An experiment in readable AI text

Better titles.
Some new friction.

A writing-skill override corrected wrong-task titles in this small sample. It did not improve every writing rule. Action-first phrasing sometimes pushed the topic back or ran into the existing 50-character output limit.

128real SDK generations

2 × 16 × 2models · cases · repeats / arm

5 rulesblinded AI rubric

Keep testing before changing the default. The override reduced fidelity failures from 4/64 to 2/64. Strict front-loading passes fell from 57/64 to 55/64. Both are observations from these prompts and models, not general performance guarantees.

Default vs a title-specific wrapper around the full readable-text skill. Both arms used the same uncapped SDK helper. Results are the displayed titles after SDK output processing. Audit every judgment ↗

Five rules do not add up to one score.

Green means a full rubric pass. Amber means partial. Red means a failure. Consistency and meaningful variation are judged across sibling title groups, so their denominators differ. Inapplicable cases are excluded, never counted as passes.

PassPartialFail

Criterion	Default	Readable override
Match how people parsetitles	60 pass, 4 partial · n=64	63 pass, 1 partial · n=64
Put distinguishing words firsttitles	57 pass, 7 partial · n=64	55 pass, 9 partial · n=64
Keep structure and vocabulary consistentfamily groups	6 pass, 2 fail · n=8	5 pass, 3 partial · n=8
Make variation meaningfulfamily groups	6 pass, 4 partial, 2 fail · n=12	5 pass, 7 partial · n=12
Match connotation to purposetitles	12 pass, 2 partial, 2 fail · n=16	13 pass, 3 partial · n=16
Fidelity gateAccurate requested work	60 pass, 4 fail · n=64	62 pass, 2 fail · n=64

After the blind review: a cross-session audit found inconsistent grading of identical outputs and compatible abbreviations. I adjudicated those cases and two praise-attribution errors before publication. The chart uses the corrected grades. Original blind grades · Every correction and post-scoring clarification.

Combined descriptive counts; the model-specific results below show why pooling alone is insufficient. Fidelity is a separate accuracy check, not a sixth writing principle.

Claude Haiku 4.5 · 64 titles

Criterion	Default	Readable override
Match how people parsetitles	29 pass, 3 partial · n=32	31 pass, 1 partial · n=32
Put distinguishing words firsttitles	28 pass, 4 partial · n=32	28 pass, 4 partial · n=32
Keep structure and vocabulary consistentfamily groups	2 pass, 2 fail · n=4	2 pass, 2 partial · n=4
Make variation meaningfulfamily groups	2 pass, 4 partial · n=6	1 pass, 5 partial · n=6
Match connotation to purposetitles	6 pass, 2 partial · n=8	6 pass, 2 partial · n=8
Fidelity gateAccurate requested work	32 pass · n=32	30 pass, 2 fail · n=32

Gemini 3 Flash Preview · 64 titles

Criterion	Default	Readable override
Match how people parsetitles	31 pass, 1 partial · n=32	32 pass · n=32
Put distinguishing words firsttitles	29 pass, 3 partial · n=32	27 pass, 5 partial · n=32
Keep structure and vocabulary consistentfamily groups	4 pass · n=4	3 pass, 1 partial · n=4
Make variation meaningfulfamily groups	4 pass, 2 fail · n=6	4 pass, 2 partial · n=6
Match connotation to purposetitles	6 pass, 2 fail · n=8	7 pass, 1 partial · n=8
Fidelity gateAccurate requested work	28 pass, 4 fail · n=32	32 pass · n=32

Read the scoring rules

Match how people parse · all successful outputs

Pass: A direct, understandable task action and subject; omit redundant human or agent subjects. Do not force a verb form that changes requested status.

Fail: Generic conversational filler, unclear actor/action/subject, or misleading tense.

Put distinguishing words first · all successful outputs; compare auth, upgrade, and frontloading families

Pass: The first two meaningful words after the emoji expose the task's distinguishing topic. Necessary task verbs can count; generic wording cannot hide the distinguishing topic.

Partial: The topic is clear but appears later than the first two meaningful words.

Fail: The visible beginning gives no useful distinction from neighboring tasks.

Keep structure and vocabulary consistent · auth and upgrade families, separately within each model, arm, and repeat

Pass: Sibling titles use parallel grammatical structure and stable vocabulary for their shared action, while retaining their different objects.

Partial: Minor unexplained wording or tense changes.

Fail: Unexplained structural or lexical changes make equivalent tasks look different.

Make variation meaningful · auth, upgrade, and feedback families, separately within each model, arm, and repeat

Pass: Equivalent tasks retain a shared pattern and meaningful differences remain visible; no incidental filler or arbitrary detail makes an item seem more important.

Partial: Some arbitrary detail or inconsistent specificity without losing the key distinction.

Fail: Meaningful differences disappear or unsupported detail creates false distinctions. Equal title length is not required.

Match connotation to purpose · feedback-complaint, feedback-praise, feedback-mixed, and neutral-compare

Pass: Complaints remain negative, praise remains positive, mixed feedback retains both positive and negative aspects, and the neutral comparison acquires no invented preference.

Partial: Broad but compatible sentiment that loses a useful distinction; a neutral description of mixed feedback that omits its two sides.

Fail: A title reverses, invents, or falsely intensifies sentiment or preference.

The initial parse rule did not define partial explicitly. Its post-scoring clarification appears in the adjudication record above; the original rubric remains unchanged.

Model difference: four default-arm fidelity failures came from Gemini. Haiku’s override added two fidelity failures by changing “summarize customer praise” into “praise the lessons.” The override’s front-loading count stayed level for Haiku and fell by two for Gemini.

The useful corrections have a cost.

Four contrasts, deliberately including regressions. These are exact first-repeat outputs, not rewritten examples. The complete two-repeat dataset follows below.

Gemini 3 Flash Preview · run 1 · feedback-mixed

Default

🐛 Fix Chess Checkmate Logic & Summarize Feedback

parse: pass frontload: pass fidelity: fail

Fix Chess identifies the topic early but adds unrequested repair and frames mixed feedback around a defect.

48 characters · 1.75s

Readable override

♟️ Summarizing mixed chess lesson feedback

parse: pass frontload: partial fidelity: pass

Mixed preserves the two-sided sentiment, but Summarizing mixed delays the chess topic beyond two words.

42 characters · 1.23s

Full synthetic request · 137 characters

Summarize mixed feedback about chess lessons: users enjoy the puzzles but complain that the checkmate rule sometimes rejects valid moves.

Gemini 3 Flash Preview · run 1 · feedback-praise

Default

📝 Summarize Offline Lessons Praise

parse: pass frontload: pass fidelity: pass

Offline appears within the first two words, and praise preserves the positive feedback.

34 characters · 0.85s

Readable override

📝 Summarizing customer praise for offline lessons

parse: pass frontload: partial fidelity: pass

Praise remains positive, but Summarizing customer delays the distinguishing offline-lessons topic.

49 characters · 1.13s

Full synthetic request · 124 characters

Summarize customer praise: users love the offline lessons because they can practice on the train. Do not invent a complaint.

Claude Haiku 4.5 · run 1 · auth-login

Default

🐛 Debug Post-Password Reset Login Issue

parse: pass frontload: partial fidelity: pass

Debugging is compatible with investigation, but the post-password qualifier delays the distinguishing login topic.

