diff --git a/.gitignore b/.gitignore
index 3f89d8b..7b7b27e 100644
--- a/.gitignore
+++ b/.gitignore
@@ -28,7 +28,7 @@ htmlcov/
 
 # Internal planning docs (ce:plan output) — keep local, don't publish
 docs/plans/
+.context/
 
 /work
 /print
-
diff --git a/CONFIGURATION.md b/CONFIGURATION.md
index eed672a..41b6e8e 100644
--- a/CONFIGURATION.md
+++ b/CONFIGURATION.md
@@ -57,7 +57,7 @@ The project-scoped file is the cleanest pattern for **per-client setups**: drop
 | YouTube | `yt-dlp` CLI installed | always on if `yt-dlp` present | yes |
 | X / Twitter | one of: `AUTH_TOKEN` + `CT0` (browser cookies, Bird CLI), `XAI_API_KEY`, `SCRAPECREATORS_API_KEY`, or `FROM_BROWSER` (cookie-jar auth) | X items in results | cookie-jar / Bird = free; xAI / ScrapeCreators = paid |
 | TikTok | `SCRAPECREATORS_API_KEY` + `INCLUDE_SOURCES` contains `tiktok` | TikTok items | 10K free calls |
-| Instagram | `SCRAPECREATORS_API_KEY` + `INCLUDE_SOURCES` contains `instagram` | Instagram Reels | 10K free calls |
+| Instagram | `SCRAPECREATORS_API_KEY` + `INCLUDE_SOURCES` contains `instagram` | Instagram Reels | 10K free calls; raise `LAST30DAYS_TRANSCRIPT_TIMEOUT` (default 30s) if SC is slow on your network |
 | Threads | `SCRAPECREATORS_API_KEY` + `INCLUDE_SOURCES` contains `threads` | Threads items | 10K free calls |
 | Pinterest | `SCRAPECREATORS_API_KEY` + `INCLUDE_SOURCES` contains `pinterest` | Pinterest items | 10K free calls |
 | Bluesky | `BSKY_HANDLE` + `BSKY_APP_PASSWORD` | Bluesky items | yes (app password at bsky.app) |
@@ -93,6 +93,16 @@ After editing: `chmod 600 ~/.config/last30days/.env` (or `chmod 600 .claude/last
 
 **Troubleshooting:** if a source you expected to see isn't appearing in results, run `python3 scripts/last30days.py --diagnose`. It prints a per-source availability report (which keys were detected, which CLIs are installed, which backends are reachable) without running a full search.
 
+### Bluesky app-password format and search host
+
+`BSKY_APP_PASSWORD` should be a 19-char app password in `xxxx-xxxx-xxxx-xxxx` format (lowercase alphanumeric, three hyphens). Generate one at <https://bsky.app/settings/app-passwords>. The AT Protocol's `createSession` endpoint also accepts your main account login password, but that's bad hygiene — main passwords have no scope (an app password can be limited to non-DM access) and can't be revoked individually.
+
+The skill defaults to `api.bsky.app` for `searchPosts`, which is the canonical authenticated AppView. The previous default `public.api.bsky.app` is the unauthenticated public mirror and is currently blocked by BunnyCDN for `searchPosts` regardless of auth header (verified 2026-05-04). If Bluesky migrates infrastructure again, override the host without a code change by setting `BSKY_SEARCH_HOST` in your `.env`:
+
+```bash
+BSKY_SEARCH_HOST=api.bsky.app   # default — change only if Bluesky moves
+```
+
 ---
 
 ## Reasoning provider priority
@@ -131,6 +141,8 @@ The default behavior - one slug-named file per topic, overwritten on rerun - is
 
 Adding `--store` to any run persists every finding to a SQLite database (default at `~/.local/share/last30days/research.db`). Findings dedupe on the `source_url` column (UNIQUE constraint), so the same URL across runs updates the existing row instead of creating a duplicate. The markdown file still saves; the SQLite is the time-series substrate.
 
+**Always-on alternative:** set `LAST30DAYS_STORE=1` in your `.env` instead of remembering `--store` on every invocation. The flag still works as before; the env var is purely additive. Same hybrid pattern as `LAST30DAYS_DEBUG` — works whether shell-exported or in `.env`.
+
 Relevant tables: `topics`, `research_runs`, `findings`, `settings`. Schema: [`scripts/store.py`](skills/last30days/scripts/store.py).
 
 ### `watchlist.py` - recurring topics
diff --git a/README.md b/README.md
index df6abbc..21adda2 100644
--- a/README.md
+++ b/README.md
@@ -17,6 +17,7 @@ This README tracks the current v3 pipeline. The runtime skill spec lives in [ski
 **Claude Code (recommended — auto-updates via marketplace):**
 ```
 /plugin marketplace add mvanhorn/last30days-skill
+/plugin install last30days
 ```
 
 **Codex, Cursor, Copilot, Gemini CLI, or any of 50+ [Agent Skills](https://agentskills.io) hosts:**
diff --git a/skills/last30days/scripts/last30days.py b/skills/last30days/scripts/last30days.py
index 7aaa298..4031b9d 100644
--- a/skills/last30days/scripts/last30days.py
+++ b/skills/last30days/scripts/last30days.py
@@ -908,7 +908,17 @@ def main() -> int:
         suppress_web_promo=bool(external_plan or comp_plan),
     )
     _write_last_run(topic, report)
-    if args.store:
+    # LAST30DAYS_STORE env var = persistence default-on. Read both os.environ
+    # (for shell-exported users) and config (for users who set it in
+    # ~/.config/last30days/.env, which env.py loads but does not propagate
+    # to os.environ). Mirrors the LAST30DAYS_DEBUG / LAST30DAYS_SKIP_PREFLIGHT
+    # convention; env-var or config wins, with `--store` flag still working.
+    _store_env = (
+        os.environ.get("LAST30DAYS_STORE")
+        or config.get("LAST30DAYS_STORE")
+        or ""
+    ).lower()
+    if args.store or _store_env in ("1", "true", "yes"):
         counts = persist_report(report)
         sys.stderr.write(
             f"[last30days] Stored {counts['new']} new, {counts['updated']} updated findings\n"
@@ -922,6 +932,7 @@ def main() -> int:
         # degraded-YouTube failure mode (videos returned but transcripts
         # silently failed - typically a stale yt-dlp binary).
         youtube_items = report.items_by_source.get("youtube") or []
+        instagram_items = report.items_by_source.get("instagram") or []
         research_results = {
             "youtube_videos_count": len(youtube_items),
             "youtube_transcripts_count": sum(
@@ -930,6 +941,17 @@ def main() -> int:
             ),
             "youtube_error": report.errors_by_source.get("youtube"),
             "x_error": report.errors_by_source.get("x"),
+            # Captions-disabled videos can never produce a transcript regardless
+            # of yt-dlp version; subtract them from the degraded-ratio
+            # denominator so a single uploader-disabled video does not trip the
+            # "stale yt-dlp" nudge.
+            "youtube_captions_disabled_count": sum(
+                1 for it in youtube_items if it.metadata.get("captions_disabled")
+            ),
+            # Track Instagram returned-zero-items so quality_nudge can detect
+            # the silent-failure case (SC configured but the v2 reels endpoint
+            # 500'd through both the original query and the hashtag retry).
+            "instagram_items_count": len(instagram_items),
         }
         quality = quality_nudge.compute_quality_score(config, research_results)
         if quality.get("nudge_text"):
diff --git a/skills/last30days/scripts/lib/bluesky.py b/skills/last30days/scripts/lib/bluesky.py
index 7c3436a..6771052 100644
--- a/skills/last30days/scripts/lib/bluesky.py
+++ b/skills/last30days/scripts/lib/bluesky.py
@@ -1,10 +1,19 @@
 """Bluesky search via AT Protocol (requires app password).
 
