feat(info): refresh episode details and show notes

app/README.md

@@ -104,6 +104,8 @@ Routes include onboarding, home, learn, briefing, settings, debug, explore, libr
 
 `StartupWorkGate` is process-scoped and releases optional cloud coordinator initialization, periodic sync scheduling, Smart Downloads scheduling/catch-up, foreground subscription sync and legacy RSS repair after the initial screen commits. Optional initialization uses a suspending mutex so ready does not release a burst of initializers. Home signals after loaded local content and two frames; other initial routes release after their first two frames. A headless worker/widget launch releases at the first idle main queue; `MainActivity` identifies a UI launch before creating content. If the first Activity stops before content is ready, background startup also releases; configuration recreation waits for the replacement UI. Recreated roots reuse the same gate. Repeated signals and cancelled initializers are safe (`StartupWorkGateTest`). Adaptive ranking telemetry also waits for readiness. Appearance flows use restored fast-cache values while background hydration fills missing DataStore keys; existing choices are never overwritten. Play update checks begin after the first two Compose frames. Playback/session restoration and preference identity remain owned by the existing composition root.
 
+- Release runtime dependency snapshots include Jsoup through `:core:catalog`, which owns shared show-notes resource and chapter parsing. No direct parser-library dependency is added here.
+
 ## Threading / lifecycle
 
 - `AppContainer` is created once from `BoxLoreApplication.onCreate` and is application-scoped.

app/dependencies/releaseRuntimeClasspath.txt

@@ -292,6 +292,7 @@ org.jetbrains.kotlinx:kotlinx-serialization-core:1.7.3
 org.jetbrains.kotlinx:kotlinx-serialization-json-jvm:1.7.3
 org.jetbrains.kotlinx:kotlinx-serialization-json:1.7.3
 org.jetbrains:annotations:23.0.0
+org.jsoup:jsoup:1.23.2
 org.jspecify:jspecify:1.0.0
 sh.calvin.reorderable:reorderable-android:2.5.1
 sh.calvin.reorderable:reorderable:2.5.1

core/catalog/README.md

@@ -36,6 +36,9 @@ Failed consent writes keep activation off, and local notification flags prevent
 
 - The two-request concurrency cap applies to publisher-feed refreshes. Podcast Index chunk sync for not-ready shows and missing-feed-URL recovery retain their independent six-request limit.
 
+- `shownotes/ShowNotesParser` interprets publisher HTML with Jsoup without fetching pages. It preserves meaningful anchor labels, resolves relative URLs only with a known base, classifies real domain boundaries and routes, filters infrastructure/media links, and deduplicates tracking variants while retaining case-sensitive paths, query values, and fragments. `EpisodeLinkTitles` removes destination-platform labels (including aliases and generic actions) so they cannot override a URL's profile handle. Social handles use platform-specific profile routes; login, sharing, and other reserved routes are not usernames. Description chapters require at least two valid timestamp/title lines; generated seek links never rewrite existing anchors, attributes, code, or prose clock times. `ChapterRepository` delegates description parsing to this shared parser.
+- `crosspromo/CrossPromotionDetector` requires explicit feed-exchange wording or corroborated introduction/description evidence; Bonus metadata and missing episode numbering do not establish a promotion. A named introduction can be corroborated by a matching listening/following invitation even in long sentences with weekday schedules, without relying on short duration or Trailer metadata. Self-show comparison uses normalized whole names (including `and`/`&`) rather than substrings. `CrossPromotionResolver` retains the full extracted name, excludes the host, rejects ambiguous matches, and prefers an unambiguous Apple Podcasts destination. A lone unnamed Apple link must resolve to a matching title before it can be used; unrelated links fall back to search. Cancellation propagates and transient search failures remain retryable. Its bounded cache expires empty searches after one minute and successful matches after one hour.
+
 ## Internal structure
 
 ```text
@@ -58,6 +61,14 @@ src/main/java/cx/aswin/boxlore/core/catalog/
   EpisodeMapper.kt
   SubscriptionRepository.kt
   ChapterRepository.kt
+  shownotes/
+    ShowNotesParser.kt
+    EpisodeLinkTitles.kt
+    DescriptionChapters.kt
+    EpisodeLinkClassifier.kt
+  crosspromo/
+    CrossPromotionDetector.kt
+    CrossPromotionResolver.kt
   TranscriptRepository.kt
   SharedAppDependencies.kt
   SubscriptionForegroundSync.kt
@@ -89,6 +100,8 @@ Main Kotlin files should remain below 1000 lines; extracted helpers keep reposit
 - Libraries: Retrofit, OkHttp, Gson, coroutines, DataStore, Firebase Database, Firebase Messaging, and Install Referrer.
 - Reverse-edge rule: catalog must not depend on playback, downloads, designsystem, analytics, or feature modules.
 
+- Show-notes HTML parsing uses Jsoup; the library stays inside `:core:catalog` and shared values remain in `:core:model`.
+
 ## Threading / lifecycle
 
 - Production repositories are application-scoped through `AppContainer` and `SharedAppDependenciesHolder`.
@@ -114,6 +127,8 @@ Main Kotlin files should remain below 1000 lines; extracted helpers keep reposit
 ./gradlew :core:catalog:testDebugUnitTest --tests 'cx.aswin.boxlore.core.catalog.PodcastRepositoryCatalogTest'
 ```
 