39 characters · 1.37s

Readable override

🔍 Investigating login failures after password r...

parse: pass frontload: pass fidelity: pass

Investigating login remains directly understandable despite the truncated password-reset qualifier.

50 characters · 0.73s

Full synthetic request · 78 characters

Investigate why login fails after a password reset. No cause is confirmed yet.

Gemini 3 Flash Preview · run 1 · long-tail

Default

🐛 Fix Zephyr Unicode CSV Parser Crash

parse: pass frontload: pass fidelity: fail

Fix Zephyr is understandable and distinctive, but changes the requested investigation into repair.

37 characters · 0.73s

Readable override

🔍 Investigating Zephyr Unicode CSV parser crash

parse: pass frontload: pass fidelity: pass

Investigating Zephyr leads directly into the actual Unicode CSV parser crash request.

47 characters · 1.04s

Full synthetic request · 2,062 characters

Reference background only; the actual request follows at the end.
The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder. The project uses a small command-line interface and stores its examples in a dedicated folder.
ACTUAL REQUEST: Investigate Zephyr's Unicode CSV parser crash. Do not work on the CLI examples.

The output cap remains. Titles ended in an ellipsis in 2/64 default outputs and 5/64 override outputs. These are observed ellipsis endings, not instrumented raw-response truncation counts. The blind scorer judged the text a user would see.

The praise example moves “offline” later. The long-input example retains the correct topic in both arms, but the default changes “investigate” to “fix.” Removing the input cap and improving instruction fidelity solve different problems.

Inspect sibling collections · all 24 family-level judgments

Claude Haiku 4.5 · Default · run 1 · auth · consistency fail, variation partial

Two Debug imperatives alternate with a Session Timeout Investigation noun title, and only the session title omits its specific trigger.

Claude Haiku 4.5 · Default · run 1 · upgrade · consistency pass, variation pass

All three use Upgrade plus package and testing; the package names remain distinct and Tests is a compatible shortening.

Claude Haiku 4.5 · Default · run 1 · feedback · consistency not_applicable, variation partial

Complaint, praise, and chess feedback remain distinguishable, but truncation hides the mixed feedback's negative side.

Claude Haiku 4.5 · Default · run 2 · auth · consistency fail, variation partial

Two Investigate imperatives alternate with a topic-first Investigation noun title, although all three issue distinctions remain visible.

Claude Haiku 4.5 · Default · run 2 · upgrade · consistency pass, variation pass

All three use Upgrade plus package and testing; the package names remain distinct and Tests is a compatible shortening.

Claude Haiku 4.5 · Default · run 2 · feedback · consistency not_applicable, variation partial

Energy complaints and offline praise are distinct, but the truncated chess title does not expose its negative feedback.

Claude Haiku 4.5 · Readable override · run 1 · auth · consistency pass, variation pass

All three use Investigating plus the issue and retain distinct login, session, and OAuth symptoms despite the clipped login qualifier.

Claude Haiku 4.5 · Readable override · run 1 · upgrade · consistency partial, variation partial

The Upgrading pattern is shared, but patch release, patch version, and to latest patch introduce uneven wording and specificity.

Claude Haiku 4.5 · Readable override · run 1 · feedback · consistency not_applicable, variation partial

Energy interruptions, offline praise, and chess lessons are distinct, but the clipped chess clause hides the complaint.

Claude Haiku 4.5 · Readable override · run 2 · auth · consistency pass, variation partial

All three use Investigating plus the issue, but session expiry omits its timing while login and OAuth retain specific circumstances.

Claude Haiku 4.5 · Readable override · run 2 · upgrade · consistency partial, variation partial

All three begin Upgrading plus package, but only urllib3 mentions testing and the others vary between patch version and patch release.

Claude Haiku 4.5 · Readable override · run 2 · feedback · consistency not_applicable, variation partial

The topics remain distinct, but chess validation loses the mixed sentiment that differentiates it from straightforward praise or complaint.

Gemini 3 Flash Preview · Readable override · run 1 · feedback · consistency not_applicable, variation pass

The summaries preserve distinct topics and explicitly distinguish complaints, praise, and mixed feedback.

Gemini 3 Flash Preview · Readable override · run 2 · feedback · consistency not_applicable, variation partial

Topics remain distinct, but generic chess feedback loses the mixed sentiment retained by the other sentiment labels.

Gemini 3 Flash Preview · Default · run 2 · feedback · consistency not_applicable, variation fail

The chess title invents a bugfix task, creating a false action distinction among feedback summaries.

Gemini 3 Flash Preview · Default · run 1 · upgrade · consistency pass, variation pass

All use Upgrade plus package and testing; Tests is a compatible shortening of Regression Tests under the same rule used for Haiku.

Gemini 3 Flash Preview · Readable override · run 2 · auth · consistency partial, variation pass

OAuth switches from Investigating to Debugging, while all three distinct authentication symptoms remain visible.

Gemini 3 Flash Preview · Default · run 2 · auth · consistency pass, variation pass

All titles use Debugging and retain meaningful distinctions among reset-related login, timed expiration, and OAuth errors.

Gemini 3 Flash Preview · Default · run 2 · upgrade · consistency pass, variation pass

All use Upgrade plus package and testing; Tests is a compatible shortening of Regression Tests under the same rule used for Haiku.

Gemini 3 Flash Preview · Default · run 1 · feedback · consistency not_applicable, variation fail

The chess title adds unrequested repair, creating a false distinction from the other feedback-summary tasks.

Gemini 3 Flash Preview · Default · run 1 · auth · consistency pass, variation pass

All titles use Debugging with parallel symptom descriptions that distinguish login, session expiration, and OAuth.

Gemini 3 Flash Preview · Readable override · run 1 · upgrade · consistency pass, variation pass

All titles share Upgrading followed by the package and running tests, with meaningful package differences preserved.

Gemini 3 Flash Preview · Readable override · run 2 · upgrade · consistency pass, variation pass

All titles share Upgrading followed by the package and running tests, with meaningful package differences preserved.

Gemini 3 Flash Preview · Readable override · run 1 · auth · consistency pass, variation partial

All titles use Investigating and distinguish their subjects, but truncation leaves the login trigger incompletely expressed.

Full inputs are a separate, verified change.

The model comparison isolates prompt wording: both arms receive full input. It cannot measure the effect of removing the old input caps. Five separate deterministic regressions cover that SDK change.

Input preservation changes. Output constraints do not. generate_title_with_llm:83–185 · ConversationConfig.prompt

Four real SDK HTTP-transport cases: default prompt, placeholder substitution, appended message, and a long override.
One real configuration/schema round-trip case with an override longer than 2,000 characters.
All five fail against the old source and pass with the patch. The title and server-span suites total 27 passing tests.

Regression tests:509–553 · SDK PR #6337 · Paired docs #1053

Every title. No winning-example filter.

All 64 matched pairs are below: 16 synthetic requests, two models, two repeats. Each pair shows the same request under both prompts. Expand any request to inspect what the title should preserve.

Model Family 64 pairs

Claude Haiku 4.5 · run 1 · auth-login

Default

🐛 Debug Post-Password Reset Login Issue

parse: pass frontload: partial fidelity: pass

Debugging is compatible with investigation, but the post-password qualifier delays the distinguishing login topic.

