UXR Intake and Evidence Collection: Working Concepts

Document status. This document establishes working principles for HXR intake. HXR now has an operational internal intake workbench and governed processing records, but this does not mean a general public intake service, finalized evidence methodology, or final technical architecture has launched. Channel availability and public access remain separate development questions.

Why intake matters. A reporting system can reproduce the harm it seeks to measure. Fax-only submission, broken web forms, voicemail-only fallback, repeated restarts, inaccessible categories, and channels that cannot accept ordinary evidence transfer the organization's operational failure back onto the person seeking help. HXR intake must itself meet HXR Blue Zone standards.

Primary pathway: empathetic conversational AI. The primary concept is a narrowly trained intake AI that accepts voice or text, lets a person describe what happened in their own words, and then asks focused questions to develop a complete report. It should support multiple languages, accessible interaction, plain language, save-and-resume continuity, and a pace appropriate to the person rather than forcing everyone through the same rigid form.

The AI's role. The intake AI should help organize and clarify, not decide whether the person deserves belief. It must avoid leading questions, distinguish the person's words from its own inference, preserve the original account, identify genuine uncertainty or conflict without manufacturing suspicion, ask for correction, and let the person review and approve the structured summary before submission.

Record provenance without suspicion. Intake should record the source of a proposition plainly. “The contributor reports…” is sufficient when the proposition comes from the contributor. When a screenshot, forwarded email, transcript, recording, or other artifact supports it, the intake record should add that support affirmatively. The system should not automatically append phrases such as “unverified,” “awaiting verification,” or equivalent credibility-reducing language merely because an artifact has not been provided. Provenance should help readers understand the evidence while preserving contributor dignity.

Alternative intake channels. HXR should consider an adaptive web form, conventional category-based forms, email forwarding, secure file and share-link upload, human-assisted telephone or video intake, live chat, submissions through a chosen caregiver or advocate, accessible paper or mail pathways, and future application programming interfaces or structured bulk submission for trusted public-interest organizations.

Contribution gradient. Different contributions can perform different evidentiary jobs, from a short account through testimony, artifacts, expertise, organization response, and reform verification. This is an internal evidence-and-motivation map, not a requirement to show the contributor a separate button for every contribution type.

Current Phase 1 boundary. The minimal Evaluation pilot uses one Add Your Experience doorway. The contributor describes what happened in ordinary language; the Intake AI or human reviewer classifies the material behind the scenes and asks only the smallest follow-up needed for a responsible candidate match, privacy/permitted-use state, or safe next action. Phase 1 reuses the existing Intake Session path and does not require follow subscriptions, automatic matching, a contributor portal, organization-specific portals, or separate contribution-type workflows. Positive outcomes, materially different experiences, corrections, workarounds, and counterexamples must be accepted through the same doorway.

Low-effort signals are not votes. “This happened to me too” may create a candidate recurrence lead. “My experience was different” may create a counterexample. Neither becomes truth, prevalence, severity, or score through count alone. Intake should seek the minimum additional context needed to determine whether experiences are materially comparable without turning a ten-second contribution into a compulsory full interview.

Phase 1 contribution summary. After pilot intake, HXR should return a plain-language summary from existing intake state: what was recorded, the candidate Evaluation match, privacy and permitted-use state, what the contribution may add, what it does not establish, and the next review step. Useful existing resolution knowledge may be returned when available. A separate persistent Contribution Impact Receipt, contributor portal, automatic notifications, and follow lifecycle remain future product hypotheses and should be implemented only if the simpler pilot summary proves insufficient.

Motivation without manipulation. Intake may show people how their contribution can help themselves, future users, institutional accuracy, public accountability, good-practice recognition, reform, and verification. It must not reward suffering, create harm leaderboards, use guilt or streak pressure, inflate the meaning of a one-tap signal, or imply that a person must disclose more than they choose in order to support reform.