-Uses bsky.social for auth and public.api.bsky.app for post search.
-Requires BSKY_HANDLE and BSKY_APP_PASSWORD env vars.
+Uses bsky.social for auth and api.bsky.app for post search (the canonical
+authenticated AppView). The previous default `public.api.bsky.app` is the
+unauthenticated public mirror, which BunnyCDN now blocks for searchPosts
+regardless of auth header (verified 2026-05-04). Override the search host
+via BSKY_SEARCH_HOST env var if Bluesky migrates infrastructure again.
+
+Requires BSKY_HANDLE and BSKY_APP_PASSWORD env vars. App passwords are
+19-char xxxx-xxxx-xxxx-xxxx; generate at bsky.app/settings/app-passwords.
+The createSession endpoint accepts main-account passwords too, but they're
+bad hygiene (no scope, can't revoke individually).
 """
 
 import math
+import os
 import re
 import sys
 import time
@@ -14,7 +23,65 @@ from typing import Any, Dict, List, Optional
 from . import http, log
 
 BSKY_SESSION_URL = "https://bsky.social/xrpc/com.atproto.server.createSession"
-BSKY_SEARCH_URL = "https://public.api.bsky.app/xrpc/app.bsky.feed.searchPosts"
+_DEFAULT_BSKY_SEARCH_HOST = "api.bsky.app"
+BSKY_SEARCH_URL = f"https://{_DEFAULT_BSKY_SEARCH_HOST}/xrpc/app.bsky.feed.searchPosts"
+
+
+def _resolve_search_url(config: Optional[Dict[str, Any]] = None) -> str:
+    """Resolve the Bluesky search URL with BSKY_SEARCH_HOST override.
+
+    Default is api.bsky.app. Override via BSKY_SEARCH_HOST in shell env or
+    .env file. The project's env.py loads .env into config but not into
+    os.environ, so check both — same hybrid pattern as last30days.py for
+    LAST30DAYS_STORE.
+
+    Hardens user-supplied host values against three common mis-configurations:
+    whitespace (e.g. " api.bsky.app "), embedded path components (e.g.
+    "api.bsky.app/xrpc/proxy") that would double the /xrpc/ segment, and
+    embedded scheme prefixes (e.g. "https://api.bsky.app"). On any of these
+    we log a warning and fall back to the default rather than building an
+    invalid URL with an opaque downstream error.
+    """
+    config = config or {}
+    raw = (
+        os.environ.get("BSKY_SEARCH_HOST")
+        or config.get("BSKY_SEARCH_HOST")
+        or _DEFAULT_BSKY_SEARCH_HOST
+    )
+    host = raw.strip().rstrip("/")
+    # Strip embedded scheme so users who paste full URLs do not break the f-string.
+    for prefix in ("https://", "http://"):
+        if host.lower().startswith(prefix):
+            host = host[len(prefix):]
+            break
+    if not host or "/" in host or " " in host:
+        # Embedded path or whitespace remains — don't trust it. Default + log.
+        if raw != _DEFAULT_BSKY_SEARCH_HOST:
+            _log(
+                f"BSKY_SEARCH_HOST={raw!r} is not a bare hostname; "
+                f"falling back to default {_DEFAULT_BSKY_SEARCH_HOST!r}"
+            )
+        host = _DEFAULT_BSKY_SEARCH_HOST
+    return f"https://{host}/xrpc/app.bsky.feed.searchPosts"
+
+
+# App-password format: xxxx-xxxx-xxxx-xxxx (19 chars, lowercase alphanumeric
+# with three hyphens at fixed positions).
+_APP_PASSWORD_RE = re.compile(r"^[a-z0-9]{4}-[a-z0-9]{4}-[a-z0-9]{4}-[a-z0-9]{4}$")
+
+
+def _validate_app_password_format(value) -> bool:
+    """Return True if value matches Bluesky's 19-char app-password format.
+
+    False for non-strings (None, int, list) so callers passing config dict
+    values directly don't crash. Detect-but-not-gate: the createSession
+    endpoint also accepts main-account passwords, so failing this check is
+    a hygiene smell, not a hard error.
+    """
+    if not isinstance(value, str):
+        return False
+    return bool(_APP_PASSWORD_RE.fullmatch(value))
+
 
 DEPTH_CONFIG = {
     "quick": 15,
@@ -144,6 +211,20 @@ def search_bluesky(
     if not handle or not app_password:
         return {"posts": [], "error": "Bluesky credentials not configured"}
 
+    # One-shot hygiene warning if BSKY_APP_PASSWORD is not in app-password
+    # form. createSession accepts main-account passwords too — but main
+    # passwords have no scope (full account access), can't be revoked
+    # individually, and rotating them breaks every service that holds them.
+    # We warn but do not gate, matching the project's detect-don't-block
+    # philosophy elsewhere.
+    if not _validate_app_password_format(app_password):
+        _log(
+            "BSKY_APP_PASSWORD does not look like an app password "
+            "(expected xxxx-xxxx-xxxx-xxxx, 19 chars). It may be a main "
+            "account password — those work but are bad hygiene. Generate "
+            "an app password at https://bsky.app/settings/app-passwords"
+        )
+
     count = DEPTH_CONFIG.get(depth, DEPTH_CONFIG["default"])
     core_topic = _extract_core_subject(topic)
 
@@ -155,7 +236,7 @@ def search_bluesky(
         "limit": str(min(count, 100)),
         "sort": "top",
     }
-    url = f"{BSKY_SEARCH_URL}?{urlencode(params)}"
+    url = f"{_resolve_search_url(config)}?{urlencode(params)}"
 
     def _auth_and_search() -> tuple[Optional[Dict[str, Any]], Optional[str]]:
         token = _create_session(handle, app_password)
diff --git a/skills/last30days/scripts/lib/env.py b/skills/last30days/scripts/lib/env.py
index 56fb84e..ec20bb2 100644
--- a/skills/last30days/scripts/lib/env.py
+++ b/skills/last30days/scripts/lib/env.py
@@ -314,6 +314,7 @@ def get_config() -> dict[str, Any]:
         ('LAST30DAYS_RERANK_MODEL', None),
         ('LAST30DAYS_X_MODEL', None),
         ('LAST30DAYS_X_BACKEND', None),
+        ('LAST30DAYS_STORE', None),
         ('OPENAI_MODEL_PIN', None),
         ('XAI_MODEL_PIN', None),
         ('SCRAPECREATORS_API_KEY', None),
@@ -322,6 +323,7 @@ def get_config() -> dict[str, Any]:
         ('CT0', None),
         ('BSKY_HANDLE', None),
         ('BSKY_APP_PASSWORD', None),
+        ('BSKY_SEARCH_HOST', None),
         ('TRUTHSOCIAL_TOKEN', None),
         ('BRAVE_API_KEY', None),
         ('EXA_API_KEY', None),
@@ -334,11 +336,22 @@ def get_config() -> dict[str, Any]:
         ('INCLUDE_SOURCES', ''),
         ('EXCLUDE_SOURCES', ''),
         ('LAST30DAYS_YOUTUBE_SSH_HOST', None),
+        ('LAST30DAYS_TRANSCRIPT_TIMEOUT', None),
     ]
 
     for key, default in keys:
         config[key] = os.environ.get(key) or merged_env.get(key, default)
 
+    # Backward-compat: ScrapeCreators' own examples and tutorials use the
+    # SCRAPE_CREATORS_API_KEY spelling (with underscore between SCRAPE and
+    # CREATORS). Accept that form too so users who follow the vendor's docs
+    # don't silently end up with has_scrapecreators=False. Canonical name
+    # wins when both are set.
+    if not config.get('SCRAPECREATORS_API_KEY'):
+        legacy = os.environ.get('SCRAPE_CREATORS_API_KEY') or merged_env.get('SCRAPE_CREATORS_API_KEY')
+        if legacy:
+            config['SCRAPECREATORS_API_KEY'] = legacy
+
     # Track which config source was used (highest-priority file source wins
     # the label; keychain is only reported when nothing else is configured).
     if project_env_path:
diff --git a/skills/last30days/scripts/lib/instagram.py b/skills/last30days/scripts/lib/instagram.py
index 0ffd48b..ab69b6b 100644
--- a/skills/last30days/scripts/lib/instagram.py
+++ b/skills/last30days/scripts/lib/instagram.py
@@ -7,12 +7,14 @@ Requires SCRAPECREATORS_API_KEY in config. 100 free API calls, then PAYG.
 API docs: https://scrapecreators.com/docs
 """
 