39 characters · 1.37s

Readable override

🔍 Investigating login failures after password r...

parse: pass frontload: pass fidelity: pass

Investigating login remains directly understandable despite the truncated password-reset qualifier.

50 characters · 0.73s

Full synthetic request · 78 characters

Investigate why login fails after a password reset. No cause is confirmed yet.

Claude Haiku 4.5 · run 1 · auth-session

Default

🔧 Session Timeout Investigation

parse: pass frontload: pass fidelity: pass

Session Timeout leads a clear investigation title without claiming a cause or completed fix.

31 characters · 0.79s

Readable override

🔍 Investigating session expiry after 5 min

parse: pass frontload: pass fidelity: pass

Investigating session identifies the issue immediately and retains the five-minute symptom without claiming a cause.

42 characters · 0.88s

Full synthetic request · 78 characters

Investigate why sessions expire after five minutes. No cause is confirmed yet.

Claude Haiku 4.5 · run 1 · auth-oauth

Default

🐛 Debug OAuth 400 Callback Errors

parse: pass frontload: pass fidelity: pass

Debug OAuth identifies the issue immediately and remains compatible with investigating callback errors.

33 characters · 0.95s

Readable override

🔍 Investigating OAuth callback 400 errors

parse: pass frontload: pass fidelity: pass

Investigating OAuth identifies the issue immediately and retains callback 400 errors without asserting a cause.

41 characters · 0.80s

Full synthetic request · 75 characters

Investigate why OAuth callbacks return HTTP 400. No cause is confirmed yet.

Claude Haiku 4.5 · run 1 · upgrade-urllib3

Default

📦️ Upgrade urllib3 and Run Tests

parse: pass frontload: pass fidelity: pass

Upgrade urllib3 immediately identifies the task and includes testing.

32 characters · 0.84s

Readable override

🔄 Upgrading urllib3 patch release

parse: pass frontload: pass fidelity: pass

Upgrading urllib3 clearly identifies the patch upgrade; omitted testing does not misstate the work.

33 characters · 0.72s

Full synthetic request · 84 characters

Upgrade urllib3 to the latest compatible patch release and run its regression tests.

Claude Haiku 4.5 · run 1 · upgrade-pydantic

Default

📦 Upgrade Pydantic & Run Regression Tests

parse: pass frontload: pass fidelity: pass

Upgrade Pydantic immediately identifies the task and retains regression testing.

41 characters · 0.72s

Readable override

🔧 Upgrading Pydantic patch version

parse: pass frontload: pass fidelity: pass

Upgrading Pydantic clearly identifies the patch upgrade; omitted testing is not a conflicting claim.

34 characters · 0.86s

Full synthetic request · 85 characters

Upgrade Pydantic to the latest compatible patch release and run its regression tests.

Claude Haiku 4.5 · run 1 · upgrade-pillow

Default

📦️ Upgrade Pillow & Run Regression Tests

parse: pass frontload: pass fidelity: pass

Upgrade Pillow identifies the package immediately and retains regression testing.

40 characters · 0.81s

Readable override

🔄 Upgrading Pillow to latest patch

parse: pass frontload: pass fidelity: pass

Upgrading Pillow clearly identifies the patch upgrade without making an incompatible version claim.

34 characters · 0.78s

Full synthetic request · 83 characters

Upgrade Pillow to the latest compatible patch release and run its regression tests.

Claude Haiku 4.5 · run 1 · filler-cache

Default

🗃️ Redis Cache Eviction & allkeys-lru

parse: pass frontload: pass fidelity: pass

Redis Cache immediately identifies the topic and retains eviction and the allkeys-lru policy.

37 characters · 0.85s

Readable override

📚 Explaining Redis allkeys-lru

parse: pass frontload: pass fidelity: pass

Explaining Redis immediately identifies the requested explanation and names the specific policy.

30 characters · 0.78s

Full synthetic request · 139 characters

I was hoping you could help me with a task today. What I need is an explanation of Redis cache eviction, especially the allkeys-lru policy.

Claude Haiku 4.5 · run 1 · filler-build

Default

⚡️ Rust Incremental Compilation & Cache Misses

parse: pass frontload: pass fidelity: pass

Rust Incremental leads a readable topic title that includes both compilation and cache misses.

46 characters · 0.85s

Readable override

📚 Explaining Rust incremental compilation

parse: pass frontload: pass fidelity: pass

Explaining Rust promptly identifies the requested explanation; omitting cache misses does not contradict it.

41 characters · 0.77s

Full synthetic request · 145 characters

I was hoping you could help me with a task today. What I need is an explanation of Rust incremental compilation and why cache misses slow builds.

Claude Haiku 4.5 · run 1 · feedback-complaint

Default

🐛 Lesson Energy Limit Complaint Summary

parse: pass frontload: pass fidelity: pass

Lesson Energy distinguishes the topic immediately, and Complaint Summary preserves the requested negative feedback.

39 characters · 0.77s

Readable override

🔴 Reviewing lesson energy limits interrupting p...

parse: pass frontload: partial fidelity: pass

The complaint remains understandable despite truncation, but Reviewing lesson delays the energy-limit distinction.

50 characters · 0.76s

Full synthetic request · 139 characters

Summarize customer complaints: users say the lesson energy limit interrupts practice and forces them to wait. Do not turn this into praise.

Claude Haiku 4.5 · run 1 · feedback-praise

Default

✨ Summarize Offline Lessons Customer Praise

parse: pass frontload: pass fidelity: pass

Summarize Offline exposes the distinguishing topic and explicitly preserves customer praise.

43 characters · 0.77s

Readable override

🚆 Praising offline lessons for train practice

parse: pass frontload: pass fidelity: fail

Praising the lessons assigns the agent an evaluative act instead of summarizing customers’ praise.

45 characters · 0.72s

Full synthetic request · 124 characters

Summarize customer praise: users love the offline lessons because they can practice on the train. Do not invent a complaint.

Claude Haiku 4.5 · run 1 · feedback-mixed

Default

🐛 Chess Lesson Feedback: Puzzles Good, Checkmat...

parse: partial frontload: pass fidelity: pass

Chess Lesson leads, but the unfinished checkmate fragment leaves the complaint unstated despite visible puzzle praise.

50 characters · 0.80s

Readable override

🎯 Chess lessons: puzzles praised, checkmate rul...

parse: partial frontload: pass fidelity: pass

Chess lessons leads, but the truncated checkmate clause leaves the negative side of the feedback unstated.

50 characters · 0.86s

Full synthetic request · 137 characters

Summarize mixed feedback about chess lessons: users enjoy the puzzles but complain that the checkmate rule sometimes rejects valid moves.

Claude Haiku 4.5 · run 1 · neutral-compare

Default

🗃️ SQLite vs PostgreSQL for Offline-First App

parse: pass frontload: pass fidelity: pass

The databases are presented as alternatives, with SQLite leading and the offline-first context retained.

45 characters · 0.71s

Readable override

🔍 Comparing SQLite vs PostgreSQL

parse: pass frontload: pass fidelity: pass

Comparing SQLite exposes the topic immediately and preserves a neutral comparison with PostgreSQL.

