Evidence Register

safety_danger_now (v0.2.0 · sensitivity high)

Question: Are you in danger right now, or is someone hurting or threatening you?

Evidence: Emergency Severity Index triage concept: danger routes to live help before any assessment

Reason kept: Routes emergencies out of assessment and into response; workflow only, never allocation.

Alternatives considered: Fold into the fleeing-violence question (rejected): Danger-now and leaving-a-situation are different facts with different responses. · Merge with the fleeing-violence question (rejected): Rejected: danger-at-this-moment is a 911 triage fact; the violence-situation question is a housing-category and routing fact; they have different responses and different consequences. · Second practitioner report that this feels too similar to the violence-situation question, suggesting keeping only one (flag #109, 2026-07-28) (undefined): Kept separate, same reasoning as before: danger-at-this-moment is a 911 triage fact, the violence situation is a category and routing fact, and they have different responses and consequences. Two independent reports now make this a named cognitive-testing item: testing will decide whether wording can differentiate them better. · No help text at all (v0.1.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION and JUDGMENT. The item that fires IA-R1, the one rule in the instrument that routes a person out of the assessment and into emergency response, was the only immediate-action item with nothing under it. Two misreads were reachable and both were plausible. "Right now" grammatically attaches to the first clause only, so a person who expects to be hurt when someone gets home tonight can read the whole question as not about them, which is an under-report on the one item where an under-report costs the most; and a person who does not want to name anyone can read the question as demanding a name. The help now anchors the period ("right now, today"), states that a threat about later today counts, and says no name and no detail is needed. Settling the boundary upward is the licensed direction here and only here: inside the safety screen an error upward produces an offer and never a denial, which is the carve-out already stated at NR-3 in the improvement register. No option, no gate and no rule changed; IA-R1 and IA-R3 read the same values they read before. Item version 0.1.0 to 0.2.0.

Known risks: None identified beyond general disclosure sensitivity.

What this answer changes: IA-R1, IA-R3 (immediate action).

Review status: pending external review · last updated 2026-08-28

safety_medical_now (v0.4.0 · sensitivity high)

Question: Do you have a medical emergency right now? This means something that needs care within minutes.

Evidence: ESI triage: urgent medical need outranks everything

Reason kept: A bleeding head wound must beat every housing question. Added after an internal review found the gap.

Alternatives considered: Rely on assessor judgment (rejected): Self-administered mode has no assessor. · "Do you need medical help right now for something serious?" (v0.1.0) (replaced): 2026-07-28: "serious" is a judgment call people make about themselves badly, and the question was catching chronic health needs. v0.2.0 asks for an emergency needing care within minutes, names concrete examples (breathing, chest pain, bleeding), and says plainly that a general health problem is not what is being asked. · Replace with questions about recent medical emergencies and severe conditions needing frequent care, since no one completes an assessment during an emergency (practitioner feedback, flag #111, 2026-07-28) (undefined): Kept, with the point recorded: recent emergencies are already captured by the emergency-room item and ongoing conditions by the disabling-condition and daily-help items. This screen's only job is live triage in assisted settings, where an assessor can be face to face with someone whose situation needs 911 before questions; removing it would remove the only path that routes an active emergency out of the assessment. · No third-person variant of the help text; the second-person text rendered under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.2.0 to 0.3.0. · “Do you have a medical emergency right now, something that needs care within minutes?”, with help text ending “...not a health problem you want care for in general” (v0.3.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). The stem scored 9.3 and the help text 8.3. The stem is now two short sentences, “Do you have a medical emergency right now? This means something that needs care within minutes.” (6.0), and the help text splits its second sentence in two (6.2). Nothing about the screen changes: the same two facts, in the same order, and the medical-emergency cliff that routes to 911 reads the same yes. Item version 0.3.0 to 0.4.0.

Known risks: None identified.

What this answer changes: IA-R1b, IA-R3 (immediate action).

Review status: pending external review · last updated 2026-08-28

safety_fleeing_violence (v0.5.0 · sensitivity high)

Question: Is someone hurting you, scaring you, or making you do things you do not want to do?

Evidence: HUD Trauma-Informed Care guidance p.7: behavior-based wording outperforms label-based; HUD Point-in-Time count gender-based violence fact sheet: covers domestic violence, sexual assault, stalking, trafficking without requiring self-identification as a victim

Reason kept: Category 4 gateway and safety routing.

Alternatives considered: "Are you fleeing domestic violence?" (rejected): Label-based; misses trafficking and people who do not name their experience violence. · "Are you leaving, or trying to leave, a situation where someone is hurting you..." (v0.2.0) (replaced): 2026-07-28: the old wording asked two things at once, whether harm is happening and whether the person is leaving, so a person still in the situation could truthfully answer no. v0.3.0 asks only about the harm; the leaving half moved to its own question, safety_leave_help, so neither answer hides the other. · "This includes violence at home, threats, or being forced or tricked into anything. You never have to give details." (v0.3.0) (replaced): 2026-08-10, practitioner flag 184 (Elisa): "violence at home" names a place most people answering this question do not have, so someone living in a car, a shelter or a friend's front room can read the question as not being about them. The relationships are what the item is actually asking about, so the help now names those (a partner, a family member, anyone you stay with) and says explicitly that where the person is staying does not matter. Tone, length and the closing reassurance are unchanged by design: this is the Category 4 gateway and a change of register here changes disclosure. Item version 0.3.0 to 0.4.0. · "It also includes threats, or being forced or tricked into anything. You never have to give details." (v0.4.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION (time anchor) and JUDGMENT (what counts), on the item that decides HUD Category 4 through HC-R3 and urgent same-day routing through IA-R2, and that gates the whole safety_routing module. Two boundaries were unsettled in opposite directions. The stem is present tense, and 24 CFR 578.3 Category 4 covers a person who "is fleeing, or is attempting to flee", so someone who left last week and is now in a shelter, with nobody currently hurting them, answers no truthfully and loses the category and the follow-up question about wanting help leaving. The help now says that having already left, with no safe way back, counts. In the other direction, "making you do things you do not want to do" is deliberately broad so that trafficking and coercion reach the item, and it can also be read onto ordinary program rules or a court order, which would produce a false Category 4 and a routing nobody asked for; the help now names those two as out of scope. Neither sentence adds an option or moves a gate. The structural half of the first finding, that the stem itself is present tense and safety_leave_help sits behind a yes, is filed as IR-31 and deliberately not built here. Item version 0.4.0 to 0.5.0.

Known risks: Disclosure in unsafe company; mitigated by ask-privately hint, skippability, quick exit.

What this answer changes: IA-R2, IA-R3 (immediate action); HC-R3 (HUD homeless category).

Review status: pending external review · last updated 2026-08-28

natural_support (v0.4.0 · sensitivity low)

Question: Is there someone you could safely stay with tonight, if you could get there?

Evidence: Diversion/problem-solving practice: a safe support tonight changes the conversation

Reason kept: Diversion discussion trigger; isolation as moderate support signal. From instrument 0.15.0-pilot the question is asked directly after the safety questions (HUD Core Elements III.C puts the diversion conversation before housing history), and a yes opens an optional stop-here interstitial: the person can try the option tonight without finishing, and stopping affects neither their place in line nor their options later.

Alternatives considered: "Is there anyone you could safely stay with tonight if the right help were available?" (v0.1.0) (replaced): 2026-07-28: "if the right help were available" was abstract, and people answered no because they were thinking about not having a ride. v0.2.0 names the obstacle concretely ("if you could get there") and the help text says the assessor can work out what it would take. · One help text in every mode, opening "The assessor can help figure out what it would take" (v0.2.0) (replaced): 2026-08-10, audit prompted by practitioner flag 199 (Elisa): third and last help string naming an assessor who does not exist in self-administered mode. The assessor-administered and proxy text is unchanged. Item version 0.2.0 to 0.3.0. · "The assessor can help figure out what it would take, like a ride or a phone call." and its self-administered twin, with no statement of what "someone" means (v0.3.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). JUDGMENT. This is the diversion gate. A no is one of the four conditions in SN-R9 and raises the support planning band to moderate; a yes raises the diversion flag and opens the diversion offer. The question asks whether there is "someone you could safely stay with tonight", and a person who reads "someone" as including a shelter or a program answers yes, loses the SN-R9 raise, and is offered a diversion conversation built on a person who does not exist. The help now opens by saying it means a person, not a shelter or a program. This is the protective direction: it prevents a wrong yes rather than encouraging one, and no sentence was added that pushes anyone toward either answer. IR-29 already records the missing third answer ("maybe, I would have to ask") and is unaffected. Item version 0.3.0 to 0.4.0.

Known risks: "Safe" is subjective; wording asks about tonight, concretely.

What this answer changes: SN-R9, SN-R12 (support planning level); ID-R6, ID-R7 (options to discuss); SS-R4 (supports worth offering); plus the diversion_candidate flag.

