Run: 2026-07-25-01KYCWKQ60J23TN3PE4JKA40HE

Run Details

Model:
Gemini 3.5 Flash Lite (Reasoning, Minimal)
Test:
Bad Writing Habits
Scenario:
Mystery: examining a crime scene
Duration:
3.6s
Cost:
$0.001757
Provider:
Google AI Studio

Evaluation summary

Score Evaluator
Details Link
100.00% Adverbs in dialogue tags (details)
100.00% AI-ism adverb frequency (details)
100.00% AI-ism character names (details)
100.00% AI-ism location names (details)
23.25% AI-ism word frequency (details)
100.00% Cliché density (details)
100.00% Emotion telling (show vs. tell) (details)
100.00% Filter word density (details)
100.00% Gibberish response detection (details)
100.00% Markdown formatting overuse (details)
100.00% Missing dialogue indicators (quotation marks) (details)
84.21% Name drop frequency (details)
100.00% Narrator intent-glossing (details)
100.00% "Not X but Y" pattern overuse (details)
100.00% Overuse of "that" (subordinate clause padding) (details)
100.00% Paragraph length variance (details)
100.00% Passive voice overuse (details)
100.00% Past progressive (was/were + -ing) overuse (details)
100.00% Em-dash & semicolon overuse (details)
100.00% Purple prose (modifier overload) (details)
100.00% Repeated phrase echo (details)
100.00% Sentence length variance (details)
71.90% Sentence opener variety (details)
0.00% Adverb-first sentence starts (details)
100.00% Pronoun-first sentence starts (details)
32.92% Subject-first sentence starts (details)
0.00% Subordinate conjunction sentence starts (details)
100.00% Technical jargon density (details)
100.00% Useless dialogue additions (details)
100.00% Dialogue tag variety (said vs. fancy) (details)
87.0756% Overall

Model input

System
You are an expert fiction writer. Always keep the following rules in mind: - Write in past tense and use General English spelling, grammar, and colloquialisms/slang. - Write in active voice - Always follow the "show, don't tell" principle. - Avoid adverbs and cliches and overused/commonly used phrases. Aim for fresh and original descriptions. - Convey events and story through dialogue. - Mix short, punchy sentences with long, descriptive ones. Drop fill words to add variety. - Skip "he/she said said" dialogue tags and convey people's actions or face expressions through their speech - Avoid mushy dialog and descriptions, have dialogue always continue the action, never stall or add unnecessary fluff. Vary the descriptions to not repeat yourself. - Put dialogue on its own paragraph to separate scene and action. - Reduce indicators of uncertainty like "trying" or "maybe" When writing text: - NEVER conclude the scene on your own, follow the beat instructions very closely. - NEVER end with foreshadowing. - NEVER write further than what I prompt you with. - AVOID imagining possible endings, NEVER deviate from the instructions. - STOP EARLY if the continuation contains what was required in the instructions. You do not need to fill out the full amount of words possible.
User/Human
Take into account the following glossary of characters/locations/items/lore... when writing your response: <codex> <character id="4416" name="Detective Harlow Quinn" aliases="Quinn, Detective Quinn, Harlow" pronouns="she/her" species="Human" storyRole="Antagonist"> <eyeColor>brown</eyeColor> <hair>closely cropped salt-and-pepper</hair> <height>5'9"</height> <age>41</age> <jawline>sharp jaw</jawline> <watch>worn leather watch on her left wrist</watch> <bearing>military precision</bearing> <description>A Metropolitan Police detective with 18 years of decorated service who suspects the clique is involved in criminal activity.</description> <backstory>She lost her partner, DS Morris, during a case three years ago under unexplained circumstances that had supernatural origins she doesn't yet understand.</backstory> </character> <character id="4414" name="Eva Kowalski" aliases="Eva, Evie" pronouns="she/her" species="Human" storyRole="Best Friend"> <eyeColor>green</eyeColor> <hair>curly red</hair> <height>5'4"</height> <age>26</age> <glasses>round glasses</glasses> <satchel>worn leather satchel full of books</satchel> <freckles>freckled complexion</freckles> <nervousHabit>tucking hair behind her left ear</nervousHabit> <description>Aurora's childhood friend and an occult researcher.</description> <backstory>Moved to London two years before Aurora and works as a research assistant at the British Museum's restricted archives. She holds a master's degree in Ancient History from Oxford.</backstory> </character> <location id="856" name="The Veil Market" aliases="Veil Market, the Market, underground market" realm="Earth"> <currentLocation>abandoned Tube station beneath Camden</currentLocation> <entryRequirement>bone token</entryRequirement> <schedule>moves locations every full moon</schedule> <description>A hidden supernatural black market that sells enchanted goods, banned alchemical substances, and information.</description> </location> <item id="2004" name="Veil Compass" aliases="the Compass, shadow compass" type="Tool" origin="The Veil Market"> <casingMaterial>small brass compass</casingMaterial> <faceMarkings>face etched with protective sigils</faceMarkings> <needle>points toward the nearest supernatural rift or portal</needle> <patina>casing has a patina of verdigris</patina> <creator>crafted by a Shade artisan</creator> <description>A compass attuned to supernatural energy.</description> </item> </codex> Continue the story using the following instructions: <instructions> Write a fully fleshed out scene of at least 1000 words in the mystery genre. A detective arrives at a crime scene in an unusual location. Something is wrong — the evidence doesn't add up. A colleague is present, offering a different interpretation. Through careful observation and deduction, the detective begins to see what others have missed. Use the codex entries above as your world reference for characters and setting. Only include the prose of the scene without any markdown or commentary. </instructions>