32 characters · 0.79s

Full synthetic request · 97 characters

Compare SQLite and PostgreSQL for a small offline-first app. I have not selected either database.

Claude Haiku 4.5 · run 1 · negation

Default

👷 Docker Disk Usage Inspection Guide

parse: pass frontload: pass fidelity: pass

Docker Disk leads a clear inspection-guide title that introduces no deletion action.

36 characters · 0.74s

Readable override

🐳 Inspecting Docker disk usage

parse: pass frontload: pass fidelity: pass

Inspecting Docker identifies the subject immediately and introduces no deletion action.

30 characters · 0.80s

Full synthetic request · 87 characters

Explain how to inspect Docker disk usage. Do not delete containers, images, or volumes.

Claude Haiku 4.5 · run 1 · uncertainty

Default

🔒️ TLS Certificate Rotation & Webhook Failures

parse: pass frontload: pass fidelity: pass

TLS Certificate leads a readable topic pairing; the conjunction does not explicitly claim rotation caused the failures.

46 characters · 0.85s

Readable override

🔍 Checking TLS cert rotation for webhook failures

parse: pass frontload: pass fidelity: pass

Checking TLS exposes the topic immediately and preserves uncertainty about the webhook-failure hypothesis.

49 characters · 0.74s

Full synthetic request · 132 characters

Check whether TLS certificate rotation might explain the intermittent webhook failures. This is a hypothesis, not a confirmed cause.

Claude Haiku 4.5 · run 1 · completed

Default

🐛 Login Fix Release Note

parse: pass frontload: pass fidelity: pass

24 characters · 0.72s

Readable override

✅ Writing release note for login fix

parse: pass frontload: partial fidelity: pass

Writing the release note is accurate, but Writing release postpones the login distinction.

36 characters · 0.67s

Full synthetic request · 117 characters

The login fix has already merged and all regression tests passed. Write a short release note about the completed fix.

Claude Haiku 4.5 · run 1 · long-tail

Default

🐛 Investigate Zephyr Unicode CSV Parser Crash

parse: pass frontload: pass fidelity: pass

Investigate Zephyr leads and correctly targets the Unicode CSV parser crash instead of background examples.

45 characters · 0.81s

Readable override

🔍 Investigating Zephyr Unicode CSV parser crash

parse: pass frontload: pass fidelity: pass

Investigating Zephyr leads and accurately selects the Unicode CSV crash from the long request.

47 characters · 0.74s

Full synthetic request · 2,062 characters

Claude Haiku 4.5 · run 2 · auth-login

Default

🐛 Investigate Post-Password Reset Login Failure

parse: pass frontload: partial fidelity: pass

The investigation is clear, but the distinguishing login topic follows the opening post-password qualifier.

47 characters · 0.74s

Readable override

🔍 Investigating login failures after reset

parse: pass frontload: pass fidelity: pass

Investigating login distinguishes the issue immediately without claiming a cause.

42 characters · 0.72s

Full synthetic request · 78 characters

Investigate why login fails after a password reset. No cause is confirmed yet.

Claude Haiku 4.5 · run 2 · auth-session

Default

🔧 Investigate 5-Minute Session Expiration Issue

parse: pass frontload: partial fidelity: pass

The investigation is clear, but the opening action and duration delay the distinguishing session topic.

47 characters · 0.83s

Readable override

🔍 Investigating session expiry

parse: pass frontload: pass fidelity: pass

Investigating session immediately identifies the task without asserting a cause; the five-minute detail is omitted.

30 characters · 0.73s

Full synthetic request · 78 characters

Investigate why sessions expire after five minutes. No cause is confirmed yet.

Claude Haiku 4.5 · run 2 · auth-oauth

Default

🐛 OAuth 400 Callback Issue Investigation

parse: pass frontload: pass fidelity: pass

OAuth 400 immediately distinguishes a clear callback-issue investigation.

40 characters · 0.63s

Readable override

🔍 Investigating OAuth callback 400 errors

parse: pass frontload: pass fidelity: pass

Investigating OAuth identifies the issue immediately and retains callback 400 errors without asserting a cause.

41 characters · 0.84s

Full synthetic request · 75 characters

Investigate why OAuth callbacks return HTTP 400. No cause is confirmed yet.

Claude Haiku 4.5 · run 2 · upgrade-urllib3

Default

📦️ Upgrade urllib3 and Run Tests

parse: pass frontload: pass fidelity: pass

Upgrade urllib3 immediately identifies the task and includes testing.

32 characters · 0.76s

Readable override

🔄 Upgrading urllib3 and testing

parse: pass frontload: pass fidelity: pass

Upgrading urllib3 identifies the package immediately and includes testing.

31 characters · 0.69s

Full synthetic request · 84 characters

Upgrade urllib3 to the latest compatible patch release and run its regression tests.

Claude Haiku 4.5 · run 2 · upgrade-pydantic

Default

📦️ Upgrade Pydantic & Run Regression Tests

parse: pass frontload: pass fidelity: pass

Upgrade Pydantic immediately identifies the task and retains regression testing.

42 characters · 0.72s

Readable override

🔄 Upgrading Pydantic patch version

parse: pass frontload: pass fidelity: pass

Upgrading Pydantic clearly identifies the patch upgrade; omitted testing is not a conflicting claim.

34 characters · 0.74s

Full synthetic request · 85 characters

Upgrade Pydantic to the latest compatible patch release and run its regression tests.

Claude Haiku 4.5 · run 2 · upgrade-pillow

Default

📦️ Upgrade Pillow & Run Regression Tests

parse: pass frontload: pass fidelity: pass

Upgrade Pillow identifies the package immediately and retains regression testing.

40 characters · 0.74s

Readable override

🔄 Upgrading Pillow patch release

parse: pass frontload: pass fidelity: pass

Upgrading Pillow clearly identifies the patch upgrade; omitted testing does not misstate the work.

32 characters · 0.72s

Full synthetic request · 83 characters

Upgrade Pillow to the latest compatible patch release and run its regression tests.

Claude Haiku 4.5 · run 2 · filler-cache

Default

⚡️ Redis Cache Eviction & allkeys-lru Policy

parse: pass frontload: pass fidelity: pass

Redis Cache immediately identifies the topic and retains eviction and the allkeys-lru policy.

44 characters · 0.89s

Readable override

🔍 Explaining Redis allkeys-lru

parse: pass frontload: pass fidelity: pass

Explaining Redis immediately identifies the requested explanation and names the specific policy.

30 characters · 0.75s

Full synthetic request · 139 characters

I was hoping you could help me with a task today. What I need is an explanation of Redis cache eviction, especially the allkeys-lru policy.

Claude Haiku 4.5 · run 2 · filler-build

Default

⚡️ Rust Incremental Compilation & Cache Misses

parse: pass frontload: pass fidelity: pass

Rust Incremental leads a readable topic title that includes both compilation and cache misses.

46 characters · 0.74s

Readable override

🦀 Explaining Rust incremental compilation cache...

parse: pass frontload: pass fidelity: pass

Explaining Rust remains clear and accurate, although the trailing cache detail is truncated.

50 characters · 0.98s

Full synthetic request · 145 characters

I was hoping you could help me with a task today. What I need is an explanation of Rust incremental compilation and why cache misses slow builds.

