diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.json
index 2f7c6eb..5574db6 100644
--- a/.claude-plugin/marketplace.json
+++ b/.claude-plugin/marketplace.json
@@ -11,7 +11,7 @@
       "name": "docent",
       "source": "./plugins/docent",
       "description": "Docent AI analysis tools for Claude Code",
-      "version": "0.1.9",
+      "version": "0.1.10",
       "author": {
         "name": "TransluceAI"
       },
diff --git a/.github/workflows/plugin-sanity.yml b/.github/workflows/plugin-sanity.yml
index 156a377..25e72ce 100644
--- a/.github/workflows/plugin-sanity.yml
+++ b/.github/workflows/plugin-sanity.yml
@@ -87,9 +87,6 @@ jobs:
           args = server.get("args")
           if not isinstance(args, list) or "--from" not in args:
               fail("docent MCP server args must include --from")
-          package = args[args.index("--from") + 1]
-          if package != "docent-python>=0.1.73":
-              fail("docent MCP server must require docent-python>=0.1.73")
 
           forbidden_names = {".mcp.local.json", "docent.env"}
           for path in plugin_dir.rglob("*"):
diff --git a/plugins/docent/.claude-plugin/plugin.json b/plugins/docent/.claude-plugin/plugin.json
index 9b26a96..9283817 100644
--- a/plugins/docent/.claude-plugin/plugin.json
+++ b/plugins/docent/.claude-plugin/plugin.json
@@ -1,5 +1,5 @@
 {
   "name": "docent",
-  "version": "0.1.9",
+  "version": "0.1.10",
   "description": "Docent AI analysis tools"
 }
diff --git a/plugins/docent/.mcp.json b/plugins/docent/.mcp.json
index a3f959c..d727d77 100644
--- a/plugins/docent/.mcp.json
+++ b/plugins/docent/.mcp.json
@@ -3,7 +3,7 @@
     "docent": {
       "type": "stdio",
       "command": "uv",
-      "args": ["tool", "run", "--from", "docent-python>=0.1.73", "docent-mcp"]
+      "args": ["tool", "run", "--from", "docent-python>=0.1.74", "docent-mcp"]
     }
   }
 }
diff --git a/plugins/docent/skills/docent/analysis.md b/plugins/docent/skills/docent/analysis.md
index 5f55aa2..50452e0 100644
--- a/plugins/docent/skills/docent/analysis.md
+++ b/plugins/docent/skills/docent/analysis.md
@@ -64,7 +64,7 @@ client = Docent.from_url("https://docent.transluce.org/dashboard/668354d8-...")
 ```
 This parses the domain and collection ID from the URL automatically.
 
-The Docent SDK can be configured by a `docent.env` file. The SDK searches from the current working directory upward through parent directories, then falls back to `~/.docent/docent.env` if no local file exists. You do not need to explicitly source `docent.env`. Config files may use INI-style `[section]` headers for multi-profile support; select a profile with `Docent(profile="my-profile")` or the `DOCENT_PROFILE` environment variable.
+The Docent SDK can be configured by a `docent.env` file. The default global config file is `~/.docent/docent.env`. The SDK also honors project-level `docent.env` files from the current working directory upward as local overrides. You do not need to explicitly source `docent.env`. Config files may use INI-style `[section]` headers for multi-profile support; select a profile with `Docent(profile="my-profile")` or the `DOCENT_PROFILE` environment variable.
 
 If you're not sure what collection the user is talking about:
 * If the user provides a Docent dashboard URL (e.g., `https://docent.transluce.org/dashboard/668354d8-...`), use `Docent.from_url()` or extract the collection ID from the last path segment (the UUID).
@@ -80,7 +80,7 @@ If you run into any issues or unexpected behavior with the Docent platform, paus
 * If authentication fails (HTTP 401) or no API key is configured, walk the user through setup:
   1. Open the API keys page for them: `open https://docent.transluce.org/settings/api-keys` (macOS) or `xdg-open https://docent.transluce.org/settings/api-keys` (Linux).
   2. Ask them to create a new API key (it will start with `dk_`).
-  3. Write the key to a local `docent.env` file or `~/.docent/docent.env`: `DOCENT_API_KEY=dk_...` (plus `DOCENT_API_URL` and `DOCENT_FRONTEND_URL` if not using the default instance).
+  3. Write the key to `~/.docent/docent.env`: `DOCENT_API_KEY=dk_...` (plus `DOCENT_API_URL` and `DOCENT_FRONTEND_URL` if not using the default instance). Use a project-level `docent.env` only when the project needs a local override.
   4. Verify connectivity by constructing a `Docent()` client — the constructor validates the API key automatically.
 * If the SDK does not match what's documented here, check whether the SDK is up to date.
 * If the Docent MCP server is available but doesn't match the tools documented here, check whether the MCP server needs an upgrade (`uv tool upgrade docent`). If an upgrade was needed, ask the user to restart the session or MCP server.
diff --git a/plugins/docent/skills/docent/dql-reference.md b/plugins/docent/skills/docent/dql-reference.md
index 1ce3444..27e196f 100644
--- a/plugins/docent/skills/docent/dql-reference.md
+++ b/plugins/docent/skills/docent/dql-reference.md
@@ -69,8 +69,8 @@ raw_rows = client.dql_result_to_dicts(result)
 | `name` | Optional transcript title. |
 | `description` | Optional description. |
 | `transcript_group_id` | Optional grouping identifier. |
-| `messages` | Binary-encoded JSON payload of message turns. |
-| `metadata_json` | Binary-encoded metadata describing the transcript. |
+| `messages` | UTF-8 bytes of a JSON array of message turns (Postgres `bytea`, not `jsonb`). Use `convert_from` before JSON operators or `jsonb_array_length` (see [Counting transcript messages](#counting-transcript-messages)). |
+| `metadata_json` | UTF-8 bytes of JSON metadata (`bytea`). Same `convert_from` pattern as `messages` when using JSON operators. |
 | `created_at` | Timestamp recorded during ingest. |
 
 ### `transcript_groups`
@@ -112,6 +112,8 @@ raw_rows = client.dql_result_to_dicts(result)
 | `dql_query` | DQL query (template readings only). |
 | `model_json` | Model configuration. |
 | `output_schema` | JSON schema for output validation. |
+| `max_new_tokens` | Maximum number of new tokens generated per LLM call. |
+| `num_rollouts` | Number of independent LLM samples generated per input row (>= 1). |
 | `source_reading_preset_id` | Optional associated preset. |
 | `created_at` | When the reading was created. |
 
@@ -151,6 +153,7 @@ For scripted readings, `arguments_dict` holds arbitrary user-supplied metadata p
 | --- | --- |
 | `reading_id` | FK to readings.id. |
 | `result_id` | FK to reading_results.id. |
+| `rollout_index` | 0-based position of this rollout within the reading's group for the same input row. Range `[0, readings.num_rollouts)`. |
 
 ## JSON Metadata Access Patterns
 
@@ -165,14 +168,20 @@ WHERE metadata_json->>'environment' = 'staging';
 
