Read the advertising out of an episode's words

A second ad detector, and it finds a class the first structurally cannot.

The audio scan compares episodes mirrored within a minute of each other, so
what it finds is advertising a CDN stitched into all of them -- validated
against the declared-versus-served length gap. A show that inserts nothing has
no gap, which by that measure means it carries no advertising, and the host-read
sponsor block in the middle of it is invisible. Every published transcript the
server can use belongs to exactly such a show, because having no inserted
advertising is the condition for using one. The two detectors do not merely fail
differently; they look at disjoint sets of episodes.

**Density, not phrases.** One marker is a mention. Measured against the 38
transcripts harvested from the live library: a single "brought to you by" is a
one-second tag inside play-by-play and skipping it would cut the game, and a
guest describing how a sponsor came forward for his yacht event says "sponsor"
four times while being the programme -- the same trap the toc=false chapter rule
already documents. A real read stacks different *kinds* of marker inside a few
seconds: who it is for, where to go, what the offer is, what the terms are. Two
kinds and fifteen seconds, the same ad unit the length-gap rule uses.

Every threshold comes from a measurement rather than a guess. The 25-second
bridge is the 17.7-second gap between "presented by Delta" and "12% off" in the
Seahawks Rewind read, with the product description in between. "Slash" needs a
domain in front of it because the corpus contains "a children's movie slash book
for adults".

Across all 38 live transcripts it reports **one** candidate: that Delta block,
41:07 to 41:33, with no programme clipped at either end. That is low recall by
design -- these are the shows with the least advertising, because they are the
only ones whose transcripts can be used at all. The shows with real ad breaks
need speech recognition, which is not built.

**Nothing is published as a skip.** The candidates are stored with the words and
the lines are dimmed on the transcript screen, so the reader can see what the
server thinks is an advertisement and disagree with it. Dimmed rather than
hidden, because a wrong call has to stay visible. Nothing in the playback path
reads them -- same posture as the signature library's shadow run, and the same
reason: a detector acting on its own unchecked first guesses makes skips into
the middle of the programme, which is worse than no skips.

Schema 18 adds the mark to the cue, defaulting to not: a line nobody judged is
not an advertisement. Migrated and checked on a device.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018AwnALZL6bD4VBtATDVv3B