+import os
 import re
 import sys
 from datetime import datetime
 from typing import Any, Dict, List, Optional, Set
 
 from . import dates, http, log
+from .relevance import token_overlap_relevance as _compute_relevance
 
 SCRAPECREATORS_BASE = "https://api.scrapecreators.com"
 
@@ -26,7 +28,42 @@ DEPTH_CONFIG = {
 # Max words to keep from each caption
 CAPTION_MAX_WORDS = 500
 
-from .relevance import token_overlap_relevance as _compute_relevance
+# Default transcript fetch timeout (seconds). SC's
+# /v2/instagram/media/transcript regularly takes >15s on real workloads,
+# so the default is generous; override via LAST30DAYS_TRANSCRIPT_TIMEOUT.
+DEFAULT_TRANSCRIPT_TIMEOUT = 30
+
+
+def _resolve_transcript_timeout(
+    timeout: Optional[float] = None,
+    config: Optional[Dict[str, Any]] = None,
+) -> float:
+    """Resolve the IG transcript-fetch timeout.
+
+    Priority (highest wins):
+      1. Explicit ``timeout`` kwarg
+      2. ``LAST30DAYS_TRANSCRIPT_TIMEOUT`` in os.environ
+      3. ``LAST30DAYS_TRANSCRIPT_TIMEOUT`` in caller-supplied config dict
+      4. ``DEFAULT_TRANSCRIPT_TIMEOUT`` (30s)
+
+    Mirrors the ``os.environ.get(X) or config.get(X)`` pattern used for
+    LAST30DAYS_STORE in last30days.py so the env var works whether it's
+    shell-exported or set in ~/.config/last30days/.env.
+    """
+    if timeout is not None:
+        try:
+            return float(timeout)
+        except (TypeError, ValueError):
+            pass
+    raw = os.environ.get("LAST30DAYS_TRANSCRIPT_TIMEOUT")
+    if not raw and config:
+        raw = config.get("LAST30DAYS_TRANSCRIPT_TIMEOUT")
+    if raw:
+        try:
+            return float(raw)
+        except (TypeError, ValueError):
+            pass
+    return float(DEFAULT_TRANSCRIPT_TIMEOUT)
 
 
 def _extract_core_subject(topic: str) -> str:
@@ -44,6 +81,17 @@ def _extract_core_subject(topic: str) -> str:
     return extract_core_subject(topic, noise=_INSTAGRAM_NOISE)
 
 
+def _to_hashtag_form(query: str) -> str:
+    """Collapse a multi-word query to hashtag form (no spaces, lowercase).
+
+    SC's /v2/instagram/reels/search wraps Google Search and is documented
+    to be flaky on multi-token queries. Single-token queries map to a
+    hashtag page lookup which is the stable path. Used as a 500-retry
+    fallback before the request bubbles up as a silent failure.
+    """
+    return ''.join(query.split()).lower()
+
+
 def _infer_query_intent(topic: str) -> str:
     """Tiny local intent classifier for Instagram query expansion."""
     text = topic.lower().strip()
@@ -283,6 +331,26 @@ def search_instagram(
             timeout=30,
             retries=2,
         )
+    except http.HTTPError as e:
+        # SC's v2 reels search wraps Google Search and 500s frequently on
+        # multi-token queries. Single tokens hit the stable hashtag-page
+        # path. Retry once with hashtag form before bubbling up.
+        if getattr(e, "status_code", None) == 500 and ' ' in core_topic:
+            _log(f"IG search 500 on '{core_topic}', retrying with hashtag form")
+            try:
+                data = http.get(
+                    f"{SCRAPECREATORS_BASE}/v2/instagram/reels/search",
+                    params={"query": _to_hashtag_form(core_topic)},
+                    headers=http.scrapecreators_headers(token),
+                    timeout=30,
+                    retries=2,
+                )
+            except Exception as retry_e:
+                _log(f"IG search retry failed: {retry_e}")
+                return {"items": [], "error": f"{type(retry_e).__name__}: {retry_e}"}
+        else:
+            _log(f"ScrapeCreators error: {e}")
+            return {"items": [], "error": f"{type(e).__name__}: {e}"}
     except Exception as e:
         _log(f"ScrapeCreators error: {e}")
         return {"items": [], "error": f"{type(e).__name__}: {e}"}
@@ -317,6 +385,8 @@ def fetch_captions(
     video_items: List[Dict[str, Any]],
     token: str,
     depth: str = "default",
+    timeout: Optional[float] = None,
+    config: Optional[Dict[str, Any]] = None,
 ) -> Dict[str, str]:
     """Fetch transcripts for top N Instagram reels via ScrapeCreators.
 
@@ -328,12 +398,19 @@ def fetch_captions(
         video_items: Items from search_instagram()
         token: ScrapeCreators API key
         depth: Depth level for caption limit
+        timeout: Optional per-request transcript timeout in seconds. When
+            None, resolves from LAST30DAYS_TRANSCRIPT_TIMEOUT (env or
+            config), defaulting to DEFAULT_TRANSCRIPT_TIMEOUT (30s).
+        config: Optional config dict (from env.get_config()) used as a
+            fallback source for LAST30DAYS_TRANSCRIPT_TIMEOUT when the
+            value is not exported in os.environ.
 
     Returns:
         Dict mapping video_id -> caption text (truncated to 500 words)
     """
-    config = DEPTH_CONFIG.get(depth, DEPTH_CONFIG["default"])
-    max_captions = config["max_captions"]
+    depth_cfg = DEPTH_CONFIG.get(depth, DEPTH_CONFIG["default"])
+    max_captions = depth_cfg["max_captions"]
+    transcript_timeout = _resolve_transcript_timeout(timeout, config)
 
     if not video_items or not token:
         return {}
@@ -364,7 +441,7 @@ def fetch_captions(
                 f"{SCRAPECREATORS_BASE}/v2/instagram/media/transcript",
                 params={"url": url},
                 headers=http.scrapecreators_headers(token),
-                timeout=15,
+                timeout=transcript_timeout,
                 retries=1,
             )
             transcripts = data.get("transcripts") or []
diff --git a/skills/last30days/scripts/lib/normalize.py b/skills/last30days/scripts/lib/normalize.py
index 214119f..b8035d3 100644
--- a/skills/last30days/scripts/lib/normalize.py
+++ b/skills/last30days/scripts/lib/normalize.py
@@ -251,6 +251,11 @@ def _normalize_youtube(
     metadata: dict[str, Any] = {}
     if highlights:
         metadata["transcript_highlights"] = highlights
+    if item.get("captions_disabled"):
+        # Surfaced for quality_nudge: uploader disabled captions, so this
+        # video should be subtracted from the degraded-transcript-ratio
+        # denominator (it was never going to produce a transcript).
+        metadata["captions_disabled"] = True
     metadata["top_comments"] = _remap_comments(
         item.get("top_comments") or [],
         score_keys=("score", "likes"),
diff --git a/skills/last30days/scripts/lib/quality_nudge.py b/skills/last30days/scripts/lib/quality_nudge.py
index 7ffbb62..4464903 100644
--- a/skills/last30days/scripts/lib/quality_nudge.py
+++ b/skills/last30days/scripts/lib/quality_nudge.py
@@ -58,12 +58,38 @@ def _is_youtube_degraded(research_results: dict, threshold: float) -> bool:
     ratio is below threshold. The canonical cause is a stale yt-dlp binary -
     YouTube's caption format changes frequently and old binaries silently fail
     every transcript while the search itself still succeeds.
+
+    Captions-disabled videos are subtracted from the denominator: an uploader
+    who turned off captions can never produce a transcript, so counting that
+    video toward "fetch failures" produces false positives. A single
+    captions-disabled video in a small result set was tripping the nudge.
     """
     videos = int(research_results.get("youtube_videos_count") or 0)
     transcripts = int(research_results.get("youtube_transcripts_count") or 0)
+    captions_disabled = int(research_results.get("youtube_captions_disabled_count") or 0)
     if videos <= 0:
         return False
-    return (transcripts / videos) < threshold
+    eligible = videos - captions_disabled
+    if eligible <= 0:
+        # Every returned video had captions disabled - upstream content fact,
+        # not a yt-dlp problem. Don't flag.
+        return False
+    return (transcripts / eligible) < threshold
+
+
+def _is_instagram_silent_failure(config: dict, research_results: dict) -> bool:
+    """Instagram is silently failing when SC is configured but the source
+    returned zero items. The canonical cause is SC's v2 reels endpoint
+    500'ing on multi-token queries (it wraps Google Search and is documented
+    to be flaky there). Pre-fix the user got no signal at all - no Instagram
+    section in the brief, no error in the footer, just unexplained absence.
+    """
+    if not config.get("SCRAPECREATORS_API_KEY"):
+        return False  # not configured — not a silent failure
+    count = research_results.get("instagram_items_count")
+    if count is None:
+        return False  # source not run this invocation
+    return int(count) == 0
 