Review status: pending external review · last updated 2026-08-28

age_category (v0.6.0 · sensitivity low)

Question: About you: which age group are you in?

Evidence: Accelerated-aging literature in homeless cohorts (geriatric conditions ~20 years early)

Reason kept: Age 60+ while homeless floors HIGH (SN-R6); youth pathway (18-24) for TH discussion; under 18 routes to youth coordinated entry and decides nothing.

Alternatives considered: Exact age (rejected): Categories suffice for every rule; exact age is more identifying. · Three bands: 18-24, 25-59, 60 or older (v0.1.0) (replaced): 2026-07-28, field practitioner feedback (flag #100): the 25-59 band hid every age distinction workers actually match on, and minors had nowhere to land. Practitioners asked for PIT-style reporting bands, for the local 55-plus and 62-plus funding cutoffs to be visible, and for an under-18 answer because school social workers assess 17-year-olds for YHDP. v0.2.0 splits the middle into decade bands and adds Under 18, 55 to 59, 60 to 61, and 62 or older. · Adopt the PIT age bands exactly (rejected): The PIT 55-64 band straddles the age-60 mortality threshold the support-need construct keys on (SN-R6). Adopting it exactly would have forced either the threshold to move without evidence or the band to be read imprecisely, so the bands split at 60 instead. · Key a rule on the 55 or 62 cutoff (rejected): Those cutoffs are local funding facts, not risk evidence. The finer split exists so a worker can see them at a glance; no HEAT rule reads 55 or 62, and adding one would import a community's funding rule into a determination. · Give the under-18 answer a determination effect (rejected): Age may never lower a determination. Under 18 surfaces as a routing flag and an offer only: youth coordinated entry pathways (YHDP, school district McKinney-Vento liaisons, youth access points) apply for minors, and this preview is designed for adults. A screening-out test proves the answer changes nothing except flags and offers. · Threshold 50 (deferred): Would floor a very large share HIGH, diluting the band; assigned to clinical review. · Capture date of birth (standard HMIS practice) (rejected): Rejected for the public tool because HEAT collects no identifying information in memory-only mode; the phase-3 HMIS-assisted mode confirms date of birth instead of asking. · No help text (v0.2.0) (replaced): 2026-08-10, practitioner flag 194 (Elisa): a parent answering an age question directly after the household questions can read it as asking about a child. One line makes whose age it is explicit. The separate half of the flag, that the item sits in the wrong place in the order, is a Reorder and went to the public register instead of being decided here. Item version 0.2.0 to 0.3.0. · Asking the age question after the two children questions (order through instrument 0.20.0-pilot) (replaced): 2026-08-11, public register item 12 (Elisa, and a second practitioner reviewer in the same round). The help text added in v0.3.0 said whose age it meant, and the position still said otherwise: a reader coming off two questions about children read the next question as being about a child, and a first-option walkthrough produced a results page describing a person under 18 with four children. The four questions about who the person is (age, adults in the household, children who will live with them, children elsewhere) now sit together directly after the safety questions and the diversion offer, age first, under a phase heading that says "Who this is about", and the stem names its own subject in both voices. Nothing about the logic moved: a byte-identical order-invariance proof over reversed and rotated instruments is in guards.test.ts. Item version 0.3.0 to 0.4.0. · “Other people in the household, including children, are counted in the household questions that follow, not here.” (v0.4.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). The help text scored 8.4, carried by “are counted in the household questions that follow”. It now says the same thing in the active voice: “The questions after this one count everyone else in the household, including children.” (7.7). The scope statement register item 12 asked for is unchanged. Item version 0.4.0 to 0.5.0. · "This asks about the age of the person this assessment is about. The questions after this one count everyone else in the household, including children." (v0.5.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). RESPONSE MAPPING at a band edge. The bands are exact, and two edges are load-bearing: under 18 routes to youth coordinated entry and closes the adult_context module, and SN-R6 raises the band for a person 60 or older who is currently homeless. A person a few weeks from a birthday, asked "which age group are you in", can reasonably give the group they are about to enter, which is the one systematic direction of error a band question has. The help now says the right answer is the age right now, not an age coming up soon. The bands themselves are untouched, which matters: NR-13 records why they split at 60 rather than following the PIT bands, and nothing here reopens that. Item version 0.5.0 to 0.6.0.

Known risks: None identified for the bands themselves. The real risk is a minor being assessed in an adult-designed preview instead of through a youth access point; the under-18 routing flag and the note on the worker card exist to catch exactly that.

What this answer changes: SN-R6, SN-R12 (support planning level); ID-R8 (options to discuss); SS-R20 (supports worth offering); XC-R7, XC-R8 (in-session consistency check); plus the minor, older_adult, youth_18_24 flags.

Review status: pending external review · last updated 2026-08-28

household_adults (v0.4.0 · sensitivity low)

Question: How many adults, including you, are in your household right now?

Evidence: Household composition drives program eligibility

Reason kept: Household sizing for options discussion.

Alternatives considered: No help text; range hint "Include yourself, so at least 1" only (v0.1.0) (replaced): 2026-08-10, practitioner flag 191 (Elisa): "adults" was undefined, so households with a 19-year-old son and households of transition-age young people were counted inconsistently. 18 or older is the HUD convention across HMIS household composition and the CoC household types, so stating it costs nothing and removes the ambiguity. The range hint also said "Include yourself" in every mode, which is wrong when someone answers for a person who is not there and is not in that household. Item version 0.1.0 to 0.2.0. · One stem for everyone, "How many adults, including you, are in your household right now?", with a minimum of 1 (v0.2.0) (replaced): 2026-08-11, illogic audit. For anyone who answered under 18 to the age question directly before this one, the stem was not awkward, it was false, and the range enforced the falsehood: this item's own help text defines an adult as anyone 18 or older, and the minimum of 1 then told a 16-year-old sleeping in a car alone that the honest answer, 0, was out of range. The only way off the screen was to enter a number the person knew was wrong, which is the exact failure that validating at the point of entry exists to prevent, and it corrupted the one number the household questions are for. Under 18 the item now drops "including you" from the stem in both voices, says plainly that people under 18 are not counted here and that 0 is right for someone on their own, and lowers the minimum to 0. Nothing on the adult path moved, and no engine reads this item at all, so no determination anywhere can change. Item version 0.2.0 to 0.3.0. · Reword the single stem for everyone ("counting yourself if you are 18 or older") (rejected): It fixes the minor's screen by adding a conditional clause to a question almost everyone answers, spending reading level for every reader to handle a case the tool already knows it is in: the age answer is in hand one question earlier. The variant is carried in the instrument as data rather than built in code, so both versions stay visible on the public questions page. · "Adults means anyone 18 or older. Count grown children who are staying with you, and anyone else 18 or older who would move with you." (v0.3.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION: the stem and the help named two different groups. The stem asks who is in the household "right now" and the help asked for anyone who "would move with you", and those are different people whenever a partner is in jail, in a hospital, or staying somewhere else this week. A respondent resolves a contradiction like that by picking one, and two respondents in the same situation pick differently, which is the definition of a wording defect rather than a respondent error. The help now states the composition test once and completes it: count anyone 18 or older who would move with you, even if you are apart right now. No rule reads this item; it is a household composition fact, so the severity is low and the fix is still worth making because an item that contradicts itself teaches a respondent that the questions are not to be read closely. The minor variant already stated both halves explicitly and is unchanged. Item version 0.3.0 to 0.4.0.

Known risks: The variant depends on the age question being asked first. That order is fixed in the instrument and pinned by an assertion in personas.test.ts, and if it ever were not, the fallback is the adult wording, which is correct for everyone the fallback can reach.

What this answer changes: Context and cross-checks only; changes no determination.

Review status: pending external review · last updated 2026-08-28

household_children (v0.6.0 · sensitivity low)

Question: How many children under 18 will live with you once you have housing?

Evidence: McKinney-Vento education rights; family program eligibility

Reason kept: Family supports and education-definition note.