17 files changed, 1646 insertions, 23 deletions

diff --git a/.coverage-baseline b/.coverage-baseline
index 1fbd300..9bfb6a0 100644
--- a/.coverage-baseline
+++ b/.coverage-baseline
@@ -1 +1 @@
-94.45
+94.36
diff --git a/app/schemas/org.terst.listen.data.AppDatabase/18.json b/app/schemas/org.terst.listen.data.AppDatabase/18.json
new file mode 100644
index 0000000..f9f7a2c
--- /dev/null
+++ b/app/schemas/org.terst.listen.data.AppDatabase/18.json
@@ -0,0 +1,1026 @@
+{
+  "formatVersion": 1,
+  "database": {
+    "version": 18,
+    "identityHash": "b832d4c76ecdc3df30db9edc8777c1c9",
+    "entities": [
+      {
+        "tableName": "subscriptions",
+        "createSql": "CREATE TABLE IF NOT EXISTS `${TABLE_NAME}` (`id` TEXT NOT NULL, `feedUrl` TEXT NOT NULL, `kind` TEXT NOT NULL, `title` TEXT NOT NULL, `description` TEXT, `author` TEXT, `imageUrl` TEXT, `link` TEXT, `addedAtMs` INTEGER NOT NULL, `etag` TEXT, `lastModified` TEXT, `lastFetchedAtMs` INTEGER, `lastError` TEXT, `playbackSpeed` REAL, `skipIntroMs` INTEGER NOT NULL, `skipOutroMs` INTEGER NOT NULL, `dripEnabled` INTEGER NOT NULL DEFAULT 0, `dripIntervalMs` INTEGER NOT NULL DEFAULT 86400000, `dripCount` INTEGER NOT NULL DEFAULT 1, `dripOldestFirst` INTEGER NOT NULL DEFAULT 1, `dripLastReleaseAtMs` INTEGER, `skipAdChapters` INTEGER NOT NULL DEFAULT 0, `midrollPendingEpisodeId` TEXT, `bedtime` INTEGER NOT NULL DEFAULT 0, PRIMARY KEY(`id`))",
+        "fields": [
+          {
+            "fieldPath": "id",
+            "columnName": "id",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "feedUrl",
+            "columnName": "feedUrl",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "kind",
+            "columnName": "kind",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "title",
+            "columnName": "title",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "description",
+            "columnName": "description",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "author",
+            "columnName": "author",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "imageUrl",
+            "columnName": "imageUrl",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "link",
+            "columnName": "link",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "addedAtMs",
+            "columnName": "addedAtMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "etag",
+            "columnName": "etag",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "lastModified",
+            "columnName": "lastModified",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "lastFetchedAtMs",
+            "columnName": "lastFetchedAtMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "lastError",
+            "columnName": "lastError",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "playbackSpeed",
+            "columnName": "playbackSpeed",
+            "affinity": "REAL"
+          },
+          {
+            "fieldPath": "skipIntroMs",
+            "columnName": "skipIntroMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "skipOutroMs",
+            "columnName": "skipOutroMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "dripEnabled",
+            "columnName": "dripEnabled",
+            "affinity": "INTEGER",
+            "notNull": true,
+            "defaultValue": "0"
+          },
+          {
+            "fieldPath": "dripIntervalMs",
+            "columnName": "dripIntervalMs",
+            "affinity": "INTEGER",
+            "notNull": true,
+            "defaultValue": "86400000"
+          },
+          {
+            "fieldPath": "dripCount",
+            "columnName": "dripCount",
+            "affinity": "INTEGER",
+            "notNull": true,
+            "defaultValue": "1"
+          },
+          {
+            "fieldPath": "dripOldestFirst",
+            "columnName": "dripOldestFirst",
+            "affinity": "INTEGER",
+            "notNull": true,
+            "defaultValue": "1"
+          },
+          {
+            "fieldPath": "dripLastReleaseAtMs",
+            "columnName": "dripLastReleaseAtMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "skipAdChapters",
+            "columnName": "skipAdChapters",
+            "affinity": "INTEGER",
+            "notNull": true,
+            "defaultValue": "0"
+          },
+          {
+            "fieldPath": "midrollPendingEpisodeId",
+            "columnName": "midrollPendingEpisodeId",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "bedtime",
+            "columnName": "bedtime",
+            "affinity": "INTEGER",
+            "notNull": true,
+            "defaultValue": "0"
+          }
+        ],
+        "primaryKey": {
+          "autoGenerate": false,
+          "columnNames": [
+            "id"
+          ]
+        },
+        "indices": [
+          {
+            "name": "index_subscriptions_feedUrl",
+            "unique": true,
+            "columnNames": [
+              "feedUrl"
+            ],
+            "orders": [],
+            "createSql": "CREATE UNIQUE INDEX IF NOT EXISTS `index_subscriptions_feedUrl` ON `${TABLE_NAME}` (`feedUrl`)"
+          }
+        ]
+      },
+      {
+        "tableName": "episodes",
+        "createSql": "CREATE TABLE IF NOT EXISTS `${TABLE_NAME}` (`id` TEXT NOT NULL, `subscriptionId` TEXT NOT NULL, `guid` TEXT NOT NULL, `title` TEXT NOT NULL, `description` TEXT, `mediaUrl` TEXT NOT NULL, `mediaType` TEXT NOT NULL, `mediaLengthBytes` INTEGER, `durationMs` INTEGER, `publishedAtMs` INTEGER, `seasonNumber` INTEGER, `episodeNumber` INTEGER, `imageUrl` TEXT, `chaptersUrl` TEXT, `discoveredAtMs` INTEGER NOT NULL, `positionMs` INTEGER NOT NULL, `playedAtMs` INTEGER, `actualDurationMs` INTEGER, `downloadState` TEXT NOT NULL, `localPath` TEXT, `downloadedAtMs` INTEGER, `audioFromServer` INTEGER NOT NULL DEFAULT 0, `dripReleasedAtMs` INTEGER, `appliedRuleId` TEXT, `appliedRuleReason` TEXT, `speedOverride` REAL, `midrollCheckedAtMs` INTEGER, `chaptersJson` TEXT, `detectedIntroMs` INTEGER, `detectedOutroMs` INTEGER, PRIMARY KEY(`id`), FOREIGN KEY(`subscriptionId`) REFERENCES `subscriptions`(`id`) ON UPDATE NO ACTION ON DELETE CASCADE )",
+        "fields": [
+          {
+            "fieldPath": "id",
+            "columnName": "id",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "subscriptionId",
+            "columnName": "subscriptionId",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "guid",
+            "columnName": "guid",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "title",
+            "columnName": "title",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "description",
+            "columnName": "description",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "mediaUrl",
+            "columnName": "mediaUrl",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "mediaType",
+            "columnName": "mediaType",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "mediaLengthBytes",
+            "columnName": "mediaLengthBytes",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "durationMs",
+            "columnName": "durationMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "publishedAtMs",
+            "columnName": "publishedAtMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "seasonNumber",
+            "columnName": "seasonNumber",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "episodeNumber",
+            "columnName": "episodeNumber",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "imageUrl",
+            "columnName": "imageUrl",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "chaptersUrl",
+            "columnName": "chaptersUrl",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "discoveredAtMs",
+            "columnName": "discoveredAtMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "positionMs",
+            "columnName": "positionMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "playedAtMs",
+            "columnName": "playedAtMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "actualDurationMs",
+            "columnName": "actualDurationMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "downloadState",
+            "columnName": "downloadState",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "localPath",
+            "columnName": "localPath",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "downloadedAtMs",
+            "columnName": "downloadedAtMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "audioFromServer",
+            "columnName": "audioFromServer",
+            "affinity": "INTEGER",
+            "notNull": true,
+            "defaultValue": "0"
+          },
+          {
+            "fieldPath": "dripReleasedAtMs",
+            "columnName": "dripReleasedAtMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "appliedRuleId",
+            "columnName": "appliedRuleId",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "appliedRuleReason",
+            "columnName": "appliedRuleReason",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "speedOverride",
+            "columnName": "speedOverride",
+            "affinity": "REAL"
+          },
+          {
+            "fieldPath": "midrollCheckedAtMs",
+            "columnName": "midrollCheckedAtMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "chaptersJson",
+            "columnName": "chaptersJson",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "detectedIntroMs",
+            "columnName": "detectedIntroMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "detectedOutroMs",
+            "columnName": "detectedOutroMs",
+            "affinity": "INTEGER"
+          }
+        ],
+        "primaryKey": {
+          "autoGenerate": false,
+          "columnNames": [
+            "id"
+          ]
+        },
+        "indices": [
+          {
+            "name": "index_episodes_subscriptionId",
+            "unique": false,
+            "columnNames": [
+              "subscriptionId"
+            ],
+            "orders": [],
+            "createSql": "CREATE INDEX IF NOT EXISTS `index_episodes_subscriptionId` ON `${TABLE_NAME}` (`subscriptionId`)"
+          },
+          {
+            "name": "index_episodes_subscriptionId_guid",
+            "unique": true,
+            "columnNames": [
+              "subscriptionId",
+              "guid"
+            ],
+            "orders": [],
+            "createSql": "CREATE UNIQUE INDEX IF NOT EXISTS `index_episodes_subscriptionId_guid` ON `${TABLE_NAME}` (`subscriptionId`, `guid`)"
+          },
+          {
+            "name": "index_episodes_publishedAtMs",
+            "unique": false,
+            "columnNames": [
+              "publishedAtMs"
+            ],
+            "orders": [],
+            "createSql": "CREATE INDEX IF NOT EXISTS `index_episodes_publishedAtMs` ON `${TABLE_NAME}` (`publishedAtMs`)"
