Validation and evidence limits
Return Path uses original fictional inputs. Software verification checks the prototype's behavior; proposed human evaluation asks whether the experience serves people. Mo owns product and program direction; AI assisted implementation and verification. No research sessions, customer endorsement or commercial outcomes are claimed.
Observed local checks
Observed on October 3, 2026, with Node 24.14.1 and the locked dependencies:
| Check | Result | Scope |
|---|---|---|
| Locked installation | Pass | npm ci;143 audited packages |
| Lint | Pass | Source, tests and configuration |
| Strict types | Pass | Production code and both test layers |
| Domain/storage tests | 40 passed | Independent timing examples, consent/preference, quiet-hour boundaries, rolling week, out-of-order reservations, immutable snapshot replay, schema rejection, raw-byte/readability binding, getter/read/write failures, bounded reset Undo and history limits |
| Production browser suite | 33 passed | Primary and recovery journeys, all three channels, opt-out/completed-goal exclusion, repeated and overlapping runs, policy history, changed selections/raw bytes, reset/Undo, read recovery disclosure, caps, document links and security metadata |
| Focus/visual regression subset | 6 passed | Subset of the same 33 journeys, rerun after placing initial dialog focus on its explanation; not six additional unique flows |
| Production build | Pass | /Return-Path/ base; product documents copied into build; Node/runtime/environment and tracked-artifact guards passed |
| Dependency audit | 0 vulnerabilities | npm audit --audit-level=high; no high/critical findings |
| Browser errors and external requests | 0 | Every browser journey asserts no console/page errors and no requests outside the local app origin |
| Layout and accessibility | Pass | All invitation controls within 1280×633 for every profile; 320/390px invitation/policy/timeline/dialogs without horizontal overflow; Tab/Shift+Tab scope, Escape/cancel return, confirmation focus and live announcements |
| Visual browser review | Pass | Actual Chrome via agent-browser; inspected desktop, 320/390px full invitation, review dialog and accepted timeline screenshots |
The production suite uses Chromium against a built static preview on port 4192. It starts and stops its own server. The independent visual check used the installed Chrome executable with a unique session. Both the visual browser and manually started preview server were closed after review. Generated browser artifacts/screenshots are excluded from public source.
The first production replay found cancellation focus loss and a Tab boundary issue. Those were repaired and the full suite passed. Visual inspection then found that focusing a bottom action scrolled the long preview past its opening explanation; initial focus now starts on the dialog heading with the review at the top. The six relevant existing journeys were rerun successfully. A short-screen layout repair made all channel/start/preview controls visible at the required desktop size.
The primary hand oracle is Aster eligible / suppressed / eligible at Oct 3, Oct 4, Oct 10 16:00 UTC under the default policy, with Birch lacking Email consent throughout. Additional independently expected examples cover Fern’s exact 72h boundary, Clover’s exact 7d rolling-window boundary, local 01:00 quiet hours for Birch and opt-out/completed-goal exclusions. See Sample Contract.
S061–S067 have implemented local behavior and evidence as mapped in PRD. S068’s public release and personal discussion are governed by the gates below. Passing these software checks does not demonstrate human comprehension, usefulness, delivery or business impact.
Proposed human evaluation
Recruit consenting learners with interrupted self-directed goals. Ask them to explain the offered value, destination, suppression reason, consent and preferred channel in their own words. Observe whether re-entry feels like continuation rather than repeated setup. Compare timing and frequency expectations, including choosing to be left alone. These are proposed methods; no participant study or retention measurement has been performed.
Public release and review gates
The first public release was independently verified on October 3, 2026: source commit 5a2df60e5fd241ab9223702aa02a909fe1430ae1, successful verification and Pages release, clean local/public main agreement, and all 10 deployed files matching the local build and GitHub release artifact byte for byte. Pages was enabled before the first push. All 33 browser journeys also passed against the public URL. Independent source review and a separate public replay checked the initial 2-eligible/4-suppressed timeline, repeated-attempt suppression, reset and bounded Undo. The public profile now includes the demo and product reviewer route.
Later documentation revisions retain the same application behavior and are verified by their own publication workflow. Personal product discussion with Mo remains a separate review step and cannot be inferred from software checks. No human research has been performed.