Model response

AI
The damp air of the abandoned Tube station clings to wool coats and rusts iron rails. Detective Harlow Quinn steps past the yellow police tape fluttering in the draft of the Northern Line tunnel. Her leather watch clicks against her wrist as she adjusts the collar of her trench coat. Salt-and-pepper cropped hair catches the harsh glare of a forensic floodlight rigged to a wooden crate. Footsteps echo against the tiled floor of the disused platform. A constable stands beside a chalk outline drawn on the soot-stained concrete. Another suicide jump. The driver hit the brakes too late. Harlow ignores the chatter of the uniform cops crowding the turnstiles. She crouches beside the chalk mark. The surface lacks the expected impact fractures. No scoring on the concrete. The dust remains undisturbed in a perimeter of three feet around the body. The bloke leaped straight from the edge without hesitation. Harlow traces a gloved finger along the edge of the platform. The soot forms a clean line. No scuff marks. No panic. People struggle on the way down. Shoes scrape concrete. Fingernails scratch tile. A shadow falls across the chalk lines as a junior officer steps closer with a clipboard. Standard London Underground protocol applies here. Case closes by morning tea. Harlow stands up. She stares down the dark tunnel toward Camden. A faint hum vibrates through the iron rails beneath her boots. Not the rumble of a midnight engineering train. Something sharper. A metallic resonance that sets her teeth on edge. The victim wore a wool trench coat and carried a worn leather satchel full of books. His glasses lie shattered a yard away. Red hair clumps around his temples. He didn't fall. Someone placed him here after the last service passed. The junior officer scoffs and flips a page on his aluminum clipboard. The driver swore he saw someone jump from the edge. The CCTV camera at the top of the stairwell confirms nobody entered the station since midnight. Harlow pulls a small brass tool from her pocket. The Veil Compass clicks open in her palm. The verdigris-crusted casing spins wildly before the needle locks hard toward the tunnel wall behind the platform billboard. Look at the dust patterns near the exit tunnel. The officer squints down at the floor. Just soot. Footprints vanish three feet inside the dark mouth of the service tunnel. No footprints lead back out. The air tastes of ozone and copper. Harlow steps past the police cordon and walks toward the dark mouth of the service tunnel. Her sharp jaw tightens. She traces the wall with her bare hand, feeling the chill of ancient brickwork give way to a slick, organic warmth. Nothing in this station stays dead for long.
created_at"2026-07-25T14:59:31.146Z" 7/25/2026, 2:59:31 PM
model"google/gemini-3.5-flash-lite-20260721"
app_id182717
external_user(null)
streamedtrue
cancelledfalse
latency298
moderation_latency(null)
generation_time3561
tokens_prompt1112
tokens_completion692
native_tokens_prompt1073
native_tokens_completion574
native_tokens_completion_images0
native_tokens_reasoning0
native_tokens_cached0
num_media_prompt(null)
num_input_audio_prompt(null)
num_media_completion0
num_search_results(null)
num_fetches(null)
web_search_engine(null)
origin"https://ncbench.com/"
is_byokfalse
finish_reason"stop"
native_finish_reason"STOP"
service_tier"default"
usage0.0017569
router(null)
provider_responses
0
endpoint_id"6bd8f433-79e1-416c-b407-1772eb796c9a"
id"U89kauaBDsOT-8YP0c7zgAo"
is_byokfalse
latency298
model_permaslug"google/gemini-3.5-flash-lite-20260721"
provider_name"Google AI Studio"
status200
user_agent"langchainjs-openai/1.0.0 ((node/v24.18.0; linux; x64))"
http_referer(null)
request_id"req-1784991571-83oKOaoV7FMSsSXMA8EF"
session_id(null)
preset_id(null)
api_type"completions"
id"gen-1784991571-peGSCMg5C90bC5nELo5U"
upstream_id"U89kauaBDsOT-8YP0c7zgAo"
total_cost0.0017569
cache_discount(null)
upstream_inference_cost0
provider_name"Google AI Studio"
response_cache_source_id(null)
data_region"global"

