fix: empty-content guard + finish_reason=length handling + reasoning budget (v0.18.1)

Reasoning models at high effort could spend the entire max_tokens budget
on thinking and return zero visible content with finish_reason=length;
the empty article flowed writer -> editor (reject) -> empty draft post,
with the cost booked as success (2026-06-11 incident: prod run acf24029,
draft post 921; diagnosis: convo pipeline-v2-phase1 seq 12).

- Empty-completion guard at the provider seam (new
  OpenRouter_Completion_Guard): empty/whitespace content is a failure;
  one retry with reasoning disabled (per-call override) when
  finish_reason=length or reasoning was active, then the stage fails.
- finish_reason=length always warning-logged (model, stage, tokens).
- Failed/retried attempts booked to generation_log with
  response_status=error + real token burn; run_decisions verdict rows
  (retry_reasoning_disabled / failed_empty_completion) for v0.18.0 runs.
- Writer-path guard in Content_Generator::execute_stage and
  belt-and-braces empty check in Publisher::create_post.
- Reasoning budget sanity: reasoning.max_tokens cap + completion
  headroom via new prautoblogger_reasoning_max_tokens setting (default
  constant 2048, 0 = legacy uncapped effort mode); cap replaces effort
  when active; governor reservations use the effective ceiling.
- Body assembly moved to Request_Builder::build_body (300-line cap).
- Tests: guard verdict matrix, builder budget shaping, writer/publisher
  guards, full provider retry path.

Agent-Session: prab-empty-draft-guard-20260611
This commit is contained in:
peptiderepo
2026-06-11 13:15:58 +00:00
parent f5658f1c97
commit 2bdc763dc0
19 changed files with 1196 additions and 34 deletions