 ```sql
 -- Retrieve nested transcript metadata
+-- `transcripts.metadata_json` is bytea (UTF-8 JSON), not jsonb — decode before JSON operators.
 -- Dots in `get_metadata_fields` output (e.g. `metadata.conversation.speaker`) indicate nested JSON objects;
 -- traverse with -> for intermediate keys and ->> for the final key.
 SELECT
   id,
-  metadata_json->'conversation'->>'speaker' AS speaker,
-  metadata_json->'conversation'->>'topic' AS topic
-FROM transcripts
-WHERE metadata_json->>'status' = 'flagged';
+  meta->'conversation'->>'speaker' AS speaker,
+  meta->'conversation'->>'topic' AS topic
+FROM (
+  SELECT
+    id,
+    convert_from(metadata_json, 'UTF8')::jsonb AS meta
+  FROM transcripts
+) AS t
+WHERE meta->>'status' = 'flagged';
 ```
 
 ```sql
@@ -185,6 +194,47 @@ WHERE metadata_json ? 'latency_ms';
 
 When querying JSON fields, comparisons default to string semantics. Cast values when you need numeric ordering or aggregation.
 
+## Counting transcript messages
+
+`transcripts.messages` is stored as `bytea` (UTF-8 JSON), not `jsonb`. You cannot use `messages -> 0` or `jsonb_array_length(messages)` directly — Postgres reports `operator does not exist: bytea -> integer`.
+
+Decode to `jsonb`, then count array elements:
+
+```sql
+jsonb_array_length(convert_from(messages, 'UTF8')::jsonb)
+```
+
+Allowed helpers: `convert_from`, `convert_to`, `jsonb_array_length`.
+
+### Agent runs with at least N messages (any transcript)
+
+```sql
+SELECT DISTINCT ar.id AS agent_run_id
+FROM agent_runs ar
+JOIN transcripts t ON t.agent_run_id = ar.id
+WHERE jsonb_array_length(convert_from(t.messages, 'UTF8')::jsonb) >= 10;
+```
+
+### Per-transcript message counts
+
+```sql
+SELECT
+  transcript_id,
+  agent_run_id,
+  message_count
+FROM (
+  SELECT
+    t.id AS transcript_id,
+    t.agent_run_id,
+    jsonb_array_length(convert_from(t.messages, 'UTF8')::jsonb) AS message_count
+  FROM transcripts t
+) AS counted
+WHERE message_count >= 10
+ORDER BY message_count DESC;
+```
+
+Express filters like “≥10 messages” in DQL with the pattern above. Do not materialize matching run IDs elsewhere and paste them into a huge `WHERE id IN (...)` clause.
+
 ## Allowed Syntax
 