+          }
+        ],
+        "foreignKeys": [
+          {
+            "table": "subscriptions",
+            "onDelete": "CASCADE",
+            "onUpdate": "NO ACTION",
+            "columns": [
+              "subscriptionId"
+            ],
+            "referencedColumns": [
+              "id"
+            ]
+          }
+        ]
+      },
+      {
+        "tableName": "queue",
+        "createSql": "CREATE TABLE IF NOT EXISTS `${TABLE_NAME}` (`episodeId` TEXT NOT NULL, `position` INTEGER NOT NULL, PRIMARY KEY(`episodeId`), FOREIGN KEY(`episodeId`) REFERENCES `episodes`(`id`) ON UPDATE NO ACTION ON DELETE CASCADE )",
+        "fields": [
+          {
+            "fieldPath": "episodeId",
+            "columnName": "episodeId",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "position",
+            "columnName": "position",
+            "affinity": "INTEGER",
+            "notNull": true
+          }
+        ],
+        "primaryKey": {
+          "autoGenerate": false,
+          "columnNames": [
+            "episodeId"
+          ]
+        },
+        "indices": [
+          {
+            "name": "index_queue_position",
+            "unique": false,
+            "columnNames": [
+              "position"
+            ],
+            "orders": [],
+            "createSql": "CREATE INDEX IF NOT EXISTS `index_queue_position` ON `${TABLE_NAME}` (`position`)"
+          }
+        ],
+        "foreignKeys": [
+          {
+            "table": "episodes",
+            "onDelete": "CASCADE",
+            "onUpdate": "NO ACTION",
+            "columns": [
+              "episodeId"
+            ],
+            "referencedColumns": [
+              "id"
+            ]
+          }
+        ]
+      },
+      {
+        "tableName": "rules",
+        "createSql": "CREATE TABLE IF NOT EXISTS `${TABLE_NAME}` (`id` TEXT NOT NULL, `name` TEXT NOT NULL, `position` INTEGER NOT NULL, `enabled` INTEGER NOT NULL, `subscriptionId` TEXT, `titleContains` TEXT, `titleExcludes` TEXT, `minDurationMs` INTEGER, `maxDurationMs` INTEGER, `maxAgeMs` INTEGER, `seasonNumber` INTEGER, `action` TEXT NOT NULL, `autoDownload` INTEGER NOT NULL, `speed` REAL, PRIMARY KEY(`id`))",
+        "fields": [
+          {
+            "fieldPath": "id",
+            "columnName": "id",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "name",
+            "columnName": "name",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "position",
+            "columnName": "position",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "enabled",
+            "columnName": "enabled",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "subscriptionId",
+            "columnName": "subscriptionId",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "titleContains",
+            "columnName": "titleContains",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "titleExcludes",
+            "columnName": "titleExcludes",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "minDurationMs",
+            "columnName": "minDurationMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "maxDurationMs",
+            "columnName": "maxDurationMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "maxAgeMs",
+            "columnName": "maxAgeMs",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "seasonNumber",
+            "columnName": "seasonNumber",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "action",
+            "columnName": "action",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "autoDownload",
+            "columnName": "autoDownload",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "speed",
+            "columnName": "speed",
+            "affinity": "REAL"
+          }
+        ],
+        "primaryKey": {
+          "autoGenerate": false,
+          "columnNames": [
+            "id"
+          ]
+        },
+        "indices": [
+          {
+            "name": "index_rules_position",
+            "unique": false,
+            "columnNames": [
+              "position"
+            ],
+            "orders": [],
+            "createSql": "CREATE INDEX IF NOT EXISTS `index_rules_position` ON `${TABLE_NAME}` (`position`)"
+          }
+        ]
+      },
+      {
+        "tableName": "show_fingerprints",
+        "createSql": "CREATE TABLE IF NOT EXISTS `${TABLE_NAME}` (`subscriptionId` TEXT NOT NULL, `kind` TEXT NOT NULL, `referenceClipPcm` BLOB, `pendingEpisodeId` TEXT, `sampleRate` INTEGER NOT NULL, `confirmedCount` INTEGER NOT NULL, `lastConfirmedEpisodeId` TEXT, `updatedAtMs` INTEGER NOT NULL, PRIMARY KEY(`subscriptionId`, `kind`), FOREIGN KEY(`subscriptionId`) REFERENCES `subscriptions`(`id`) ON UPDATE NO ACTION ON DELETE CASCADE )",
+        "fields": [
+          {
+            "fieldPath": "subscriptionId",
+            "columnName": "subscriptionId",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "kind",
+            "columnName": "kind",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "referenceClipPcm",
+            "columnName": "referenceClipPcm",
+            "affinity": "BLOB"
+          },
+          {
+            "fieldPath": "pendingEpisodeId",
+            "columnName": "pendingEpisodeId",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "sampleRate",
+            "columnName": "sampleRate",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "confirmedCount",
+            "columnName": "confirmedCount",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "lastConfirmedEpisodeId",
+            "columnName": "lastConfirmedEpisodeId",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "updatedAtMs",
+            "columnName": "updatedAtMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          }
+        ],
+        "primaryKey": {
+          "autoGenerate": false,
+          "columnNames": [
+            "subscriptionId",
+            "kind"
+          ]
+        },
+        "indices": [
+          {
+            "name": "index_show_fingerprints_subscriptionId",
+            "unique": false,
+            "columnNames": [
+              "subscriptionId"
+            ],
+            "orders": [],
+            "createSql": "CREATE INDEX IF NOT EXISTS `index_show_fingerprints_subscriptionId` ON `${TABLE_NAME}` (`subscriptionId`)"
+          }
+        ],
+        "foreignKeys": [
+          {
+            "table": "subscriptions",
+            "onDelete": "CASCADE",
+            "onUpdate": "NO ACTION",
+            "columns": [
+              "subscriptionId"
+            ],
+            "referencedColumns": [
+              "id"
+            ]
+          }
+        ]
+      },
+      {
+        "tableName": "show_midrolls",
+        "createSql": "CREATE TABLE IF NOT EXISTS `${TABLE_NAME}` (`subscriptionId` TEXT NOT NULL, `id` TEXT NOT NULL, `referenceClipPath` TEXT NOT NULL, `sampleRate` INTEGER NOT NULL, `confirmedCount` INTEGER NOT NULL, `lastConfirmedEpisodeId` TEXT, `updatedAtMs` INTEGER NOT NULL, PRIMARY KEY(`subscriptionId`, `id`), FOREIGN KEY(`subscriptionId`) REFERENCES `subscriptions`(`id`) ON UPDATE NO ACTION ON DELETE CASCADE )",
+        "fields": [
+          {
+            "fieldPath": "subscriptionId",
+            "columnName": "subscriptionId",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "id",
+            "columnName": "id",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "referenceClipPath",
+            "columnName": "referenceClipPath",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "sampleRate",
+            "columnName": "sampleRate",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "confirmedCount",
+            "columnName": "confirmedCount",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "lastConfirmedEpisodeId",
+            "columnName": "lastConfirmedEpisodeId",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "updatedAtMs",
+            "columnName": "updatedAtMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          }
+        ],
+        "primaryKey": {
+          "autoGenerate": false,
+          "columnNames": [
+            "subscriptionId",
+            "id"
+          ]
+        },
+        "indices": [
+          {
+            "name": "index_show_midrolls_subscriptionId",
+            "unique": false,
+            "columnNames": [
+              "subscriptionId"
+            ],
+            "orders": [],
+            "createSql": "CREATE INDEX IF NOT EXISTS `index_show_midrolls_subscriptionId` ON `${TABLE_NAME}` (`subscriptionId`)"
+          }
+        ],
+        "foreignKeys": [
+          {
+            "table": "subscriptions",
+            "onDelete": "CASCADE",
+            "onUpdate": "NO ACTION",
+            "columns": [
+              "subscriptionId"
+            ],
+            "referencedColumns": [
+              "id"
+            ]
+          }
+        ]
+      },
+      {
+        "tableName": "episode_midrolls",
+        "createSql": "CREATE TABLE IF NOT EXISTS `${TABLE_NAME}` (`episodeId` TEXT NOT NULL, `candidateId` TEXT NOT NULL, `startMs` INTEGER NOT NULL, `endMs` INTEGER NOT NULL, `kind` TEXT NOT NULL DEFAULT 'ad', `partners` INTEGER NOT NULL DEFAULT 0, PRIMARY KEY(`episodeId`, `candidateId`), FOREIGN KEY(`episodeId`) REFERENCES `episodes`(`id`) ON UPDATE NO ACTION ON DELETE CASCADE )",
+        "fields": [
+          {
+            "fieldPath": "episodeId",
+            "columnName": "episodeId",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "candidateId",
+            "columnName": "candidateId",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "startMs",
+            "columnName": "startMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "endMs",
+            "columnName": "endMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "kind",
+            "columnName": "kind",
+            "affinity": "TEXT",
+            "notNull": true,
+            "defaultValue": "'ad'"
+          },
+          {
+            "fieldPath": "partners",
+            "columnName": "partners",
+            "affinity": "INTEGER",
+            "notNull": true,
+            "defaultValue": "0"
+          }
+        ],
+        "primaryKey": {
+          "autoGenerate": false,
+          "columnNames": [
+            "episodeId",
+            "candidateId"
+          ]
+        },
+        "indices": [
+          {
+            "name": "index_episode_midrolls_episodeId",
+            "unique": false,
+            "columnNames": [
+              "episodeId"
+            ],
+            "orders": [],
+            "createSql": "CREATE INDEX IF NOT EXISTS `index_episode_midrolls_episodeId` ON `${TABLE_NAME}` (`episodeId`)"
+          }
+        ],
+        "foreignKeys": [
+          {
+            "table": "episodes",
+            "onDelete": "CASCADE",
+            "onUpdate": "NO ACTION",