Alternatives considered: "How many children under 18 are with you right now?" (v0.1.0) (replaced): 2026-07-28: "right now" counted the family as it is at its worst moment, so parents whose children were temporarily staying elsewhere were sized for a smaller unit than they need. v0.2.0 asks who will live with them once they have housing, and the help text names children currently with family, friends, or foster care. · No third-person variant of the help text; the second-person text rendered under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.2.0 to 0.3.0. · The same help text for a minor as for anyone else (v0.3.0) (replaced): 2026-08-11, illogic audit. Under 18 with children in the household is uncommon and completely real: parenting youth are a population HUD names and funds for, and the combination had to keep working without comment. It does, and it always did, but the audit asked what a 17-year-old parent actually reads on this screen, and the answer was a question about children that says nothing about whether their situation is expected here. Under 18 the help text now adds two sentences saying that some people under 18 are parents, that entering children and entering 0 are both ordinary answers, and that neither changes the result. Nothing is blocked, nothing is confirmed, nothing is flagged, and no message appears after the answer is given: the note sits on the screen before the question is answered, so it can never read as doubt about the answer. Item version 0.3.0 to 0.4.0. · A confirm step after the answer ("you entered under 18 and 2 children; continue if that is right") (rejected): The app has exactly one point-of-entry message pattern, the error summary, and it blocks Continue. Reusing it here would put a stop in front of a true answer given by the person least able to absorb one, and building a second, softer interruption for a real family is a worse idea than not interrupting. A note that arrives before the answer costs the reader nothing and cannot be read as doubt about what they typed. · “Count children staying with family, friends, or foster care right now who would come back to live with you when you have a place.” (v0.4.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). One 24-word sentence, grade 8.0. It is two sentences now, and the two things it asks the reader to do are one each: find the children, then apply the test. “Count children who are staying with family, friends, or in foster care right now. Count them if they would come back to live with you when you have a place.” (4.0). The minor variant carries the same first two sentences and keeps its parenting-minor paragraph, now in shorter sentences. Item version 0.4.0 to 0.5.0. · "Count children who are staying with family, friends, or in foster care right now. Count them if they would come back to live with you when you have a place." and its minor and proxy variants (v0.5.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION. The help listed the non-obvious inclusion and never mentioned the obvious one. The stem is future tense ("will live with you once you have housing"), which is already an unusual thing to be asked, and the help then talked only about children who are elsewhere right now, so a careful literal reader can count only those and leave out the children asleep beside them. A count above zero raises the family_with_children flag, which is how a household reaches family-specific resources, so the direction of the error matters. All three variants now begin by counting the children who are with the household now. This is the item probe (d) of the cognitive interview script is written for, and the probe is unchanged: the foster-care question it asks is about harm, not comprehension, and no wording fixes that. IR-28 records the ceiling of 15 and is unaffected. Item version 0.5.0 to 0.6.0.

Known risks: The added sentences make the screen longer for every minor, including the majority who will enter 0. That is the price of the combination never being questioned after the fact, and it is the right trade.

What this answer changes: plus the family_with_children flag. No determination reads this answer.

Review status: pending external review · last updated 2026-08-28

children_elsewhere (v0.6.0 · sensitivity moderate)

Question: Do you have children under 18 who are not staying with you right now?

Evidence: Family reunification practice

Reason kept: Offer-only reunification support. Excluded from the printable header (minimum necessary).

Alternatives considered: Ask custody details (rejected): Invasive; no determination needs it. · Remove as redundant with the reworded household-children question (2026-07-28) (undefined): They measure different sets: the household question counts children who will live in the unit once housed; this one includes children who are separated and may not return. The reunification support it drives is offer-only. Help text added (v0.2.0) to make the distinction explicit after practitioner feedback prompted the household-question rewording. · No third-person variant of the help text; the second-person text rendered under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.2.0 to 0.3.0. · "Include anyone currently under the age of 18 for whom you have not lost parental rights" (practitioner alternate, register item 11) (rejected): 2026-08-11. A parental-rights frame turns a question about offering family-connection help into a question about the person's legal standing as a parent, in a housing record, and it invites an answer nobody here is qualified to interpret. Declined in any form, on the owner's ruling. · "This can include children you counted in the last question who would come back once you have housing, and children who would not..." (v0.3.0) (replaced): 2026-08-11, public register item 11 (Elisa). The old help asked the reader to carry the previous question in their head to work out who counts. v0.4.0 says the scope directly, in one sentence: every child under 18 not staying with them right now, both the children who would come back once housed and the children who would not. The reassurance is unchanged and still whole (asked only so help with family connection can be offered, never a requirement, never affects priority). Item version 0.3.0 to 0.4.0. · “...right now: children who would come back to live with you once you have housing, and children who would not... It is never a requirement, and it never affects priority.” (v0.4.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). Grade 8.2. The colon that carried the scope is now a full stop and “That means...”, and “never affects priority”, which is our word rather than the reader's, is now “never affects who gets housing first” (5.8). The reassurance is unchanged and still whole: offer-only, never required, never affects who gets housing first. Item version 0.4.0 to 0.5.0. · "Count every child under 18 who is not staying with you right now." (v0.5.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). RESPONSE MAPPING. The help opened with a counting instruction on a question whose only answers are yes and no, which is a leftover from the counted item next to it and asks the respondent to do something the screen will not let them do. It now reads "This includes every child under 18 ...", which states the same scope as an inclusion rather than as an instruction. Nothing else in the help changed, including the two sentences that say the answer is offered-only and never affects who gets housing first. The item feeds the children_elsewhere flag and SS-R11, both offer-only. Item version 0.5.0 to 0.6.0.

Known risks: Highly sensitive; never printed, offer-only.

What this answer changes: SS-R11 (supports worth offering); plus the children_elsewhere flag. No determination reads this answer.

Review status: pending external review · last updated 2026-08-28

living_situation (v0.6.0 · sensitivity low)

Question: Where did you stay last night?

Evidence: 24 CFR 578.3 qualifying situations

Reason kept: Gateway fact for HUD category, chronicity present-tense test, and exposure-based support rules.

Alternatives considered: "Outside, in a vehicle, or somewhere not meant for living in" (replaced): 2026-07-28, field practitioner feedback (Daniel Gore): same phrase change as the chronicity items, so the choice a person picks and the questions that follow use identical words. · Ask where the household stayed LAST NIGHT instead of tonight (practitioner feedback, flag #110, 2026-07-28) (undefined): Adopted (v0.3.0). Programs record the night before, the HMIS chronic-status decision table keys on the night before, and a night that already happened is easier to answer than a night that has not. · Help text without examples of "a place not meant for people to live in" (v0.3.0) (replaced): 2026-08-10, practitioner flag 189 (Elisa): the phrase is precise and still abstract, and people asked what counts. Examples added from 24 CFR 578.3 Category 1(i) (car, abandoned building, park, bus or train station) plus the vehicle, shed, garage and storage-unit cases practitioners meet. The same sentence is used in every item carrying the phrase. Separately, the answer "My own housing" read in the first person under a third-person question in proxy mode; it now has a third-person label. Item version 0.3.0 to 0.4.0. · “Last night, not tonight: programs record where a person actually stayed the night before, and it is easier to answer about something that already happened.” plus the one-sentence definition of a place not meant for people to live in (v0.4.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). Grade 12.7, the third worst string in the instrument, and the 34-word definition sentence was most of it. That definition appears in seven help texts and is now two sentences everywhere it appears, split at the point where a list of places stops being one breath. The help text is 5.4. No place was added to or removed from the list. Item version 0.4.0 to 0.5.0. · "... It also means a storage unit, an abandoned building, a park, or a bus or train station." as the end of the help (v0.5.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). JUDGMENT, on the highest-fan-out item in the instrument: it decides the HUD category through HC-R1 to HC-R7, the present-tense test for chronic homelessness through CH-R8, the unsheltered inputs to SN-R4, SN-R5 and SN-R6, whether the housed wording of four other questions is used, and which modules open. The help defined every place a person could sleep and said nothing about the person who did not sleep anywhere. "Where did you stay last night" has no true answer for someone who walked all night, and the honest-looking choice is "Somewhere else", which returns an undetermined HUD category (HC-R7) and an unknown current situation to the chronicity engine. The help now says that being awake outside all night, with nowhere to sleep, counts as outside. No option was added or reworded; the person is being directed to the answer that was already correct for them. IR-25 records the options that really are missing (transitional housing, a safe haven, a motel) and is unaffected. Item version 0.5.0 to 0.6.0.

Known risks: None identified.

What this answer changes: HC-R1, HC-R5, HC-R7 (HUD homeless category); CH-R8, CH-R10 (chronic homelessness); SN-R4, SN-R5, SN-R6, SN-R12 (support planning level); ID-R5 (options to discuss); SS-R5 (supports worth offering); RA-R8 (reassessment guidance); XC-R4, XC-R5 (in-session consistency check); plus the unsheltered flag.

Review status: pending external review · last updated 2026-08-28

housing_trajectory (v0.8.0 · sensitivity moderate)

Question: Where were you staying just before this time without housing began?

Evidence: Institutional-discharge pathway literature; CMS discharge planning concept

Reason kept: Distinguishes eviction, institutional exit, shelter/transitional entry, unsheltered entry, and other pathways; drives transition supports and moderate signal.