 | Feature |
@@ -194,7 +244,7 @@ When querying JSON fields, comparisons default to string semantics. Cast values
 | `WITH` (CTEs) |
 | `UNION [ALL]`, `INTERSECT`, `EXCEPT` |
 | `GROUP BY`, `HAVING` |
-| Aggregations (`COUNT`, `AVG`, `MIN`, `MAX`, `SUM`, `STDDEV_POP`, `STDDEV_SAMP`, `VAR_POP`, `VAR_SAMP`, `ARRAY_AGG`, `STRING_AGG`, `JSON_AGG`, `JSONB_AGG`, `JSON_OBJECT_AGG`, `PERCENTILE_CONT`, `PERCENTILE_DISC` with `WITHIN GROUP`) |
+| Aggregations (`COUNT`, `AVG`, `MIN`, `MAX`, `SUM`, `STDDEV_POP`, `STDDEV_SAMP`, `VAR_POP`, `VAR_SAMP`, `ARRAY_AGG`, `STRING_AGG`, `JSON_AGG`, `JSONB_AGG`, `JSON_OBJECT_AGG`, `MODE`, `PERCENTILE_CONT`, `PERCENTILE_DISC` with `WITHIN GROUP`) |
 | Window functions (`ROW_NUMBER`, `RANK`, `DENSE_RANK`, `NTILE`, `LAG`, `LEAD`, `FIRST_VALUE`, `LAST_VALUE`, `NTH_VALUE`, `PERCENT_RANK`, `CUME_DIST`) |
 | `ORDER BY`, `LIMIT`, `OFFSET` |
 | Conditional & null helpers (`CASE`, `COALESCE`, `NULLIF`) |
@@ -202,7 +252,7 @@ When querying JSON fields, comparisons default to string semantics. Cast values
 | Comparison operators (`=`, `!=`, `<`, `<=`, `>`, `>=`, `IS`, `IS NOT`, `IS DISTINCT FROM`, `IN`, `BETWEEN`, `LIKE`, `ILIKE`, `EXISTS`, `SIMILAR TO`, `~`, `~*`, `!~`, `!~*`) |
 | Arithmetic & math (`+`, `-`, `*`, `/`, `%`, `POWER`, `ABS`, `SIGN`, `SQRT`, `LN`, `LOG`, `EXP`, `GREATEST`, `LEAST`, `FLOOR`, `CEIL`, `ROUND`, `RANDOM`) |
 | String helpers (`SUBSTRING`, `LEFT`, `RIGHT`, `LENGTH`, `UPPER`, `LOWER`, `INITCAP`, `TRIM`, `REPLACE`, `SPLIT_PART`, `POSITION`, `CONCAT`, `CONCAT_WS`, `STRING_AGG`) |
-| JSON operators & functions (`->`, `->>`, `#>`, `#>>`, `@>`, `?`, `?|`, `?&`, `jsonb_build_object`, `jsonb_build_array`, `json_agg`, `jsonb_agg`, `json_object_agg`, `jsonb_set`, `jsonb_path_query`, `jsonb_path_exists`) |
+| JSON operators & functions (`->`, `->>`, `#>`, `#>>`, `@>`, `?`, `?|`, `?&`, `jsonb_build_object`, `jsonb_build_array`, `jsonb_array_length`, `json_agg`, `jsonb_agg`, `json_object_agg`, `jsonb_set`, `jsonb_path_query`, `jsonb_path_exists`, `convert_from`, `convert_to`) |
 | Date/time basics (`CURRENT_DATE`, `CURRENT_TIME`, `CURRENT_TIMESTAMP`, `NOW()`, `EXTRACT`, `DATE_TRUNC`, `AGE`, `AT TIME ZONE`, `timezone()`) |
 | Interval arithmetic (`timestamp +/- INTERVAL`, `INTERVAL` literals, `MAKE_INTERVAL`, `JUSTIFY_DAYS`, `JUSTIFY_HOURS`, `JUSTIFY_INTERVAL`) |
 | Construction & conversion (`MAKE_DATE`, `MAKE_TIME`, `MAKE_TIMESTAMP`, `MAKE_TIMESTAMPTZ`, `TO_CHAR`, `TO_DATE`, `TO_TIMESTAMP`, `DATE_PART`) |
@@ -281,6 +331,92 @@ ORDER BY rr.id DESC
 LIMIT 50;
 ```
 
+### Rollouts and Self-Consistency
+
+When a reading is configured with `num_rollouts > 1`, each input row produces multiple
+independent LLM samples. Rollouts are stored as separate `reading_results` rows joined
+to the reading via `reading_result_links`, with `reading_result_links.rollout_index`
+recording the 0-based position within the reading. Samples are fungible: a single
+result row may be linked by multiple readings (at potentially different rollout
+positions) when a cached sample is reused.
+
+**Always filter out pending and failed rollouts before aggregating outputs.** The
+canonical predicate is:
+
+```sql
+rr.output IS NOT NULL AND (rr.error IS NULL OR rr.error::text = 'null')
+```
+