+            "columns": [
+              "episodeId"
+            ],
+            "referencedColumns": [
+              "id"
+            ]
+          }
+        ]
+      },
+      {
+        "tableName": "break_verdicts",
+        "createSql": "CREATE TABLE IF NOT EXISTS `${TABLE_NAME}` (`episodeId` TEXT NOT NULL, `startMs` INTEGER NOT NULL, `endMs` INTEGER NOT NULL, `verdict` TEXT NOT NULL, `decidedAtMs` INTEGER NOT NULL, `kind` TEXT NOT NULL DEFAULT 'ad', `partners` INTEGER NOT NULL DEFAULT 0, PRIMARY KEY(`episodeId`, `startMs`, `endMs`))",
+        "fields": [
+          {
+            "fieldPath": "episodeId",
+            "columnName": "episodeId",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "startMs",
+            "columnName": "startMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "endMs",
+            "columnName": "endMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "verdict",
+            "columnName": "verdict",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "decidedAtMs",
+            "columnName": "decidedAtMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "kind",
+            "columnName": "kind",
+            "affinity": "TEXT",
+            "notNull": true,
+            "defaultValue": "'ad'"
+          },
+          {
+            "fieldPath": "partners",
+            "columnName": "partners",
+            "affinity": "INTEGER",
+            "notNull": true,
+            "defaultValue": "0"
+          }
+        ],
+        "primaryKey": {
+          "autoGenerate": false,
+          "columnNames": [
+            "episodeId",
+            "startMs",
+            "endMs"
+          ]
+        },
+        "indices": [
+          {
+            "name": "index_break_verdicts_episodeId",
+            "unique": false,
+            "columnNames": [
+              "episodeId"
+            ],
+            "orders": [],
+            "createSql": "CREATE INDEX IF NOT EXISTS `index_break_verdicts_episodeId` ON `${TABLE_NAME}` (`episodeId`)"
+          }
+        ]
+      },
+      {
+        "tableName": "ad_reports",
+        "createSql": "CREATE TABLE IF NOT EXISTS `${TABLE_NAME}` (`episodeId` TEXT NOT NULL, `startMs` INTEGER NOT NULL, `endMs` INTEGER NOT NULL, `reportedAtMs` INTEGER NOT NULL, `correctionWindowMs` INTEGER NOT NULL, `endCorrected` INTEGER NOT NULL DEFAULT 0, PRIMARY KEY(`episodeId`, `startMs`))",
+        "fields": [
+          {
+            "fieldPath": "episodeId",
+            "columnName": "episodeId",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "startMs",
+            "columnName": "startMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "endMs",
+            "columnName": "endMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "reportedAtMs",
+            "columnName": "reportedAtMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "correctionWindowMs",
+            "columnName": "correctionWindowMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "endCorrected",
+            "columnName": "endCorrected",
+            "affinity": "INTEGER",
+            "notNull": true,
+            "defaultValue": "0"
+          }
+        ],
+        "primaryKey": {
+          "autoGenerate": false,
+          "columnNames": [
+            "episodeId",
+            "startMs"
+          ]
+        },
+        "indices": [
+          {
+            "name": "index_ad_reports_episodeId",
+            "unique": false,
+            "columnNames": [
+              "episodeId"
+            ],
+            "orders": [],
+            "createSql": "CREATE INDEX IF NOT EXISTS `index_ad_reports_episodeId` ON `${TABLE_NAME}` (`episodeId`)"
+          }
+        ]
+      },
+      {
+        "tableName": "episode_transcripts",
+        "createSql": "CREATE TABLE IF NOT EXISTS `${TABLE_NAME}` (`episodeId` TEXT NOT NULL, `source` TEXT NOT NULL, `cues` INTEGER NOT NULL, `note` TEXT NOT NULL, `fetchedAtMs` INTEGER NOT NULL, PRIMARY KEY(`episodeId`), FOREIGN KEY(`episodeId`) REFERENCES `episodes`(`id`) ON UPDATE NO ACTION ON DELETE CASCADE )",
+        "fields": [
+          {
+            "fieldPath": "episodeId",
+            "columnName": "episodeId",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "source",
+            "columnName": "source",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "cues",
+            "columnName": "cues",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "note",
+            "columnName": "note",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "fetchedAtMs",
+            "columnName": "fetchedAtMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          }
+        ],
+        "primaryKey": {
+          "autoGenerate": false,
+          "columnNames": [
+            "episodeId"
+          ]
+        },
+        "foreignKeys": [
+          {
+            "table": "episodes",
+            "onDelete": "CASCADE",
+            "onUpdate": "NO ACTION",
+            "columns": [
+              "episodeId"
+            ],
+            "referencedColumns": [
+              "id"
+            ]
+          }
+        ]
+      },
+      {
+        "tableName": "transcript_cues",
+        "createSql": "CREATE TABLE IF NOT EXISTS `${TABLE_NAME}` (`episodeId` TEXT NOT NULL, `startMs` INTEGER NOT NULL, `endMs` INTEGER NOT NULL, `text` TEXT NOT NULL, `advertising` INTEGER NOT NULL DEFAULT 0, PRIMARY KEY(`episodeId`, `startMs`), FOREIGN KEY(`episodeId`) REFERENCES `episodes`(`id`) ON UPDATE NO ACTION ON DELETE CASCADE )",
+        "fields": [
+          {
+            "fieldPath": "episodeId",
+            "columnName": "episodeId",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "startMs",
+            "columnName": "startMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "endMs",
+            "columnName": "endMs",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "text",
+            "columnName": "text",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "advertising",
+            "columnName": "advertising",
+            "affinity": "INTEGER",
+            "notNull": true,
+            "defaultValue": "0"
+          }
+        ],
+        "primaryKey": {
+          "autoGenerate": false,
+          "columnNames": [
+            "episodeId",
+            "startMs"
+          ]
+        },
+        "indices": [
+          {
+            "name": "index_transcript_cues_episodeId",
+            "unique": false,
+            "columnNames": [
+              "episodeId"
+            ],
+            "orders": [],
+            "createSql": "CREATE INDEX IF NOT EXISTS `index_transcript_cues_episodeId` ON `${TABLE_NAME}` (`episodeId`)"
+          }
+        ],
+        "foreignKeys": [
+          {
+            "table": "episodes",
+            "onDelete": "CASCADE",
+            "onUpdate": "NO ACTION",
+            "columns": [
+              "episodeId"
+            ],
+            "referencedColumns": [
+              "id"
+            ]
+          }
+        ]
+      }
+    ],
+    "setupQueries": [
+      "CREATE TABLE IF NOT EXISTS room_master_table (id INTEGER PRIMARY KEY,identity_hash TEXT)",
+      "INSERT OR REPLACE INTO room_master_table (id,identity_hash) VALUES(42, 'b832d4c76ecdc3df30db9edc8777c1c9')"
+    ]
+  }
+}
\ No newline at end of file
diff --git a/app/src/androidTest/java/org/terst/listen/data/MigrationTest.kt b/app/src/androidTest/java/org/terst/listen/data/MigrationTest.kt
index 52f07eb..49d3406 100644
--- a/app/src/androidTest/java/org/terst/listen/data/MigrationTest.kt
+++ b/app/src/androidTest/java/org/terst/listen/data/MigrationTest.kt
@@ -803,4 +803,57 @@ class MigrationTest {
             assertEquals("the transcript row outlived its episode", 0, it.getInt(0))
         }
     }
+
+    /**
+     * Cues gain a mark saying the server read them as advertising, and existing ones default to
+     * not. A mark on the words, never a range to skip -- nothing in the playback path reads it.
+     */
+    @Test
+    fun migrate17To18MarksAdvertisingWithoutDisturbingExistingCues() {
+        helper.createDatabase(name, 17).apply {
+            execSQL(
+                """INSERT INTO subscriptions
+                   (id, feedUrl, kind, title, addedAtMs, skipIntroMs, skipOutroMs, dripEnabled,
+                    dripIntervalMs, dripCount, dripOldestFirst, skipAdChapters, bedtime)
+                   VALUES ('s1', 'https://example.com/feed.xml', 'PODCAST', 'A Show', 1000, 0, 0,
+                           0, 86400000, 1, 1, 1, 0)"""
+            )
+            execSQL(
+                """INSERT INTO episodes
+                   (id, subscriptionId, guid, title, mediaUrl, mediaType, discoveredAtMs,
+                    positionMs, downloadState)
+                   VALUES ('e1', 's1', 'g1', 'One', 'https://cdn/1.mp3', 'audio/mpeg', 1000,
+                           161803, 'NONE')"""
+            )
+            execSQL(
+                "INSERT INTO episode_transcripts (episodeId, source, cues, note, fetchedAtMs) " +
+                    "VALUES ('e1', 'published', 1, '', 99)"
+            )
+            execSQL(
+                "INSERT INTO transcript_cues (episodeId, startMs, endMs, text) " +
+                    "VALUES ('e1', 0, 3120, 'Welcome back.')"
+            )
+            close()
+        }
+
+        val db = helper.runMigrationsAndValidate(name, 18, true, AppDatabase.MIGRATION_17_18)
+
+        db.query("SELECT text, advertising FROM transcript_cues WHERE episodeId = 'e1'").use {
+            it.moveToFirst()
+            assertEquals("Welcome back.", it.getString(0))
+            assertEquals("a line nobody judged is not an advertisement", 0, it.getInt(1))
+        }
+        db.execSQL(
+            "INSERT INTO transcript_cues (episodeId, startMs, endMs, text, advertising) " +
+                "VALUES ('e1', 4000, 9000, 'Our sponsor this week.', 1)"
+        )
+        db.query("SELECT advertising FROM transcript_cues WHERE startMs = 4000").use {
+            it.moveToFirst()
+            assertEquals(1, it.getInt(0))
+        }
+        db.query("SELECT positionMs FROM episodes WHERE id = 'e1'").use {
+            it.moveToFirst()
+            assertEquals("the crown jewel must survive", 161803L, it.getLong(0))
+        }
+    }
 }
diff --git a/app/src/main/java/org/terst/listen/data/AppDatabase.kt b/app/src/main/java/org/terst/listen/data/AppDatabase.kt
index 96a25b9..5b9219d 100644
--- a/app/src/main/java/org/terst/listen/data/AppDatabase.kt
+++ b/app/src/main/java/org/terst/listen/data/AppDatabase.kt
@@ -34,7 +34,7 @@ class Converters {
         EpisodeTranscriptEntity::class,
         TranscriptCueEntity::class,
     ],
-    version = 17,
+    version = 18,
     exportSchema = true,
 )
 @TypeConverters(Converters::class)
@@ -469,6 +469,21 @@ abstract class AppDatabase : RoomDatabase() {
             }
         }
 