Alternatives considered: "Just before this, where were you living?" (v0.1.0) (replaced): 2026-07-28, field practitioner feedback (Elisa): "just before this" gave no anchor, so people answered about different points in time. v0.2.0 names the anchor (the place lost or left right before the current period of homelessness began). · v0.3.0 adds shelter_or_transitional and unsheltered (2026-08-03, owner flag #138 'missing-option') (adopted): HMIS Prior Living Situation (element 3.917) treats 'place not meant for habitation' and 'emergency shelter / transitional housing' as distinct prior situations; both previously fell through to 'Somewhere else', erasing the two most operationally distinct pathways (shelter-to-shelter movement and continuing unsheltered homelessness). ENGINE-NEUTRAL by design: no scoring rule keys on the new values until evidence supports one; the gain is honest capture, not weight. · A 'fleeing violence or abuse' option in this list (rejected): HMIS models DV as its own element (4.11), not a prior-residence category: disclosure inside a where-did-you-stay list is a safety risk (shoulder-surfing, coerced screens) and orthogonal to residence type (one can flee violence from any of the listed situations). If DV screening enters HEAT it must be a dedicated, sensitivity-high, telemetry-off item with its own review. · No third-person help text or answer label; both rendered in the first or second person under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.3.0 to 0.4.0. · Remove the shelter and unsheltered options, which read as a contradiction (register item 9) (rejected): 2026-08-11, public register item 9 (Elisa): "how can someone have been on the street just before becoming homeless?" The options are right and the question was not explaining itself. A period of homelessness can end and start again after a break, and where someone was immediately before THIS period is what the chronic homelessness arithmetic reads and what the discharge pathways are built on. Two sentences of help text say so now, in both voices; the options are untouched and no rule changed. Item version 0.4.0 to 0.5.0. · One present-tense wording for everyone, whose stem asserts a current period of homelessness (v0.5.0) (replaced): 2026-08-11, follow-up flagged by the v0.50 release, owner ruling (a verified problem with a known solution gets fixed): the three timeline questions stopped presupposing a current street or shelter period and this one did not, so a parent staying with a friend was still asked where they stayed “just before this current period of homelessness began”, having answered two questions earlier that they have a place tonight and never stayed outside or in a shelter. Same defect, same mechanism as the timeline items: a past-tense variant for someone with a place tonight (“where were you staying just before you came to where you are staying now?”), in both voices, with help text saying that naming a shelter or the street here is never counted as time without housing. NOT gated, unlike the start-date question, and the reason is what the rules actually read: this item is a deciding item for housing support need (SN-R9 raises the band on an institutional exit) and the trigger for the transition-support (SS-R6), medical-respite (SS-R17) and tenancy-counselling (SS-R13) offers, every one of which applies to somebody who left a hospital or jail and is now in a relative's spare room, which is the ordinary shape of a discharge. Verified by counterfactual rather than asserted: marking the item not-asked for exactly that person drops the support-need band from moderate to lower, turns the determination from determined to provisional, and deletes both discharge-planning offers. The chronic homelessness arithmetic does not read this item at all (it reads the situation, the start date, the occasion count and the cumulative months), so no determination moves either way, and the housed help text says what is true rather than repeating the v0.5.0 sentence about the arithmetic. Item version 0.5.0 to 0.6.0. · “Where were you staying just before this current period of homelessness began?”, with help text ending “...is what discharge planning and support planning read” (v0.6.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). The stem scored 8.8 and the help text 10.9. “This current period of homelessness” is three abstractions in a row for the one thing the reader is living in; the stem now says “this time without housing” (5.9), which is the phrase the results page already uses. The help text drops the field's own words: “discharge planning and support planning read” is now “is what the next plan is built on” (3.9). The housed variant loses the same jargon. The presupposition each version carries is unchanged, and guards.test.ts still pins it. Item version 0.6.0 to 0.7.0. · "That means the place you lost or left right before staying where you are now." (v0.7.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION: the help's first sentence pointed at a different place than the stem. The stem asks where the person was staying "just before this time without housing began", which is the pathway into homelessness and is what the rules read; the help's opening sentence asked for the place left right before the current one. For anyone who has moved between shelters those are two different answers, and the difference is not cosmetic: SN-R9 raises the band on an institutional exit, SS-R13 offers tenancy counselling only when the place before this period was the person's own, and SS-R17 offers medical respite on a facility exit. A person cycling between a shelter and a relative's spare room would answer "an emergency shelter" under one reading and "my own housing, and I lost it" under the other, and would lose the offer either way it went wrong. The first sentence now names the same place the stem does. The rest of the help, which explains that a time without housing can end and start again, was already written for the stem's reading and is unchanged. The stem itself is untouched, including the phrase a guard pins it to, and the housed variant, which asks a genuinely different question of a genuinely different reader, is untouched. Item version 0.7.0 to 0.8.0.

Known risks: None identified.

What this answer changes: SN-R9, SN-R12 (support planning level); SS-R6, SS-R13, SS-R17 (supports worth offering).

Review status: pending external review · last updated 2026-08-28

months_since_permanent_housing (v0.6.0 · sensitivity low)

Question: About how many months has it been since you had a permanent place to stay?

Evidence: HMIS 3.917 structure

Reason kept: Cross-checks episode dates; contradiction prevention.

Alternatives considered: "...since you had a permanent place OF YOUR OWN?" (v0.1.0) (replaced): 2026-07-28, field practitioner feedback (Elisa): "of your own" excluded people whose settled housing was a family member's or a partner's, so long-term doubled-up histories were undercounted. v0.2.0 counts any settled place and tells people who have never had one to choose "I do not know" and say so, instead of guessing a number. · One help text in every mode, ending "choose “I do not know” and tell the assessor" (v0.3.0) (replaced): 2026-08-10, audit prompted by practitioner flag 199 (Elisa). Same defect as pet_paperwork: the help instructed a person answering alone to tell an assessor who is not there. The assessor-administered text is unchanged; a self-administered variant drops the instruction and a proxy variant carries the third-person voice the question stem already used. Item version 0.3.0 to 0.4.0. · Telling anyone who has never had a permanent place to choose “I do not know” (v0.4.0) (replaced): 2026-08-28, practitioner flag #225, owner ruling: this is the trigger case for the honest-answer sweep of the whole instrument. All three help texts carried the instruction, and the engines treat “I do not know” as MISSING, so the record said nobody knew about a person who had just said so plainly. It cost them twice: the item was reported as missing information and counted against how complete their record was, and the chronicity documentation questions, which open at twelve months or more since a permanent place, stayed shut for the one person whose answer is the strongest form of what that gate looks for. The item now carries a real answer, “I have never had a permanent place”, offered next to the number box, and the instruction is gone from every voice. The answer is not a number, so no arithmetic invents one and both cross-checks that compare it to a number fall silent. For any rule that asks whether it has been at least so many months, it counts as more than any number the box accepts, which is what opens the documentation questions. Nothing a person could already answer changed: proven by a 263-session sweep whose payload digests were taken before the change and still match byte for byte (packages/test-scenarios/src/honestAnswers.test.ts). The self-administered help text was also retired rather than rewritten: it existed only to drop “and tell the assessor”, and that clause existed only to make the workaround work, because the record could not hold the fact. It can now. Item version 0.4.0 to 0.5.0. · A ceiling of 600 months (v0.4.0) (replaced): 2026-08-28, honest-answer sweep, same owner ruling. 600 months is fifty years. A person in the 62 or older band whose last permanent place was longer ago than that, and who therefore cannot truthfully answer “never” either, had no choice but to enter a number they knew was too small: the same defect as the trigger case, arriving through a range instead of through help text. The ceiling is 1200. The lifetime cross-check (XC-R7) was extended with it: its bands stopped at 44 on the stated ground that no answer a 45-year-old could physically enter reached their bound, and that ground moved, so 45 to 54, 55 to 59 and 60 to 61 now have bounds. 62 or older has no top and so can have no bound, which is a limit stated rather than an omission. Item version 0.4.0 to 0.5.0 (with the change above). · "Permanent means a place you could stay as long as you wanted, with or without your name on a lease. Any settled place counts, even if it was not your own." with no statement of what does not count (v0.5.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). JUDGMENT. The definition said what counts and never what does not, and the omitted cases are the ones a person without housing has actually lived in. A two-year shelter stay, a long jail sentence and a residential treatment program are all places a person stayed continuously and could describe as settled, and counting any of them shortens the answer, which is the direction that closes the chronicity documentation module: it opens at 12 months or more since a permanent place, and that module holds disabling_condition, the item CH-R2 reads to end or continue the chronic pathway. The help now says a shelter, a jail or a treatment program is not a permanent place. Nothing was added to what counts, so no answer that was correct before is incorrect now. Probes 4 to 6 in the cognitive interview script are written for exactly these boundaries and stay in the script: the help settles three of them and testing decides the rest. Item version 0.5.0 to 0.6.0.

Known risks: Recall error; tolerated with 1-2 month slack in cross-checks.

Answers that are not a number: “I have never had a permanent place” (answering for someone not present: “They have never had a permanent place”): A real answer, never a missing one: no output may list this item as missing or count the person as less complete for giving it. No rule reads a number from it, so nothing that needs a number changes. Where a rule asks whether it has been 12 months or more since a permanent place, this answer meets that threshold, because never having had one is longer than any number the box accepts. That is the chronicity documentation gate, so this person is now asked the disabling-condition question that a high number would already have opened for them.