View File

@@ -0,0 +1,240 @@
<?php
/**
* Tests for PRAutoBlogger_OpenRouter_Completion_Guard (v0.18.1).
*
* Validates the empty-completion verdict matrix: healthy pass-through,
* the single reasoning-disabled retry on length/reasoning empties, the
* never-retry-twice rule, and the hard failure for unexplained empties.
*
* @package PRAutoBlogger\Tests\Providers
*/
namespace PRAutoBlogger\Tests\Providers;
use PRAutoBlogger\Tests\BaseTestCase;
use Brain\Monkey\Functions;
class OpenRouterCompletionGuardTest extends BaseTestCase {
protected function setUp(): void {
parent::setUp();
\PRAutoBlogger_Run_Context::clear();
\PRAutoBlogger_Audit_Writer::flush_cache();
$this->stub_get_option( [
'prautoblogger_log_level' => 'error',
] );
// Cost_Tracker::log_api_call() (failed-attempt booking) inserts a
// generation_log row and prices the burn via the pricing chain.
$wpdb = $this->create_mock_wpdb();
$wpdb->method( 'insert' )->willReturn( 1 );
$wpdb->insert_id = 7;
$GLOBALS['wpdb'] = $wpdb;
}
protected function tearDown(): void {
unset( $GLOBALS['wpdb'] );
\PRAutoBlogger_Run_Context::clear();
\PRAutoBlogger_Audit_Writer::flush_cache();
parent::tearDown();
}
/**
* Build a parsed-response fixture in the Response_Parser shape.
*
* @param string $content Visible content.
* @param string $finish finish_reason value.
* @return array
*/
private function parsed( string $content, string $finish ): array {
return [
'content' => $content,
'model' => 'deepseek/deepseek-v4-flash',
'prompt_tokens' => 1200,
'completion_tokens' => 4000,
'reasoning_tokens' => 4000,
'total_tokens' => 5200,
'finish_reason' => $finish,
'reasoning_content' => 'thinking…',
];
}
/**
* Healthy completion passes through untouched; the retry callable is
* never invoked.
*/
public function test_healthy_completion_passes_through(): void {
$parsed = $this->parsed( 'A full article.', 'stop' );
$invoked = 0;
$result = ( new \PRAutoBlogger_OpenRouter_Completion_Guard() )->finalize(
$parsed,
[ 'stage' => 'draft' ],
false,
function () use ( &$invoked ) {
++$invoked;
return [];
}
);
$this->assertSame( $parsed, $result );
$this->assertSame( 0, $invoked );
}
/**
* The incident shape — empty content + finish_reason=length — retries
* exactly once with reasoning disabled and the empty_retry marker set,
* and returns the retry's result.
*/
public function test_empty_plus_length_retries_once_with_reasoning_disabled(): void {
$recovered = $this->parsed( 'Recovered article.', 'stop' );
$captured_opts = null;
$result = ( new \PRAutoBlogger_OpenRouter_Completion_Guard() )->finalize(
$this->parsed( '', 'length' ),
[ 'stage' => 'draft', 'prompt_key' => 'content.single_pass' ],
true,
function ( array $retry_options ) use ( &$captured_opts, $recovered ) {
$captured_opts = $retry_options;
return $recovered;
}
);
$this->assertSame( $recovered, $result );
$this->assertSame( [ 'enabled' => false ], $captured_opts['reasoning'] );
$this->assertTrue( $captured_opts['empty_retry'] );
$this->assertSame( 'draft', $captured_opts['stage'] );
}
/**
* Whitespace-only content counts as empty (the guard trims).
*/
public function test_whitespace_only_content_counts_as_empty(): void {
$invoked = 0;
( new \PRAutoBlogger_OpenRouter_Completion_Guard() )->finalize(
$this->parsed( " \n\t ", 'length' ),
[ 'stage' => 'polish' ],
true,