+        /**
+         * A cue gains a mark saying the server read it as advertising.
+         *
+         * A mark on the words, not a range to skip. Existing cues default to not: a line
+         * nobody has judged is not an advertisement, which is the same three-state care the
+         * transcript verdict itself takes.
+         */
+        val MIGRATION_17_18 = object : Migration(17, 18) {
+            override fun migrate(db: SupportSQLiteDatabase) {
+                db.execSQL(
+                    "ALTER TABLE `transcript_cues` ADD COLUMN `advertising` INTEGER NOT NULL DEFAULT 0"
+                )
+            }
+        }
+
         fun get(context: Context): AppDatabase = instance ?: synchronized(this) {
             instance ?: build(context.applicationContext).also { instance = it }
         }
@@ -482,6 +497,7 @@ abstract class AppDatabase : RoomDatabase() {
                     MIGRATION_6_7, MIGRATION_7_8, MIGRATION_8_9, MIGRATION_9_10,
                     MIGRATION_10_11, MIGRATION_11_12, MIGRATION_12_13, MIGRATION_13_14,
                     MIGRATION_14_15, MIGRATION_15_16, MIGRATION_16_17,
+                    MIGRATION_17_18,
                 )
                 .addCallback(SEED_DEFAULT_FILTERS)
                 // No fallbackToDestructiveMigration. A listener's playback positions and play
diff --git a/app/src/main/java/org/terst/listen/data/Entities.kt b/app/src/main/java/org/terst/listen/data/Entities.kt
index 2cfd15f..5a81e5d 100644
--- a/app/src/main/java/org/terst/listen/data/Entities.kt
+++ b/app/src/main/java/org/terst/listen/data/Entities.kt
@@ -629,4 +629,14 @@ data class TranscriptCueEntity(
     val startMs: Long,
     val endMs: Long,
     val text: String,
+    /**
+     * Whether the server read this line as part of an advertisement.
+     *
+     * A mark on the words, not a range to skip. It is stored on the cue rather than as its own
+     * table because it is a property of the line as displayed, and the only thing that reads it
+     * is the screen drawing that line. Nothing in the playback path sees it, deliberately: the
+     * detector that produced it has never been checked against anything, and a skip into the
+     * middle of the programme is worse than no skip.
+     */
+    @ColumnInfo(defaultValue = "0") val advertising: Boolean = false,
 )
diff --git a/app/src/main/java/org/terst/listen/server/ServerSkips.kt b/app/src/main/java/org/terst/listen/server/ServerSkips.kt
index d7fb4f5..8dcd115 100644
--- a/app/src/main/java/org/terst/listen/server/ServerSkips.kt
+++ b/app/src/main/java/org/terst/listen/server/ServerSkips.kt
@@ -107,12 +107,19 @@ class ServerSkips(
                 note = "",
                 fetchedAtMs = System.currentTimeMillis(),
             ),
-            fetched.cues.map {
+            fetched.cues.map { cue ->
                 TranscriptCueEntity(
                     episodeId = episode.id,
-                    startMs = it.startMs,
-                    endMs = it.endMs,
-                    text = it.text,
+                    startMs = cue.startMs,
+                    endMs = cue.endMs,
+                    text = cue.text,
+                    // Flattened onto the cue here rather than kept as ranges. The only thing
+                    // that reads it is the screen drawing that line, and a range the playback
+                    // path could reach is a skip waiting to be made out of a guess that has
+                    // never been checked against anything.
+                    advertising = fetched.ads.any { ad ->
+                        cue.startMs >= ad.startMs && cue.endMs <= ad.endMs
+                    },
                 )
             },
         )
diff --git a/app/src/main/java/org/terst/listen/server/SkipServer.kt b/app/src/main/java/org/terst/listen/server/SkipServer.kt
index 0980aff..a05eaf7 100644
--- a/app/src/main/java/org/terst/listen/server/SkipServer.kt
+++ b/app/src/main/java/org/terst/listen/server/SkipServer.kt
@@ -97,12 +97,29 @@ data class ServerCue(
     val text: String = "",
 )
 