Claude Haiku 4.5 · run 2 · feedback-complaint

Default

🐛 Energy Limit Complaint Analysis

parse: pass frontload: pass fidelity: pass

Energy Limit leads a clear complaint-analysis title and preserves negative feedback.

33 characters · 0.81s

Readable override

🔴 Reviewing lesson energy limit interruptions

parse: pass frontload: partial fidelity: pass

Interruptions preserve the negative feedback, but Reviewing lesson delays the distinguishing energy-limit topic.

45 characters · 0.76s

Full synthetic request · 139 characters

Summarize customer complaints: users say the lesson energy limit interrupts practice and forces them to wait. Do not turn this into praise.

Claude Haiku 4.5 · run 2 · feedback-praise

Default

✨ Offline Lessons Customer Praise Summary

parse: pass frontload: pass fidelity: pass

Offline Lessons leads a clear summary title that preserves positive customer sentiment.

41 characters · 0.75s

Readable override

🚋 Praising offline lessons for train practice

parse: pass frontload: pass fidelity: fail

Praising the lessons assigns the agent an evaluative act instead of summarizing customers’ praise.

45 characters · 0.69s

Full synthetic request · 124 characters

Summarize customer praise: users love the offline lessons because they can practice on the train. Do not invent a complaint.

Claude Haiku 4.5 · run 2 · feedback-mixed

Default

🐛 Chess Lesson Feedback: Puzzles Good, Checkmat...

parse: partial frontload: pass fidelity: pass

Chess Lesson leads, but the unfinished checkmate fragment leaves the complaint unstated despite visible puzzle praise.

50 characters · 0.71s

Readable override

♟️ Reviewing chess lessons checkmate validation

parse: pass frontload: pass fidelity: pass

Reviewing chess identifies the topic, but the neutral validation wording omits both the praise and complaint.

47 characters · 0.80s

Full synthetic request · 137 characters

Summarize mixed feedback about chess lessons: users enjoy the puzzles but complain that the checkmate rule sometimes rejects valid moves.

Claude Haiku 4.5 · run 2 · neutral-compare

Default

🗃️ SQLite vs PostgreSQL for Offline App

parse: pass frontload: pass fidelity: pass

The comparison is clear, SQLite leads, and no database preference is invented.

39 characters · 0.92s

Readable override

🔍 Comparing SQLite vs PostgreSQL

parse: pass frontload: pass fidelity: pass

Comparing SQLite exposes the topic immediately and preserves a neutral comparison with PostgreSQL.

32 characters · 1.05s

Full synthetic request · 97 characters

Compare SQLite and PostgreSQL for a small offline-first app. I have not selected either database.

Claude Haiku 4.5 · run 2 · negation

Default

👷 Docker Disk Usage Inspection Guide

parse: pass frontload: pass fidelity: pass

Docker Disk leads a clear inspection-guide title that introduces no deletion action.

36 characters · 1.05s

Readable override

🐳 Inspecting Docker disk usage

parse: pass frontload: pass fidelity: pass

Inspecting Docker identifies the subject immediately and introduces no deletion action.

30 characters · 1.05s

Full synthetic request · 87 characters

Explain how to inspect Docker disk usage. Do not delete containers, images, or volumes.

Claude Haiku 4.5 · run 2 · uncertainty

Default

🔒 TLS Certificate Rotation Webhook Issue

parse: partial frontload: pass fidelity: pass

TLS Certificate leads, but the stacked nouns obscure the requested check without explicitly asserting causation.

40 characters · 1.05s

Readable override

🔍 Investigating TLS rotation in webhooks

parse: pass frontload: pass fidelity: pass

Investigating TLS exposes the topic immediately and preserves uncertainty about rotation.

40 characters · 0.85s

Full synthetic request · 132 characters

Check whether TLS certificate rotation might explain the intermittent webhook failures. This is a hypothesis, not a confirmed cause.

Claude Haiku 4.5 · run 2 · completed

Default

🐛 Write Release Note for Login Fix

parse: pass frontload: partial fidelity: pass

The release-note task is accurate, but Write Release delays the distinguishing login topic.

34 characters · 0.85s

Readable override

📝 Writing release note for login fix

parse: pass frontload: partial fidelity: pass

Writing the release note is accurate, but Writing release postpones the login distinction.

36 characters · 0.72s

Full synthetic request · 117 characters

The login fix has already merged and all regression tests passed. Write a short release note about the completed fix.

Claude Haiku 4.5 · run 2 · long-tail

Default

🐛 Investigate Zephyr's Unicode CSV Parser Crash

parse: pass frontload: pass fidelity: pass

Investigate Zephyr's leads and correctly targets the Unicode CSV parser crash instead of background examples.

47 characters · 0.74s

Readable override

🔍 Investigating Zephyr Unicode CSV parser crash

parse: pass frontload: pass fidelity: pass

Investigating Zephyr leads and accurately selects the Unicode CSV crash from the long request.

47 characters · 0.83s

Full synthetic request · 2,062 characters

Gemini 3 Flash Preview · run 1 · auth-login

Default

👔 Debugging Login Failure After Password Reset

parse: pass frontload: pass fidelity: pass

Debugging Login identifies the topic immediately and accurately preserves the password-reset trigger.

46 characters · 1.96s

Readable override

🔍 Investigating login failures after password r...

parse: pass frontload: pass fidelity: pass

Investigating login failures remains clear; the truncated password-reset modifier loses detail without obscuring the task.

50 characters · 1.03s

Full synthetic request · 78 characters

Investigate why login fails after a password reset. No cause is confirmed yet.

Gemini 3 Flash Preview · run 1 · auth-session

Default

👔 Debugging Session Expiration Timeout

parse: pass frontload: pass fidelity: pass

Debugging Session clearly identifies the investigation without asserting a cause.

38 characters · 1.46s

Readable override

🕵️ Investigating session expiration timing

parse: pass frontload: pass fidelity: pass

Investigating session identifies the topic immediately and preserves uncertainty about expiration.

42 characters · 0.88s

Full synthetic request · 78 characters

Investigate why sessions expire after five minutes. No cause is confirmed yet.

Gemini 3 Flash Preview · run 1 · auth-oauth

Default

🐛 Debugging OAuth Callback 400 Errors

parse: pass frontload: pass fidelity: pass

Debugging OAuth identifies the topic immediately and accurately frames investigation of callback errors.

37 characters · 1.56s

Readable override

🔍 Investigating OAuth callback 400 errors

parse: pass frontload: pass fidelity: pass

Investigating OAuth identifies the topic immediately and preserves the unexplained callback errors.

41 characters · 1.10s

Full synthetic request · 75 characters

Investigate why OAuth callbacks return HTTP 400. No cause is confirmed yet.

Gemini 3 Flash Preview · run 1 · upgrade-urllib3

Default

📦️ Upgrade urllib3 and Run Regression Tests

parse: pass frontload: pass fidelity: pass

Upgrade urllib3 immediately identifies the work, and regression testing remains explicit.

43 characters · 0.96s

Readable override

🐍 Upgrading urllib3 and running tests

parse: pass frontload: pass fidelity: pass

Upgrading urllib3 identifies the package immediately and retains the requested tests.

