feat: polish raw query diff output

docs/query-diffs.md

@@ -37,6 +37,8 @@ from a UTF-8 file, and choose keyed or unkeyed comparison explicitly:
 sqb diff \
   --left-query 'SELECT order_id, amount_cents FROM analytics.orders_before' \
   --right-query 'SELECT order_id, amount_cents FROM analytics.orders_after' \
+  --left-label before \
+  --right-label after \
   --key order_id
 ```
 
@@ -56,6 +58,9 @@ Query mode performs a full comparison when no comparison mode flag is present. `
 also available. `--bounded` is model-specific; filter both raw queries directly when a bounded
 query comparison is needed.
 
+Raw-query discovery loads only project configuration, target configuration, and the selected
+adapter. It does not import unrelated project sinks, hooks, loaders, or providers.
+
 ### Sampling, tolerances, and exclusions
 
 Keyed query comparisons support deterministic union-key sampling and the same bounded mismatch
@@ -85,6 +90,31 @@ sqb diff \
 `--exhaustive` disables inherited or requested sampling. Unkeyed comparison is always exhaustive
 and exact, so it does not accept sampling or numeric tolerances.
 
+### Bounded example values
+
+Comparison remains exact regardless of how examples are rendered. By default, long values are
+limited to 160 characters and show context around the first differing character. Every shortened
+value is marked as truncated and includes its original length.
+
+Control evidence independently from the number of collected examples:
+
+```bash
+# Increase the displayed context without changing comparison scope.
+sqb diff --left-query-file before.sql --right-query-file after.sql \
+  --key order_id --max-value-length 500
+
+# Retain keys and aggregate counts but suppress example values.
+sqb diff --left-query-file before.sql --right-query-file after.sql \
+  --key order_id --no-example-values
+
+# Explicitly opt in to complete values. --verbose alone remains bounded.
+sqb diff --left-query-file before.sql --right-query-file after.sql \
+  --key order_id --full-example-values
+```
+
+Ordinary keys are kept complete. Exceptionally long keys are shortened with an explicit length and
+stable digest so an operator can still identify them without flooding terminal output.
+
 ## Stable query results and cleanup
 
 SQLBuild materializes each raw query once in the selected target's default schema before comparing
@@ -115,6 +145,20 @@ error.
 
 ## Structured output and exit behavior
 
-Use `--json-output PATH` in either mode. Query results include `input_kind = "query"`; model results
-use `input_kind = "model"`. Exit code `0` means no differences were found. Exit code `1` means
-differences were found or the comparison could not execute successfully.
+Use `--json-output PATH` in either mode, or `--json` to emit one JSON document to stdout while all
+lifecycle progress remains on stderr. Query results include `input_kind = "query"`, named left and
+right inputs, `query_comparisons`, sparse nonzero `column_mismatches`, lifecycle timings, and
+explicit example truncation metadata. Existing `models` output remains available for compatibility.
+
+Raw-query automation distinguishes complete findings from incomplete evidence and execution
+failure:
+
+| Exit | Outcome |
+|---:|---|
+| `0` | Complete comparison with no differences |
+| `1` | Complete comparison with findings |
+| `2` | Safety, schema, or key preflight prevented complete value evidence |
+| `3` | Setup, connection, materialization, comparison execution, ownership, or cleanup failed |
+
+Human output and JSON use the same outcome. Lifecycle messages include elapsed connection,
+reconciliation, inspection, materialization, comparison, cleanup, and total durations.

src/sqlbuild/.agents/skills/sqlbuild/SKILL.md

@@ -7780,6 +7780,8 @@ SQLBuild templates such as `ref()` are not expanded.
 sqb diff \
   --left-query 'SELECT order_id, amount_cents FROM analytics.orders_before' \
   --right-query 'SELECT order_id, amount_cents FROM analytics.orders_after' \
+  --left-label before \
+  --right-label after \
   --key order_id
 
 sqb diff \
@@ -7793,6 +7795,9 @@ sqb diff \
 Query mode defaults to a full comparison and also supports `--schema-only`. It does not accept
 `--bounded`; put the desired filter in both SQL inputs instead.
 
+Raw-query mode loads configuration and the selected adapter without importing unrelated project
+sinks, hooks, loaders, or providers.
+
 ### Comparison modes
 
 Model mode requires exactly one mode. Raw-query mode defaults to full when no mode is supplied:
@@ -7979,19 +7984,47 @@ sqb diff prod:dev --full --select customer_status_snapshot --max-column-examples
 Example limits only control diagnostic values printed after comparison. They are separate from
 `row_diff_sample_rows`, which controls how many unique keys receive the wide value comparison.
 
+Long values default to 160 rendered characters. SQLBuild shows context around the first difference,
+marks truncation, and reports the original length without changing exact comparison counts. Control
+value evidence independently from example counts:
+
+```bash
+sqb diff \
+  --left-query-file queries/orders_before.sql \
+  --right-query-file queries/orders_after.sql \
+  --key order_id \
+  --max-value-length 500
+
+sqb diff \
+  --left-query-file queries/orders_before.sql \
+  --right-query-file queries/orders_after.sql \
+  --key order_id \
+  --no-example-values
+
+sqb diff \
+  --left-query-file queries/orders_before.sql \
+  --right-query-file queries/orders_after.sql \
+  --key order_id \
+  --full-example-values
+```
+
+`--verbose` increases the example count but does not remove the value-length bound.
+
 ### Structured output
 
 Use `--json-output PATH` to write a stable structured result while retaining the normal terminal
-summary:
+summary, or `--json` to write one JSON document to stdout with progress on stderr:
 