Response-state continuity. Contributors should be able to learn whether an involved organization acknowledged the issue, corrected a fact, explained a constraint, supplied a remedy, proposed reform, demonstrated reform, declined to respond, or did not respond after a documented opportunity. Nonresponse is worth recording under defined contact conditions, but silence remains ambiguous and must not be represented as automatic proof of guilt, agreement, indifference, motive, or inability to reform.

Cognitive-load threshold and immediate exit. Intake must watch for signs that the process itself is exhausting, confusing, repetitive, distressing, or unsafe. A person should be able to switch immediately to a simple structured form, pause, or request human assistance without re-entering information, re-uploading evidence, defending the request, or starting over. The system should carry forward the existing narrative, answers, attachments, consent choices, and unresolved questions.

Channel continuity. The long-term system should let a person save progress, move between supported channels without starting over, add evidence later, export their own submission, review what HXR recorded, correct mistakes, understand what happens next, and track applicable status. Phase 1 does not promise a public account portal, automatic case identifier, cross-channel transfer, or notification service. The immediate requirement is that the private Intake Session preserve completed work and allow correction or supplementation without forcing the contributor to restart. A failed offered channel must not be represented as functioning.

Evidence requests. The process should request verifiable support when reasonably available and useful, including forwarded emails, screenshots, photographs, recordings, saved voicemails, chat logs, receipts, notices, system exports, authenticated share links, integrity tokens, timestamps, or direct records supplied by the responsible organization. The request for supporting material should communicate that the material adds support; it should not imply that the contributor's account has no value without it.

Corrections and supplemental evidence. Contributors should be able to refine dates, wording, amounts, sequence, interpretation, and other details as they remember more or obtain records. Intake should preserve the original account privately while carrying the best-supported current facts forward into the working case. Supplying better evidence should be treated as improving the record, not as creating a credibility penalty for the contributor.

Public correction without contributor callout. When later evidence establishes a better-supported fact, the public case should normally be rewritten around that fact and the superseded public detail removed. HXR does not need to tell public readers that a contributor originally remembered ordinary wording, timing, or another detail differently unless the discrepancy itself materially affects the case. The internal working record should preserve the provenance and correction history needed for auditability.

Evidence can change the case in either direction. New evidence may strengthen a proposition, weaken it, narrow it, contradict it, change responsibility, or require a different case boundary. Intake and case-management systems should support those changes without making correction feel punitive. The objective is a more accurate case, not a performance evaluation of the contributor's memory.

Evidence remains under human control. People must understand what they are providing, how it may be used, who may receive it, what will be public, what will be redacted, how long it will be retained, and how to withdraw or restrict material where possible. HXR should collect only what serves a defined purpose.

Reports without documentary proof. A report remains evidence even when the person lacks an artifact. HXR may seek corroboration, request records from another source, compare independently submitted accounts, examine public information, or preserve the report for future matching and investigation. The absence of documentary proof should not trigger language that treats the contributor as presumptively unreliable. Accumulated independent evidence can support pattern detection, but volume does not automatically establish every proposition.

Identity confidence and claim confidence. HXR may need ways to estimate whether submissions originate from distinct people rather than bots, paid campaigns, or duplicate accounts. Those controls must remain separate from evaluating the claim and must not require public identity, forced surveillance, or unnecessary personal data. Protected and anonymous reporting must remain possible where risk justifies it.

Human assistance and escalation. People must have a route out of the AI pathway when disability, distress, complexity, safety, language, technical failure, or personal preference makes human assistance necessary. Immediate danger, urgent medical needs, threats of violence, and other time-sensitive conditions require clear routing beyond ordinary research intake.

Continuity after submission. Intake should not end with a receipt number and silence. The future system should preserve the contributor's connection to the report, support correction and supplementation, provide understandable status information, and connect the person to relevant resolution knowledge where appropriate. The separate development document on contributor continuity defines this intended layer without fixing its final workflow.

