Editor-selected existing cases and prompts, separate from AI news. Selection dates use Hong Kong time.
Five editorial adaptations use fictional fixtures to check pack rounding, document meaning, tempo conversion, overnight lighting and grid boundaries. Not newly discovered real-world cases; prompts are untested. Each card includes human checks, its original case and service limits.
Use fictional stock and replacement packs to check whole-pack rounding, totals and unapproved substitutions.
Prompt
From [item targets, stock, pack sizes, unit prices, currency, included fees, cap, substitution rules], draft a purchase ledger. Show deficits, whole-pack ceilings, resulting quantities and totals; ask about unknown units. Recalculate [one parameter change] and identify invalidated decisions. Stop on over-budget or unapproved substitutions; do not reduce requirements, switch products, edit carts or pay.
Fictional fixture, not a real usage record: Target: 3 L oat drink; stock: 0.6 L. Original 1 L packs are unavailable. Approved replacement: 0.8 L per pack at 18 currency units, cap 55 excluding delivery. Dairy milk is 1 L at 15 but is not an approved substitute. Variant: change only replacement size to 0.7 L, retaining price.
Evidence limits and cautions
The author describes a CLI/MCP workflow for Ukraine’s Silpo, explicitly leaving checkout to humans. This establishes no HK/Macau/Taiwan supermarket support. We rebuild no cart and adopt no time-saving estimate. Source published: 2026-09-24; checked: 2026-11-11. The original source provides inspiration, not validation of this exercise. Its toolchain, regional services and plans have not been reproduced. This card uses only the documented Gemini web text environment.
Original editorial adaptation, not a newly discovered use case or the source author’s verbatim workflow. Source-reported outcomes do not guarantee this exercise. Prompts were not tested. Adapted from this existing case: https://hkfrees.com/ai/case/reported-silpo-restock/?lang=en Adds a controlled pack-size change that exposes a rounding-induced budget failure. Open this card’s title for the official environment-source links and human acceptance steps on its full case page.
- Platform and region: ordinary Gemini web text chat. Official lists include Hong Kong, Macau, Taiwan, Chinese variants and English. Documentation checked 2026-11-11; local connections were not tested.
- Plan and account: prefer an adult personal account. Standard free access is a starting point, with variable usage caps and no required subscription. Taiwan advertises NT$0/month free access; Hong Kong/Macau paid prices are unverified. No particular model or prompt count is guaranteed. Work, school and child accounts need separate checks.
- No original toolchain is connected. Web availability does not establish mobile, API, CLI or original-automation access. Never supply confidential data. Activity-off chats can still remain for 72 hours and receive safety review.
- Use fictional items and amounts, not shopping history, addresses or payment data.
Original source: Dmytro Bilukha: Real-World Agentic E-Commerce
Compare fictional originals and converted text for changed negation, version conditions and defaults.
Prompt
Compare [numbered original] with [numbered conversion] against [required semantics and version fields]. Separate omissions, meaning changes, unsupported additions and harmless formatting. Preserve negation, conditions, units and scope. Cite both line IDs and propose minimal textual repairs; invent no source content. Review supplied text only, without crawling pages or validating software behavior.
Fictional fixture, not a real usage record: S1: Without approval, cache deletion is prohibited. S2: In v2, timeout defaults to 0, meaning no timeout. S3: Only v3 supports retry. Converted C1: Approval permits cache deletion. C2: timeout defaults to timing out after 0 seconds. C3: retry is supported. C4: Automatic backup is supported, with no matching original.
Evidence limits and cautions
The author describes local Pulpie MCP extraction and one timed 60-page crawl. This is self-report, not a fidelity or speed guarantee. This card compares supplied short text without crawling, installation or original-tool validation. Source published: 2026-10-05; checked: 2026-11-11. The original source provides inspiration, not validation of this exercise. Its toolchain, regional services and plans have not been reproduced. This card uses only the documented Gemini web text environment.
