No sessions have been conducted and no participants, quotes or comprehension findings are claimed. The first six candidate treatments are still undergoing implementation review. This protocol will also cover the remaining pilot and material new treatments; it does not declare the full pilot accepted.
Scope and recruitment#
Seek willing readers who resemble the guide audiences: a product/design reader new to these implementation details and an engineer with the prerequisite architecture context. Ask which reading setup they actually use and whether any accessibility or input accommodation would help. Do not collect demographic or health information to fill a template. Record recruitment and sampling limits; a small convenience sample is qualitative evidence, not a population estimate.
The task owner may participate or identify willing reviewers. Obtain agreement before recording observations; do not contact anyone or publish identifying information merely because this protocol exists. Use participant codes. Record only task-relevant observations and obtain permission before retaining quotes.
Freeze the stimulus before each session#
Record guide and slide IDs, source pin, authored slide hash, HTML/style hashes, render or PDF hash, narration hash if used, and the exact ordered pilot slides. Record viewport, effective embedded size, expanded/collapsed navigation, normal browser zoom, device pixel ratio, fonts, input mode and any enlargement. A revised slide is a different stimulus. Never attach old reader findings to a new artifact solely because the ID stayed the same.
Let the reader look naturally; do not impose a speed score. First show slides without coaching or narration. Then allow the intended narrated/interactive experience, including Explain and sources, recording whether assistance or additional context changed their interpretation. Keep slide-only and supported reading outcomes separate.
Neutral tasks and source-backed interpretations#
| Stimulus | Prompt, asked without showing the expected interpretation | Interpretation to assess afterward |
|---|---|---|
v1-web-pwa-boundary |
What is this slide saying? How do its parts relate? | Route, release, capability and outcome need distinct evidence. They are not a funnel or a guarantee that a preceding state proves the next. |
| Same slide | A local queue accepted an action. What can you conclude, and what remains unknown? | Queue acceptance alone does not establish an acknowledged owning-server result or completed member journey. |
member-day |
Describe one visit that this picture allows. | The reader may select any eligible activity, not compulsory Tara → Arete → Nyx → Nisaba steps. No activity is required. |
| Same slide | The person leaves immediately. What does the picture say about a later visit? | Arrival can recur without completing an activity. Library/history and permitted memory name potential continuity, not automatic capture or fabricated saved state. |
| Both in guide order | How does the second explanation affect what you would claim about the member experience? | An illustrative journey clarifies the intended experience but does not itself supply outcome evidence or prove consumer integration. |
Do not penalize equivalent phrasing. Note hesitation, unsupported arrows or mandatory-order inferences, missed consent/release qualifiers, unreadable labels, confusing emphasis, and trouble locating supporting detail. Ask “What led you to that interpretation?” before explaining. Avoid correcting a reader until the initial observation is recorded.
The runtime decision candidate adds these neutral tasks. Its exact version must be frozen before a willing-reader session; these are not observed results.
| Stimulus | Neutral prompt | Source-backed interpretation to assess afterward |
|---|---|---|
psyche-runtime-fallback |
A voice kill switch and a blocking policy are supplied together while the current mode is avatar. What does this resolver return, and what must the caller still decide? | The kill switch wins and produces a voice candidate. The caller must still apply independent protective constraints and capability; the diagram does not authorize weaker safety or prove a media switch. |
| Same slide | The first supplied signal is malformed. Trace what happens next. | Validation throws; it does not try a lower-priority signal. Invalid current-mode input throws even earlier, outside the depicted selection. |
| Same slide | No signal is supplied. What can you conclude from this result? | The resolver retains the current mode, not proof that every policy, provider or actual session is healthy. |
The quantitative/table candidate adds these version-bound tasks, not results:
| Stimulus | Neutral prompt | Source-backed interpretation to assess afterward |
|---|---|---|
psyche-latency-targets |
What does the ASR p99 entry tell you? | No p99 target is configured. It is not zero, a measured latency or evidence of meeting a target. |
| Same slide | What can these numbers establish about production performance? | They are exact configured millisecond targets for distinct intervals, not sampled production results. An empty sample's ok assessment does not demonstrate them. |
room-maturity, on mobile |