+- `ShowNotesParserTest`, `CrossPromotionDetectorTest`, and `CrossPromotionResolverTest` cover malformed/relative anchors, platform-label/handle separation, reserved profile routes, destination purpose and spoof domains, URL deduplication, safe timestamp linking, chapter boundaries, weekday-scheduled full-length introductions, false promotions and shorter-name prefixes, full-title matching, ambiguous/host matches, verified unnamed Apple links, retry, cancellation, and cache expiry.
+
 ## CI relevance
 
 - `unit-tests.yml` runs catalog JVM tests.

core/catalog/build.gradle.kts

@@ -95,6 +95,7 @@ dependencies {
 
     // JSON Streaming
     implementation(libs.gson)
+    implementation(libs.jsoup)
     implementation(libs.okhttp)
     // Firebase (database and messaging — SubscriptionRepository uses firebase.database + messaging)
     implementation(platform(libs.firebase.bom))

core/catalog/dependencies/releaseRuntimeClasspath.txt

@@ -124,4 +124,5 @@ org.jetbrains.kotlinx:kotlinx-serialization-core:1.6.3
 org.jetbrains.kotlinx:kotlinx-serialization-json-jvm:1.6.3
 org.jetbrains.kotlinx:kotlinx-serialization-json:1.6.3
 org.jetbrains:annotations:23.0.0
+org.jsoup:jsoup:1.23.2
 org.jspecify:jspecify:1.0.0

core/catalog/src/main/java/cx/aswin/boxlore/core/catalog/ChapterRepository.kt

@@ -3,6 +3,7 @@ package cx.aswin.boxlore.core.catalog
 import cx.aswin.boxlore.core.model.Chapter
 import cx.aswin.boxlore.core.model.Episode
 import java.net.URL
+import kotlinx.coroutines.CancellationException
 import kotlinx.coroutines.Dispatchers
 import kotlinx.coroutines.withContext
 import org.json.JSONObject
@@ -61,6 +62,8 @@ object ChapterRepository {
                 cache[normalizedUrl] = chapters
             }
             chapters
+        } catch (cancelled: CancellationException) {
+            throw cancelled
         } catch (e: Exception) {
             android.util.Log.w("ChapterRepo", "Failed to fetch chapters: $normalizedUrl", e)
             emptyList()
@@ -141,72 +144,8 @@ object ChapterRepository {
      * Parses chapter timestamps from the episode description.
      * Supports both hh:mm:ss and mm:ss formats, and detects timestamps at either start or end of lines.
      */
-    fun parseChaptersFromDescription(htmlDescription: String?): List<Chapter> {
-        if (htmlDescription.isNullOrEmpty()) return emptyList()
-
-        try {
-            // Replace HTML paragraph, line break, and list tags with newlines
-            val cleanText = htmlDescription
-                .replace("(?i)<p>".toRegex(), "\n")
-                .replace("(?i)</p>".toRegex(), "\n")
-                .replace("(?i)<br\\s*/?>".toRegex(), "\n")
-                .replace("(?i)<li>".toRegex(), "\n")
-                .replace("<[^>]+>".toRegex(), "") // Strip all other HTML tags
-
-            val lines = cleanText.split("\n")
-            val chapters = mutableListOf<Chapter>()
-
-            // Regex to find timestamps (e.g. 12:34 or 1:12:34)
-            val timestampRegex = """\b(?:(\d{1,2}):)?(\d{1,2}):(\d{2})\b""".toRegex()
-
-            for (line in lines) {
-                val trimmed = line.trim()
-                val match = timestampRegex.find(trimmed) ?: continue
-
-                val hours = match.groups[1]?.value?.toIntOrNull() ?: 0
-                val minutes = match.groups[2]?.value?.toIntOrNull() ?: 0
-                val seconds = match.groups[3]?.value?.toIntOrNull() ?: 0
-
-                if (minutes >= 60 || seconds >= 60) continue
-
-                // Determine whether the title is before or after the timestamp
-                val matchStart = match.range.first
-                val matchEnd = match.range.last + 1
-
-                val rawTitle = if (matchStart > trimmed.length - matchEnd) {
-                    trimmed.substring(0, matchStart)
-                } else {
-                    trimmed.substring(matchEnd)
-                }
-
-                // Clean up any remaining braces, hyphens, colons, or whitespace around the title
-                val cleanTitle = rawTitle.trim()
-                    .replace("^[^\\p{L}\\p{N}]+".toRegex(), "") // Strip leading non-alphanumeric chars
-                    .replace("[^\\p{L}\\p{N}]+$".toRegex(), "") // Strip trailing non-alphanumeric chars
-                    .trim()
-
-                if (cleanTitle.isNotEmpty()) {
-                    val startTime = hours * 3600.0 + minutes * 60.0 + seconds.toDouble()
-                    chapters.add(
-                        Chapter(
-                            startTime = startTime,
-                            title = cleanTitle
-                        )
-                    )
-                }
-            }
-
-            // Safeguard: only return if we find at least 2 valid chapters (avoid random standalone timestamp false positives)
-            return if (chapters.size >= 2) {
-                chapters.sortedBy { it.startTime }
-            } else {
-                emptyList()
-            }
-        } catch (e: Exception) {
-            android.util.Log.w("ChapterRepo", "Failed to parse chapters from description", e)
-            return emptyList()
-        }
-    }
+    fun parseChaptersFromDescription(htmlDescription: String?): List<Chapter> =
+        cx.aswin.boxlore.core.catalog.shownotes.ShowNotesParser.parse(htmlDescription).chapters
 
     private fun JSONObject.optStringOrNull(name: String, fallback: String? = null): String? {
         if (isNull(name)) return fallback