Original editorial adaptation, not a newly discovered use case or the source author’s verbatim workflow. Source-reported outcomes do not guarantee this exercise. Prompts were not tested. Adapted from this existing case: https://hkfrees.com/ai/case/reported-docs-extraction-fidelity/?lang=en Adds known-answer logical reversals and invented-content checks, beyond file existence or word counts. Open this card’s title for the official environment-source links and human acceptance steps on its full case page.
- Platform and region: ordinary Gemini web text chat. Official lists include Hong Kong, Macau, Taiwan, Chinese variants and English. Documentation checked 2026-11-11; local connections were not tested.
- Plan and account: prefer an adult personal account. Standard free access is a starting point, with variable usage caps and no required subscription. Taiwan advertises NT$0/month free access; Hong Kong/Macau paid prices are unverified. No particular model or prompt count is guaranteed. Work, school and child accounts need separate checks.
- No original toolchain is connected. Web availability does not establish mobile, API, CLI or original-automation access. Never supply confidential data. Activity-off chats can still remain for 72 hours and receive safety review.
- Use fictional documentation, not internal APIs, keys or restricted documents.
Original source: Documentation crawler test by Jessica Doering
Check beat-to-second conversion with a fictional arrangement, separating arithmetic duration, audio tails and unheard quality.
Prompt
From [bars per section, beats per bar, BPM beat unit, tempo, overlap/rest rules], list beats and start/end seconds. Show formulas and precision; do not invent sections. Recalculate [tempo change] while retaining beat counts. Separate score duration, rendering tails and actual file length. Produce planning text only; claim no audio generation, listening or measurement.
Fictional fixture, not a real usage record: Constant quarter-note tempo: 120 BPM, 4/4. Section A: 4 bars; B: 6 bars, consecutive with no overlap, rests or intro; 4 beats per bar. Variant: change tempo only to 90 BPM. No audio file or tail settings are supplied.
Evidence limits and cautions
The author shows code-generated music with Claude Code and also reports vocal and sampling failures. We neither listen to, synthesize nor measure audio or reproduce that code workflow. Correct arithmetic does not establish musical quality. Source published: 2026-02-04; checked: 2026-11-11. The original source provides inspiration, not validation of this exercise. Its toolchain, regional services and plans have not been reproduced. This card uses only the documented Gemini web text environment.
Original editorial adaptation, not a newly discovered use case or the source author’s verbatim workflow. Source-reported outcomes do not guarantee this exercise. Prompts were not tested. Adapted from this existing case: https://hkfrees.com/ai/case/reported-code-audio-arrangement/?lang=en Adds beat-count invariance and timing recalculation after a tempo change, catching stale timestamps. Open this card’s title for the official environment-source links and human acceptance steps on its full case page.
- Platform and region: ordinary Gemini web text chat. Official lists include Hong Kong, Macau, Taiwan, Chinese variants and English. Documentation checked 2026-11-11; local connections were not tested.
- Plan and account: prefer an adult personal account. Standard free access is a starting point, with variable usage caps and no required subscription. Taiwan advertises NT$0/month free access; Hong Kong/Macau paid prices are unverified. No particular model or prompt count is guaranteed. Work, school and child accounts need separate checks.
- No original toolchain is connected. Web availability does not establish mobile, API, CLI or original-automation access. Never supply confidential data. Activity-off chats can still remain for 72 hours and receive safety review.
- Use original fictional beat plans, not unlicensed recordings; this exercise synthesizes no audio.
Original source: Claude composer
Use fictional rules and timestamps to check overnight intervals and the precedence of a manual override.
Prompt
Using [rules, timezone, dates, interval inclusivity, precedence, test timestamps], list expected brightness, governing rule and reason. Resolve overnight intervals; ask rather than invent missing precedence. Check start/end boundaries and override expiry. This is a paper test: create no automation, control no devices and claim no execution. Second test: remove the manual-override precedence setting and reassess Nov 11 23:55.
Fictional fixture, not a real usage record: One fixed timezone, no daylight saving. Daily 22:00 inclusive to next-day 06:00 exclusive: 20% brightness; otherwise 70%. Manual override takes precedence: Nov 11 23:50 inclusive to Nov 12 00:20 exclusive holds 60%. Check Nov 11 21:59, 22:00, 23:55 and Nov 12 00:20, 05:59, 06:00. Second test: remove the manual-override precedence setting and reassess Nov 11 23:55.