+`error` is JSONB, so SQL `NULL` and JSON `null` are both possible.
+
+#### Per-row rollouts side by side
+
+```sql
+SELECT
+  rr.arguments_dict->'agent_run'->>'id' AS agent_run_id,
+  rrl.rollout_index,
+  rr.output->>'answer' AS answer
+FROM reading_results rr
+JOIN reading_result_links rrl ON rrl.result_id = rr.id
+WHERE rrl.reading_id = '<reading-uuid>'
+  AND rr.output IS NOT NULL
+ORDER BY agent_run_id, rrl.rollout_index;
+```
+
+#### Per-row self-consistency and modal vote
+
+`COUNT(DISTINCT ...)` measures spread; `MODE() WITHIN GROUP` picks the majority answer.
+
+```sql
+SELECT
+  rr.arguments_dict->'agent_run'->>'id' AS agent_run_id,
+  COUNT(rr.id) AS n_completed,
+  COUNT(DISTINCT rr.output->>'answer') AS n_distinct_answers,
+  MODE() WITHIN GROUP (ORDER BY rr.output->>'answer') AS modal_answer
+FROM reading_results rr
+JOIN reading_result_links rrl ON rrl.result_id = rr.id
+WHERE rrl.reading_id = '<reading-uuid>'
+  AND rr.output IS NOT NULL
+GROUP BY agent_run_id;
+```
+
+To compare two readings, wrap this query (selecting `agent_run_id, modal_answer`) as a
+subquery per reading and join the two on `agent_run_id`.
+
+#### Reading-level self-consistency rate
+
+```sql
+SELECT
+  reading_id,
+  ROUND(CAST(AVG(CASE WHEN n_distinct = 1 THEN 1.0 ELSE 0.0 END) AS NUMERIC), 3)
+    AS unanimous_row_fraction,
+  ROUND(CAST(AVG(n_distinct) AS NUMERIC), 3) AS avg_distinct_per_row
+FROM (
+  SELECT
+    rrl.reading_id AS reading_id,
+    rr.cache_key_hash AS row_key,
+    COUNT(DISTINCT rr.output->>'answer') AS n_distinct
+  FROM reading_results rr
+  JOIN reading_result_links rrl ON rrl.result_id = rr.id
+  WHERE rrl.reading_id IN ('<reading-A>', '<reading-B>')
+    AND rr.output IS NOT NULL
+  GROUP BY rrl.reading_id, rr.cache_key_hash
+) AS row_stats
+GROUP BY reading_id;
+```
+
+**Counting semantics.** Because cached samples are pooled across readings, a single
+`reading_results` row may appear in multiple readings via different links. Choose:
+
+- `COUNT(rr.id)` or `COUNT(rrl.result_id)` — counts links (i.e. rollouts as seen by
+  this reading set). What you usually want.
+- `COUNT(DISTINCT rr.id)` — counts unique LLM calls (i.e. the underlying sample pool).
+
+**Rollout pairing caveat.** `rollout_index` is per-link, not per-result: rollout #2 of
+reading A and rollout #2 of reading B are not paired draws. Avoid joining across
+readings on `(input, rollout_index)` for paired tests — fungible samples have no
+positional identity across readings.
+
 ## Restrictions and Best Practices
 
 - **Read-only**: Only `SELECT`-style queries are permitted.
@@ -290,6 +426,7 @@ LIMIT 50;
 - **Limit enforcement**: Every query is capped at 10,000 rows. Use pagination (`OFFSET`/`LIMIT`) for larger row collections.
 - **JSON performance**: Heavy JSON traversal across large collections can be slow. Prefer top-level fields when available.
 - **Type awareness**: Cast values explicitly when precision matters.
+- **Reading results: filter by completion.** Querying `reading_results` will include pending and failed rollouts by default. Add `WHERE rr.output IS NOT NULL AND (rr.error IS NULL OR rr.error::text = 'null')` to any aggregation that should ignore them.
 
 ## DQL quirks
 
@@ -339,6 +476,9 @@ FROM (...) AS subq
 GROUP BY task
 ```
 