Systems not yet defined. HXR is intentionally not yet fixing evidence weights, confidence scales, identity-assurance methods, AI evaluation tests, human-review thresholds, retention periods, moderation rules, integration standards, status taxonomies, or scoring consequences. Those decisions depend on the evidence, severity, participation-integrity, privacy, governance, and domain frameworks now under development.

Next development stage. The next stage is to continue exercising the internal intake workbench and regression suite against representative scenarios, including stressful narratives, incomplete evidence, conflicting records, accessibility needs, anonymous reporting, channel failure, cognitive overload, requests to switch channels, later artifacts that refine earlier remembered details, and attempts to manipulate the system. Testing should measure both information quality and the human burden created by intake itself, including whether contributors feel safe correcting the record, while public-service launch remains a separate readiness decision.

Accept First, Enrich Second

A coherent report that identifies a recognizable human experience is enough to begin preserving intake and candidate case meaning. Follow-up questions should normally be framed as optional ways to strengthen, connect, clarify, protect, or route the record, not as conditions for belief or acceptance.

When HXR recognizes a likely match to an existing Incident family or Evaluation, the preferred posture is: “We have enough to understand and record this. If you have more, it may strengthen the case or help connect it to others.” Deeper interview mode remains appropriate when the contributor explicitly requests it or when a genuinely material boundary, safety, privacy, permitted-use, or target ambiguity cannot responsibly be resolved otherwise.

Source-Aware Intake

Intake may identify an existing HXR Source, create a new Source candidate, or preserve a contributor-supplied source. A Source identifies where information came from. Evidence Items record the distinct observations the source supports. Proposition Source Resolution remains the intake-stage provenance mechanism. Supported Evidence Contributions become the durable final graph edges explaining why an Evidence Item matters to a particular target.

The intake agent must not copy one source URL and summary into every related Case or Evaluation. It should create or reuse one canonical Source, create the atomic Evidence Item or Items warranted by that Source, and create separate supported Contribution edges for each target relationship.

Bounded External Source Discovery

A sufficiently coherent report is accepted before external corroboration. When public research can materially strengthen, narrow, challenge, contextualize, or connect the resulting Incident or Evaluation, HXR should then perform a bounded source-discovery pass rather than assigning that work to the contributor.

The research question must stay tied to the actual experience: the product or service, interface, action sequence, displayed language, version/date context, operating environment, and requested human outcome. A narrow interface failure should not automatically become a study of every technically adjacent operation, every product in the category, or the entire underlying platform. Broader mechanisms may deserve related Evaluations, but they should not displace the issue the contributor actually brought.

Useful research lenses include:

• exact error messages, control labels, action sequences, and ordinary-language descriptions of the same behavior;
• current first-party documentation describing the feature, setting, expected behavior, limitation, or official procedure;
• official bug trackers, feature-request systems, support communities, staff responses, known-issue pages, release notes, and fix or regression history;
• independent professional communities, forums, Q&A sites, issue repositories, and long-lived discussions containing similar reports, materially different experiences, workaround outcomes, or counterexamples;
• operating-system, API, standards, or protocol documentation when a claimed technical restriction or feasibility question actually matters; and
• comparison evidence only when it directly tests a proposed acceptance condition or claimed constraint without silently broadening the Evaluation.

Each retained source should show its URL, title, platform or publisher, source class, author or account role when supported, date, review date, applicable product/version context, reported or documented behavior, what it supports, what it does not support, independence or dependency, and any organization response, workaround, fix, or regression state.

A company-hosted user post remains user testimony. A staff or product-team response on that page is separately attributable organization-authored evidence. Official documentation establishes what an organization represents about its product, not what happened in one contributor's occurrence. Similar accounts may corroborate recurrence, but likes, votes, views, “me too” counts, and raw report totals do not establish population prevalence, severity, causation, intent, or one universal technical explanation.

Search snippets are discovery aids, not final evidence. HXR should inspect and preserve the underlying source, record contradictions and inaccessible or changed pages, and avoid counting a source ledger or synthesis as an additional independent report. Durable research follows the normalized path: Source → Source Snapshot when needed → atomic Evidence Item → evidence-specific Evidence Contribution.