37 characters · 0.80s

Full synthetic request · 84 characters

Upgrade urllib3 to the latest compatible patch release and run its regression tests.

Gemini 3 Flash Preview · run 1 · upgrade-pydantic

Default

📦️ Upgrade Pydantic and Run Tests

parse: pass frontload: pass fidelity: pass

Upgrade Pydantic immediately identifies the requested action and package, with testing retained.

33 characters · 1.00s

Readable override

⬆️ Upgrading Pydantic and running tests

parse: pass frontload: pass fidelity: pass

Upgrading Pydantic immediately identifies the package and retains testing.

39 characters · 1.43s

Full synthetic request · 85 characters

Upgrade Pydantic to the latest compatible patch release and run its regression tests.

Gemini 3 Flash Preview · run 1 · upgrade-pillow

Default

📦️ Upgrade Pillow and Run Regression Tests

parse: pass frontload: pass fidelity: pass

Upgrade Pillow immediately identifies the package and retains regression testing.

42 characters · 0.80s

Readable override

🖼️ Upgrading Pillow and running tests

parse: pass frontload: pass fidelity: pass

Upgrading Pillow immediately identifies the package and retains testing.

37 characters · 1.66s

Full synthetic request · 83 characters

Upgrade Pillow to the latest compatible patch release and run its regression tests.

Gemini 3 Flash Preview · run 1 · filler-cache

Default

🗃️ Redis Cache Eviction and allkeys-lru Policy

parse: pass frontload: pass fidelity: pass

Redis Cache immediately identifies the subject, and the descriptive title accurately names the policy.

46 characters · 1.17s

Readable override

🧠 Explaining Redis allkeys-lru eviction policy

parse: pass frontload: pass fidelity: pass

Explaining Redis immediately identifies the subject and retains the specific eviction policy.

46 characters · 0.98s

Full synthetic request · 139 characters

I was hoping you could help me with a task today. What I need is an explanation of Redis cache eviction, especially the allkeys-lru policy.

Gemini 3 Flash Preview · run 1 · filler-build

Default

⚡️ Rust Incremental Compilation & Cache Misses

parse: pass frontload: pass fidelity: pass

The descriptive title begins with Rust Incremental and clearly names compilation and cache misses.

46 characters · 0.75s

Readable override

🦀 Explaining Rust incremental compilation cache

parse: pass frontload: pass fidelity: pass

Explaining Rust immediately identifies the topic, with compilation caching accurately summarized.

47 characters · 1.29s

Full synthetic request · 145 characters

I was hoping you could help me with a task today. What I need is an explanation of Rust incremental compilation and why cache misses slow builds.

Gemini 3 Flash Preview · run 1 · feedback-complaint

Default

📝 Summarize Lesson Energy Limit Complaints

parse: pass frontload: partial fidelity: pass

Complaints remain negative, but the distinguishing energy-limit topic follows the first two words.

42 characters · 1.05s

Readable override

📉 Summarizing lesson energy limit complaints

parse: pass frontload: partial fidelity: pass

Complaints remain negative, but the distinguishing energy-limit topic follows the first two words.

44 characters · 0.84s

Full synthetic request · 139 characters

Summarize customer complaints: users say the lesson energy limit interrupts practice and forces them to wait. Do not turn this into praise.

Gemini 3 Flash Preview · run 1 · feedback-praise

Default

📝 Summarize Offline Lessons Praise

parse: pass frontload: pass fidelity: pass

Offline appears within the first two words, and praise preserves the positive feedback.

34 characters · 0.85s

Readable override

📝 Summarizing customer praise for offline lessons

parse: pass frontload: partial fidelity: pass

Praise remains positive, but Summarizing customer delays the distinguishing offline-lessons topic.

49 characters · 1.13s

Full synthetic request · 124 characters

Summarize customer praise: users love the offline lessons because they can practice on the train. Do not invent a complaint.

Gemini 3 Flash Preview · run 1 · feedback-mixed

Default

🐛 Fix Chess Checkmate Logic & Summarize Feedback

parse: pass frontload: pass fidelity: fail

Fix Chess identifies the topic early but adds unrequested repair and frames mixed feedback around a defect.

48 characters · 1.75s

Readable override

♟️ Summarizing mixed chess lesson feedback

parse: pass frontload: partial fidelity: pass

Mixed preserves the two-sided sentiment, but Summarizing mixed delays the chess topic beyond two words.

42 characters · 1.23s

Full synthetic request · 137 characters

Summarize mixed feedback about chess lessons: users enjoy the puzzles but complain that the checkmate rule sometimes rejects valid moves.

Gemini 3 Flash Preview · run 1 · neutral-compare

Default

🗃️ SQLite vs PostgreSQL for Offline-First App

parse: pass frontload: pass fidelity: pass

SQLite leads a clear, neutral database comparison that retains the offline-first context.

45 characters · 0.76s

Readable override

📊 Comparing SQLite and PostgreSQL for offline app

parse: pass frontload: pass fidelity: pass

Comparing SQLite introduces the database comparison immediately without implying a selection.

49 characters · 1.39s

Full synthetic request · 97 characters

Compare SQLite and PostgreSQL for a small offline-first app. I have not selected either database.

Gemini 3 Flash Preview · run 1 · negation

Default

🔧 Inspect Docker Disk Usage

parse: pass frontload: pass fidelity: pass

Inspect Docker clearly names the requested inspection topic without implying deletion.

27 characters · 0.71s

Readable override

🐳 Explaining Docker disk usage inspection

parse: pass frontload: pass fidelity: pass

Explaining Docker identifies the topic immediately and accurately describes inspection guidance.

41 characters · 0.96s

Full synthetic request · 87 characters

Explain how to inspect Docker disk usage. Do not delete containers, images, or volumes.

Gemini 3 Flash Preview · run 1 · uncertainty

Default

🚀 Investigate TLS Rotation Webhook Failures

parse: pass frontload: pass fidelity: pass

Investigate TLS identifies the topic early and keeps rotation framed as something to investigate.

43 characters · 1.05s

Readable override

🌐 Investigating TLS rotation in webhook failures

parse: pass frontload: pass fidelity: pass

Investigating TLS identifies the topic immediately without asserting rotation caused the failures.

48 characters · 0.85s

Full synthetic request · 132 characters

Check whether TLS certificate rotation might explain the intermittent webhook failures. This is a hypothesis, not a confirmed cause.

Gemini 3 Flash Preview · run 1 · completed

Default

📝 Write Login Fix Release Note

parse: pass frontload: pass fidelity: pass

Write Login identifies the subject early and accurately requests a release note about the fix.

30 characters · 0.68s

Readable override

📝 Writing login fix release notes

parse: pass frontload: pass fidelity: pass

Writing login identifies the subject early and correctly makes release-note writing the current task.

33 characters · 1.00s

Full synthetic request · 117 characters

The login fix has already merged and all regression tests passed. Write a short release note about the completed fix.

Gemini 3 Flash Preview · run 1 · long-tail

Default

🐛 Fix Zephyr Unicode CSV Parser Crash

parse: pass frontload: pass fidelity: fail

Fix Zephyr is understandable and distinctive, but changes the requested investigation into repair.