What this answer changes: XC-R3, XC-R7 (in-session consistency check). No determination reads this answer.

Review status: pending external review · last updated 2026-08-28

episode_start (v0.8.0 · sensitivity low)

Question: About when did you start staying outside, in an emergency shelter, or in a place not meant for people to live in? This means the time you are in now.

Evidence: 24 CFR 578.3 continuous-episode criterion; HMIS 3.917 partial-month rule

Reason kept: Continuous-pathway chronicity arithmetic, shown to the person.

Alternatives considered: Exact date entry (rejected): Month/year approximation matches how people remember; exact dates create false precision. · "somewhere not meant for living" (replaced): 2026-07-28, field practitioner feedback (Daniel Gore): the phrase read as jargon and people asked what it meant. Replaced with "a place not meant for people to live in" everywhere it appears (episode_start, occasions_3yr, months_homeless_3yr, the living_situation choice, and inst_prior_situation) so every question uses one phrasing. · Consolidate the duration questions (practitioner feedback, Josh Johnson, 2026-07-28) (undefined): Each question is a distinct quantity 24 CFR 578.3 requires: episode start drives the continuous pathway, occasion count and cumulative months drive the episodic pathway, and months-since-permanent-housing is context and a cross-check. Collapsing them would lose the arithmetic the determination must show, and the cross-checks that catch contradictions depend on asking them separately. · Help text without examples of "a place not meant for people to live in" (v0.3.0) (replaced): 2026-08-10, practitioner flag 189 (Elisa): same examples sentence added wherever the phrase appears. Item version 0.3.0 to 0.4.0. · Asking everyone when “this current period on the streets, in a shelter, or a place not meant for people to live in” started (v0.4.0) (replaced): 2026-08-11, v0.49 audit finding, owner ruling (obvious fix, shipped without a register round): the question presupposed a street, shelter, or uninhabitable period the person is IN, and it was asked of everyone, including people who had just answered that last night was spent with family or friends or in their own housing. For them the stem asserts something false, and the neighbouring items' help text excludes the place they actually are. Two changes shipped. First, a gate (askWhen): someone who has a place tonight is asked this only when they reported at least one street or shelter time in the past three years, and otherwise is not asked at all. A skipped question counts as not asked, never as missing, and the completeness denominator drops with it. Second, a past-tense wording for that group (“about when did you most recently start staying on the streets, in a shelter, or a place not meant for people to live in?”), whose answer is history for the plan and is NOT fed to the continuous-pathway arithmetic: elapsed time since a period that ended is not the length of anything, and feeding it in produced both a false arithmetic line and a false “answers conflict”. No determination for anyone whose situation can still be a qualifying one changed, proven by an instrument-stripping inertness test over a full scenario sweep. Item version 0.4.0 to 0.5.0. · “About when did this current period of staying on the streets, in a shelter, or a place not meant for people to live in start?” (v0.5.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). Grade 8.8 in the second person and 9.1 in the third, with a 25-word run between the verb and its object. It is two sentences now: the question, then the scope. “About when did you start staying outside, in an emergency shelter, or in a place not meant for people to live in? This means the time you are in now.” (5.2). Two other changes ride with it: “on the streets” becomes “outside”, which is the word the answer list and the results page already use, and “in a shelter” becomes “in an emergency shelter”, which is what this item always meant and what its two neighbours already said. The housed variant is split the same way. Item version 0.5.0 to 0.6.0. · No way to say the period never happened (v0.6.0) (replaced): 2026-08-28, honest-answer sweep, owner ruling, same class as flag #225. The gate added in 0.23.0-pilot spares anyone with a place tonight who reports no street or shelter time, but it does not reach everyone: a person in a hospital, jail or treatment facility is still asked when their current time outside, in an emergency shelter, or in a place not meant for people to live in began, and so is anyone whose situation the answers do not settle. Someone in a facility who has never stayed in any of those places had no true answer on the screen, and the only way past it was “I do not know”, which the engines treat as missing. The item now carries the answer, “I have never stayed in any of those places”. It holds no date, so the continuous-pathway arithmetic is unchanged and still cannot be met, which is the correct result rather than a side effect of the value. What changes is that the record stops listing the question as unanswered, stops counting the person as less complete for it, and stops telling the worker to reassess when the missing answer arrives (RA-R5), because nothing is coming. Item version 0.6.0 to 0.7.0. · "An estimate is fine. Short stays housed (under 7 nights) or in a facility (under 90 days) do not restart this period." (v0.7.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION, and an asymmetry between two variants of one item. This date is the input to the continuous pathway: months between it and today, and 12 or more meets the criterion at CH-R4. The past-tense variant written for a person with a place tonight already said "This is about the most recent of those times, not the first one." The main help, read by everyone actually outside or in a shelter, did not, so the sentence existed for the group whose answer does not drive the arithmetic and was missing for the group whose answer does. A person homeless on and off for years, asked when they started staying outside, can give the year they first became homeless, and the continuous pathway then reads years where the correct answer might be four months. The sentence is now in both. The stem's own qualifier ("This means the time you are in now") is unchanged and so is the named answer added in 0.26.0-pilot. Item version 0.7.0 to 0.8.0.

Known risks: Recall error; approximate by design.

Answers that are not a number: “I have never stayed in any of those places” (answering for someone not present: “They have never stayed in any of those places”): A real answer, never a missing one: no output may list this item as missing or count the person as less complete for giving it. There is no date in it, so the continuous-pathway arithmetic is unchanged and still cannot be met, which is the correct result rather than a side effect. Reachable because the question is still asked of anyone in a facility, or in a situation the answers do not settle, and a person there may never have stayed outside or in a shelter at all.

What this answer changes: CH-R4, CH-R6, CH-R7 (chronic homelessness); XC-R1, XC-R3, XC-R8 (in-session consistency check).

Review status: pending external review · last updated 2026-08-28

occasions_3yr (v0.8.0 · sensitivity low)

Question: Think about the past three years. How many separate times have you stayed outside, in an emergency shelter, or in a place not meant for people to live in? Count now as one of them.

Evidence: 24 CFR 578.3 episodic criterion (4 occasions / 12 months / 3 years); HUD FAQ 2752: no minimum occasion length

Reason kept: Episodic-pathway chronicity.

Alternatives considered: "somewhere not meant for living" (replaced): 2026-07-28, field practitioner feedback (Daniel Gore): replaced with "a place not meant for people to live in" across every item that carries the street/shelter phrase. · Help text without examples of "a place not meant for people to live in" (v0.3.0) (replaced): 2026-08-10, practitioner flag 189 (Elisa): same examples sentence added wherever the phrase appears. Item version 0.3.0 to 0.4.0. · One present-tense wording for everyone: “how many separate times have you stayed ... including now?” (v0.4.0) (replaced): 2026-08-11, v0.49 audit finding, owner ruling (obvious fix, shipped without a register round): “including now” is a counting instruction that is wrong for anyone who is not on the streets or in a shelter tonight, and the item's own help text tells the same person that where they are staying does not count. A past-tense variant is now shown to people who have a place tonight (staying with family or friends, or their own housing), asked with “if any” and stating that 0 is an ordinary answer, because for that group it usually is and an unlabelled 0 reads like a failed question. Everyone whose situation can still be a qualifying one, including anyone in a facility, sees the original wording unchanged. The arithmetic is untouched. Item version 0.4.0 to 0.5.0. · “In the past three years, how many separate times have you stayed on the streets, in an emergency shelter, or a place not meant for people to live in, including now?” (v0.5.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). Grade 12.1, one 31-word sentence holding a window, a definition and an instruction. Three sentences now, one each: “Think about the past three years. How many separate times have you stayed outside, in an emergency shelter, or in a place not meant for people to live in? Count now as one of them.” (4.1). “Including now” survives as its own instruction, which is the half most often missed. The help text's “A time apart is separated by 7 or more nights somewhere housed” becomes “Two times count as separate when 7 or more nights housed came between them”: same 7-night rule, said forwards. Item version 0.5.0 to 0.6.0. · A ceiling of 36 separate times (v0.6.0) (replaced): 2026-08-28, honest-answer sweep, owner ruling. 36 is the number of MONTHS in the three-year window, borrowed for a question that counts TIMES. The item's own help text says two times count as separate when seven or more nights housed came between them, so one separate time costs eight nights at minimum and up to 136 of them fit in three years. Someone who cycles between a shelter and a weekend at a relative's could truthfully be above 36 and had to enter a number they knew was too low. The ceiling is 136, which is the arithmetic rather than a round number. No determination reads a count above four, so no result moves for anyone; what changes is that the question stops asking a person to under-report. Item version 0.6.0 to 0.7.0. · "Time staying with family or friends does not count here. Two times count as separate when 7 or more nights housed came between them." and the same pair in the housed variant (v0.7.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION. The help defined the break and never defined the non-break. Four or more separate times, with 12 or more cumulative months, is the episodic pathway to chronic homelessness at CH-R4, and the natural reading of "how many separate times have you stayed outside, in an emergency shelter, or in a place not meant for people to live in" counts moves: a person who went shelter, street, different shelter, camp, different camp says five and has said something false, because none of those moves had seven housed nights in it. Five clears the pathway; one does not. The help now says that moving from one shelter or camp to another is still the same time, in the main help and in both voices of the housed variant. The seven-night rule is unchanged and no ceiling moved: 136 stands, with the derivation recorded in the 0.26.0-pilot entry below. Item version 0.7.0 to 0.8.0.

