Methods and evaluation
How these practice cases are built, scored, and would be evaluated
SA HealthPath is an independent, formative practice exercise using fictional residents. It is not a validated clinical triage instrument, not a competency measure, and not a certification. Federal public-health frameworks are used here as public references only; citing them is not approval or endorsement by the agencies or organizations that publish them.
What this is, and what it is not
This is a browser-based teaching exercise. A learner reads a fictional case, chooses three priority questions, marks the risks the case actually supports, assigns an urgency, plans actions, practices a handoff, and then reads feedback against that case's answer key.
- Formative practice with feedback — not an assessment of anyone's professional readiness.
- Fictional residents. No real client or patient information should ever be entered.
- Not a diagnostic tool, not clinical decision support, not an emergency-response system, and not continuing-education credit of any kind.
- Feedback describes agreement with one case's answer key, written by the author from public sources.
Public-health practice crosswalk
The case design is crosswalked to selected CDC Essential Public Health Services. The EPHS describe public health activities that communities should undertake and place equity at the center of public-health practice. These entries are paraphrased as a design reference only; this project is not an assessment of a workforce, agency, program, or curriculum, and CDC has not reviewed, approved, or endorsed it.
EPHS 1 — Assess community needs, assets, and health-influencing factors
Cases name barriers that come from systems — bus coverage, rental conditions, clinic hours — rather than attributing outcomes to a resident's character, culture, or group.
EPHS 3 — Communicate clearly about health and the factors that influence it
Each case asks what the resident says matters to them and which existing supports they already use.
EPHS 3 — Use respectful, understandable communication
Preferred language and communication preferences are confirmed with the resident, never inferred.
EPHS 4 — Strengthen partnerships with communities and organizations
Every debrief asks for one upstream response that would affect more people than this household.
EPHS 9 — Evaluate and improve services and public-health functions
The rubric and the measures below separate process (did practice happen), learning (did answers change), and health outcomes (unmeasured here).
EPHS 5 and 7 — Plan policies and systems that support equitable access to services
Cases require cross-sector coordination — transit, food, housing, legal aid, health system — and ask who benefits and who is left out.
EPHS 7 — Link people to needed services while naming access limits
Consent before a referral, qualified language-access wording, and teach-back are practiced as explicit selectable steps.
EPHS 8 — Stay inside role boundaries while supporting a skilled navigation workforce
Practice includes coordinating across services and staying inside a navigator's role.
EPHS 10 — Build infrastructure that supports public-health work
The relationship diagram below is the systems view of a single case's reinforcing constraints.
Social-determinant domains, and the line between a need and a determinant
Cases map to the five social-determinant domains described by Healthy People. An immediate social need is what one household lacks right now — no fare for the bus this week. A structural determinant is the population-level condition that produces those needs repeatedly — where routes run and how often.
| Domain | Population-level example |
|---|---|
| Economic stability | The share of households in a ZIP code spending more than half of income on rent, which shapes how much is left for medication or a bus fare. |
| Education access and quality | Availability of summer learning and adult literacy programs, which affects how written health instructions land. |
| Health care access and quality | Distance and travel time to the nearest clinic with evening hours, and whether interpretation is routinely offered. |
| Neighborhood and built environment | Tree canopy, air conditioning, housing conditions, and sidewalk or transit coverage across neighborhoods. |
| Social and community context | Whether people have someone who can check on them during a heat advisory or accompany them to an appointment. |
A case only establishes the facts it actually states. Immigration status, identity, and household circumstances are never inferred from a ZIP code, a name, or a preferred language. Where a case states facts in a domain, the mapping is presented as established by the case. Where a domain is only a plausible pathway the case does not establish, it is shown in the case material labeled as a possible mechanism, not established — so the difference between a stated fact and an inference stays visible instead of the domain being dropped or asserted.
Healthy People — Social Determinants of Health
See the full mapping: which answers sit in which domain, and the points they carry
How practice is scored
This exercise uses the following weights and rules. Current rubric version in source: 2026-09-12.v2.
Priority questions — 25 points
Full credit when all three selected questions are the ones marked essential in this case's answer key. Useful questions earn partial credit; low-yield questions earn none.
Supported risk identification — 25 points
Credit for each risk the case actually supports. Selecting a risk the case does not establish subtracts 3 points, because recording assumptions as findings causes real harm.
Urgency judgment — 20 points
Full credit for the answer-key urgency. Choosing a lower urgency than the key loses 8 points per level, because under-response is the more dangerous error. Choosing higher loses 5 per level.
Action and referral fit — 20 points
Credit for each essential action. Supportive actions are neither rewarded nor penalized. Actions marked not indicated subtract 3 points each.