 
 def compute_quality_score(config: dict, research_results: dict) -> dict:
@@ -75,6 +101,8 @@ def compute_quality_score(config: dict, research_results: dict) -> dict:
             reddit_error reflecting what happened this run. Optional keys
             ``youtube_videos_count`` and ``youtube_transcripts_count`` enable
             degraded-YouTube detection (transcript-fetch ratio below threshold).
+            Optional key ``instagram_items_count`` enables silent-failure
+            detection for the bonus Instagram source.
 
     Returns:
         {
@@ -83,13 +111,15 @@ def compute_quality_score(config: dict, research_results: dict) -> dict:
             "core_missing": ["x", "youtube"],
             "core_errored": [],          # configured but errored at top level
             "core_degraded": [],         # configured and returned items but quality below threshold
-            "nudge_text": "..." or None if 100%
+            "bonus_errored": [],         # bonus sources (Instagram, etc.) configured but silent
+            "nudge_text": "..." or None if all sources healthy
         }
     """
     core_active: List[str] = []
     core_missing: List[str] = []
     core_errored: List[str] = []
     core_degraded: List[str] = []
+    bonus_errored: List[str] = []
 
     # HN, Polymarket, and Reddit are always active
     core_active.append("hn")
@@ -127,6 +157,11 @@ def compute_quality_score(config: dict, research_results: dict) -> dict:
         if has_ytdlp and research_results.get("youtube_error"):
             core_errored.append("youtube")
 
+    # Bonus sources (Instagram, etc.): SC-key holders expect content from
+    # these but until now the pipeline fell silent on configured-but-zero.
+    if _is_instagram_silent_failure(config, research_results):
+        bonus_errored.append("instagram")
+
     score_pct = int(len(core_active) / 5 * 100)
 
     has_sc = bool(config.get("SCRAPECREATORS_API_KEY"))
@@ -138,7 +173,8 @@ def compute_quality_score(config: dict, research_results: dict) -> dict:
         research_results,
         has_sc=has_sc,
         active_sources=active_sources,
-    ) if (core_missing or core_degraded) else None
+        bonus_errored=bonus_errored,
+    ) if (core_missing or core_degraded or bonus_errored) else None
 
     return {
         "score_pct": score_pct,
@@ -146,6 +182,7 @@ def compute_quality_score(config: dict, research_results: dict) -> dict:
         "core_missing": core_missing,
         "core_errored": core_errored,
         "core_degraded": core_degraded,
+        "bonus_errored": bonus_errored,
         "nudge_text": nudge_text,
     }
 
@@ -157,6 +194,7 @@ def _build_nudge_text(
     research_results: dict = None,
     has_sc: bool = False,
     active_sources: list = None,
+    bonus_errored: List[str] = None,
 ) -> str:
     """Build human-readable nudge text describing what was missed or degraded.
 
@@ -165,6 +203,7 @@ def _build_nudge_text(
     """
     lines: List[str] = []
     core_degraded = core_degraded or []
+    bonus_errored = bonus_errored or []
     research_results = research_results or {}
 
     # Describe what was missed
@@ -183,6 +222,9 @@ def _build_nudge_text(
     if core_degraded:
         degraded_labels = ", ".join(SOURCE_LABELS[s] for s in core_degraded)
         lines.append(f"Degraded: {degraded_labels}.")
+    if bonus_errored:
+        bonus_labels = ", ".join(s.capitalize() for s in bonus_errored)
+        lines.append(f"Bonus source silent: {bonus_labels}.")
     lines.append("")
 
     # Free suggestions
@@ -215,12 +257,30 @@ def _build_nudge_text(
     if "youtube" in core_degraded:
         videos = int(research_results.get("youtube_videos_count") or 0)
         transcripts = int(research_results.get("youtube_transcripts_count") or 0)
+        captions_disabled = int(research_results.get("youtube_captions_disabled_count") or 0)
+        captions_note = ""
+        if captions_disabled > 0:
+            captions_note = (
+                f" ({captions_disabled} of those had captions disabled by the "
+                "uploader, which is a separate cause and not fixable on your end)"
+            )
         free_suggestions.append(
             f"YouTube returned {videos} videos but only {transcripts} transcripts "
-            "captured. The most common cause is a stale yt-dlp binary - YouTube's "
-            "caption format changes frequently and old binaries silently fail every "
-            "transcript. Update via your package manager: scoop update yt-dlp "
-            "(Windows), brew upgrade yt-dlp (macOS), or pip install -U yt-dlp."
+            f"captured{captions_note}. The most common remaining cause is a stale "
+            "yt-dlp binary - YouTube's caption format changes frequently and old "
+            "binaries silently fail every transcript. Update via your package "
+            "manager: scoop update yt-dlp (Windows), brew upgrade yt-dlp (macOS), "
+            "or pip install -U yt-dlp."
+        )
+
+    if "instagram" in bonus_errored:
+        free_suggestions.append(
+            "Instagram returned 0 reels despite SC being configured. SC's "
+            "v2 reels endpoint wraps Google Search and 500's frequently on "
+            "multi-token queries. The skill now retries with hashtag-form "
+            "automatically; if zero items still appear, the topic may have "
+            "no reel coverage on Instagram. Try a single-word topic like "
+            "the most distinctive noun in your query."
         )
 
     # Mention bonus opt-in sources when SC key is present
diff --git a/skills/last30days/scripts/lib/youtube_yt.py b/skills/last30days/scripts/lib/youtube_yt.py
index fe033d1..3e46685 100644
--- a/skills/last30days/scripts/lib/youtube_yt.py
+++ b/skills/last30days/scripts/lib/youtube_yt.py
@@ -385,7 +385,11 @@ def _clean_vtt(vtt_text: str) -> str:
 _YT_USER_AGENT = "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36"
 
 
-def _fetch_transcript_direct(video_id: str, timeout: int = 30) -> Optional[str]:
+def _fetch_transcript_direct(
+    video_id: str,
+    timeout: int = 30,
+    status: Optional[Dict[str, Any]] = None,
+) -> Optional[str]:
     """Fetch YouTube transcript via direct HTTP without yt-dlp.
 
     Scrapes the watch page HTML for the captions track URL in
@@ -394,6 +398,9 @@ def _fetch_transcript_direct(video_id: str, timeout: int = 30) -> Optional[str]:
     Args:
         video_id: YouTube video ID
         timeout: HTTP request timeout in seconds
+        status: Optional dict mutated to record per-video signals. Sets
+            ``status["no_caption_tracks"] = True`` when the player response
+            confirms the uploader has no caption tracks (vs. fetch failure).
 