+### Do not inline large lists of precomputed IDs
+When a filter depends on transcript shape (message count, metadata, joins), compute it in DQL with `JOIN`/`WHERE`/`GROUP BY` — not by pasting hundreds of UUIDs into `WHERE agent_runs.id IN ('…', '…', …)`. That pattern is hard to maintain, blows query size limits, and usually means the real filter belongs in SQL (see [Counting transcript messages](#counting-transcript-messages)).
+
 ### Avoid Dynamic IN Clauses with String Interpolation
 Building IN clauses with f-strings is dangerous:
 - Task names containing `::` can be parsed as PostgreSQL type casts
diff --git a/plugins/docent/skills/docent/ingestion.md b/plugins/docent/skills/docent/ingestion.md
index d4da82b..e0de68b 100644
--- a/plugins/docent/skills/docent/ingestion.md
+++ b/plugins/docent/skills/docent/ingestion.md
@@ -36,7 +36,7 @@ Before Python work, use an existing virtual environment if present. If no enviro
 
 Collect only what is needed to plan:
 
-- API key: prefer `$DOCENT_API_KEY` or an SDK-discovered `docent.env` (current directory upward, then `~/.docent/docent.env`); ask only if neither is available.
+- API key: prefer `$DOCENT_API_KEY` or an SDK-discovered config file. The default global file is `~/.docent/docent.env`; project-level `docent.env` files are supported as local overrides.
 - Data path: the file or directory to ingest.
 - Optional context: what produced the data and what analysis the user wants to do in Docent.
 
diff --git a/plugins/docent/skills/docent/readings-reference.md b/plugins/docent/skills/docent/readings-reference.md
index 990a475..aeef211 100644
--- a/plugins/docent/skills/docent/readings-reference.md
+++ b/plugins/docent/skills/docent/readings-reference.md
@@ -298,6 +298,17 @@ Glob filter rules:
 * Common pitfall: do not set `transcript_group_names=GlobFilter(include=("*",))` when the user asks to render only a specific transcript name. Including all transcript groups makes all visible descendants render, so it can override the intended narrow transcript selection. In that case, make `transcript_group_names` exclude-all and set only `transcript_names=GlobFilter(include=("<requested transcript name>",))`.
 * Transcript group filtering is path-scoped. Including a nested group makes that group and its visible descendants render, and any ancestors needed to reach it may render as wrappers. It does not make sibling branches visible. For example, if `G1` contains both `G2 -> G3` and `G2-prime`, including `G3` can render wrapper groups `G1` and `G2`, but `G2-prime` remains hidden unless it or one of its descendants is independently included.
 
+### Multiple rollouts
+
+If the user asks for multiple rollouts, you can use the `num_rollouts` parameter. Leave it unset (defaults to 1) unless the user explicitly asks.
+```python
+reading = client.read(
+    prompt_template=["Summarize: ", rows.transcript.as_type("transcript")],
+    model="openai/gpt-5.4-mini",
+    num_rollouts=3
+)
+```
+
 ### `client.step_group(label) -> StepGroupContext`
 Opens a labeled step group in the session UI. Use as a context manager to auto-close the group scope:
 ```python