Known risks: Occasion counting is genuinely hard; cross-checks catch impossible combinations without blaming.

What this answer changes: CH-R4, CH-R5, CH-R6, CH-R7 (chronic homelessness); SS-R13 (supports worth offering); RA-R4 (reassessment guidance); XC-R2, XC-R4, XC-R5, XC-R6 (in-session consistency check); plus the first_time_homeless flag.

Review status: pending external review · last updated 2026-08-28

months_homeless_3yr (v0.7.0 · sensitivity low)

Question: Think about the past three years. About how many months in all have you stayed outside, in an emergency shelter, or in a place not meant for people to live in?

Evidence: 24 CFR 578.3; 80 FR 75791 break rules (7+ nights housed incl. doubled-up)

Reason kept: Cumulative chronicity months and exposure input to support rules.

Alternatives considered: "Time without your own housing" (rejected): Would have counted doubled-up time toward chronicity, inflating determinations. Caught pre-pilot; on the public changelog. · "somewhere not meant for living" (replaced): 2026-07-28, field practitioner feedback (Daniel Gore): replaced with "a place not meant for people to live in" across every item that carries the street/shelter phrase. · Help text without examples of "a place not meant for people to live in" (v0.3.0) (replaced): 2026-08-10, practitioner flag 189 (Elisa): same examples sentence added wherever the phrase appears. Item version 0.3.0 to 0.4.0. · Merge or drop one of the three timeline questions (register item 10) (rejected): 2026-08-11, public register item 10 (Elisa): part way through the assessment the three timeline questions were repetitive enough that she could not remember what she had already answered. They are not redundant. HUD's chronic homelessness definition (24 CFR 578.3) is arithmetic over three different numbers: when this period started, how many separate times there have been, and how many months they add up to. Dropping any one of them makes the determination unprovable. The answer shipped is the explanation, one sentence at the head of the help text on this item, rather than a change to the questions. Item version 0.4.0 to 0.5.0. · One present-tense wording for everyone, whose help text names “the three timeline questions” (v0.5.0) (replaced): 2026-08-11, v0.49 audit finding, owner ruling (obvious fix, shipped without a register round): same defect as the occasion count, plus one more: the help text described three timeline questions to a reader who is not asked the third one. The past-tense variant for people with a place tonight names two, says a partial month still counts as a whole month, states that time with family or friends or in their own place is not counted, and names 0 as an ordinary answer. Nothing about the arithmetic or the thresholds moved. Item version 0.5.0 to 0.6.0. · “Adding it all together, about how many months in the past three years have you stayed on the streets, in an emergency shelter, or a place not meant for people to live in?”, with help text opening “The three timeline questions do overlap...is arithmetic over...” (v0.6.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). The stem scored 12.7 (33 words in one sentence) and the help text 9.7. The stem is now “Think about the past three years. About how many months in all have you stayed outside...” (5.7); the present perfect is deliberate and pinned by guards.test.ts, because this stem covers time that is still running and only the housed variant may use the past tense. In the help text “arithmetic” becomes “math”, “a partial month” becomes “part of a month”, and the citation (24 CFR 578.3) is untouched. Item version 0.6.0 to 0.7.0.

Known risks: Recall error.

What this answer changes: CH-R4, CH-R5, CH-R6, CH-R7 (chronic homelessness); SN-R4, SN-R5, SN-R9, SN-R12 (support planning level); ID-R1 (options to discuss); XC-R1, XC-R2, XC-R4, XC-R5, XC-R6 (in-session consistency check).

Review status: pending external review · last updated 2026-08-28

income_any (v0.4.0 · sensitivity low)

Question: Do you have any money coming in each month?

Evidence: HUD CE Equity Initiative: income items carry differential-disclosure bias

Reason kept: RRH program-fit discussion ONLY. Excluded from the vulnerability construct entirely; a test suite proves it cannot change any determination.

Alternatives considered: Use income in support-need (rejected): Not vulnerability; disclosure bias; CPD-17-01 II.B.4 adjacency. · "Do you have any money coming in right now, from any source?" (v0.1.0) (replaced): 2026-07-28, field practitioner feedback (Elisa): "from any source" read as formal income, so informal money (recycling, family, people helping out) went unreported. v0.2.0 asks per month and names those examples in the help text. · No third-person variant of the help text; the second-person text rendered under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.2.0 to 0.3.0. · "Any way at all counts: aid, a job, recycling cans, family, or people helping you out." (v0.3.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION of the time anchor. The stem asks about money coming in "each month", and the help listed every source and never addressed the regularity the stem asserts. Day labor, seasonal work, recycling and family help are exactly the sources the list names and exactly the sources that do not arrive every month, so the list and the stem pull against each other and a person with irregular income can honestly answer no. A no raises the zero_income flag, which is a match-relevant fact a community may prioritise on. The help now says money that comes some months and not others still counts. IR-21, the household-versus-individual ambiguity for minors in families, is a different finding and is unaffected. Item version 0.3.0 to 0.4.0.

Known risks: Perceived as gatekeeping if misused; the screening-out suite is the guard.

What this answer changes: ID-R4 (options to discuss); SS-R8, SS-R9 (supports worth offering); plus the zero_income flag.

Review status: pending external review · last updated 2026-08-28

housing_worked_before (v0.4.0 · sensitivity low)

Question: Have you ever had a place to stay that was stable for six months or longer?

Evidence: Strengths-based assessment practice; protective-factor identification (SAMHSA trauma-informed principles)

Reason kept: The one question that asks what already works. Feeds support planning display and a build-on-what-worked offer (SS-R18) only; never prioritization. Replaced employment_any, whose program-fit signal income_any already carries.

Alternatives considered: Keep employment_any alongside it (rejected): Question budget; employment was redundant with income for every rule that read it. · "Was there ever a time when you had housing that worked well for you..." (v0.1.0) (replaced): 2026-07-28, field practitioner feedback (Elisa): "worked well" invited a judgment about the whole period and people said no about housing that had in fact been stable. v0.2.0 asks the plain fact (stable for six months or longer) and the help text says any place counts, even one that was not their own. · No third-person variant of the help text; the second-person text rendered under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.2.0 to 0.3.0. · Bound the question to a recent window, for example the past three years (register item 1) (rejected): 2026-08-11, public register item 1 (a review group member) with the same worry raised in the field round by Elisa ("what if it was when I was a kid, or in jail?"). Narrowing the gate would erase the strength for exactly the people who have been homeless longest, which is the opposite of what a protective factor is for. "Ever" stays. The recency concern is answered by DISPLAY instead: a new unscored follow-up, housing_worked_when, asks roughly when that stable time was and prints it in the what-worked planning text, so a worker can tell last year from childhood without anyone's answer being narrowed away. · "Any place counts, even if it was not your own. This is about what has worked, not what went wrong." (v0.3.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). JUDGMENT. "Have you ever had a place to stay that was stable for six months or longer?" gates the protective_history module and triggers SS-R18, the one offer that starts a plan from what already worked. Asked in a housing assessment, an adult hears the question as being about adult housing, so a person whose only long stability was a childhood home answers no and the module never opens. That module contains housing_worked_when, which carries "When I was a child or teenager" as an answer, so the instrument was offering a choice it had already made unreachable one question earlier. The help now says growing up in one place counts too. Never prioritisation, never a determination: the necessity test recorded in this item's purpose is unchanged. Item version 0.3.0 to 0.4.0.

Known risks: A no answer could feel discouraging; help text frames the question as optional and blame-free.

What this answer changes: SS-R18 (supports worth offering). No determination reads this answer.

Review status: pending external review · last updated 2026-08-28

functional_support_need (v0.5.0 · sensitivity moderate)

Question: Say you had housing tomorrow. Would you need regular help from another person to keep it?

Evidence: Supportive-housing practice: in-home support need predicts housing sustainment

Reason kept: Housing-sustainment construct (SN-R7). Concrete capability question, never a diagnosis. Fair-housing guardrail in the rule statement: plans support, never limits housing choice.