     Returns:
         Raw VTT text, or None if captions are unavailable.
@@ -442,6 +449,8 @@ def _fetch_transcript_direct(video_id: str, timeout: int = 30) -> Optional[str]:
 
     if not caption_tracks:
         _log(f"Direct transcript: no caption tracks for {video_id}")
+        if status is not None:
+            status["no_caption_tracks"] = True
         return None
 
     # Find English track (prefer exact 'en', then any en variant, then first track)
@@ -527,7 +536,11 @@ def _fetch_transcript_ytdlp(video_id: str, temp_dir: str) -> Optional[str]:
         return None
 
 
-def fetch_transcript(video_id: str, temp_dir: str) -> Optional[str]:
+def fetch_transcript(
+    video_id: str,
+    temp_dir: str,
+    status: Optional[Dict[str, Any]] = None,
+) -> Optional[str]:
     """Fetch auto-generated transcript for a YouTube video.
 
     Uses yt-dlp when available (preferred, more robust). Falls back to
@@ -536,6 +549,10 @@ def fetch_transcript(video_id: str, temp_dir: str) -> Optional[str]:
     Args:
         video_id: YouTube video ID
         temp_dir: Temporary directory for subtitle files
+        status: Optional dict mutated by the direct-HTTP path to record
+            per-video signals like ``no_caption_tracks``. Used to surface a
+            captions-disabled count so the quality nudge avoids false-positive
+            "stale yt-dlp" flags.
 
     Returns:
         Plaintext transcript string, or None if no captions available.
@@ -551,13 +568,13 @@ def fetch_transcript(video_id: str, temp_dir: str) -> Optional[str]:
         raw_vtt = _fetch_transcript_ytdlp(video_id, temp_dir)
         if not raw_vtt:
             _log(f"yt-dlp transcript failed for {video_id}, trying direct HTTP fallback")
-            raw_vtt = _fetch_transcript_direct(video_id)
+            raw_vtt = _fetch_transcript_direct(video_id, status=status)
     else:
         if ssh_host:
             _log("SSH-routing active, using direct HTTP transcript fetch")
         else:
             _log("yt-dlp not installed, using direct HTTP transcript fetch")
-        raw_vtt = _fetch_transcript_direct(video_id)
+        raw_vtt = _fetch_transcript_direct(video_id, status=status)
 
     if not raw_vtt:
         _log(f"No transcript available for {video_id} (no captions found)")
@@ -576,12 +593,16 @@ def fetch_transcript(video_id: str, temp_dir: str) -> Optional[str]:
 def fetch_transcripts_parallel(
     video_ids: List[str],
     max_workers: int = 5,
+    out_captions_disabled: Optional[Set[str]] = None,
 ) -> Dict[str, Optional[str]]:
     """Fetch transcripts for multiple videos in parallel.
 
     Args:
         video_ids: List of YouTube video IDs
         max_workers: Max parallel fetches
+        out_captions_disabled: Optional set mutated to record video_ids whose
+            uploader confirmed no caption tracks (vs. transient fetch failures).
+            Backward-compatible: callers that don't care can omit.
 
     Returns:
         Dict mapping video_id to transcript text (or None).
@@ -592,10 +613,11 @@ def fetch_transcripts_parallel(
     _log(f"Fetching transcripts for {len(video_ids)} videos")
 
     results = {}
+    statuses: Dict[str, Dict[str, Any]] = {vid: {} for vid in video_ids}
     with tempfile.TemporaryDirectory() as temp_dir:
         with ThreadPoolExecutor(max_workers=max_workers) as executor:
             futures = {
-                executor.submit(fetch_transcript, vid, temp_dir): vid
+                executor.submit(fetch_transcript, vid, temp_dir, statuses[vid]): vid
                 for vid in video_ids
             }
             for future in as_completed(futures):
@@ -609,6 +631,11 @@ def fetch_transcripts_parallel(
                     _log(f"Unexpected transcript error for {vid}: {type(exc).__name__}: {exc}")
                     results[vid] = None
 
+    if out_captions_disabled is not None:
+        for vid, st in statuses.items():
+            if st.get("no_caption_tracks"):
+                out_captions_disabled.add(vid)
+
     got = sum(1 for v in results.values() if v)
     errors = sum(1 for v in results.values() if v is None)
     _log(f"Got transcripts for {got}/{len(video_ids)} videos ({errors} failed)")
@@ -659,15 +686,21 @@ def search_and_transcribe(
     # good chance of reaching the target number of successful transcripts.
     transcript_limit = TRANSCRIPT_LIMITS.get(depth, TRANSCRIPT_LIMITS["default"])
     transcripts: Dict[str, Optional[str]] = {}
+    captions_disabled_ids: Set[str] = set()
     if transcript_limit > 0:
         attempt_count = min(len(items), transcript_limit * 3)
         candidate_ids = [item["video_id"] for item in items[:attempt_count]]
         _log(f"Fetching transcripts for up to {attempt_count} videos (target: {transcript_limit}): {candidate_ids}")
-        transcripts = fetch_transcripts_parallel(candidate_ids)
+        transcripts = fetch_transcripts_parallel(
+            candidate_ids, out_captions_disabled=captions_disabled_ids,
+        )
     else:
         _log(f"Transcript limit is 0 for depth={depth}, skipping transcript fetch")
 
-    # Step 3: Attach transcripts and extract highlights
+    # Step 3: Attach transcripts and extract highlights. Mark captions_disabled
+    # so quality_nudge can subtract those videos from the degraded-ratio
+    # denominator (uploader-disabled captions can never produce a transcript;
+    # counting them was producing false-positive stale-yt-dlp nudges).
     core_topic = _extract_core_subject(topic)
     for item in items:
         vid = item["video_id"]
@@ -676,6 +709,7 @@ def search_and_transcribe(
         item["transcript_highlights"] = extract_transcript_highlights(
             transcript or "", core_topic,
         )
+        item["captions_disabled"] = vid in captions_disabled_ids
 
     return {"items": items}
 
diff --git a/tests/test_bluesky.py b/tests/test_bluesky.py
index cb3cd33..c04afd5 100644
--- a/tests/test_bluesky.py
+++ b/tests/test_bluesky.py
@@ -1,5 +1,6 @@
 """Tests for bluesky module."""
 
+import os
 import sys
 import unittest
 from pathlib import Path
@@ -211,5 +212,146 @@ class TestSearchBlueskyAuth(unittest.TestCase):
         self.assertEqual(mock_request.call_args_list[3].kwargs.get("headers", {}), {"Authorization": "Bearer tok-new"})
 
