetrySet) ready(key string, now time.Time) bool {
+	state, found := retries[key]
+	return !found || !now.Before(state.notBefore)
+}
+
+func (retries backfillRetrySet) failed(key string, now time.Time) {
+	state := retries[key]
+	state.failures++
+	delay := 5 * time.Second
+	for attempt := uint(1); attempt < state.failures && delay < maxBackfillRetryDelay; attempt++ {
+		delay = min(delay*2, maxBackfillRetryDelay)
+	}
+	state.notBefore = now.Add(delay)
+	retries[key] = state
+}
+
+func (retries backfillRetrySet) succeeded(key string) {
+	delete(retries, key)
+}
+
+func (retries backfillRetrySet) waitDelay(now time.Time, maximum time.Duration) time.Duration {
+	delay := maximum
+	for _, state := range retries {
+		if !now.Before(state.notBefore) {
+			return 0
+		}
+		delay = min(delay, state.notBefore.Sub(now))
+	}
+	return delay
+}
+
+func backfillBatchWaitDelay(
+	attempted int, retries backfillRetrySet, now time.Time,
+) time.Duration {
+	if attempted != 0 || len(retries) == 0 {
+		return 100 * time.Millisecond
+	}
+	return retries.waitDelay(now, 10*time.Second)
+}

cmd/docbank/backfill_retry_test.go

@@ -0,0 +1,38 @@
+package main
+
+import (
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestBackfillRetrySetQuarantinesFailuresWithoutBlockingOtherTargets(t *testing.T) {
+	retries := newBackfillRetrySet()
+	now := time.Date(2026, time.August, 24, 12, 0, 0, 0, time.UTC)
+	assert.True(t, retries.ready("bad", now))
+	assert.True(t, retries.ready("later", now))
+
+	retries.failed("bad", now)
+	assert.False(t, retries.ready("bad", now))
+	assert.True(t, retries.ready("later", now))
+	assert.True(t, retries.ready("bad", now.Add(5*time.Second)))
+
+	retries.failed("bad", now.Add(5*time.Second))
+	assert.False(t, retries.ready("bad", now.Add(14*time.Second)))
+	assert.True(t, retries.ready("bad", now.Add(15*time.Second)))
+
+	retries.succeeded("bad")
+	assert.True(t, retries.ready("bad", now))
+}
+
+func TestBackfillBatchWaitHonorsRetryWhenEveryTargetIsSkipped(t *testing.T) {
+	retries := newBackfillRetrySet()
+	now := time.Date(2026, time.August, 25, 12, 0, 0, 0, time.UTC)
+	retries.failed("deferred", now)
+
+	assert.Equal(t, 5*time.Second,
+		backfillBatchWaitDelay(0, retries, now))
+	assert.Equal(t, 100*time.Millisecond,
+		backfillBatchWaitDelay(1, retries, now))
+}

cmd/docbank/cli_test.go

@@ -686,13 +686,25 @@ func TestJobsShowsDaemonStatus(t *testing.T) {
 	assert.Contains(t, out, "extract:plain-text")
 	assert.Contains(t, out, "running")
 
-	out, err = runCLI(t, "jobs", "--json")
-	require.NoError(t, err)
 	var got api.JobList
-	require.NoError(t, json.Unmarshal([]byte(out), &got))
-	require.Len(t, got.Items, 1)
+	require.Eventually(t, func() bool {
+		out, err = runCLI(t, "jobs", "--json")
+		if err != nil || json.Unmarshal([]byte(out), &got) != nil || len(got.Items) != 4 {
+			return false
+		}
+		return got.Items[1].Name == "extract:source-metadata" &&
+			got.Items[2].Name == "maintenance:auxiliary-checksums" && got.Items[2].Status == "completed" &&
+			got.Items[3].Name == "process:vector-indexes" && got.Items[3].Status == "running"
+	}, 5*time.Second, 25*time.Millisecond)
+	require.Len(t, got.Items, 4)
 	assert.Equal(t, "extract:plain-text", got.Items[0].Name)
 	assert.Equal(t, "running", got.Items[0].Status)
+	assert.Equal(t, "extract:source-metadata", got.Items[1].Name)
+	assert.Equal(t, "running", got.Items[1].Status)
+	assert.Equal(t, "maintenance:auxiliary-checksums", got.Items[2].Name)
+	assert.Equal(t, "completed", got.Items[2].Status)
+	assert.Equal(t, "process:vector-indexes", got.Items[3].Name)
+	assert.Equal(t, "running", got.Items[3].Status)
 }
 
 func TestConfiguredAutomaticPackingPacksAndKeepsDaemonAlive(t *testing.T) {
@@ -734,10 +746,14 @@ func TestConfiguredAutomaticPackingPacksAndKeepsDaemonAlive(t *testing.T) {
 	require.NoError(t, err)
 	var got api.JobList
 	require.NoError(t, json.Unmarshal([]byte(out), &got))
-	require.Len(t, got.Items, 2)
+	require.Len(t, got.Items, 5)
 	assert.Equal(t, "extract:plain-text", got.Items[0].Name)
-	assert.Equal(t, "storage:pack", got.Items[1].Name)
-	assert.Equal(t, "running", got.Items[1].Status)
+	assert.Equal(t, "extract:source-metadata", got.Items[1].Name)
+	assert.Equal(t, "maintenance:auxiliary-checksums", got.Items[2].Name)
+	assert.Equal(t, "process:vector-indexes", got.Items[3].Name)
+	assert.Equal(t, "running", got.Items[3].Status)
+	assert.Equal(t, "storage:pack", got.Items[4].Name)
+	assert.Equal(t, "running", got.Items[4].Status)
 
 	time.Sleep(100 * time.Millisecond)
 	_, _, found, err := client.Find(t.Context(), home)
@@ -778,11 +794,15 @@ func TestConfiguredWatchIngestsStableFilesAndRemainsObservable(t *testing.T) {
 	require.NoError(t, err)
 	var got api.JobList
 	require.NoError(t, json.Unmarshal([]byte(out), &got))
-	require.Len(t, got.Items, 2)
+	require.Len(t, got.Items, 5)
 	assert.Equal(t, "extract:plain-text", got.Items[0].Name)
 	assert.Equal(t, "running", got.Items[0].Status)
-	assert.Equal(t, "watch:sessions", got.Items[1].Name)
-	assert.Equal(t, "running", got.Items[1].Status)
+	assert.Equal(t, "extract:source-metadata", got.Items[1].Name)
+	assert.Equal(t, "maintenance:auxiliary-checksums", got.Items[2].Name)
+	assert.Equal(t, "process:vector-indexes", got.Items[3].Name)
+	assert.Equal(t, "running", got.Items[3].Status)
+	assert.Equal(t, "watch:sessions", got.Items[4].Name)
+	assert.Equal(t, "running", got.Items[4].Status)
 
 	out, err = runCLI(t, "watch", "list", "--json")
 	require.NoError(t, err)