 ```bash
 sqb diff prod:dev --bounded 14d --sample-rows 50000 --select order_lines --json-output diff.json
+sqb diff --left-query-file before.sql --right-query-file after.sql --key order_id --json
 ```
 
 Each model records `schema_only`, `exhaustive`, or `sampled` comparison scope; requested and observed
 cursor coverage; bounded population and compared key counts; seed and configured limit; row result
-counts; and changed-column counts. The top-level status distinguishes `no_differences_found` from
-`differences_found`.
+counts; and changed-column counts. Query output additionally includes named inputs,
+`query_comparisons`, sparse nonzero `column_mismatches`, lifecycle timings, and explicit evidence
+truncation metadata. Existing `models` output remains for compatibility.
 
 ### Invocation safety limits
 
@@ -8022,9 +8055,10 @@ sqb diff prod:dev --full --select tag:acceptance
 
 ### Exit codes
 
-`sqb diff` returns exit code `0` when no differences are found, and `1` when any input has schema or
-row differences or the comparison cannot execute successfully. This makes it usable in CI pipelines
-as a validation gate.
+Complete comparisons return `0` when no differences are found and `1` for findings. Raw-query mode
+returns `2` when safety, schema, or key preflight prevents complete value evidence, and `3` for
+setup, connection, materialization, comparison execution, ownership publication, or cleanup
+failure. Human and JSON output carry the same outcome.
 
 ## Execution Observability
 
@@ -14541,17 +14575,23 @@ and falls back to a full row comparison when no cursor is configured.
 | `--sample-seed` | Override the deterministic sample seed |
 | `--exhaustive` | Disable inherited sampling for this invocation |
 | `--json-output` | Write structured comparison scope, coverage, and results to a JSON file |
+| `--json` | Write one structured JSON document to stdout; progress remains on stderr |
 | `--max-models` | Fail when the selected scope contains more models than this limit |
 | `--max-columns` | Fail when either side of a model has more columns than this limit |
 | `--left-query` | Literal read-only SQL for the left side |
 | `--left-query-file` | UTF-8 SQL file for the left side |
 | `--right-query` | Literal read-only SQL for the right side |
 | `--right-query-file` | UTF-8 SQL file for the right side |
+| `--left-label` | Display and structured-output label for the left query |
+| `--right-label` | Display and structured-output label for the right query |
 | `--target` | Target connection for raw-query mode; defaults to the active target |
 | `--key` | Row identity column; repeat for a composite key or to override model configuration |
 | `--unkeyed` | Compare exact full-row multiplicities without a key |
 | `--exclude-column` | Column to omit from row comparison; repeat as needed |
 | `--tolerance` | Per-column numeric rule such as `amount:absolute=0.01`; repeat as needed |
+| `--max-value-length` | Maximum rendered characters per example value (default: 160) |
+| `--no-example-values` | Retain example keys and counts while suppressing values |
+| `--full-example-values` | Explicitly opt in to complete example values |
 | `--no-sql-analysis` | Disable compile-time SQL analysis (`--no-sql-validation` is an alias) |
 | `--select`, `-s` | Select specific models to diff (required in v1) |
 | `--exclude` | Exclude specific models from diffing |
@@ -14578,6 +14618,8 @@ sqb diff prod:dev --bounded 14d --exhaustive --select order_lines
 sqb diff \
   --left-query-file queries/orders_before.sql \
   --right-query-file queries/orders_after.sql \
+  --left-label before \
+  --right-label after \
   --key account_id \
   --key order_id
 ```