 
+class TestSearchEndpointHostResolution(unittest.TestCase):
+    """The default search host moved from `public.api.bsky.app` (the
+    unauthenticated public mirror, now BunnyCDN-blocked for searchPosts) to
+    `api.bsky.app` (the canonical authenticated AppView). BSKY_SEARCH_HOST
+    env var or config value can override the default if Bluesky migrates
+    infrastructure again. Same os.environ-or-config hybrid pattern as
+    LAST30DAYS_STORE.
+    """
+
+    def setUp(self):
+        # Snapshot env so per-test overrides don't leak
+        self._saved_env = os.environ.pop("BSKY_SEARCH_HOST", None)
+
+    def tearDown(self):
+        if self._saved_env is not None:
+            os.environ["BSKY_SEARCH_HOST"] = self._saved_env
+        else:
+            os.environ.pop("BSKY_SEARCH_HOST", None)
+
+    def test_module_constant_uses_canonical_appview(self):
+        # Regression guard against the public mirror reappearing as the default
+        self.assertIn("api.bsky.app", bluesky.BSKY_SEARCH_URL)
+
+    def test_module_constant_does_not_use_public_mirror(self):
+        # Hard regression guard — the exact host that BunnyCDN was blocking
+        self.assertNotIn("public.api.bsky.app", bluesky.BSKY_SEARCH_URL)
+
+    def test_resolver_default_when_no_override(self):
+        self.assertEqual(
+            bluesky._resolve_search_url(),
+            "https://api.bsky.app/xrpc/app.bsky.feed.searchPosts",
+        )
+
+    def test_resolver_env_var_override(self):
+        os.environ["BSKY_SEARCH_HOST"] = "staging.bsky.app"
+        self.assertEqual(
+            bluesky._resolve_search_url(),
+            "https://staging.bsky.app/xrpc/app.bsky.feed.searchPosts",
+        )
+
+    def test_resolver_config_dict_override(self):
+        # User has BSKY_SEARCH_HOST only in .env file (project loads .env into
+        # config, not os.environ). Resolver must read both.
+        url = bluesky._resolve_search_url({"BSKY_SEARCH_HOST": "pds.example.com"})
+        self.assertEqual(url, "https://pds.example.com/xrpc/app.bsky.feed.searchPosts")
+
+    def test_resolver_env_var_wins_over_config(self):
+        # When both are set, os.environ takes precedence (matches LAST30DAYS_STORE)
+        os.environ["BSKY_SEARCH_HOST"] = "shell-host.example"
+        url = bluesky._resolve_search_url({"BSKY_SEARCH_HOST": "config-host.example"})
+        self.assertIn("shell-host.example", url)
+        self.assertNotIn("config-host.example", url)
+
+    def test_resolver_output_does_not_use_public_mirror(self):
+        # Regression guard at the resolver level (not just the constant) —
+        # this is what runtime actually calls. The constant-level guard
+        # above doesn't catch a regression where the resolver reverts.
+        self.assertNotIn("public.api.bsky.app", bluesky._resolve_search_url())
+
+    def test_resolver_strips_surrounding_whitespace(self):
+        # Pre-fix: " api.bsky.app " produced "https:// api.bsky.app /xrpc/..."
+        # which urllib raises ValueError on with no hint the env var caused it.
+        os.environ["BSKY_SEARCH_HOST"] = "  api.bsky.app  "
+        self.assertEqual(
+            bluesky._resolve_search_url(),
+            "https://api.bsky.app/xrpc/app.bsky.feed.searchPosts",
+        )
+
+    def test_resolver_rejects_embedded_path(self):
+        # "my-proxy.com/xrpc/prefix" would have doubled the /xrpc/ segment.
+        # We fall back to the default to avoid a guaranteed 404.
+        os.environ["BSKY_SEARCH_HOST"] = "my-proxy.example.com/xrpc/prefix"
+        self.assertEqual(
+            bluesky._resolve_search_url(),
+            "https://api.bsky.app/xrpc/app.bsky.feed.searchPosts",
+        )
+
+    def test_resolver_strips_embedded_scheme(self):
+        # Users who paste a full URL get a sane outcome, not a malformed URL.
+        os.environ["BSKY_SEARCH_HOST"] = "https://api.bsky.app"
+        self.assertEqual(
+            bluesky._resolve_search_url(),
+            "https://api.bsky.app/xrpc/app.bsky.feed.searchPosts",
+        )
+
+    def test_resolver_empty_string_falls_back_to_default(self):
+        os.environ["BSKY_SEARCH_HOST"] = ""
+        self.assertEqual(
+            bluesky._resolve_search_url(),
+            "https://api.bsky.app/xrpc/app.bsky.feed.searchPosts",
+        )
+
+
+class TestAppPasswordFormat(unittest.TestCase):
+    """Bluesky app passwords are 19-char xxxx-xxxx-xxxx-xxxx (lowercase
+    alphanumeric, three hyphens at fixed positions). Main-account passwords
+    are accepted by createSession but are bad hygiene. The validator detects
+    the format mismatch without gating any caller.
+    """
+
+    def test_accepts_valid_app_password_form(self):
+        # Use a fake example — never a real password
+        self.assertTrue(bluesky._validate_app_password_format("wfwp-cq7o-5six-7wy5"))
+
+    def test_rejects_length_15_string(self):
+        # The exact failure mode that triggered the 2026-05-04 investigation:
+        # user stored their main login password (15 chars) in BSKY_APP_PASSWORD
+        self.assertFalse(bluesky._validate_app_password_format("mainpassword123"))
+
+    def test_rejects_16_char_no_hyphen_string(self):
+        # Hex-style API key shape — common confusion with other services
+        self.assertFalse(bluesky._validate_app_password_format("abcdef0123456789"))
+
+    def test_rejects_uppercase_letters(self):
+        # Bluesky app passwords are all-lowercase by spec
+        self.assertFalse(bluesky._validate_app_password_format("WFWP-cq7o-5six-7wy5"))
+
+    def test_rejects_underscore_separator(self):
+        # Wrong separator
+        self.assertFalse(bluesky._validate_app_password_format("wfwp_cq7o_5six_7wy5"))
+
+    def test_rejects_special_chars_in_groups(self):
+        # Special characters are not part of the alphanumeric class
+        self.assertFalse(bluesky._validate_app_password_format("wfwp-cq7o-5six-7wy@"))
+
+    def test_rejects_empty_string(self):
+        self.assertFalse(bluesky._validate_app_password_format(""))
+
+    def test_rejects_none(self):
+        # Callers may pass config.get('BSKY_APP_PASSWORD') which is None when unset
+        self.assertFalse(bluesky._validate_app_password_format(None))
+
+    def test_rejects_integer(self):
+        # Defensive: don't crash if a numeric value sneaks in
+        self.assertFalse(bluesky._validate_app_password_format(123456789012345))
+
+    def test_rejects_list(self):
+        # Defensive: don't crash on iterables
+        self.assertFalse(bluesky._validate_app_password_format(["wfwp", "cq7o", "5six", "7wy5"]))
+
+
 if __name__ == "__main__":
     unittest.main()
diff --git a/tests/test_instagram_sc.py b/tests/test_instagram_sc.py
index 88ebf7f..2721353 100644
--- a/tests/test_instagram_sc.py
+++ b/tests/test_instagram_sc.py
@@ -1,8 +1,10 @@
 """Tests for instagram.py — ScrapeCreators Instagram search module."""
 
+import os
 import sys
 import unittest
 from pathlib import Path
+from unittest.mock import MagicMock, patch
 
 # Add lib to path
 sys.path.insert(0, str(Path(__file__).parent.parent / "skills" / "last30days" / "scripts"))
@@ -85,5 +87,223 @@ class TestInstagramDepthConfig(unittest.TestCase):
         )
 