function ( array $retry_options ) use ( &$invoked ) {
++$invoked;
return $this->parsed( 'ok', 'stop' );
}
);
$this->assertSame( 1, $invoked );
}
/**
* An attempt already marked empty_retry is never retried again — the
* guard fails the call instead (no infinite retry loops).
*/
public function test_already_retried_empty_fails_without_second_retry(): void {
$invoked = 0;
try {
( new \PRAutoBlogger_OpenRouter_Completion_Guard() )->finalize(
$this->parsed( '', 'length' ),
[ 'stage' => 'draft', 'empty_retry' => true ],
false,
function () use ( &$invoked ) {
++$invoked;
return [];
}
);
$this->fail( 'Expected RuntimeException for retried-and-still-empty completion.' );
} catch ( \RuntimeException $e ) {
$this->assertStringContainsString( 'empty content', $e->getMessage() );
$this->assertStringContainsString( 'after a reasoning-disabled retry', $e->getMessage() );
}
$this->assertSame( 0, $invoked );
}
/**
* Empty content with finish_reason=stop and no reasoning active: a
* reasoning-disabled retry cannot plausibly help — fail immediately.
*/
public function test_unexplained_empty_fails_without_retry(): void {
$invoked = 0;
try {
( new \PRAutoBlogger_OpenRouter_Completion_Guard() )->finalize(
$this->parsed( '', 'stop' ),
[ 'stage' => 'review' ],
false,
function () use ( &$invoked ) {
++$invoked;
return [];
}
);
$this->fail( 'Expected RuntimeException for unexplained empty completion.' );
} catch ( \RuntimeException $e ) {
$this->assertStringContainsString( 'stage "review"', $e->getMessage() );
$this->assertStringContainsString( 'finish_reason=stop', $e->getMessage() );
}
$this->assertSame( 0, $invoked );
}
/**
* Empty content with reasoning active retries even when finish_reason
* is not 'length' (reasoning can eat the budget without a length flag).
*/
public function test_empty_with_reasoning_active_retries_on_any_finish_reason(): void {
$invoked = 0;
( new \PRAutoBlogger_OpenRouter_Completion_Guard() )->finalize(
$this->parsed( '', 'stop' ),
[ 'stage' => 'draft' ],
true,
function ( array $retry_options ) use ( &$invoked ) {
++$invoked;
return $this->parsed( 'ok', 'stop' );
}
);
$this->assertSame( 1, $invoked );
}
/**
* The failed attempt is booked to generation_log with its REAL status:
* response_status='error', the stage, the model, and the token burn.
*/
public function test_failed_attempt_books_generation_log_row_with_error_status(): void {
$captured_row = null;
$wpdb = $this->create_mock_wpdb();
$wpdb->method( 'insert' )->willReturnCallback(
function ( $table, $row ) use ( &$captured_row ) {
if ( false !== strpos( (string) $table, 'generation_log' ) ) {
$captured_row = $row;
}
return 1;
}
);
$wpdb->insert_id = 9;
$GLOBALS['wpdb'] = $wpdb;
try {
( new \PRAutoBlogger_OpenRouter_Completion_Guard() )->finalize(
$this->parsed( '', 'stop' ),
[ 'stage' => 'draft' ],
false,
function () {
return [];
}
);
$this->fail( 'Expected RuntimeException.' );
} catch ( \RuntimeException $e ) {
$this->assertIsArray( $captured_row );
$this->assertSame( 'error', $captured_row['response_status'] );
$this->assertSame( 'draft', $captured_row['stage'] );
$this->assertSame( 'deepseek/deepseek-v4-flash', $captured_row['model'] );
$this->assertSame( 4000, $captured_row['completion_tokens'] );
$this->assertStringContainsString( 'Empty completion', $captured_row['error_message'] );
}
}
}