src/sqlbuild/cli/commands/_helpers/diff/evidence.py

@@ -0,0 +1,108 @@
+"""Bounded, diff-aware rendering for row-diff evidence."""
+
+from __future__ import annotations
+
+import hashlib
+
+from sqlbuild.cli.commands.models import DiffExampleRenderOptions, RenderedDiffExampleValue
+
+_MAX_RENDERED_KEY_LENGTH: int = 256
+
+
+def render_example_pair(
+    *,
+    left_value: object,
+    right_value: object,
+    options: DiffExampleRenderOptions,
+) -> tuple[RenderedDiffExampleValue, RenderedDiffExampleValue]:
+    """Render two unequal values around their first differing character."""
+
+    left_text: str = str(left_value)
+    right_text: str = str(right_value)
+    first_difference: int | None = _first_difference(left=left_text, right=right_text)
+    return (
+        _render_value(text=left_text, first_difference=first_difference, options=options),
+        _render_value(text=right_text, first_difference=first_difference, options=options),
+    )
+
+
+def display_example_value(value: RenderedDiffExampleValue) -> str:
+    """Render one bounded value for terminal output."""
+
+    if value.suppressed:
+        return f"<value suppressed; {value.original_length:,} chars>"
+    rendered: str = value.text or ""
+    if value.truncated:
+        return f"{rendered} [truncated; {value.original_length:,} chars]"
+    return rendered
+
+
+def render_key_value(value: object) -> str:
+    """Keep ordinary keys complete and identify exceptionally long shortened keys."""
+
+    text: str = str(value)
+    if len(text) <= _MAX_RENDERED_KEY_LENGTH:
+        return text
+    digest: str = hashlib.sha256(text.encode("utf-8")).hexdigest()[:12]
+    return f"{text[:_MAX_RENDERED_KEY_LENGTH]}… [truncated; {len(text):,} chars; sha256:{digest}]"
+
+
+def example_value_payload(value: RenderedDiffExampleValue) -> dict[str, object]:
+    """Build stable structured evidence for one rendered value."""
+
+    return {
+        "value": value.text,
+        "original_length": value.original_length,
+        "truncated": value.truncated,
+        "suppressed": value.suppressed,
+        "first_difference": value.first_difference,
+    }
+
+
+def _render_value(
+    *,
+    text: str,
+    first_difference: int | None,
+    options: DiffExampleRenderOptions,
+) -> RenderedDiffExampleValue:
+    original_length: int = len(text)
+    if options.suppress_values:
+        return RenderedDiffExampleValue(
+            text=None,
+            original_length=original_length,
+            truncated=False,
+            suppressed=True,
+            first_difference=first_difference,
+        )
+    limit: int | None = options.max_value_length
+    if limit is None or original_length <= limit:
+        return RenderedDiffExampleValue(
+            text=text,
+            original_length=original_length,
+            truncated=False,
+            suppressed=False,
+            first_difference=first_difference,
+        )
+    difference: int = first_difference or 0
+    context_before: int = min(limit // 3, max(limit - 1, 0))
+    start: int = min(max(difference - context_before, 0), original_length - limit)
+    end: int = start + limit
+    fragment: str = text[start:end]
+    rendered: str = f"{'…' if start else ''}{fragment}{'…' if end < original_length else ''}"
+    return RenderedDiffExampleValue(
+        text=rendered,
+        original_length=original_length,
+        truncated=True,
+        suppressed=False,
+        first_difference=first_difference,
+    )
+
+
+def _first_difference(*, left: str, right: str) -> int | None:
+    index: int
+    for index, (left_char, right_char) in enumerate(zip(left, right, strict=False)):
+        if left_char != right_char:
+            return index
+    if len(left) != len(right):
+        return min(len(left), len(right))
+    return None