+/**
+ * A stretch of an episode whose words read as advertising.
+ *
+ * A candidate, never a skip. It is shown beside the transcript so the reader can see what the
+ * server thinks is an advertisement, and nothing in the playback path reads it -- the same
+ * posture the signature library's shadow run takes, because a detector acting on its own first
+ * guesses makes skips into the middle of the programme.
+ */
+@Serializable
+data class ServerAdCandidate(
+    @SerialName("startMs") val startMs: Long = 0,
+    @SerialName("endMs") val endMs: Long = 0,
+    /** Which kinds of marker qualified it: sponsor, code, url, offer, legal. */
+    val signals: List<String> = emptyList(),
+)
+
 /** An episode's words as the server holds them. */
 @Serializable
 data class ServerTranscript(
     val source: String = "",
     @SerialName("languageCode") val languageCode: String? = null,
     val cues: List<ServerCue> = emptyList(),
+    val ads: List<ServerAdCandidate> = emptyList(),
 )
 
 /** What the server knows about one episode. */
@@ -322,6 +339,7 @@ class SkipServer(
             // to seek to. Same filtering the breaks get, because the player acts on these
             // numbers too -- a tap on a cue seeks to its start.
             cues = parsed.cues.filter { it.endMs > it.startMs && it.text.isNotBlank() },
+            ads = parsed.ads.filter { it.endMs > it.startMs },
         )
     }
 
diff --git a/app/src/main/java/org/terst/listen/ui/TranscriptScreen.kt b/app/src/main/java/org/terst/listen/ui/TranscriptScreen.kt
index c6d6e52..81c1b84 100644
--- a/app/src/main/java/org/terst/listen/ui/TranscriptScreen.kt
+++ b/app/src/main/java/org/terst/listen/ui/TranscriptScreen.kt
@@ -100,6 +100,9 @@ fun TranscriptScreen(
     }
 }
 
+/** Advertising is dimmed, not hidden: a wrong call has to stay visible to be disagreed with. */
+private const val ADVERTISING_ALPHA = 0.45f
+
 @Composable
 internal fun CueList(
     cues: List<TranscriptCueEntity>,
@@ -119,10 +122,15 @@ internal fun CueList(
                 cue.text,
                 style = MaterialTheme.typography.bodyMedium,
                 fontWeight = if (speaking) FontWeight.Bold else FontWeight.Normal,
-                color = if (speaking) {
-                    MaterialTheme.colorScheme.onSurface
-                } else {
-                    MaterialTheme.colorScheme.onSurfaceVariant
+                // Advertising is dimmed rather than hidden or struck through. The reader can
+                // see what the server thinks is an advertisement and disagree with it, which
+                // matters because nothing has ever checked this detector against anything --
+                // and hiding the lines would make a wrong call invisible.
+                color = when {
+                    cue.advertising -> MaterialTheme.colorScheme.onSurfaceVariant
+                        .copy(alpha = ADVERTISING_ALPHA)
+                    speaking -> MaterialTheme.colorScheme.onSurface
+                    else -> MaterialTheme.colorScheme.onSurfaceVariant
                 },
                 modifier = Modifier
                     .fillMaxWidth()
diff --git a/app/src/test/java/org/terst/listen/server/ServerSkipsTest.kt b/app/src/test/java/org/terst/listen/server/ServerSkipsTest.kt
index ae6a23b..3b0d671 100644
--- a/app/src/test/java/org/terst/listen/server/ServerSkipsTest.kt
+++ b/app/src/test/java/org/terst/listen/server/ServerSkipsTest.kt
@@ -453,4 +453,67 @@ class ServerSkipsTest {
         assertEquals(1, transcripts.cueRows.size)
         assertEquals(1, transcripts.rows.size)
     }
+
+    /**
+     * Advertising the server read in the words is marked on the lines it covers.
+     *
+     * Flattened onto the cue rather than stored as ranges, deliberately. The only thing that
+     * reads it is the screen drawing that line; a range the playback path could reach would be
+     * a skip made out of a guess nothing has ever checked.
+     */
+    @Test
+    fun `marks the lines the server read as advertising`() = runTest {
+        val transcripts = FakeTranscriptDao()
+        val transport = StubTransport(
+            null,
+            alsoServing = mapOf("api/transcript" to """
+                {"source":"published","cues":[
+                  {"startMs":0,"endMs":3000,"text":"Back in a moment."},
+                  {"startMs":10000,"endMs":15000,"text":"Sponsored by a mattress company."},
+                  {"startMs":15000,"endMs":26000,"text":"Use promo code LISTEN for 20% off."},
+                  {"startMs":30000,"endMs":34000,"text":"Right, where were we?"}],
+                 "ads":[{"startMs":10000,"endMs":26000,"signals":["sponsor","code"]}]}
+            """.trimIndent()),
+        )
+
+        subject(transport, Fakes(transcripts = transcripts)).record(
+            episode,
+            ServerEpisode(
+                ready = true, audioUrl = "x", durationMs = null, introMs = null, outroMs = null,
+                detected = true,
+                transcript = ServerTranscriptNote(
+                    source = "published", cues = 4,
+                    url = "https://listen.terst.org/api/transcript?episode=abc",
+                ),
+            ),
+        )
+
+        val marked = transcripts.cueRows.filter { it.advertising }.map { it.startMs }
+        assertEquals(listOf(10_000L, 15_000L), marked.sorted())
+    }
+
+    /** No candidates means no marks, rather than everything or nothing. */
+    @Test
+    fun `leaves every line unmarked when the server found no advertising`() = runTest {
+        val transcripts = FakeTranscriptDao()
+        val transport = StubTransport(
+            null,
+            alsoServing = mapOf("api/transcript" to
+                """{"source":"asr","cues":[{"startMs":0,"endMs":3000,"text":"Only this."}]}"""),
+        )
+
+        subject(transport, Fakes(transcripts = transcripts)).record(
+            episode,
+            ServerEpisode(
+                ready = true, audioUrl = "x", durationMs = null, introMs = null, outroMs = null,
+                detected = true,
+                transcript = ServerTranscriptNote(
+                    source = "asr", cues = 1,
+                    url = "https://listen.terst.org/api/transcript?episode=abc",
+                ),
+            ),
+        )
+
+        assertEquals(emptyList<Long>(), transcripts.cueRows.filter { it.advertising }.map { it.startMs })
+    }
 }
diff --git a/server/cmd/listend/main.go b/server/cmd/listend/main.go
index 0e787e4..495ed03 100644
--- a/server/cmd/listend/main.go
+++ b/server/cmd/listend/main.go
@@ -345,7 +345,7 @@ func (w *worker) pollOne(ctx context.Context, feedURL string) {
 // every poll would be work done for ever to reach the same answer -- and the refusal is itself
 // the record of which episodes speech recognition would have to do.
 func (w *worker) harvestTranscripts(ctx context.Context, show *store.Show, candidates []*store.Episode) {
-	taken, declined := 0, 0
+	taken, declined, spotted := 0, 0, 0
 	for _, e := range candidates {
 		if ctx.Err() != nil {
 			return
@@ -372,6 +372,7 @@ func (w *worker) harvestTranscripts(ctx context.Context, show *store.Show, candi
 			e.TranscriptAtUnix = time.Now().Unix()
 			e.TranscriptNote = ""
 			taken++
+			spotted += result.Ads
 		case result.Note != "":
 			e.TranscriptNote = result.Note
 			e.TranscriptAtUnix = time.Now().Unix()
@@ -379,7 +380,10 @@ func (w *worker) harvestTranscripts(ctx context.Context, show *store.Show, candi
 		}
 	}
 	if taken > 0 || declined > 0 {
-		log.Printf("%s: transcripts -- took %d, declined %d", show.Title, taken, declined)
+		// The ad count is reported, not acted on. These are candidates read out of the words,
+		// and nothing publishes them as ranges to skip -- see transcript.Spot.
+		log.Printf("%s: transcripts -- took %d, declined %d, %d ad candidate(s) in the words",
+			show.Title, taken, declined, spotted)
 	}
 }
 