View File

@@ -166,4 +166,153 @@ class OpenRouterProviderTest extends BaseTestCase {
$this->expectException( \RuntimeException::class );
$provider->send_chat_completion( $messages, 'model/test', $options );
}
/**
* Wire the provider for a full send_chat_completion run: decryptable
* API key, reasoning settings, HTTP stubs that replay scripted
* responses, and a wpdb mock for the guard's failure bookkeeping.
*
* @param array $option_overrides Extra get_option values.
* @param array $responses Scripted JSON-decodable response bodies.
* @param array $requests (by ref) captures decoded request bodies.
*/
private function wire_full_provider( array $option_overrides, array $responses, array &$requests ): void {
\PRAutoBlogger_Run_Context::clear();
$encrypted_key = \PRAutoBlogger_Encryption::encrypt( 'sk-or-v1-unit-test-key-0000' );
$this->stub_get_option( array_merge( [
'prautoblogger_openrouter_api_key' => $encrypted_key,
'prautoblogger_log_level' => 'error',
], $option_overrides ) );
Functions\when( 'absint' )->alias( function ( $val ) {
return abs( (int) $val );
} );
Functions\when( 'wp_remote_post' )->alias(
function ( $url, $args ) use ( &$requests, $responses ) {
$requests[] = json_decode( $args['body'], true );
$body = $responses[ count( $requests ) - 1 ];
return [
'body' => $body,
'response' => [ 'code' => 200 ],
];
}
);
Functions\when( 'wp_remote_retrieve_response_code' )->alias(
function ( $response ) {
return $response['response']['code'];
}
);
Functions\when( 'wp_remote_retrieve_body' )->alias(
function ( $response ) {
return $response['body'];
}
);
// The guard books failed attempts through Cost_Tracker + Logger.
$wpdb = $this->create_mock_wpdb();
$wpdb->method( 'insert' )->willReturn( 1 );
$wpdb->insert_id = 11;
$GLOBALS['wpdb'] = $wpdb;
}
/**
* v0.18.1 incident shape, end to end: reasoning enabled (uncapped
* effort mode), the first response burns the whole completion budget
* on thinking and returns empty content with finish_reason=length —
* the provider retries exactly once with reasoning disabled and
* returns the recovered completion.
*/
public function test_empty_length_completion_retries_once_with_reasoning_disabled(): void {
$requests = [];
$empty_response = json_encode( [
'model' => 'deepseek/deepseek-v4-flash',
'choices' => [ [ 'message' => [ 'content' => '', 'reasoning' => 'endless thinking…' ], 'finish_reason' => 'length' ] ],
'usage' => [ 'prompt_tokens' => 1200, 'completion_tokens' => 4000, 'reasoning_tokens' => 4000, 'total_tokens' => 5200 ],
] );
$good_response = json_encode( [
'model' => 'deepseek/deepseek-v4-flash',
'choices' => [ [ 'message' => [ 'content' => 'Recovered article.' ], 'finish_reason' => 'stop' ] ],
'usage' => [ 'prompt_tokens' => 1200, 'completion_tokens' => 900, 'total_tokens' => 2100 ],
] );
$this->wire_full_provider(
[
'prautoblogger_reasoning_enabled' => '1',
'prautoblogger_reasoning_effort' => 'xhigh',
'prautoblogger_reasoning_max_tokens' => 0,
],
[ $empty_response, $good_response ],
$requests
);
try {
$provider = new \PRAutoBlogger_OpenRouter_Provider();
$result = $provider->send_chat_completion(
[ [ 'role' => 'user', 'content' => 'Write the article.' ] ],
'deepseek/deepseek-v4-flash',
[ 'temperature' => 0.7, 'max_tokens' => 4000, 'stage' => 'draft' ]
);
$this->assertSame( 'Recovered article.', $result['content'] );
$this->assertCount( 2, $requests );
// First attempt: global reasoning, uncapped effort mode.
$this->assertSame( [ 'enabled' => true, 'effort' => 'xhigh' ], $requests[0]['reasoning'] );
$this->assertSame( 4000, $requests[0]['max_tokens'] );
// Retry: reasoning disabled per call.
$this->assertSame( [ 'enabled' => false ], $requests[1]['reasoning'] );
// Caller metadata never reaches the HTTP body.
$this->assertArrayNotHasKey( 'stage', $requests[0] );
$this->assertArrayNotHasKey( 'empty_retry', $requests[1] );
} finally {
unset( $GLOBALS['wpdb'] );
\PRAutoBlogger_Run_Context::clear();
}
}
/**
* v0.18.1 reasoning budget: with the token cap active (default), the
* outgoing request caps thinking via reasoning.max_tokens and raises
* max_tokens by the cap — and a healthy completion needs no retry.
*/
public function test_reasoning_cap_shapes_the_outgoing_request(): void {
$requests = [];
$good_response = json_encode( [
'model' => 'deepseek/deepseek-v4-flash',
'choices' => [ [ 'message' => [ 'content' => 'Article text.' ], 'finish_reason' => 'stop' ] ],
'usage' => [ 'prompt_tokens' => 1200, 'completion_tokens' => 900, 'total_tokens' => 2100 ],
] );
$this->wire_full_provider(
[
'prautoblogger_reasoning_enabled' => '1',
'prautoblogger_reasoning_effort' => 'xhigh',
'prautoblogger_reasoning_max_tokens' => 2048,
],
[ $good_response ],
$requests
);
try {
$provider = new \PRAutoBlogger_OpenRouter_Provider();
$result = $provider->send_chat_completion(
[ [ 'role' => 'user', 'content' => 'Write the article.' ] ],
'deepseek/deepseek-v4-flash',
[ 'temperature' => 0.7, 'max_tokens' => 4000, 'stage' => 'draft' ]
);
$this->assertSame( 'Article text.', $result['content'] );
$this->assertCount( 1, $requests );
$this->assertSame( 2048, $requests[0]['reasoning']['max_tokens'] );
$this->assertArrayNotHasKey( 'effort', $requests[0]['reasoning'] );
$this->assertSame( 6048, $requests[0]['max_tokens'] );
} finally {
unset( $GLOBALS['wpdb'] );
\PRAutoBlogger_Run_Context::clear();
}
}
}

