Select a scenario
Inspect its evidence, assumptions, and a short future-incident storyboard.
Step 1 · Guided conversation
Let’s understand the idea together.
SafeBARS asks one question at a time, reflects back what it heard, and gradually widens the view from intended benefit to people, data, values, and consequences.
Step 2 · Ground
Nine lenses, each traceable to literature.
Coverage describes how explicitly your plan reasons about a concern. It is not a measure of how ethical the project is.
Evidence coverage
Waiting for analysis
No dimensions match this filter
Choose another coverage state to continue.
Step 3 · Perspective probe
Walk through futures the project plan does not describe.
Bounded agents create concrete, inspectable scenarios from different positions around the proposed app. They do not vote on ethics and they do not speak for real communities.
separate bounded probes
Step 4 · Dissonance map
See where a declared value and a design choice pull apart.
The map makes an argument inspectable: your commitment, the plan evidence, a plausible consequence, and the person who may experience it.
No dissonance paths yet
Run the literature and perspective analysis to construct the map.
Multimodal mirror
See your design through three views—then notice what you missed.
The mirror does not tell you what is wrong. It renders the people and futures your plan touches so you can discover the gap yourself.
Self-discovery
…
① Stakeholder map
Green = groups your plan already names. Red (pulsing) = groups the mirror infers you have not yet accounted for.
Named in plan
Inferred, needs attention
Inferred, lower priority
② Value-conflict heatmap
Each ethics lens × its evidence coverage. Redder cells mean the concern is named but not yet reasoned through.
③ Scenario branch tree
Each tension as a fork: the design as built versus the design with a safeguard you add.
Step 5 · Revision sandbox
Change the design—or explain why the argument does not hold.
The mirror surfaced tensions between your plan and each ethics lens. For every one, choose a reasoned response—revise, add a safeguard, contest with evidence, or consult the people affected. You stay the author; the tool only holds the space for that reflection.
Your realization from Step 4
What you noticed
The mirror only rendered this gap — turning it into a design change is yours to do in the editor below. The AI may suggest, but it cannot edit your plan for you.
Select a tension to respond
Each response is recorded with the relevant commitment, plan passage, scenario, and literature source.
Optional plan edit
Keep or revise in context
You do not have to change the plan. Edit only when a tension reveals a useful design change. To keep the wording, choose Contest with evidence or Consult real people above and record why.
Step 6 · Revision evidence
What changed, what remains, and why.
The ledger preserves the chain from original plan evidence to possible consequence, researcher response, and revised design. It does not turn reflection into a pass/fail certificate.
Locked. This step opens after you respond to at least one tension in Step 5 — choose Revise, Add safeguard, Contest, or Consult on a tension in the map.
No before / after evidence yet
Resolve at least one tension and replay the scenarios to build the ledger.