Alternatives considered: Activities-of-daily-living (ADL/IADL) battery (open): More precise but costs questions; construct-scope decision pending owner + clinical review. · "...such as managing bills, medications, appointments, or everyday tasks?" (v0.2.0) (replaced): 2026-07-28, field practitioner feedback (Elisa): the examples inside the question read as a checklist people had to match, so partial needs went unreported. v0.3.0 keeps the question short and moves the examples to the help text with "help with anything at all counts." · Replace the yes/no with a list of support types (register item 2) (rejected): 2026-08-11, public register item 2 (a review group member). The yes/no is the assessed item and stays exactly as it is: it is a capability question the support-planning rules read, and turning it into a menu would turn a housing assessment into a disability profile held in a housing record. The usefulness the reviewer asked for ships as a separate unscored follow-up, support_types_needed, asked only on a yes and printed on the case manager's planning card as a list to talk about. No rule reads it, and a byte-identical inertness proof over every combination of it is in screeningOut.test.ts. · “If you had housing tomorrow, would you need regular help from another person to keep it?”, with help text “Help with anything at all counts: money and bills, medications, appointments, cleaning, or anything else.” (v0.3.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). The stem scored 8.3 and the help text 10.7, the latter almost entirely on “medications” and the colon splice. The stem is now “Say you had housing tomorrow. Would you need regular help from another person to keep it?” (5.2), which keeps the hypothetical and keeps “regular”, the word the construct turns on. “Medications” becomes “medicine” and the list gets its own sentence (6.3). No rule reading this item changes. Item version 0.3.0 to 0.4.0. · "Help with anything at all counts. That could be money and bills, medicine, appointments, cleaning, or anything else." (v0.4.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION, on the item behind SN-R7, which raises the support planning band to high on its own and opens the support_kinds module. The construct is a concrete capability question: would another person's regular help be needed to keep housing. The old example list opened with "money and bills", which reads as needing money at least as readily as it reads as needing help managing money, and a person holding or hoping for a rent subsidy answers yes to a question about a voucher and is recorded as needing high service coordination. The example is now "managing money and bills", and a closing sentence says this means help from a person, not money from a program. The construct is unchanged and deliberately still broad: help with anything at all still counts, and this is still never a diagnosis, per CPD-17-01. IR-18, the near-auto-yes for minors, is a separate finding and is unaffected. Item version 0.4.0 to 0.5.0.

Known risks: Could be misread as independence screening; guarded in statement, docs, and tests.

What this answer changes: SN-R7, SN-R12 (support planning level); SS-R7 (supports worth offering).

Review status: pending external review · last updated 2026-08-28

crisis_contacts_12mo (v0.2.0 · sensitivity moderate)

Question: In the past 12 months, about how many times have you been to an emergency room?

Evidence: Frequent-user literature: high ED utilization marks unmet need and elevated mortality; CPD-16-11 severity factor

Reason kept: High-utilizer signal (SN-R8 at 4+; moderate at 1+); also drives the health-care-connection soft support. Asks only about emergency room visits, over the same 12-month window the other recency items use.

Alternatives considered: Threshold 3 (rejected): Too sensitive; a small number of visits is within ordinary range for unsheltered life. · Threshold 5-6 (rejected): Self-report undercounts; too specific. · "In the past six months... emergency room or another crisis-related visit" (crisis_contacts_6mo v0.2.0) (replaced): 2026-07-28, field practitioner feedback (Elisa): mixed lookback windows across the instrument (6 months here, 12 months for the health items, 3 years for the chronicity items) produced unreliable answers from people who struggle with timekeeping, and "another crisis-related visit" meant different things to different people. The item was retired and replaced by crisis_contacts_12mo: one window, one concrete event. 4+ ED visits in 12 months aligns with common emergency-department high-utilizer definitions. This is deliberately more sensitive than the old 4-in-6-months rule, and the widening is accepted on purpose: recall error was judged the larger problem. · Ask at household level, “you or someone in your household” (practitioner feedback, Josh Johnson, 2026-07-28) (undefined): The mortality evidence behind this item (emergency-department utilization) is individual-level; a household aggregate would dilute the signal this construct exists to capture. Other household members receive their own assessments. · "Any reason counts. An estimate is fine." (v0.1.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). JUDGMENT. The help settled the reason and never settled the place, on an item where four visits raises the band at SN-R8 and one visit is a condition of SN-R9. Urgent care, a walk-in clinic, a psychiatric crisis centre, a detox unit and an ambulance ride that ended in the driveway are all things a person calls going to emergency, and a person who counts urgent-care visits can cross four without ever having been to an emergency room. The help now says it means a hospital emergency room and that urgent care and a regular clinic do not count, and keeps both of the old sentences. This narrows the item to what it has always asked and what the threshold was derived against; that the construct named in the item id is broader than the question is a real finding and is filed as IR-33 rather than fixed by widening a threshold nobody re-derived. Item version 0.1.0 to 0.2.0.

Known risks: PROXY RISK, explicitly tracked: ER access differs by community (rural distance, urgent-care substitution, immigration-related avoidance). The 2026-07-28 move from a 6-month to a 12-month window widens the net, so more people clear the threshold; fairness monitoring (VALIDATION_PLAN section 8) tests exactly this, and the threshold is policy-editable per CoC.

What this answer changes: SN-R8, SN-R9, SN-R12 (support planning level); plus the high_utilizer flag.

Review status: pending external review · last updated 2026-08-28

health_cold_injury (v0.6.0 · sensitivity high)

Question: Have you ever needed medical care because you could not get warm? Frostbite or hypothermia would count.

Evidence: Hwang 1998 (JAMA): cold-injury history OR ~8 for mortality among homeless adults

Reason kept: Strongest single mortality marker available by self-report (SN-R1 floors HIGH).

Alternatives considered: "Ever ... frostbite" unscoped (replaced): A mountaineering frostbite is not homelessness risk; v0.2.0 scopes to 'because you could not get warm'. · 12-month window (rejected): The evidence base is lifetime cold-injury history; recency scoping would discard the validated marker. · Help text "You can skip this." (replaced): 2026-07-28: "skip" named no control on the screen, so people looked for a skip button that does not exist. Replaced everywhere with the exact label of the choice that is there: "If you prefer not to answer, choose 'I prefer not to answer.'" · No third-person variant of the help text; the second-person text rendered under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.3.0 to 0.4.0. · “Have you ever needed medical care because you could not get warm, like frostbite or hypothermia?” (v0.4.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). Grade 8.3. The examples now stand as their own sentence: “Have you ever needed medical care because you could not get warm? Frostbite or hypothermia would count.” (5.1). The lifetime window, the mortality marker it feeds (SN-R11) and the two clinical terms are unchanged. Item version 0.4.0 to 0.5.0. · Help text that began "Cold injuries ..." and went straight to "Answering helps match the right level of support." (v0.5.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION of the stem's own verb. The stem asks whether medical care was ever NEEDED, which is deliberate and correct, because the population this item is about is the population least able to reach care; but a person reads "needed medical care" as "was treated", and a person who lay in a doorway until they could feel their hands again answers no. The answer sets a floor of high at SN-R1 and is one of the four mortality events SN-R4 compounds, so the under-report is on the item where an under-report is least affordable. The help now says it counts even if a doctor could not be reached. The scoping that keeps a sports or workplace injury out of the item is in the stem and is unchanged, and so is the sentence about declining to answer. Event, never a diagnosis. Item version 0.5.0 to 0.6.0.

Known risks: Coachable binary; mitigation is assessor practice (no leading), never documentation demands.

What this answer changes: SN-R1, SN-R12 (support planning level); SS-R17 (supports worth offering).

Review status: pending external review · last updated 2026-08-28

health_heat_injury (v0.4.0 · sensitivity high)

Question: Have you ever needed medical care because you could not get out of the heat? Heat stroke, heat exhaustion, or dehydration would count.

Evidence: Heat-death surveillance (e.g. Maricopa County annual heat reports): people experiencing homelessness carry an outsized share of heat-associated deaths

Reason kept: Heat counterpart of the cold-injury mortality marker (SN-R11 floors HIGH). Added 2026-08-01 from an owner flag: the instrument asked about deadly cold and not deadly heat, and assessment happens in hot communities too.

Alternatives considered: Fold into the cold-injury question ("too hot or too cold") (rejected): A double-barreled question cannot be answered no to one half and yes to the other, and the two exposures carry separate evidence bases. · 12-month window (rejected): Same reasoning as the cold item: the marker is lifetime event history; recency scoping would discard it. · "could not get cool" (exact mirror of the cold wording) (rejected): The owner's flag names the real situation: nowhere cool to be. "Could not get out of the heat" says that plainly and scopes out sports and workplace heat illness the way "could not get warm" scopes out mountaineering frostbite. · No third-person variant of the help text; the second-person text rendered under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.1.0 to 0.2.0. · “Have you ever needed medical care because you could not get out of the heat, like heat stroke, heat exhaustion, or dehydration?” (v0.2.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). Grade 9.6, the worst of the health items, because three clinical terms sat in the tail of a 22-word question. Split the same way as its cold twin: “Have you ever needed medical care because you could not get out of the heat? Heat stroke, heat exhaustion, or dehydration would count.” (5.3). All three terms are kept. Item version 0.2.0 to 0.3.0. · Help text that began "Heat illness ..." and went straight to "Answering helps match the right level of support." (v0.3.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION of the stem's own verb. The stem asks whether medical care was ever NEEDED, which is deliberate and correct, because the population this item is about is the population least able to reach care; but a person reads "needed medical care" as "was treated", and a person who lay in a doorway until they could feel their hands again answers no. The answer sets a floor of high at SN-R11 and is one of the four mortality events SN-R4 compounds, so the under-report is on the item where an under-report is least affordable. The help now says it counts even if a doctor could not be reached. The scoping that keeps a sports or workplace injury out of the item is in the stem and is unchanged, and so is the sentence about declining to answer. Event, never a diagnosis. Item version 0.3.0 to 0.4.0.