Evidence limits and cautions
Tom’s Guide describes Alexa+ natural-language lighting/thermostat routines and inconsistent agent actions. We adapt only rule planning, without validating Alexa+ regional availability or historical pricing and without connecting devices. Source published: 2025-11-02; checked: 2026-11-11. The original source provides inspiration, not validation of this exercise. Its toolchain, regional services and plans have not been reproduced. This card uses only the documented Gemini web text environment.
Original editorial adaptation, not a newly discovered use case or the source author’s verbatim workflow. Source-reported outcomes do not guarantee this exercise. Prompts were not tested. Adapted from this existing case: https://hkfrees.com/ai/case/reported-alexa-routine-preview/?lang=en Adds overnight endpoint tests and a missing-precedence variant that must remain unresolved. Open this card’s title for the official environment-source links and human acceptance steps on its full case page.
- Platform and region: ordinary Gemini web text chat. Official lists include Hong Kong, Macau, Taiwan, Chinese variants and English. Documentation checked 2026-11-11; local connections were not tested.
- Plan and account: prefer an adult personal account. Standard free access is a starting point, with variable usage caps and no required subscription. Taiwan advertises NT$0/month free access; Hong Kong/Macau paid prices are unverified. No particular model or prompt count is guaranteed. Work, school and child accounts need separate checks.
- No original toolchain is connected. Web availability does not establish mobile, API, CLI or original-automation access. Never supply confidential data. Activity-off chats can still remain for 72 hours and receive safety review.
- Use fictional lighting rules only, not locks, alarms, heating or real household routines.
Original source: I’ve been using Alexa+ for 3 months — here are my 5 favorite features (and one I’m not a fan of)
Compare fixed-dead and wrapping boundaries on a fictional grid, preserving neighbour counts and synchronous updates.
Prompt
From [grid dimensions, coordinates, initial live cells, neighbour rule, boundary rule, update count], tabulate old-state neighbour counts and next states. Compute all cells from the same old grid, never partially updated cells. Change only [boundary rule] for comparison and explain added or removed live cells. Execute no code and claim no simulator deployment.
Fictional fixture, not a real usage record: 5×5 grid, 1-based (row,column). Only (1,2), (1,3), (1,4) are alive. Eight neighbours; live cells survive with 2 or 3, dead cells are born with exactly 3. Update all cells synchronously once. First use fixed-dead exterior, then change only to toroidal wrapping in both dimensions.
Evidence limits and cautions
Cognition’s vendor article shows Devin building and deploying a Game of Life site, not independent validation of this exercise. We build or deploy no game and promise no development or benchmark performance. Source published: 2024-03-12; checked: 2026-11-11. The original source provides inspiration, not validation of this exercise. Its toolchain, regional services and plans have not been reproduced. This card uses only the documented Gemini web text environment.
Original editorial adaptation, not a newly discovered use case or the source author’s verbatim workflow. Source-reported outcomes do not guarantee this exercise. Prompts were not tested. Adapted from this existing case: https://hkfrees.com/ai/case/reported-cellular-simulation-step/?lang=en Adds a known-answer topology-only comparison to distinguish boundary mistakes from update-order errors. Open this card’s title for the official environment-source links and human acceptance steps on its full case page.
- Platform and region: ordinary Gemini web text chat. Official lists include Hong Kong, Macau, Taiwan, Chinese variants and English. Documentation checked 2026-11-11; local connections were not tested.
- Plan and account: prefer an adult personal account. Standard free access is a starting point, with variable usage caps and no required subscription. Taiwan advertises NT$0/month free access; Hong Kong/Macau paid prices are unverified. No particular model or prompt count is guaranteed. Work, school and child accounts need separate checks.
- No original toolchain is connected. Web availability does not establish mobile, API, CLI or original-automation access. Never supply confidential data. Activity-off chats can still remain for 72 hours and receive safety review.
- This is a fictional game-rule exercise, not a biological or real-world prediction.
Original source: Introducing Devin, the first AI software engineer