You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Research protocol: docs/research/PRODUCT_DISCOVERY_PROTOCOL.md in PR #17.
Problem
APP-01 has a strong security/architecture foundation but no verified primary user, differentiated top task or evidence that a public MCP browser client is the right first product form.
Implementation activity is not product evidence. A controlled internal client, gateway-mediated application, integration into an existing host, pivot or stop are valid outcomes.
Pre-work
Before the first evaluative session, commit:
one candidate primary persona and anti-persona;
one bounded read-only task;
current alternative/workaround;
target task time and success rate;
consent-comprehension threshold;
explicit falsification and go/pivot/stop criteria.
Method
Problem evidence
5–8 structured interviews or documented proxy evaluations.
Ask about recent observed work, frequency, consequence and alternatives.
Record raw observations separately from interpretation.
Capture non-adoption and contradictory evidence.
Task/consent prototype
Mock-backed and non-production:
endpoint/operator identity;
data/privacy boundary;
untrusted capabilities;
one visible read-only request;
explicit approval/rejection;
timeout, cancel and normalized errors;
bounded result and clear-session behavior.
Comprehension measures
Without showing the consent text again, verify the participant can state:
No runtime feature expansion and no closure of this issue without linked evidence. The chosen task must be compatible with an endpoint/deployment model that can pass APP-01 security, privacy, accessibility and operational gates.
Execution artifact
Research protocol:
docs/research/PRODUCT_DISCOVERY_PROTOCOL.mdin PR #17.Problem
APP-01 has a strong security/architecture foundation but no verified primary user, differentiated top task or evidence that a public MCP browser client is the right first product form.
Implementation activity is not product evidence. A controlled internal client, gateway-mediated application, integration into an existing host, pivot or stop are valid outcomes.
Pre-work
Before the first evaluative session, commit:
Method
Problem evidence
Task/consent prototype
Mock-backed and non-production:
Comprehension measures
Without showing the consent text again, verify the participant can state:
Architecture fit
Feed evidence into #13/ADR-003:
Exit criteria
Gate
No runtime feature expansion and no closure of this issue without linked evidence. The chosen task must be compatible with an endpoint/deployment model that can pass APP-01 security, privacy, accessibility and operational gates.