Evaluation details

Result Evaluator Details Meta Data
100.00% Adverbs in dialogue tags
Target: ≤10% dialogue tags with adverbs
totalTags0
100.00% AI-ism adverb frequency
Target: <2% AI-ism adverbs (58 tracked)
wordCount456
totalAiIsmAdverbs0
found(empty)
highlights(empty)
100.00% AI-ism character names
Target: 0 AI-default names (17 tracked, −20% each)
codexExemptions(empty)
found(empty)
100.00% AI-ism location names
Target: 0 AI-default location names (33 tracked, −20% each)
codexExemptions(empty)
found(empty)
23.25% AI-ism word frequency
Target: <2% AI-ism words (290 tracked)
wordCount456
totalAiIsms7
found
0
word"footsteps"
count1
1
word"echo"
count1
2
word"standard"
count1
3
word"resonance"
count1
4
word"shattered"
count1
5
word"chill"
count1
6
word"warmth"
count1
highlights
0"footsteps"
1"echo"
2"standard"
3"resonance"
4"shattered"
5"chill"
6"warmth"
100.00% Cliché density
Target: ≤1 cliche(s) per 800-word window
totalCliches0
maxInWindow0
found(empty)
highlights(empty)
100.00% Emotion telling (show vs. tell)
Target: ≤3% sentences with emotion telling
emotionTells0
narrationSentences51
matches(empty)
100.00% Filter word density
Target: ≤3% sentences with filter/hedge words
filterCount1
hedgeCount0
narrationSentences51
filterMatches
0"watch"
hedgeMatches(empty)
100.00% Gibberish response detection
Target: ≤1% gibberish-like sentences (hard fail if a sentence exceeds 800 words)
analyzedSentences51
gibberishSentences0
adjustedGibberishSentences0
longSentenceCount0
runOnParagraphCount0
giantParagraphCount0
wordSaladCount0
repetitionLoopCount0
controlTokenCount0
repeatedSegmentCount0
maxSentenceWordsSeen21
ratio0
matches(empty)
100.00% Markdown formatting overuse
Target: ≤5% words in markdown formatting
markdownSpans0
markdownWords0
totalWords456
ratio0
matches(empty)
100.00% Missing dialogue indicators (quotation marks)
Target: ≤10% speech attributions without quotation marks
totalAttributions0
unquotedAttributions0
matches(empty)
84.21% Name drop frequency
Target: ≤1.0 per-name mentions per 100 words
totalMentions15
wordCount456
uniqueNames10
maxNameDensity1.32
worstName"Harlow"
maxWindowNameDensity2
worstWindowName"Harlow"
discoveredNames
Tube1
Harlow6
Quinn1
Northern1
Line1
London1
Underground1
Camden1
Veil1
Compass1
persons
0"Harlow"
1"Quinn"
2"Compass"
places
0"London"
1"Camden"
globalScore0.842
windowScore1
100.00% Narrator intent-glossing
Target: ≤2% narration sentences with intent-glossing patterns
analyzedSentences38
glossingSentenceCount0
matches(empty)
100.00% "Not X but Y" pattern overuse
Target: ≤1 "not X but Y" per 1000 words
totalMatches0
per1kWords0
wordCount456
matches(empty)
100.00% Overuse of "that" (subordinate clause padding)
Target: ≤2% sentences with "that" clauses
thatCount0
totalSentences51
matches(empty)
100.00% Paragraph length variance
Target: CV ≥0.5 for paragraph word counts
totalParagraphs21
mean21.71
std16.02
cv0.738
sampleLengths
066
122
210
342
49
522
612
716
811
941
1029
1112
1212
1326
1435
159
167
172
1817
1948
208
100.00% Passive voice overuse