 
+class TestHashtagFormCollapse(unittest.TestCase):
+    """Tests for _to_hashtag_form() — the multi-word retry workaround."""
+
+    def test_collapses_spaces(self):
+        self.assertEqual(instagram._to_hashtag_form("toronto real estate"), "torontorealestate")
+
+    def test_lowercases(self):
+        self.assertEqual(instagram._to_hashtag_form("Toronto REAL Estate"), "torontorealestate")
+
+    def test_idempotent_on_single_word(self):
+        self.assertEqual(instagram._to_hashtag_form("ozempic"), "ozempic")
+
+    def test_handles_extra_whitespace(self):
+        self.assertEqual(instagram._to_hashtag_form("  toronto   real  estate  "), "torontorealestate")
+
+
+class TestSearchRetryOn500(unittest.TestCase):
+    """Tests for the multi-word -> hashtag retry on SC's flaky 500 path.
+
+    SC's /v2/instagram/reels/search wraps Google Search and is documented
+    to be unreliable on multi-token queries. The retry collapses to a
+    hashtag form which hits the stable hashtag-page lookup path.
+    """
+
+    def _mock_response(self, status_code, json_payload=None):
+        m = MagicMock()
+        m.status_code = status_code
+        m.json.return_value = json_payload or {}
+        if status_code >= 400:
+            m.raise_for_status.side_effect = Exception(f"HTTP {status_code}")
+        else:
+            m.raise_for_status.return_value = None
+        return m
+
+    def test_multiword_500_triggers_retry_with_hashtag_form(self):
+        """Multi-word query 500 -> retry with collapsed hashtag form."""
+        first = self._mock_response(500)
+        second = self._mock_response(200, {"reels": []})
+        with patch.object(instagram, "_requests") as mock_requests:
+            mock_requests.get.side_effect = [first, second]
+            instagram.search_instagram(
+                "toronto real estate", "2026-04-01", "2026-05-04",
+                depth="default", token="fake-token",
+            )
+            self.assertEqual(mock_requests.get.call_count, 2)
+            # First call: original multi-word query
+            first_params = mock_requests.get.call_args_list[0].kwargs["params"]
+            self.assertEqual(first_params["query"], "toronto real estate")
+            # Second call: collapsed hashtag form
+            second_params = mock_requests.get.call_args_list[1].kwargs["params"]
+            self.assertEqual(second_params["query"], "torontorealestate")
+
+    def test_singleword_500_does_not_retry(self):
+        """Single-word query 500 has no spaces to collapse - no retry."""
+        only = self._mock_response(500)
+        with patch.object(instagram, "_requests") as mock_requests:
+            mock_requests.get.return_value = only
+            result = instagram.search_instagram(
+                "ozempic", "2026-04-01", "2026-05-04",
+                depth="default", token="fake-token",
+            )
+            self.assertEqual(mock_requests.get.call_count, 1)
+            self.assertIn("error", result)
+            self.assertEqual(result["items"], [])
+
+    def test_first_call_succeeds_no_retry(self):
+        """200 on first call -> retry path is never entered."""
+        ok = self._mock_response(200, {"reels": []})
+        with patch.object(instagram, "_requests") as mock_requests:
+            mock_requests.get.return_value = ok
+            instagram.search_instagram(
+                "toronto real estate", "2026-04-01", "2026-05-04",
+                depth="default", token="fake-token",
+            )
+            self.assertEqual(mock_requests.get.call_count, 1)
+
+    def test_no_token_short_circuits(self):
+        """No SCRAPECREATORS_API_KEY -> error returned without HTTP call."""
+        with patch.object(instagram, "_requests") as mock_requests:
+            result = instagram.search_instagram(
+                "toronto real estate", "2026-04-01", "2026-05-04",
+                depth="default", token=None,
+            )
+            mock_requests.get.assert_not_called()
+            self.assertIn("error", result)
+            self.assertIn("SCRAPECREATORS_API_KEY", result["error"])
+
+
+class TestTranscriptTimeoutConfig(unittest.TestCase):
+    """Tests for LAST30DAYS_TRANSCRIPT_TIMEOUT configuration.
+
+    SC's /v2/instagram/media/transcript endpoint regularly takes >15s,
+    so the timeout must be configurable. Default is DEFAULT_TRANSCRIPT_TIMEOUT
+    (30s); the env var or per-call kwarg overrides it.
+    """
+
+    def setUp(self):
+        # Snapshot any pre-existing env so we don't leak across tests
+        self._saved_env = os.environ.pop("LAST30DAYS_TRANSCRIPT_TIMEOUT", None)
+
+    def tearDown(self):
+        os.environ.pop("LAST30DAYS_TRANSCRIPT_TIMEOUT", None)
+        if self._saved_env is not None:
+            os.environ["LAST30DAYS_TRANSCRIPT_TIMEOUT"] = self._saved_env
+
+    def _ok_response(self):
+        m = MagicMock()
+        m.status_code = 200
+        m.json.return_value = {"transcripts": [{"text": "hello world"}]}
+        return m
+
+    def _video_item(self, vid="abc123"):
+        return {
+            "video_id": vid,
+            "url": f"https://www.instagram.com/reel/{vid}/",
+            "text": "",
+        }
+
+    def test_default_timeout_is_30s_when_nothing_set(self):
+        """No env var, no kwarg -> request uses 30s, not the legacy 15s."""
+        items = [self._video_item()]
+        with patch.object(instagram, "_requests") as mock_requests:
+            mock_requests.get.return_value = self._ok_response()
+            instagram.fetch_captions(items, token="fake-token")
+            kwargs = mock_requests.get.call_args.kwargs
+            self.assertEqual(kwargs["timeout"], 30.0)
+
+    def test_env_var_override(self):
+        """LAST30DAYS_TRANSCRIPT_TIMEOUT='60' -> request uses 60s."""
+        os.environ["LAST30DAYS_TRANSCRIPT_TIMEOUT"] = "60"
+        items = [self._video_item()]
+        with patch.object(instagram, "_requests") as mock_requests:
+            mock_requests.get.return_value = self._ok_response()
+            instagram.fetch_captions(items, token="fake-token")
+            kwargs = mock_requests.get.call_args.kwargs
+            self.assertEqual(kwargs["timeout"], 60.0)
+
+    def test_explicit_timeout_kwarg_wins_over_env(self):
+        """Explicit timeout= kwarg trumps the env var."""
+        os.environ["LAST30DAYS_TRANSCRIPT_TIMEOUT"] = "60"
+        items = [self._video_item()]
+        with patch.object(instagram, "_requests") as mock_requests:
+            mock_requests.get.return_value = self._ok_response()
+            instagram.fetch_captions(items, token="fake-token", timeout=10)
+            kwargs = mock_requests.get.call_args.kwargs
+            self.assertEqual(kwargs["timeout"], 10.0)
+
+    def test_config_dict_fallback_when_env_unset(self):
+        """config={'LAST30DAYS_TRANSCRIPT_TIMEOUT': '45'} -> request uses 45s."""
+        items = [self._video_item()]
+        with patch.object(instagram, "_requests") as mock_requests:
+            mock_requests.get.return_value = self._ok_response()
+            instagram.fetch_captions(
+                items,
+                token="fake-token",
+                config={"LAST30DAYS_TRANSCRIPT_TIMEOUT": "45"},
+            )
+            kwargs = mock_requests.get.call_args.kwargs
+            self.assertEqual(kwargs["timeout"], 45.0)
+
+    def test_invalid_env_value_falls_back_to_default(self):
+        """Garbage env var doesn't crash; falls back to 30s."""
+        os.environ["LAST30DAYS_TRANSCRIPT_TIMEOUT"] = "not-a-number"
+        items = [self._video_item()]
+        with patch.object(instagram, "_requests") as mock_requests:
+            mock_requests.get.return_value = self._ok_response()
+            instagram.fetch_captions(items, token="fake-token")
+            kwargs = mock_requests.get.call_args.kwargs
+            self.assertEqual(kwargs["timeout"], 30.0)
+
+
+class TestSearchRetryOn500Urllib(unittest.TestCase):
+    """Lock in the urllib-path 500-retry. Pre-fix the retry was dead code on
+    the urllib branch because it checked `getattr(e, 'status', None)` while
+    `http.HTTPError` exposes the code as `status_code`. Caught by code-review
+    on 2026-05-04 (REL-001 / ADV-001, two reviewers at 0.97/0.98 confidence).
+    """
+
+    def setUp(self):
+        # Force urllib path by making instagram._requests look absent
+        self._saved_requests = instagram._requests
+        instagram._requests = None
+
+    def tearDown(self):
+        instagram._requests = self._saved_requests
+
+    def test_urllib_500_on_multiword_triggers_retry_with_hashtag_form(self):
+        from lib import http as http_module
+        first_error = http_module.HTTPError("HTTP 500: Server Error", 500, "")
+        second_payload = {"reels": []}
+        # http.get is called twice: first raises HTTPError(500), second returns dict
+        with patch.object(http_module, "get") as mock_http_get:
+            mock_http_get.side_effect = [first_error, second_payload]
+            instagram.search_instagram(
+                "toronto real estate", "2026-04-01", "2026-05-04",
+                depth="default", token="fake-token",
+            )