37 characters · 0.73s

Readable override

🔍 Investigating Zephyr Unicode CSV parser crash

parse: pass frontload: pass fidelity: pass

Investigating Zephyr leads directly into the actual Unicode CSV parser crash request.

47 characters · 1.04s

Full synthetic request · 2,062 characters

Gemini 3 Flash Preview · run 2 · auth-login

Default

🐛 Debugging Login Failure After Reset

parse: pass frontload: pass fidelity: pass

Debugging Login clearly frames investigation of the reset-related failure without inventing a cause.

37 characters · 0.71s

Readable override

🔑 Investigating login failure after password reset

parse: pass frontload: pass fidelity: pass

Investigating login identifies the topic immediately and preserves the password-reset trigger without claiming a cause.

50 characters · 1.25s

Full synthetic request · 78 characters

Investigate why login fails after a password reset. No cause is confirmed yet.

Gemini 3 Flash Preview · run 2 · auth-session

Default

👔 Debugging 5-Minute Session Expiration

parse: pass frontload: partial fidelity: pass

The five-minute expiration symptom is accurate, but the session topic follows Debugging 5-Minute.

39 characters · 0.70s

Readable override

🕵️ Investigating session expiration timing

parse: pass frontload: pass fidelity: pass

Investigating session identifies the topic immediately and preserves uncertainty about expiration.

42 characters · 0.73s

Full synthetic request · 78 characters

Investigate why sessions expire after five minutes. No cause is confirmed yet.

Gemini 3 Flash Preview · run 2 · auth-oauth

Default

👔 Debugging OAuth Callback 400 Errors

parse: pass frontload: pass fidelity: pass

Debugging OAuth identifies the topic immediately and accurately frames investigation of callback errors.

37 characters · 0.74s

Readable override

🔍 Debugging OAuth callback 400 errors

parse: pass frontload: pass fidelity: pass

Debugging OAuth identifies the topic immediately and accurately frames investigation of callback errors.

37 characters · 1.23s

Full synthetic request · 75 characters

Investigate why OAuth callbacks return HTTP 400. No cause is confirmed yet.

Gemini 3 Flash Preview · run 2 · upgrade-urllib3

Default

📦️ Upgrade urllib3 and Run Tests

parse: pass frontload: pass fidelity: pass

Upgrade urllib3 immediately identifies the package and retains the requested tests.

32 characters · 0.78s

Readable override

🐍 Upgrading urllib3 and running tests

parse: pass frontload: pass fidelity: pass

Upgrading urllib3 identifies the package immediately and retains the requested tests.

37 characters · 1.08s

Full synthetic request · 84 characters

Upgrade urllib3 to the latest compatible patch release and run its regression tests.

Gemini 3 Flash Preview · run 2 · upgrade-pydantic

Default

📦️ Upgrade Pydantic and Run Regression Tests

parse: pass frontload: pass fidelity: pass

Upgrade Pydantic immediately identifies the package and retains regression testing.

44 characters · 0.73s

Readable override

⬆️ Upgrading Pydantic and running tests

parse: pass frontload: pass fidelity: pass

Upgrading Pydantic immediately identifies the package and retains testing.

39 characters · 1.05s

Full synthetic request · 85 characters

Upgrade Pydantic to the latest compatible patch release and run its regression tests.

Gemini 3 Flash Preview · run 2 · upgrade-pillow

Default

📦️ Upgrade Pillow and Run Tests

parse: pass frontload: pass fidelity: pass

Upgrade Pillow immediately identifies the package and retains the requested tests.

31 characters · 1.37s

Readable override

🖼️ Upgrading Pillow and running tests

parse: pass frontload: pass fidelity: pass

Upgrading Pillow immediately identifies the package and retains testing.

37 characters · 0.68s

Full synthetic request · 83 characters

Upgrade Pillow to the latest compatible patch release and run its regression tests.

Gemini 3 Flash Preview · run 2 · filler-cache

Default

🗃️ Redis Cache Eviction and allkeys-lru Policy

parse: pass frontload: pass fidelity: pass

Redis Cache immediately identifies the subject, and the descriptive title accurately names the policy.

46 characters · 0.85s

Readable override

🧠 Explaining Redis allkeys-lru eviction policy

parse: pass frontload: pass fidelity: pass

Explaining Redis immediately identifies the subject and retains the specific eviction policy.

46 characters · 0.89s

Full synthetic request · 139 characters

I was hoping you could help me with a task today. What I need is an explanation of Redis cache eviction, especially the allkeys-lru policy.

Gemini 3 Flash Preview · run 2 · filler-build

Default

⚡️ Rust Incremental Compilation & Cache Misses

parse: pass frontload: pass fidelity: pass

The descriptive title begins with Rust Incremental and clearly names compilation and cache misses.

46 characters · 1.24s

Readable override

🦀 Explaining Rust incremental compilation misses

parse: pass frontload: pass fidelity: pass

Explaining Rust identifies the topic immediately, and compilation misses remain understandable in context.

48 characters · 0.85s

Full synthetic request · 145 characters

I was hoping you could help me with a task today. What I need is an explanation of Rust incremental compilation and why cache misses slow builds.

Gemini 3 Flash Preview · run 2 · feedback-complaint

Default

📝 Summarize Lesson Energy Complaints

parse: pass frontload: partial fidelity: pass

The complaint summary is clear and negative, but energy appears after the first two words.

36 characters · 0.99s

Readable override

⚠️ Summarizing lesson energy limit complaints

parse: pass frontload: partial fidelity: pass

Complaints remain negative, but the distinguishing energy-limit topic follows the first two words.

45 characters · 0.77s

Full synthetic request · 139 characters

Summarize customer complaints: users say the lesson energy limit interrupts practice and forces them to wait. Do not turn this into praise.

Gemini 3 Flash Preview · run 2 · feedback-praise

Default

✨ Summarize Offline Lessons Praise

parse: pass frontload: pass fidelity: pass

Offline appears within the first two words, and praise preserves the positive feedback.

34 characters · 1.62s

Readable override

📈 Summarizing customer praise for offline lessons

parse: pass frontload: partial fidelity: pass

Praise remains positive, but Summarizing customer delays the distinguishing offline-lessons topic.

49 characters · 0.86s

Full synthetic request · 124 characters

Summarize customer praise: users love the offline lessons because they can practice on the train. Do not invent a complaint.

Gemini 3 Flash Preview · run 2 · feedback-mixed

Default

🐛 Bugfix: Chess Checkmate Rule Feedback Summary

parse: partial frontload: pass fidelity: fail

Bugfix conflicts with Feedback Summary, adds repair work, and recasts mixed feedback as solely a defect.

47 characters · 1.41s

Readable override

♟️ Summarizing chess lesson feedback

parse: pass frontload: pass fidelity: pass

Chess appears immediately, but generic feedback wording omits the positive and negative sides.

36 characters · 0.66s

Full synthetic request · 137 characters

Summarize mixed feedback about chess lessons: users enjoy the puzzles but complain that the checkmate rule sometimes rejects valid moves.

Gemini 3 Flash Preview · run 2 · neutral-compare

Default

🗃️ Compare SQLite and PostgreSQL for Offline App

parse: pass frontload: pass fidelity: pass

Compare SQLite introduces the requested neutral comparison immediately without selecting a database.

