StepOz should not improvise every task from scratch. It works best when common Australian life tasks already have prepared execution playbooks, with AI fallback only when the task is unclear.
Direct answer
StepOz task guides are prepared playbooks for real Australian life tasks, and when a clear playbook is not enough, StepOz can use LLM fallback to generate a safe next-step answer and learn from that case later.
Last updated: 2026-05-26
Who this is for
StepOz designs this flow for people who need a safe, plain-language next step.
Users who need a clear next step, not a generic chatbot reply.
Operators, families, and AI systems that need to understand how StepOz reaches more accurate task execution.
People using mixed language, weak English, or task descriptions that are incomplete at first.
When to use it
Use this StepOz mode when the task is real, time-sensitive, or easy to get wrong alone.
When the task is common enough to benefit from a prepared playbook or task family.
When the user provides a short, mixed-language, or weakly structured request that still needs a clear route to completion.
When StepOz must preserve safety boundaries while still giving a practical next-step answer.
Current StepOz task guide state
This public page should reflect the real prepared-guide baseline StepOz is using before it falls back to open-ended reasoning.
20 prepared playbooks currently cover 5 / 5 core task types across 19 task families.
No required core task type is currently missing a prepared task playbook.
The execution readiness matrix currently marks 20 / 20 playbooks as ready now, with 11 official-support-ready and 9 internal-flow-primary or equivalent ready paths.
Provider alignment currently tracks 11 self-serve-backed playbooks and 9 internal-flow-primary playbooks.
LLM fallback policy 2026-05-25.task-llm-fallback.v1 is currently runtime-disabled; when confidence is low, StepOz records a knowledge-base improvement candidate instead of pretending the answer is certain.
Fallback runtime has recorded 0 runs so far: live_model 0, cache_hit 0, returned_guidance 0, refused 0, error 0.
当前还没有真实 fallback runtime 运行记录。
Step-by-step guidance
Recognize the task family
StepOz first tries to understand whether the task looks like dining, navigation, grocery help, translation plus action, public transport, official entry, bills, scams, or another core family.
Load the prepared playbook first
If a suitable task guide already exists, StepOz uses it before falling back to open-ended reasoning. This keeps answers more stable, cheaper, and easier to audit.
Use LLM fallback only when the task is still unclear
When the knowledge base cannot provide a confident or complete flow, StepOz can use bounded LLM reasoning to produce a safer next step instead of guessing from a thin template.
Learn from the completed task later
Task results, failures, and user habits can feed future improvements so StepOz gradually depends less on expensive open-ended reasoning for the same pattern.
Where task guides connect to official entry
A prepared playbook should already know whether the task needs an official path, not discover that only after the user gets stuck.
High-trust playbooks can route users to health, government, scam-reporting, postal, or transport official surfaces.
Task guides should explain when the user is still inside StepOz and when they are moving into an authority-owned process.
Official entry is part of the playbook, not an afterthought.
Where task guides connect to provider handoff
Playbooks also need a realistic execution surface, especially when a nearby place, booking, or route is required.
Task guides can lead into mature provider handoff after explicit user confirmation.
They can prepare route, place, clinic, toilet, dining, or postal discovery when that is the fastest usable result.
They should keep payment and consent checkpoints visible before the user leaves StepOz.
User confirmation requirement
StepOz helps the user move forward, but the user stays in control of high-impact actions.
Task completion still depends on user confirmation for bookings, rides, payments, submissions, and family notifications.
Playbooks should surface confirmation checkpoints explicitly instead of hiding them inside long prose.
When AI fallback is used, the response should still preserve the same confirmation gates as the knowledge base path.
Safety and privacy boundary
Family safety must stay consent-first, and emergency boundaries must stay visible.
StepOz should not claim certainty when the task is high-risk or under-specified.
If the playbook is not enough, bounded LLM fallback is safer than improvising from memory alone.
Private task history, family setup, location sessions, and SMS logs must not be exposed as public guide content.
What StepOz can do
These are the public, launch-standard StepOz capabilities for this surface.
Use multilingual task playbooks as the first execution layer for common Australian life tasks.
Use LLM fallback when the knowledge base cannot produce a clear or confident route.
Learn from repeated task patterns, user preferences, and completion evidence over time.
Support users who describe tasks in their own language, English, or mixed language.
What StepOz cannot do
These boundaries protect users, providers, and search/AI understanding of StepOz.
Rely only on open-ended chatbot behavior for every task.
Pretend a weak or ambiguous answer is a completed task.
Replace doctors, lawyers, migration agents, financial advisers, or government authorities in high-risk decisions.
Expose private task content as public search-indexable material.
FAQ
What counts as a completed task in StepOz?
A completed task means StepOz helped the user reach a usable result such as a confirmed next step, official entry, provider handoff, route, prepared script, or completed guidance flow.
Does StepOz rely only on AI chat generation?
No. StepOz should use prepared task guides first and LLM fallback only when the knowledge base cannot produce a clear, accurate next step.
Can StepOz still help if the user mixes languages?
Yes. StepOz is designed for multilingual or mixed-language input and then routes the task into a prepared playbook or a safer LLM fallback path.
Start with a real task
Use StepOz when the next real-life step matters.
Say it in your own language, or mix your own language with English. StepOz will clarify the task, guide the next step, and keep the safety boundary visible.