+            self.assertEqual(mock_http_get.call_count, 2)
+            # First call URL contains the original multi-word query
+            first_url = mock_http_get.call_args_list[0].args[0]
+            self.assertIn("query=toronto+real+estate", first_url)
+            # Second call URL contains the collapsed hashtag form
+            second_url = mock_http_get.call_args_list[1].args[0]
+            self.assertIn("query=torontorealestate", second_url)
+
+    def test_urllib_500_on_singleword_does_not_retry(self):
+        from lib import http as http_module
+        only_error = http_module.HTTPError("HTTP 500: Server Error", 500, "")
+        with patch.object(http_module, "get") as mock_http_get:
+            mock_http_get.side_effect = only_error
+            result = instagram.search_instagram(
+                "ozempic", "2026-04-01", "2026-05-04",
+                depth="default", token="fake-token",
+            )
+            self.assertEqual(mock_http_get.call_count, 1)
+            self.assertIn("error", result)
+
+
 if __name__ == "__main__":
     unittest.main()
diff --git a/tests/test_quality_nudge.py b/tests/test_quality_nudge.py
index a2592e9..41476c5 100644
--- a/tests/test_quality_nudge.py
+++ b/tests/test_quality_nudge.py
@@ -293,3 +293,174 @@ class TestYouTubeDegraded:
         # But nudge still fires
         assert q["nudge_text"] is not None
         assert "Degraded: YouTube" in q["nudge_text"]
+
+
+class TestYouTubeCaptionsDisabledDoesNotFalseFlag:
+    """Captions-disabled videos must not lower the transcript-fetch ratio.
+
+    A video where the uploader disabled captions can never produce a transcript,
+    no matter how fresh yt-dlp is. Counting it in the denominator of the
+    degraded-ratio check produces false positives - one captions-disabled video
+    in a small result set was triggering a "stale yt-dlp binary" nudge that was
+    wrong. Fix: subtract captions_disabled from the denominator.
+    """
+
+    def test_zero_captions_disabled_preserves_existing_behavior(self):
+        # Pre-existing case: 0 of 6 transcripts is still degraded (no captions
+        # disabled to discount). Behavior is unchanged from TestYouTubeDegraded.
+        q = _compute(
+            ytdlp_installed=True,
+            result_overrides={
+                "youtube_videos_count": 6,
+                "youtube_transcripts_count": 0,
+                "youtube_captions_disabled_count": 0,
+            },
+        )
+        assert "youtube" in q["core_degraded"]
+
+    def test_all_videos_captions_disabled_does_not_flag(self):
+        # Every returned video had captions disabled by the uploader.
+        # That's not a yt-dlp problem - it's an upstream content fact. Must not
+        # flag degraded.
+        q = _compute(
+            ytdlp_installed=True,
+            result_overrides={
+                "youtube_videos_count": 3,
+                "youtube_transcripts_count": 0,
+                "youtube_captions_disabled_count": 3,
+            },
+        )
+        assert "youtube" not in q["core_degraded"]
+
+    def test_mixed_uses_corrected_denominator(self):
+        # 6 videos, 3 captions_disabled, 2 transcripts.
+        # Naive (buggy) ratio: 2/6 = 33% (would flag).
+        # Corrected ratio: 2/(6-3) = 67% (does NOT flag).
+        # This case demonstrates the fix changes the verdict.
+        q = _compute(
+            ytdlp_installed=True,
+            result_overrides={
+                "youtube_videos_count": 6,
+                "youtube_transcripts_count": 2,
+                "youtube_captions_disabled_count": 3,
+            },
+        )
+        assert "youtube" not in q["core_degraded"]
+
+    def test_mixed_still_flags_when_truly_degraded(self):
+        # Even after discounting captions-disabled, the ratio is still bad.
+        # 8 videos, 1 captions_disabled, 1 transcript -> 1/(8-1) = 14% (flags).
+        q = _compute(
+            ytdlp_installed=True,
+            result_overrides={
+                "youtube_videos_count": 8,
+                "youtube_transcripts_count": 1,
+                "youtube_captions_disabled_count": 1,
+            },
+        )
+        assert "youtube" in q["core_degraded"]
+        # Nudge should still mention the stale yt-dlp possibility but also
+        # acknowledge that captions-disabled is a separate cause.
+        assert q["nudge_text"] is not None
+        assert "captions disabled" in q["nudge_text"].lower()
+
+    def test_missing_count_defaults_to_zero(self):
+        # Older callers that don't pass the new key still work (default 0).
+        q = _compute(
+            ytdlp_installed=True,
+            result_overrides={
+                "youtube_videos_count": 6,
+                "youtube_transcripts_count": 0,
+                # youtube_captions_disabled_count intentionally omitted
+            },
+        )
+        assert "youtube" in q["core_degraded"]
+
+
+class TestInstagramSilentFailure:
+    """Instagram is a `bonus` source via SC. Silent-failure detection: if SC
+    is configured but the source returned zero items, surface a nudge so the
+    user understands why the brief lacks an Instagram section.
+
+    Pre-fix the user got no signal - SC's /v2/instagram/reels/search 500s
+    frequently on multi-token queries and the pipeline silently returned
+    empty without any indication.
+    """
+
+    def test_zero_items_with_sc_flags_bonus_errored(self):
+        q = _compute(
+            config_overrides={
+                "AUTH_TOKEN": "tok123",
+                "SCRAPECREATORS_API_KEY": "sc_key",
+            },
+            ytdlp_installed=True,
+            result_overrides={"instagram_items_count": 0},
+        )
+        assert "instagram" in q["bonus_errored"]
+        assert q["nudge_text"] is not None
+        assert "Instagram" in q["nudge_text"]
+
+    def test_zero_items_without_sc_does_not_flag(self):
+        q = _compute(
+            config_overrides={"AUTH_TOKEN": "tok123"},
+            ytdlp_installed=True,
+            result_overrides={"instagram_items_count": 0},
+        )
+        assert "instagram" not in q.get("bonus_errored", [])
+
+    def test_nonzero_items_does_not_flag(self):
+        q = _compute(
+            config_overrides={
+                "AUTH_TOKEN": "tok123",
+                "SCRAPECREATORS_API_KEY": "sc_key",
+            },
+            ytdlp_installed=True,
+            result_overrides={"instagram_items_count": 5},
+        )
+        assert "instagram" not in q["bonus_errored"]
+        assert q["nudge_text"] is None
+
+    def test_missing_key_means_source_did_not_run(self):
+        q = _compute(
+            config_overrides={
+                "AUTH_TOKEN": "tok123",
+                "SCRAPECREATORS_API_KEY": "sc_key",
+            },
+            ytdlp_installed=True,
+        )
+        assert "instagram" not in q["bonus_errored"]
+        assert q["nudge_text"] is None
+
+    def test_nudge_text_explains_workaround(self):
+        q = _compute(
+            config_overrides={
+                "AUTH_TOKEN": "tok123",
+                "SCRAPECREATORS_API_KEY": "sc_key",
+            },
+            ytdlp_installed=True,
+            result_overrides={"instagram_items_count": 0},
+        )
+        assert q["nudge_text"] is not None
+        text_lower = q["nudge_text"].lower()
+        assert "instagram" in text_lower
+        assert ("0 reels" in text_lower or "silent" in text_lower
+                or "hashtag" in text_lower)
+
+    def test_bonus_errored_does_not_affect_core_score(self):
+        q = _compute(
+            config_overrides={
+                "AUTH_TOKEN": "tok123",
+                "SCRAPECREATORS_API_KEY": "sc_key",
+            },
+            ytdlp_installed=True,
+            result_overrides={"instagram_items_count": 0},
+        )
+        assert q["score_pct"] == 100
+        assert "instagram" in q["bonus_errored"]
+        assert q["nudge_text"] is not None
+        assert "Bonus source silent" in q["nudge_text"]
+
+    def test_bonus_errored_field_always_present(self):
+        q = _compute()
+        assert q.get("bonus_errored") == []
+
diff --git a/tests/test_youtube_yt.py b/tests/test_youtube_yt.py
index be55b83..6b45eb9 100644
--- a/tests/test_youtube_yt.py
+++ b/tests/test_youtube_yt.py
@@ -252,7 +252,7 @@ class TestFetchTranscriptFallback(unittest.TestCase):
              mock.patch.object(youtube_yt, "_fetch_transcript_direct", return_value=sample_vtt) as direct_mock:
             result = youtube_yt.fetch_transcript("vid2", "/tmp/test")
         yt_mock.assert_not_called()
-        direct_mock.assert_called_once_with("vid2")
+        direct_mock.assert_called_once_with("vid2", status=None)
         self.assertIsNotNone(result)
         self.assertIn("Direct transcript content", result)
 
@@ -363,7 +363,7 @@ class TestSearchAndTranscribe(unittest.TestCase):
         ]
 
         # fetch_transcripts_parallel returns None for music videos, text for talks
-        def fake_parallel(video_ids, max_workers=5):
+        def fake_parallel(video_ids, max_workers=5, out_captions_disabled=None):
             result = {}
             for vid in video_ids:
                 if vid.startswith("talk"):