diff --git a/server/internal/safefetch/live_check_test.go b/server/internal/safefetch/live_check_test.go
index a10d162..cc62c43 100644
--- a/server/internal/safefetch/live_check_test.go
+++ b/server/internal/safefetch/live_check_test.go
@@ -20,7 +20,7 @@ func TestTheLiveSubscriptionsAreAllReachable(t *testing.T) {
 		Feeds []string `json:"feeds"`
 		Shows map[string]struct {
 			Episodes []struct {
-				EnclosureURL  string `json:"enclosureUrl"`
+				EnclosureURL string `json:"enclosureUrl"`
 			} `json:"episodes"`
 		} `json:"shows"`
 	}
diff --git a/server/internal/safefetch/safefetch_test.go b/server/internal/safefetch/safefetch_test.go
index bcb7e20..ec3c84d 100644
--- a/server/internal/safefetch/safefetch_test.go
+++ b/server/internal/safefetch/safefetch_test.go
@@ -32,14 +32,14 @@ func TestLoopbackAndPrivateAddressesAreRefused(t *testing.T) {
 		"0.0.0.0", "::",
 		// Special-purpose ranges that are not "private" but are not the open internet
 		// either, and several of which route somewhere on a real network.
-		"0.1.2.3",        // "this network"
-		"100.64.0.1",     // carrier-grade NAT
-		"192.0.0.1",      // IETF protocol assignments
-		"192.0.2.1",      // documentation
-		"198.18.0.1",     // benchmarking
-		"198.51.100.1",   // documentation
-		"203.0.113.1",    // documentation
-		"240.0.0.1",      // reserved
+		"0.1.2.3",         // "this network"
+		"100.64.0.1",      // carrier-grade NAT
+		"192.0.0.1",       // IETF protocol assignments
+		"192.0.2.1",       // documentation
+		"198.18.0.1",      // benchmarking
+		"198.51.100.1",    // documentation
+		"203.0.113.1",     // documentation
+		"240.0.0.1",       // reserved
 		"255.255.255.255", // broadcast
 		"64:ff9b::7f00:1", // NAT64, carrying 127.0.0.1
 		"2001:db8::1",     // documentation
diff --git a/server/internal/transcript/ads.go b/server/internal/transcript/ads.go
new file mode 100644
index 0000000..f9f9a12
--- /dev/null
+++ b/server/internal/transcript/ads.go
@@ -0,0 +1,145 @@
+package transcript
+
+import (
+	"regexp"
+	"strings"
+)
+
+/*
+Advertising read out of an episode's words.
+
+A second detector, and it finds a class the audio one structurally cannot. The audio scan
+compares episodes mirrored within a minute of each other, so what it finds is the advertising a
+CDN stitched into all of them -- validated against the declared-versus-served length gap. A show
+that inserts nothing has no gap, which by that measure means no advertising, and the host-read
+sponsor block in the middle of it is invisible.
+
+Every published transcript this server can use belongs to exactly such a show, because having no
+inserted advertising is the condition for using one. So the two detectors do not merely fail
+differently; they look at disjoint sets of episodes.
+
+Nothing here is published as a skip. These are candidates, recorded and reported, and the
+listener's labels are what can confirm them -- the same posture the signature library's shadow
+run takes, and for the same reason: a detector that acts on its own first guesses makes
+untrustworthy skips, and a skip into the middle of the programme is worse than no skip.
+*/
+
+// Candidate is a stretch of an episode whose words read as advertising.
+type Candidate struct {
+	StartMs int64 `json:"startMs"`
+	EndMs   int64 `json:"endMs"`
+	// Signals are the kinds of marker that qualified it, so the candidate can be weighed
+	// rather than taken on faith -- the same reason a published break carries its partner count.
+	Signals []string `json:"signals"`
+}
+
+const (
+	// bridgeMs is how far apart two markers may be and still belong to one read. Measured on
+	// the Seahawks Rewind block, where "presented by" and "12% off" are 17.7 seconds apart
+	// with the product description between them; 25 gives that room without reaching the
+	// programme on either side.
+	bridgeMs = 25_000
+	// minCandidateMs is the shortest thing worth calling an advertisement. Advertising is sold
+	// in units of roughly fifteen seconds, the same figure the length-gap rule uses, and
+	// anything briefer is a mention rather than a break -- skipping one would clip the show.
+	minCandidateMs = 15_000
+	// minKinds is how many different kinds of marker must appear. Two, because one kind
+	// repeated says one thing: a host saying "sponsor" four times while describing a
+	// sponsorship deal is the programme, which is the same trap the toc=false chapter rule
+	// documents. Real examples of both are in the tests.
+	minKinds = 2
+)
+
+// marker is one kind of advertising language. Kind rather than phrase, so several wordings of
+// the same thing -- "sponsored by", "brought to you by", "presented by" -- count once.
+type marker struct {
+	kind    string
+	pattern *regexp.Regexp
+}
+
+var markers = []marker{
+	{"sponsor", regexp.MustCompile(
+		`sponsored by|brought to you by|presented by|our sponsor|today's sponsor|this episode is sponsored`)},
+	{"code", regexp.MustCompile(
+		`promo code|promotional code|use code|discount code|code [a-z0-9]+ at checkout|at checkout`)},
+	// "dot com" rather than ".com" because this is spoken text: a transcript writes the URL
+	// the way it was said. The campaign path has to follow a domain -- a bare "slash" matches
+	// ordinary speech, and the corpus has "a children's movie slash book for adults" to prove
+	// it.
+	{"url", regexp.MustCompile(
+		`dot com|dot co dot uk|\.com[/ ]|(dot com|\.com) slash |visit [a-z0-9]+ dot `)},
+	{"offer", regexp.MustCompile(
+		`\d+% off|\d+ percent off|free trial|first month free|save up to|limited time|` +
+			`for a limited|your first order|sign up today|enroll in`)},
+	{"legal", regexp.MustCompile(
+		`terms and conditions|terms apply|restrictions apply|see site for details|while supplies last`)},
+}
+
+// Spot reads an episode's words and reports the stretches that sound like advertising.
+//
+// Density rather than any single phrase. One marker is a mention -- a show is routinely
+// "brought to you by" somebody in a single line of play-by-play, and a guest describing a
+// sponsorship deal says "sponsor" repeatedly -- and both of those are the programme. A real
+// read stacks different kinds of marker within a few seconds: who it is for, where to go, what
+// the offer is, and what the terms are.
+func Spot(cues []Cue) []Candidate {
+	type hit struct {
+		cue   Cue
+		kinds []string
+	}
+	var hits []hit
+	for _, cue := range cues {
+		if kinds := kindsIn(cue.Text); len(kinds) > 0 {
+			hits = append(hits, hit{cue: cue, kinds: kinds})
+		}
+	}
+
+	var found []Candidate
+	for start := 0; start < len(hits); {
+		end := start
+		// Grow the run while the next marker is near enough to belong to the same read.
+		for end+1 < len(hits) && hits[end+1].cue.StartMs-hits[end].cue.EndMs <= bridgeMs {
+			end++
+		}
+
+		kinds := map[string]bool{}
+		for _, h := range hits[start : end+1] {
+			for _, k := range h.kinds {
+				kinds[k] = true
+			}
+		}
+		candidate := Candidate{
+			StartMs: hits[start].cue.StartMs,
+			EndMs:   hits[end].cue.EndMs,
+			Signals: sortedKinds(kinds),
+		}
+		if len(kinds) >= minKinds && candidate.EndMs-candidate.StartMs >= minCandidateMs {
+			found = append(found, candidate)
+		}
+		start = end + 1
+	}
+	return found
+}
+
+func kindsIn(text string) []string {
+	lower := strings.ToLower(text)
+	var kinds []string
+	for _, m := range markers {
+		if m.pattern.MatchString(lower) {
+			kinds = append(kinds, m.kind)
+		}
+	}
+	return kinds
+}
+
+// sortedKinds returns the kinds in the fixed order markers are declared in, so a candidate
+// reads the same way every time it is reported.
+func sortedKinds(present map[string]bool) []string {
+	var out []string
+	for _, m := range markers {
+		if present[m.kind] {
+			out = append(out, m.kind)
+		}
+	}
+	return out
+}
diff --git a/server/internal/transcript/ads_test.go b/server/internal/transcript/ads_test.go
new file mode 100644
index 0000000..5512ff9
--- /dev/null
+++ b/server/internal/transcript/ads_test.go
@@ -0,0 +1,211 @@
+package transcript
+
+import "testing"
+
+/*
+Advertising read out of an episode's words.
+
+A second detector, finding a class the audio one structurally cannot. The audio scan works by
+comparing episodes mirrored within a minute of each other, which finds the advertising a CDN
+stitched into all of them -- and the declared-versus-served length gap validates it. A show that
+inserts nothing has no gap, so by that measure it carries no advertising at all, and the
+host-read sponsor block in the middle of it is invisible. Every published transcript the server
+can use belongs to exactly such a show, because that is the condition for using one.
+
+The cues below are real, from the live corpus harvested 2026-10-28, trimmed.
+*/
+
+// A 25-second host-read block from Seahawks Rewind, which the length gap cannot see because
+// that show inserts nothing: served 4938s against 4938s declared.
+var deltaBlock = []Cue{
+	{2461200, 2463800, "you know, you give away too many turnovers, you get little chance to win."},
+	{2467500, 2470200, "All right, today's game is presented by Delta."},
+	{2470200, 2473400, "12 status is back for its 10th anniversary, rewarding Seahawks"},
+	{2473400, 2476600, "fans all season long. Enroll in 12 status and earn"},
+	{2476600, 2481500, "one SkyMile for every Seahawk passing yard this season. 12"},
+	{2481500, 2485200, "status members also receive priority boarding on Delta flights out"},
+	{2485200, 2488600, "of Seattle-Tacoma International Airport and 12% off the Seahawks Pro"},
+	{2488600, 2492800, "Shop on the 12th of every month. Sign up today, 12status. com."},
+	{2492800, 2495300, "We got B. Walt back from TV. Did you guys"},
+}
+
+func TestFindsAHostReadAdBlock(t *testing.T) {
+	got := Spot(deltaBlock)
+
+	if len(got) != 1 {
+		t.Fatalf("got %d candidates, want 1: %+v", len(got), got)
+	}
+	if got[0].StartMs != 2467500 {
+		t.Errorf("starts at %d, want 2467500 -- the line that opens the read", got[0].StartMs)
+	}
+	if got[0].EndMs != 2492800 {
+		t.Errorf("ends at %d, want 2492800 -- the line that closes it", got[0].EndMs)
+	}
+}
+
+/*
+One mention is not a break.
+
+"The call of the game is brought to you by Beacon Plumbing" is a single line inside play-by-play,
+and the programme carries straight on. Skipping it would cut the game. Real, from the same show.
+*/
+func TestASingleSponsorMentionIsNotABreak(t *testing.T) {
+	cues := []Cue{
+		{3288000, 3290900, "First down."},
+		{3290900, 3292500, "The call of the game is brought to you by"},
+		{3292500, 3296700, "Beacon Plumbing. Stop freaking call Beacon. Calm yourself down, Michael."},
+		{3296700, 3298600, "Let's pause 10 seconds to give our network stations a"},
+		{3298600, 3302400, "chance to identify themselves here on the Columbia Bank Radio Network."},
+	}
+
+	if got := Spot(cues); len(got) != 0 {
+		t.Errorf("got %+v, want nothing -- one tag line is not an ad break", got)
+	}
+}
+
+/*
+A show talking about sponsorship is the programme.
+
+Real, from Wards Way Podcast: a guest describing how a sponsor came forward for his yacht event.
+The same trap the toc=false chapter rule already documents -- a show discussing its own
+sponsorship has a section called "Sponsors" which is the actual content.
+*/
+func TestTalkingAboutASponsorIsNotAnAdvertisement(t *testing.T) {
+	cues := []Cue{
+		{607500, 611700, "We're just gonna run a little conference in a hotel and look from across the water."},
+		{611700, 612100, "Sure."},
+		{612100, 618300, "And he kindly came forward and said, I'd love to get involved, and I'd like to sponsor it."},
+		{618300, 621900, "Why don't you host an event on my super mega yacht?"},
+		{621900, 626200, "To which I said, I'm quite happy to have 150 people on anybody's yacht for a party."},
+	}
+
+	if got := Spot(cues); len(got) != 0 {
+		t.Errorf("got %+v, want nothing -- this is two people talking", got)
+	}
+}
+
+// Two different kinds of marker, not two instances of one: a host saying "sponsor" four times
+// while discussing a sponsorship deal is still the programme.
+func TestOneKindOfMarkerRepeatedIsNotEnough(t *testing.T) {
+	cues := []Cue{
+		{0, 5000, "Our sponsor had a lot to say about that."},
+		{5000, 10000, "The sponsor, as you know, sponsors the whole series."},
+		{10000, 15000, "And that sponsor has been with us a long time."},
+		{15000, 20000, "Anyway, a sponsor is a sponsor."},
+		{20000, 25000, "Let's get back to it."},
+	}
+
+	if got := Spot(cues); len(got) != 0 {
+		t.Errorf("got %+v, want nothing -- one marker repeated says one thing", got)
+	}
+}
+
+// A promo code, a URL and an offer inside half a minute is an advertisement.
+func TestFindsACodeAndOfferRead(t *testing.T) {
+	cues := []Cue{
+		{600000, 604000, "Back in a moment."},
+		{610000, 615000, "This episode is sponsored by a mattress company."},
+		{615000, 620000, "Go to mattress dot com slash listen and use promo code LISTEN"},
+		{620000, 626000, "for 20% off your first order. Terms and conditions apply."},
+		{630000, 634000, "Right, where were we?"},
+	}
+
+	got := Spot(cues)
+	if len(got) != 1 {
+		t.Fatalf("got %d candidates, want 1: %+v", len(got), got)
+	}
+	if got[0].StartMs != 610000 || got[0].EndMs != 626000 {
+		t.Errorf("got %d-%d, want 610000-626000", got[0].StartMs, got[0].EndMs)
+	}
+}
+
+// What was found, so a candidate can be weighed rather than taken on faith -- the same reason a
+// published break carries its partner count.
+func TestACandidateSaysWhatItFound(t *testing.T) {
+	got := Spot(deltaBlock)
+
+	if len(got) != 1 {
+		t.Fatalf("got %d candidates, want 1", len(got))
+	}
+	if len(got[0].Signals) < 2 {
+		t.Errorf("signals = %v, want at least the two kinds that qualified it", got[0].Signals)
+	}
+}
+
+/*
+Two breaks in one episode are two candidates.
+
+They must not be bridged into one. The span between them is the programme, and a single
+candidate covering both would be a skip over the middle of the show.
+*/
+func TestTwoSeparateBreaksStaySeparate(t *testing.T) {
+	cues := []Cue{
+		{60000, 65000, "This episode is sponsored by a mattress company."},
+		{65000, 72000, "They have been making mattresses for forty years and they are very good."},
+		{72000, 78000, "Go to mattress dot com and use promo code LISTEN for 20% off."},
+		{300000, 305000, "The programme carries on for several minutes here."},
+		{900000, 905000, "Today's show is brought to you by a razor company."},
+		{905000, 912000, "Five blades, which is more blades than anyone has ever had."},
+		{912000, 918000, "Visit razors dot com slash listen, terms and conditions apply."},
+	}
+
+	got := Spot(cues)
+	if len(got) != 2 {
+		t.Fatalf("got %d candidates, want 2: %+v", len(got), got)
+	}
+	if got[0].EndMs >= got[1].StartMs {
+		t.Errorf("the two candidates overlap: %+v", got)
+	}
+}
+
+// A blip is not a break. Nothing shorter than an advertisement is worth skipping, and skipping
+// it would clip the programme on either side.
+func TestSomethingTooShortToBeAnAdvertisementIsNotOne(t *testing.T) {
+	cues := []Cue{
+		{0, 2000, "Sponsored by a mattress company, mattress dot com."},
+		{2000, 6000, "Anyway."},
+	}
+
+	if got := Spot(cues); len(got) != 0 {
+		t.Errorf("got %+v, want nothing -- two seconds is not an ad read", got)
+	}
+}
+
+func TestAnEmptyTranscriptHasNoAdvertising(t *testing.T) {
+	if got := Spot(nil); len(got) != 0 {
+		t.Errorf("got %+v", got)
+	}
+}
+
+/*
+"Slash" is an ordinary English word.
+
+Real, from the corpus: "is a children's movie slash book for four or adults". A campaign path is
+read aloud after a domain, so that is what the pattern requires -- a bare slash matches people
+talking.
+*/
+func TestSlashInOrdinarySpeechIsNotAURL(t *testing.T) {
+	cues := []Cue{
+		{0, 6000, "It is a children's movie slash book for adults, really."},
+		{6000, 12000, "Our sponsor had opinions about that, as you can imagine."},
+		{12000, 20000, "Anyway, the sponsor of this whole discussion is beside the point."},
+	}
+
+	if got := Spot(cues); len(got) != 0 {
+		t.Errorf("got %+v, want nothing -- \"movie slash book\" is not a web address", got)
+	}
+}
+
+// The real form, which does follow a domain.
+func TestACampaignPathAfterADomainIsAURL(t *testing.T) {
+	cues := []Cue{
+		{0, 5000, "This week we are brought to you by a mattress company."},
+		{5000, 11000, "They have been making mattresses for forty years and they are very good."},
+		{11000, 18000, "Go to mattress dot com slash listen for your first order."},
+	}
+
+	got := Spot(cues)
+	if len(got) != 1 {
+		t.Fatalf("got %d candidates, want 1: %+v", len(got), got)
+	}
+}
diff --git a/server/internal/transcript/harvest.go b/server/internal/transcript/harvest.go
index 6369ea3..64130fd 100644
--- a/server/internal/transcript/harvest.go
+++ b/server/internal/transcript/harvest.go
@@ -30,6 +30,8 @@ type Offer struct {
 type Result struct {
 	Stored bool
 	Cues   int
+	// Ads is how many stretches of the stored transcript read as advertising. Candidates only.
+	Ads int
 	// Note is why nothing was stored, and is empty when something was. It is a verdict about
 	// this copy of the episode, so it is recorded rather than retried -- and for an episode
 	// carrying inserted advertising it is also the signal that this is one recognition has to do.
@@ -76,11 +78,19 @@ func (h Harvester) Harvest(ctx context.Context, offer Offer) (Result, error) {
 		return Result{Note: verdict.Why}, nil
 	}
 