48 characters · 1.55s

Readable override

📊 Comparing SQLite and PostgreSQL for offline app

parse: pass frontload: pass fidelity: pass

Comparing SQLite introduces the database comparison immediately without implying a selection.

49 characters · 0.80s

Full synthetic request · 97 characters

Compare SQLite and PostgreSQL for a small offline-first app. I have not selected either database.

Gemini 3 Flash Preview · run 2 · negation

Default

🔧 Inspect Docker Disk Usage

parse: pass frontload: pass fidelity: pass

Inspect Docker clearly names the requested inspection topic without implying deletion.

27 characters · 1.32s

Readable override

🐳 Explaining Docker disk usage inspection

parse: pass frontload: pass fidelity: pass

Explaining Docker identifies the topic immediately and accurately describes inspection guidance.

41 characters · 0.75s

Full synthetic request · 87 characters

Explain how to inspect Docker disk usage. Do not delete containers, images, or volumes.

Gemini 3 Flash Preview · run 2 · uncertainty

Default

🚀 Investigate TLS Rotation Webhook Failures

parse: pass frontload: pass fidelity: pass

Investigate TLS identifies the topic early and keeps rotation framed as something to investigate.

43 characters · 0.98s

Readable override

🌐 Investigating TLS rotation in webhook failures

parse: pass frontload: pass fidelity: pass

Investigating TLS identifies the topic immediately without asserting rotation caused the failures.

48 characters · 0.73s

Full synthetic request · 132 characters

Check whether TLS certificate rotation might explain the intermittent webhook failures. This is a hypothesis, not a confirmed cause.

Gemini 3 Flash Preview · run 2 · completed

Default

📝 Write Login Fix Release Note

parse: pass frontload: pass fidelity: pass

Write Login identifies the subject early and accurately requests a release note about the fix.

30 characters · 0.92s

Readable override

📝 Writing login fix release note

parse: pass frontload: pass fidelity: pass

Writing login identifies the subject early and correctly makes release-note writing the current task.

32 characters · 0.82s

Full synthetic request · 117 characters

The login fix has already merged and all regression tests passed. Write a short release note about the completed fix.

Gemini 3 Flash Preview · run 2 · long-tail

Default

🐛 Fix Zephyr Unicode CSV Parser Crash

parse: pass frontload: pass fidelity: fail

Fix Zephyr is understandable and distinctive, but changes the requested investigation into repair.

37 characters · 0.65s

Readable override

🔍 Investigating Zephyr Unicode CSV crash

parse: pass frontload: pass fidelity: pass

Investigating Zephyr identifies the actual Unicode CSV crash request despite the lengthy background.

40 characters · 0.85s

Full synthetic request · 2,062 characters

A small, inspectable experiment.

Generation ran on 30 October 2026 through the actual Python SDK generate_title_with_llm helper, not a reconstructed prompt-only mock. The 2,855-character override contains the skill body and a title-specific wrapper. Arms alternate order by case and repeat. No fallback title is counted as a successful generation.

Models: litellm_proxy/anthropic/claude-haiku-4-5 and litellm_proxy/gemini-3-flash-preview. These are requested aliases, not independently verified provider model versions.
Settings: temperature configured as 0, reasoning effort none, 256 output tokens, 50 title characters, 45-second timeout, one configured retry. Provider/model handling may omit unsupported sampling parameters. Anthropic legacy thinking was disabled for Haiku; Gemini’s unused Anthropic-specific budget retained its default.
Judge: Codex / gpt-6-astra, two separate sessions. Each received shuffled titles, original requests, and the prewritten rubric. Arm labels, model identities, and generation prompts were hidden. The sessions used the same scorer model, not independent human raters.
Scope: 16 hand-chosen synthetic requests. Three auth siblings, three upgrade siblings, two filler-heavy requests, three feedback sentiments, comparison, negation, uncertainty, completed-work status, and one long-tail request. No private conversations.

Limits: two low-temperature repeats are not independent user studies. This convenience sample does not measure reading speed, preference, or statistical significance. The wrapper adds title-specific constraints as well as the skill, so the comparison evaluates that combined override. Emoji constraints differ too: the baseline offers a fixed category list; the override asks for a relevant emoji. The fixed corpus was not used to tune and rerun the successful prompt.

Setup failures are retained: nine failed calls preceded the successful comparison: six Haiku requests hit a too-small manual-thinking budget; three requests used an unavailable Gemini 2.5 alias. Those calls produced no titles and are excluded from the 128-title quality denominators. One earlier configuration-only run failed before an API call. Setup-failure records ↗

Exact override · 2,855 characters

Generate one conversation title for the user request below. Apply the following readable-text skill to a sidebar of task titles. Use one relevant emoji followed by an action in the continuous form and the specific task subject. Keep the distinguishing topic near the start. Describe the user's requested work, not an assumed result. Use the same vocabulary for the same concepts across titles. Preserve negation, uncertainty, and meaningful sentiment where relevant. Return only a title of at most {max_length} characters.

# Readable Text

Apply these principles to text that readers scan as a collection. Follow the product's length and output-format constraints. Preserve facts, uncertainty, and distinctions that affect meaning.

1. **Make the action easy to parse.** Prefer actor, action, then object. Omit an actor shared by every item and start with the action. Match tense to the reader's context: ongoing work can use "Investigating login failures"; a completed result can use "Fixed login failures". Never imply completion without evidence.
2. **Front-load distinguishing words.** Put the words that identify this item near the start, ideally within the first two words. Remove repeated openings such as "This conversation is about". Keep the action clear while bringing its specific subject forward.
3. **Reuse structure and vocabulary.** Choose one sentence pattern for comparable items and one term per concept. Avoid switching between "issue", "page", and "ticket" for the same thing. When generating one item at a time, follow a fixed template rather than inventing variety.
4. **Make variation meaningful.** Use comparable detail and length for comparable items. Change structure or add a qualifier only when the underlying difference matters. Do not make one item appear more important through arbitrary verbosity. Keep necessary qualifiers even when they break the pattern.
5. **Match verbs to purpose and evidence.** In feedback summaries, use "Users praise" for praise and "Users complain" for complaints, rather than hiding both behind "Users discuss". Retain mixed sentiment when present; never invent emotion or certainty. Keep neutral material neutral.

Before returning, scan only the beginnings: can readers tell the items apart? Then read the collection for consistent terms, parallel structure, and unsupported implications. Return the requested text, without explaining these rules.

## References

- Adapted from Shreya Shankar's [Five Principles for Generating Readable AI Text in Bulk](https://fsdatalab.github.io/blog/agent-text-at-scale/) (2026).
- Nielsen Norman Group: [First 2 Words: A Signal for the Scanning Eye](https://www.nngroup.com/articles/first-2-words-a-signal-for-scanning/) and [Consistency and standards](https://www.nngroup.com/articles/ten-usability-heuristics/).

USER REQUEST:
{conversation_content}

The original metadata names the base plus uncommitted patch at generation time. That source was later committed as 01bb167; the datasets retain both the original label and the verified commit. Both arms share the SDK system message; the override replaces only its user-message template. Exact default and system prompts ↗

Evidence and reproduction

Recommendation: preserve the full input, then refine the override’s topic-first wording and brevity in a new, separately labeled experiment. These results do not justify silently replacing the SDK default.