Target: ≤2% passive sentences
passiveCount0
totalSentences51
matches(empty)
100.00% Past progressive (was/were + -ing) overuse
Target: ≤2% past progressive verbs
pastProgressiveCount0
totalVerbs62
matches(empty)
100.00% Em-dash & semicolon overuse
Target: ≤2% sentences with em-dashes/semicolons
emDashCount0
semicolonCount0
flaggedSentences0
totalSentences51
ratio0
matches(empty)
100.00% Purple prose (modifier overload)
Target: <4% adverbs, <2% -ly adverbs, no adj stacking
wordCount460
adjectiveStacks0
stackExamples(empty)
adverbCount5
adverbRatio0.010869565217391304
lyAdverbCount1
lyAdverbRatio0.002173913043478261
100.00% Repeated phrase echo
Target: ≤20% sentences with echoes (window: 2)
totalSentences51
echoCount0
echoWords(empty)
100.00% Sentence length variance
Target: CV ≥0.4 for sentence word counts
totalSentences51
mean8.94
std4.87
cv0.545
sampleLengths
016
118
216
316
410
512
63
77
811
96
107
115
1213
139
1411
156
163
172
186
193
203
2116
226
235
243
258
2611
278
282
299
3016
317
326
333
349
3512
3610
3716
389
398
4018
419
427
432
4412
455
467
4716
484
4921
71.90% Sentence opener variety
Target: ≥60% unique sentence openers
consecutiveRepeats5
diversityRatio0.49019607843137253
totalSentences51
uniqueOpeners25
0.00% Adverb-first sentence starts
Target: ≥3% sentences starting with an adverb
adverbCount0
totalSentences48
matches(empty)
ratio0
100.00% Pronoun-first sentence starts
Target: ≤30% sentences starting with a pronoun
pronounCount7
totalSentences48
matches
0"Her leather watch clicks against"
1"She crouches beside the chalk"
2"She stares down the dark"
3"His glasses lie shattered a"
4"He didn't fall."
5"Her sharp jaw tightens."
6"She traces the wall with"
ratio0.146
32.92% Subject-first sentence starts
Target: ≤72% sentences starting with a subject
subjectCount41
totalSentences48
matches
0"The damp air of the"
1"Detective Harlow Quinn steps past"
2"Her leather watch clicks against"
3"Salt-and-pepper cropped hair catches the"
4"Footsteps echo against the tiled"
5"A constable stands beside a"
6"Another suicide jump."
7"The driver hit the brakes"
8"Harlow ignores the chatter of"
9"She crouches beside the chalk"
10"The surface lacks the expected"
11"The dust remains undisturbed in"
12"The bloke leaped straight from"
13"Harlow traces a gloved finger"
14"The soot forms a clean"
15"People struggle on the way"
16"Shoes scrape concrete."
17"Fingernails scratch tile."
18"A shadow falls across the"
19"Case closes by morning tea."
ratio0.854
0.00% Subordinate conjunction sentence starts
Target: ≥2% sentences starting with a subordinating conjunction
subConjCount0
totalSentences48
matches(empty)
ratio0
100.00% Technical jargon density
Target: ≤6% sentences with technical-jargon patterns
analyzedSentences19
technicalSentenceCount0
matches(empty)
100.00% Useless dialogue additions
Target: ≤5% dialogue tags with trailing filler fragments
totalTags0
uselessAdditionCount0
matches(empty)
100.00% Dialogue tag variety (said vs. fancy)
Target: ≤10% fancy dialogue tags
totalTags0
87.0756%