Safety and access items — 10 points
Credit for the specific safety, communication, consent, and language-access items listed for this case. Every case has at least one; there is no free credit for a case with none.
Exact point rules
- Priority questions: an essential question counts 1, a useful question counts 0.5, a low-yield question counts 0; the total is divided by three and multiplied by 25.
- Supported risks: the share of supported risks selected, multiplied by 25, minus 3 points for each risk the case does not establish.
- Urgency: 20 points for the key answer; 8 points lost per level of under-response and 5 per level of over-response, because under-response is treated as the more dangerous error.
- Actions: the share of essential actions selected, multiplied by 20, minus 3 points for each action marked not indicated. Supportive actions neither add nor subtract.
- Safety and access: the share of that case's safety, consent, communication, and language-access items selected, multiplied by 10.
- Every category is rounded to a whole number and clamped to its own range, so the total is always bounded between 0 and 100. Repeated selections are collapsed before scoring and cannot inflate a category.
Safety and access items are also questions, risks, or actions, so they can be credited twice. That double weighting is a deliberate teaching choice, not a statistical property.
What the feedback flags separately
- Critical omissions: an item this case marks as essential to safety that was not selected.
- Unsafe selected actions: choices outside a navigator's role or contrary to the cited guidance.
- Under-triage: a lower urgency than the key.
- Severe over-triage: a much higher urgency, shown as a separate warning rather than a safety failure.
A clear training-item status means only that the items specified for this exercise were selected. It is never clinical clearance, and it never means a real situation is safe.
Choosing three questions is a limited prioritization exercise, not comprehensive screening. The reflection and handoff steps are unscored, and the action checkboxes do not assess real-world sequencing or timing. An attempt with invalid input is left unscored rather than guessed at.
Conceptual logic model
Each stage is only a hypothesis about the next. Nothing to the right of practice has been measured here.
Inputs
Desk-reviewed fictional cases and public resources; independent review still pending
Activities
Practice attempts with itemized feedback
Short-term
Observable learning within the exercise
Hypothesized
Retained learning on unseen cases
Long-term
Navigation and health outcomes — needs separate evaluation
Systems relationship diagram
A conceptual, hypothesized loop that varies by context. It is not empirical proof of local causation, and no numerical effect is claimed.
Proposed evaluation
These six phases follow the CDC program evaluation framework, mapped to this application. Collaboration with the people affected, fairness in how findings are produced and reported, and actually using the findings run through every phase rather than sitting in one step. This is a proposal only: no pilot has been run, no consented pilot data exist, and no independent clinical review has taken place.
1. Understand the context
Assess local needs and assets with intended users, and co-design case content with community partners and frontline navigators.
2. Describe the program
State inputs, activities, and intended outcomes explicitly — the conceptual logic model above is that description.
3. Focus the evaluation design
Set the questions in advance, choose alternate cases and follow-up points, and justify any comparison condition and sample size before collecting data.
4. Gather credible evidence
Clinical and crisis-protocol review, language-access review, usability testing including keyboard and screen-reader use, then consented pilot data.
5. Interpret and support conclusions
Report uncertainty, attrition between practice and follow-up, and differences between groups without over-reading small numbers.
6. Act on the findings
Revise cases and scoring, communicate what was learned including negative results, and re-evaluate after changes.
Proposed measures
Each measure names its numerator, denominator, time frame, and the thing it cannot tell you. None of these are being collected today.
| Measure | Numerator | Denominator | Time frame | Limitation |
|---|---|---|---|---|
| Case completion | Practice attempts submitted with a complete, valid answer set | Practice attempts started | Per training period | Completion is a process measure. It says nothing about learning or about resident outcomes. |
| Essential safety-action recognition | Valid attempts in which every essential safety action for that case was selected and no action flagged as unsafe or contraindicated was selected | Valid attempts on cases that contain an essential safety action | Per training period | Excluding unsafe selections stops “select everything” from counting as recognition. Even so, recognizing an item in a written case is not the same as acting on it under pressure with a real person. |
| Retained learning on an alternate case | Learners meeting the prespecified threshold on a held-out case not seen during practice | Learners who completed practice and attempted the alternate case | At a follow-up interval fixed before data collection begins | A single alternate case is a narrow sample of judgment, and attrition between practice and follow-up biases the result. |
| Real-world referral connection (confirmed receipt) | Consented referrals with confirmed receipt of the service by the resident | Consented referrals made | Within a follow-up window fixed in advance | A service acknowledging a referral is not closure. Unresolved barriers with a named follow-up person and date are a separate tracked status and are never counted as successful connections. |
| Voluntary equity disaggregation (completion rate by preferred-language group) | Consenting learners who completed an assigned case, within one voluntarily reported preferred-language group | Consenting learners in that same preferred-language group who started that case during the training period | Per training period, reported only when both counts meet a minimum fixed in advance | Rates are compared between groups; a minimum count and small-cell suppression are prespecified, and groups below it are left unreported rather than estimated. These are learner counts and support no inference about any population. |
This demonstration does not ask for resident data. Practice attempts are saved in the browser on the device being used; result links contain selected answers and may appear in browser history or hosting logs. The optional external AI review is suspended. That local record is not population analytics and cannot support any of the measures above.