Read the last column for Nyx. What evidence is still needed? | The row still identifies Nyx after scrolling; its additional proof boundary is external feeds and provider synchronization. The table does not assert demonstrated integration. |
The ownership candidate adds these neutral tasks, not observed results:
| Stimulus | Neutral prompt | Source-backed interpretation to assess afterward |
|---|---|---|
architecture-read-handoffs |
Who starts the reads, and what must happen before the result returns? | The shared Tara adapter starts all three reads; all must fulfill before it passes their values to composition and returns the result. |
| Same slide | One read rejects while two remain pending. What does the caller receive? | The call rejects without a partial-failure envelope; the other started reads are not canceled by this method. |
| Overview and detail | What can this establish about a deployed BFF hop or an authenticated member? | The lanes identify code responsibilities. An injected API adapter or the package HTTP client can supply reads; neither the diagram nor a userId parameter proves deployed routing or authentication. |
The exact-source candidate adds these neutral tasks, not observed results:
| Stimulus | Neutral prompt | Source-backed interpretation to assess afterward |
|---|---|---|
pwa-install-update |
What makes this function return false? What has happened when it returns true? | No waiting worker means false and no message. True follows posting the request without a synchronous exception; it does not wait for activation. |
| Same slide | What can you conclude about the page or worker after true is returned? | The helper has no activation acknowledgement or reload; real worker/page effects need separate evidence. |
| Same slide with Explain available | Which parts are original code, which are explanation, and where would you check the complete function? | Source bands preserve the named pinned lines; adjacent annotations are interpretation. The signature and final brace are outside this excerpt and the full source is in Explain. |
The remaining pilot needs tasks for further branching/failure, other artifacts and the complete guide's narrative arc. Write each task and expected source-backed interpretation before its sessions, once the specific explanation and stimulus are ready. Do not substitute a successful two-slide task for this wider coverage.
Record, revise and retest#
For each task retain participant code, stimulus identity, viewing setup, unaided interpretation, observed difficulty, assistance, subsequent interpretation and a bounded result (understood, material misunderstanding, inaccessible or inconclusive). Separate direct observations from the author's interpretation and proposed fix. Aesthetic preference is useful but cannot override an incorrect inferred relationship.
Turn material findings into concrete defects. Recompose or clarify, rerun the affected automated/source/visual checks, then retest the changed explanation and a suitable transfer example. Record unresolved findings explicitly. Only after material factual, accessibility, optical and reader findings are resolved may a treatment enter the accepted benchmark collection.
Unchanged applications may inherit evidence about the treatment with an explicit scope and per-guide review. New semantics, meaningfully different density, audiences, fonts or viewing sizes can invalidate that inheritance. Automated tests, model inspection, normal-speed listening and real reader sessions are separate categories. A generated deck does not become human-tested by using a previously reviewed treatment.
Eve completion candidate · 8 September 2026#
These additional tasks apply to the final artifacts in
verification/eve-completion-2026-09-08/. They are prompts for a future willing
reader session, not observations or successful comprehension scores. Keep the
answer rubric hidden until the reader responds; record stimulus hash, viewport,
prior familiarity, exact response, hesitation, assistance and resulting
revision. The current user has already discussed several of these examples, so
their future responses cannot automatically count as unaided first exposure.
| Stimulus | Neutral reader task | Rubric, withheld during the task |
|---|---|---|
| Work board and recovery map | The item is in-review and the review is rejected. What can happen next, and what would you inspect? | Reopen-from-review returns ready; inspect preserved attempt/evidence and current eligibility. It does not imply shipped or verified. |
| Decision supersession | Which record replaces which, and what evidence would establish that replacement? | The accepted successor links back to the earlier superseded ADR; a proposed document alone cannot supersede it. |
| Studio dossier | What can you learn about the content from this response, and what requires another read? | Revision metadata and claim count are present; script/claim text and execution cannot be inferred. |
| Readiness evaluation | What would you tell the person asking whether to launch? | 75% counts passed/waived gates, but a blocking safety failure and missing release-manager signoff yield no_go; no release decision was persisted. |
| Final Enrollment account | What is established, and what is still missing? | Link the actual artifact, report and exact condition verdict; distinguish structural coverage, current member checks and publication/acceptance. |