View File

@@ -0,0 +1,189 @@
<?php
/**
* Tests for PRAutoBlogger_OpenRouter_Request_Builder::build_body() (v0.18.1).
*
* Validates body assembly, the global-vs-per-call reasoning precedence, the
* reasoning token cap + completion headroom, and that caller-metadata keys
* never leak into the HTTP body.
*
* @package PRAutoBlogger\Tests\Providers
*/
namespace PRAutoBlogger\Tests\Providers;
use PRAutoBlogger\Tests\BaseTestCase;
use Brain\Monkey\Functions;
class OpenRouterRequestBuilderTest extends BaseTestCase {
private const MESSAGES = [
[ 'role' => 'system', 'content' => 'sys' ],
[ 'role' => 'user', 'content' => 'usr' ],
];
protected function setUp(): void {
parent::setUp();
Functions\when( 'absint' )->alias( function ( $val ) {
return abs( (int) $val );
} );
}
/**
* Reasoning enabled globally + cap set: thinking is capped via
* reasoning.max_tokens, effort is dropped, and the completion budget
* is raised by the cap so content keeps its full allowance.
*/
public function test_reasoning_cap_replaces_effort_and_raises_headroom(): void {
$this->stub_get_option( [
'prautoblogger_reasoning_enabled' => '1',
'prautoblogger_reasoning_effort' => 'xhigh',
'prautoblogger_reasoning_max_tokens' => 2048,
] );
$body = ( new \PRAutoBlogger_OpenRouter_Request_Builder() )
->build_body( self::MESSAGES, 'deepseek/deepseek-v4-flash', [ 'max_tokens' => 4000 ] );
$this->assertSame( 2048, $body['reasoning']['max_tokens'] );
$this->assertArrayNotHasKey( 'effort', $body['reasoning'] );
$this->assertSame( 6048, $body['max_tokens'] );
$this->assertTrue( $body['reasoning']['enabled'] );
}
/**
* Cap setting 0 = pure effort mode: the historical body shape is
* preserved exactly (effort kept, no reasoning.max_tokens, no headroom).
*/
public function test_cap_zero_keeps_pure_effort_mode(): void {
$this->stub_get_option( [
'prautoblogger_reasoning_enabled' => '1',
'prautoblogger_reasoning_effort' => 'xhigh',
'prautoblogger_reasoning_max_tokens' => 0,
] );
$body = ( new \PRAutoBlogger_OpenRouter_Request_Builder() )
->build_body( self::MESSAGES, 'deepseek/deepseek-v4-flash', [ 'max_tokens' => 4000 ] );
$this->assertSame( [ 'enabled' => true, 'effort' => 'xhigh' ], $body['reasoning'] );
$this->assertSame( 4000, $body['max_tokens'] );
}
/**
* Per-call reasoning override (the guard's retry) beats the global
* setting; a disabled override gets no cap and no headroom.
*/
public function test_per_call_disabled_override_beats_global_setting(): void {
$this->stub_get_option( [
'prautoblogger_reasoning_enabled' => '1',
'prautoblogger_reasoning_effort' => 'xhigh',
'prautoblogger_reasoning_max_tokens' => 2048,
] );
$body = ( new \PRAutoBlogger_OpenRouter_Request_Builder() )
->build_body(
self::MESSAGES,
'deepseek/deepseek-v4-flash',
[
'max_tokens' => 4000,
'reasoning' => [ 'enabled' => false ],
]
);
$this->assertSame( [ 'enabled' => false ], $body['reasoning'] );
$this->assertSame( 4000, $body['max_tokens'] );
}
/**
* Global reasoning off + no override: no reasoning key at all and no
* budget math (historical default path).
*/
public function test_reasoning_off_produces_no_reasoning_key(): void {
$this->stub_get_option( [
'prautoblogger_reasoning_enabled' => '0',
] );
$body = ( new \PRAutoBlogger_OpenRouter_Request_Builder() )
->build_body(
self::MESSAGES,
'google/gemini-2.5-flash-lite',
[
'temperature' => 0.7,
'max_tokens' => 4000,
'response_format' => [ 'type' => 'json_object' ],
]
);
$this->assertArrayNotHasKey( 'reasoning', $body );
$this->assertSame( 0.7, $body['temperature'] );
$this->assertSame( 4000, $body['max_tokens'] );
$this->assertSame( [ 'type' => 'json_object' ], $body['response_format'] );
}
/**
* An explicit caller reasoning config with its own max_tokens is
* respected (no cap overwrite) but still gets completion headroom.
*/
public function test_caller_reasoning_max_tokens_respected_with_headroom(): void {
$this->stub_get_option( [
'prautoblogger_reasoning_enabled' => '0',
'prautoblogger_reasoning_max_tokens' => 2048,
] );
$body = ( new \PRAutoBlogger_OpenRouter_Request_Builder() )
->build_body(
self::MESSAGES,
'deepseek/deepseek-v4-flash',
[
'max_tokens' => 1000,
'reasoning' => [ 'enabled' => true, 'max_tokens' => 512 ],
]
);
$this->assertSame( 512, $body['reasoning']['max_tokens'] );
$this->assertSame( 1512, $body['max_tokens'] );
}
/**
* Caller-metadata keys (stage, prompt_key, empty_retry) must never
* appear in the HTTP body.
*/
public function test_caller_metadata_keys_never_leak_into_body(): void {
$this->stub_get_option( [
'prautoblogger_reasoning_enabled' => '0',
] );
$body = ( new \PRAutoBlogger_OpenRouter_Request_Builder() )
->build_body(
self::MESSAGES,
'google/gemini-2.5-flash-lite',
[
'max_tokens' => 4000,
'stage' => 'draft',
'prompt_key' => 'content.single_pass',
'empty_retry' => true,
]
);
$this->assertArrayNotHasKey( 'stage', $body );
$this->assertArrayNotHasKey( 'prompt_key', $body );
$this->assertArrayNotHasKey( 'empty_retry', $body );
$this->assertSame( [ 'model', 'messages', 'max_tokens' ], array_keys( $body ) );
}
/**
* Reasoning enabled but the call has no completion ceiling: nothing to
* protect, body passes through with effort untouched.
*/
public function test_no_max_tokens_means_no_budget_math(): void {
$this->stub_get_option( [
'prautoblogger_reasoning_enabled' => '1',
'prautoblogger_reasoning_effort' => 'high',
'prautoblogger_reasoning_max_tokens' => 2048,
] );
$body = ( new \PRAutoBlogger_OpenRouter_Request_Builder() )
->build_body( self::MESSAGES, 'deepseek/deepseek-v4-flash', [] );
$this->assertSame( [ 'enabled' => true, 'effort' => 'high' ], $body['reasoning'] );
$this->assertArrayNotHasKey( 'max_tokens', $body );
}
}