-	stored := Transcript{EpisodeID: offer.EpisodeID, Source: SourcePublished, Cues: cues}
+	stored := Transcript{
+		EpisodeID: offer.EpisodeID,
+		Source:    SourcePublished,
+		Cues:      cues,
+		// Read for advertising now rather than on demand: it is cheap against text, and a
+		// candidate computed at harvest is a candidate that exists offline on the phone with
+		// the words it belongs to.
+		Ads: Spot(cues),
+	}
 	if err := h.Store.Put(stored); err != nil {
 		return Result{}, err
 	}
-	return Result{Stored: true, Cues: len(cues)}, nil
+	return Result{Stored: true, Cues: len(cues), Ads: len(stored.Ads)}, nil
 }
 
 func (h Harvester) download(ctx context.Context, url string) (string, error) {
diff --git a/server/internal/transcript/harvest_test.go b/server/internal/transcript/harvest_test.go
index 23a192a..25327ae 100644
--- a/server/internal/transcript/harvest_test.go
+++ b/server/internal/transcript/harvest_test.go
@@ -152,3 +152,51 @@ func TestAnEpisodeOfferingNoTranscriptIsNotFetched(t *testing.T) {
 // something only it can reach and hand the result to the phone as an episode's words. A
 // httptest server is on loopback by definition, so the tests supply the Client seam.
 func loopbackClient() *http.Client { return &http.Client{Timeout: 30 * time.Second} }
+
+/*
+A stored transcript carries what its words say about advertising.
+
+Computed at harvest rather than on demand: it is cheap against text, and a candidate worked out
+here is one that exists on the phone, offline, beside the words it belongs to.
+
+Candidates, never skips. See Spot.
+*/
+func TestAHarvestedTranscriptCarriesItsAdCandidates(t *testing.T) {
+	const withAnAd = `1
+00:10:00,000 --> 00:10:05,000
+This episode is sponsored by a mattress company.
+
+2
+00:10:05,000 --> 00:10:11,000
+They have been making mattresses for forty years and they are good.
+
+3
+00:10:11,000 --> 00:10:18,000
+Go to mattress dot com slash listen and use promo code LISTEN for 20% off.
+
+4
+00:20:00,000 --> 00:20:04,000
+Right, where were we?
+`
+	server := serving(t, 200, withAnAd, "application/srt")
+	store := Store{Dir: t.TempDir()}
+	harvester := Harvester{Store: store, Client: loopbackClient()}
+
+	got, err := harvester.Harvest(context.Background(), Offer{
+		EpisodeID: "ep1", URL: server.URL, Type: "application/srt",
+		ServedMs: 1_217_000, DeclaredMs: 1_216_000,
+	})
+	if err != nil {
+		t.Fatalf("Harvest: %v", err)
+	}
+	if got.Ads != 1 {
+		t.Fatalf("Ads = %d, want 1", got.Ads)
+	}
+	held, err := store.Get("ep1")
+	if err != nil {
+		t.Fatalf("Get: %v", err)
+	}
+	if len(held.Ads) != 1 || held.Ads[0].StartMs != 600_000 {
+		t.Errorf("stored ads = %+v", held.Ads)
+	}
+}
diff --git a/server/internal/transcript/store.go b/server/internal/transcript/store.go
index d7e86a4..8cf9419 100644
--- a/server/internal/transcript/store.go
+++ b/server/internal/transcript/store.go
@@ -28,6 +28,10 @@ type Transcript struct {
 	// LanguageCode is empty when nobody said. Recognition reports what it detected.
 	LanguageCode string `json:"languageCode,omitempty"`
 	Cues         []Cue  `json:"cues"`
+	// Ads are the stretches whose words read as advertising. Candidates, not skips: they are
+	// stored with the transcript and shown beside it, and nothing publishes them as ranges to
+	// jump over. See Spot for why, and for what the listener's labels are for.
+	Ads []Candidate `json:"ads,omitempty"`
 }
 
 // LastEndMs is where the words stop.