Governance and review status
- Public-source desk review
- September 12, 2026. Desk review of publicly available web pages on 2026-09-12. Not verified by telephone or with providers. Hours, coverage areas, eligibility rules, and program availability change — confirm directly before relying on any listing.
- Targeted official-source check
- September 14, 2026. Targeted desk check of two official pages on 2026-09-14: the CDC NIOSH heat-related illnesses page (heat stroke first aid — call 911, stay with the person, move them to a cooler area, and cool them promptly while emergency medical services are on the way) and the VIA Link page (a ride service operating within designated zones, booked in advance rather than a fixed route).
- AI-assisted internal review
- September 14, 2026. Every case was walked through with an AI assistant using twelve intended-user perspectives and six subject-matter lenses, and the cited pages were re-read. These are analytical lenses, not participants: no person was recruited, observed, or interviewed, no expert accepted or endorsed this content, and the walkthrough is not validation. A second source check on 2026-09-14 re-read the CDC heat pages, CDC NIOSH heat-related illnesses, NHLBI asthma attacks, the 988 Lifeline pages, CDC Hear Her (including its urgent maternal warning signs handout), the HHS National CLAS Standards, and the VIA Link page. Three things were corrected as a result: CDC Hear Her says to get medical care immediately and does not itself direct a 911 call; the CLAS Standards require language assistance at no cost and discourage untrained interpreters but contain no statement that language assistance must not delay emergency care, so that sequencing is described here as a design choice of this exercise; and the VIA Link page was re-read, confirming designated zones and advance booking alongside the stop-to-stop wording recorded on 12 September. The San Antonio Food Bank page could not be opened during this check, so its listing still rests on the September 12, 2026 desk review. Reading published pages only: no provider was contacted and nothing here is clinically validated. The independent human reviews and the pilot described below remain incomplete.
- Narrow correction pass
- 2026-09-21. On 2026-09-21 the heat-without-symptoms case was corrected so that it no longer presents an electric fan as cooling. CDC's heat page says to use a fan only when the indoor temperature is below 90°F, because at higher indoor temperatures a fan can increase body temperature. The indoor temperature in that fictional case is unknown, so a fan cannot be counted as cooling there. Provenance: the CDC page (dated July 20, 2026) was read on September 21, 2026 by the supervising reviewer, who supplied that wording and the page date; a direct request from this project environment was refused by automated access controls, so the wording here is quoted from that supervised reading rather than from a fetch made here. The symptom-based urgent response in both heat cases is unchanged, and nothing in either case advises on medicines. The same pass made every fact that earns points in a case visible in the case brief, so no learner can lose points for evidence that only an optional question would have revealed. Version history for this project lives in this Lovable project alone; there is no external code repository for it.
- Known limits of the exercise format
- Three design choices make the exercise scoreable and do not reflect practice. Each case asks for exactly three questions and exactly one urgency level, while real screening is iterative and often ends in uncertainty. Each case has one answer key, so the debrief also lists defensible alternatives rather than presenting the key as the only correct approach. Safety information — including the emergency response a case is built around — is stated on the case page before you answer, because withholding safety guidance to make a training item harder is not acceptable; the detailed learning objectives are held back until after submission instead.
- Paired cases
- The ten cases sit in five contrasting pairs. Within a pair the subject area is held constant and one decision changes — for example, heat exposure with confusion reported against heat exposure with no symptoms and a safe cooler destination. The intent is that learners practice recognizing when a situation is an emergency and when it is not, rather than learning to answer from the topic label.
- Author and content owner
- Michael Richard Martin, MPH.
- Proposed review schedule
- Monthly check of provider links, quarterly educational review of case content and rubric, and a safety review triggered by any change to a high-risk case or to cited emergency guidance. This is a proposed schedule, not evidence of completed expert governance.
- Reviews not yet completed
- Independent clinical review, crisis-protocol review, community review, language-access review, and accessibility testing with assistive-technology users have all not been completed.
- Standards used as targets
- HHS National CLAS StandardsSection508.gov accessibility laws and policiesThese are the targets the project aims at. No conformance, compliance, or certification is claimed.
Every source cited across the project is listed on the resources page.