Public Discovery as Intake Access

Affected people rarely begin by wanting to inspect HXR as a project. They begin with a current problem, question, benefit, workaround, organization response, or search phrase. Because canonical Notion records are private, the Shopify projection is the public doorway through which many future contributors will first recognize that HXR understands their experience.

Public Source, Evidence, Incident, Evaluation, Dataset, and Reform pages should therefore use the ordinary language, examples, limitations, and relationship context that help affected people find the correct issue and decide that contributing is worth their effort. This is not permission to manufacture thin search pages. The public page must answer the person’s need and show why HXR’s structured synthesis adds value beyond the individual sources.

What Others Have Tried

When a recurring issue has fragmented community troubleshooting, HXR may collect suggested actions and source-reported outcomes as structured evidence. The intake agent may invite a contributor to mark whether an action Worked, Did Not Work, Worked Temporarily, Helped Partially, Was Unavailable, Was Unacceptable, or Was Not Tried.

These entries support recognition and aggregate experiential research. They are not HXR recommendations unless separately evaluated and approved. “Did not work for me” is an outcome contribution, not a debunking contest or a credibility judgment.

Community Return

When HXR later develops a material synthesis, obtains an organization response, verifies a workaround, documents reform, corrects an error, or observes regression, it may return that useful information to the originating community through a governed Source Engagement.

Community return must be human-reviewed, on-topic, affiliation-disclosed, platform-compliant, non-duplicative, and useful even to a reader who does not follow the HXR link. It must not become automated promotion or backlink harvesting. Responses and corrections generated by the community re-enter the normal intake and evidence process.

Evidence-informed search-language preservation

External source discovery has a second legitimate purpose beyond corroboration and technical context: learning the vocabulary people naturally use when they encounter the same problem. For new researchable, non-urgent case families, preserve the Intake Session first and then perform this discovery before finalizing durable Incident/Evaluation naming and visitor/search-facing language whenever practical. The sequencing improves terminology and discoverability without making corroboration a condition for preserving testimony.

When affected people use materially accurate recurring language, preserve exact error phrases, ordinary action descriptions, informal problem labels, question forms, and the words they use to seek help. Keep source/provenance attached to that language. When an Evaluation becomes public, use the strongest accurate forms of that vocabulary to inform its Public / Search Title, Public Subtitle, opening explanation, headings, and body copy so a person searching for the problem in their own words has a realistic path to finding the Evaluation.

This is not an SEO override. HXR should not keyword-stuff, manufacture search variants, repeat phrases unnaturally, adopt unsupported claims because they rank well, or broaden an Evaluation merely to capture adjacent searches. Public usefulness and evidentiary accuracy remain controlling. Discoverability should arise from accurately using the language of the problem and the people who experience it.

Search-language findings can also expose boundary differences. If people use the same visible message for materially different mechanisms or failure states, preserve that ambiguity rather than merging the Evaluations for search convenience.

Verified transient binary-evidence relay

The August 18, 2026 binary capability test established a controlled fallback when direct ChatGPT-to-Notion binary transfer is unavailable.

For evidence whose privacy and permitted-use state allows the temporary transport risk, the workbench may stage the current-task original in Shopify Files, resolve Shopify originalSource, verify the original byte length against the supplied source, import that original-source URL into the intended Notion Evidence Item, read back the durable Notion-hosted artifact, delete the temporary Shopify asset, and verify that the Shopify node no longer resolves. Processed Shopify CDN renditions are not acceptable substitutes for the canonical original because the test demonstrated byte transformation.

The Shopify asset is transport only. It must not be attached to a product, Article, public HXR projection, or other storefront content. Direct Notion upload remains preferred. Highly sensitive evidence must not automatically use a relay that creates a temporarily addressable staging URL. If the imported binary is durably present in the Notion Evidence Item but the connector still cannot populate the structured Evidence Files property, record that structured-binding limitation separately rather than reporting that the binary itself was not preserved.