Known risks: Coachable binary; same mitigation as the cold item (assessor practice, never documentation demands). The evidence base is surveillance-level rather than a cohort odds ratio, so the rule is tiered Indirect where the cold rule is Direct.

What this answer changes: SN-R11, SN-R12 (support planning level).

Review status: pending external review · last updated 2026-08-28

health_overdose_year (v0.7.0 · sensitivity high)

Question: In the past 12 months, did you overdose? Or did anyone use Narcan on you, or dial 911 because of your drinking or drug use?

Evidence: Overdose is a leading cause of death in homeless cohorts; SAMHSA Overdose Prevention Toolkit

Reason kept: Active mortality risk (SN-R2 floors HIGH); triggers naloxone and treatment-access offers, never barriers.

Alternatives considered: Lifetime window (rejected): Punishes recovery; recency tracks active risk. · "emergency help" alone (replaced): Lay naloxone from a friend is often not counted as emergency help; v0.2.0 adds 'someone have to revive you'. · "...did you get emergency help, or did someone have to revive you..." (v0.2.0) (replaced): 2026-07-28, field practitioner feedback (Elisa): "revive" was not the word people use, and "emergency help" was read as an ambulance ride only. v0.3.0 names what actually happens: Narcan, a 911 call, or someone getting them help. The skip line was also replaced with the exact wording of the decline choice on screen. · "...did anyone use Narcan on you, call 911 for you, or get you emergency help after drinking or using drugs?" (v0.3.0) (replaced): 2026-07-28, field practitioner feedback (flag #102): the "get you emergency help" clause floated free of the substance-use frame, so emergency help for an unrelated reason could be answered yes and inflate a mortality-risk signal. v0.4.0 names the overdose directly and ties every clause to the person's drinking or drug use. Help text unchanged. · No third-person variant of the help text; the second-person text rendered under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.4.0 to 0.5.0. · "...did you overdose, or did anyone use Narcan on you or call 911 because of your drinking or drug use?" (v0.5.0) (replaced): 2026-08-10, consistency audit arising from flag 202 (Elisa). The product spoke about the emergency number two ways: the site banner and the results page said dial 911, while the safety-stop screen and this question said call 911. One product, one verb, so every surface now says dial. Nothing else about the question changed: same event, same 12-month window, every clause still tied to the person's drinking or drug use. Item version 0.5.0 to 0.6.0. · Soften to "hospitalization", also capture being at risk of overdose, make it a dropdown, and ask whether the person is seeking treatment (register item 4) (rejected): 2026-08-11, public register item 4 (a review group member). Declined on the owner's ruling, each part for its own reason. "Hospitalization" is not softer, it is narrower: most overdoses that matter here never reach a hospital, which is why the item names Narcan and a 911 call. "At risk of overdose" is a judgment, not an event, and this item exists precisely because an event can be counted without anyone assessing the person. A dropdown hides the choices behind a control that is worse on a phone and worse with a screen reader. Asking whether someone is seeking treatment puts a treatment-compliance question in a housing record, which is the exact shape of thing HEAT will not collect: housing is never conditioned on treatment. What WAS adopted from the concern behind it had already shipped: the help text says plainly that the answer never blocks any housing option and that treatment is never required, the decline choice is quoted in the words on screen, and v0.6.0 changed "call 911" to "dial 911" so the sentence reads as one product. Re-read on 2026-08-11 against the humane-tone test and left unchanged: no cheap improvement was available that did not change the meaning. · “In the past 12 months, did you overdose, or did anyone use Narcan on you or dial 911 because of your drinking or drug use?” (v0.6.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). Grade 8.8 across 25 words with two “or” branches a reader has to hold at once. It is two questions now, and the second still opens on “Or” so that one yes still covers either: “In the past 12 months, did you overdose? Or did anyone use Narcan on you, or dial 911 because of your drinking or drug use?” (3.9). The 2026-07-28 retie to drinking or drug use is untouched, and so is the reassurance under it. Item version 0.6.0 to 0.7.0.

Known risks: Disclosure fear despite never-a-barrier framing; help text states it plainly.

What this answer changes: SN-R2, SN-R12 (support planning level); SS-R14, SS-R15, SS-R17 (supports worth offering).

Review status: pending external review · last updated 2026-08-28

health_suicide_attempt_year (v0.3.0 · sensitivity high)

Question: In the past 12 months, have you tried to end your life?

Evidence: Prior attempt is the strongest single predictor of suicide death; NIMH ASQ: asking directly is safe and recommended

Reason kept: Active mortality risk (SN-R3 floors HIGH); pairs with follow-up + trusted-person offer (SAMHSA SAFE-T concept), 988 in help text, acute-vs-history note on results.

Alternatives considered: Euphemism ("hurt yourself") (rejected): NIMH guidance: direct wording is safer and clearer; euphemisms conflate self-harm and attempts. · Help text "You can skip this." (replaced): 2026-07-28: "skip" named no control on the screen, so people looked for a skip button that does not exist. Replaced everywhere with the exact label of the choice that is there: "If you prefer not to answer, choose 'I prefer not to answer.'" · Also ask about suicidal thoughts or wanting to die (ideation) (rejected): Rejected for a public self-administered tool with no clinician present; attempt history is the strongest single mortality predictor, 988 is surfaced universally, and ideation screening belongs to clinical settings (C-SSRS decision, on record). · Add further trauma-informed framing (practitioner feedback, Josh Johnson, 2026-07-28) (undefined): Reviewed against the current item: it already carries 988 in the help text, an explicit statement that the answer never counts against the person, clear decline instructions, and an ask-privately assessor hint. No gap identified; open to specific wording proposals through the feedback channel. · No third-person variant of the help text; the second-person text rendered under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.2.0 to 0.3.0.

Known risks: Emotional weight; asked late, privately, skippable, never printed.

What this answer changes: SN-R3, SN-R12 (support planning level); SS-R16 (supports worth offering).

Review status: pending external review · last updated 2026-08-28

barrier_missing_id (v0.4.0 · sensitivity low)

Question: Do you have your ID or identity documents with you, or could you get them within a few days?

Evidence: Field practice: documents block program entry and leases

Reason kept: Document-recovery offer; practical unblocking.

Alternatives considered: "...with you or somewhere you can get them?" (v0.1.0) (replaced): 2026-07-28: "somewhere you can get them" was answered yes by people whose documents were hundreds of miles away or would cost money to replace, which hid the barrier. v0.2.0 bounds it ("within a few days") and the help text says to answer no when money, travel, or ordering records is involved. · "If getting them would take money, travel, or ordering records from far away, answer no. Help recovering documents can be offered; it is never required." (v0.3.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). RESPONSE MAPPING on a compound question. "Do you have your ID or identity documents with you, or could you get them within a few days?" is two questions and a plural noun answered by one yes or no, and the help already carried one mapping instruction for the hard case, which is why the remaining gap was worth closing: it said nothing about a person holding some documents and not others, which is the ordinary state of the population. A state ID with no birth certificate is a documentation barrier and was being recorded as none. The help now says having some of them, but not all, is still a no. The item feeds the id_documents_missing flag and a documentation-recovery offer, never a determination, so the direction of this fix costs nobody anything and buys an offer for the people who need it. Item version 0.3.0 to 0.4.0.

Known risks: None identified.

What this answer changes: SS-R1 (supports worth offering); plus the id_documents_missing flag. No determination reads this answer.

Review status: pending external review · last updated 2026-08-28

pets_any (v0.3.0 · sensitivity low)

Question: Do you have any pets or other animals that stay with you?

Evidence: Field practitioner report (Elisa, 2026-07-28): pets are among the strongest housing-search barriers, and people refuse placements rather than leave an animal; Fair Housing Act, 42 U.S.C. 3604(f)(3)(B), and Section 504 of the Rehabilitation Act for federally assisted housing: the reasonable-accommodation duty an assistance-animal request runs on. HUD explained how to assess such a request in FHEO Notice 2020-01, which HUD withdrew on 2025-09-17 (Notice of the Withdrawal of FHEO Guidance Documents, hud.gov). The statutory duty is unchanged; the federal explanation of how to document it is not currently in force

Reason kept: Drives an offer-only support (SS-R19) and a match-relevant fact on the worker card. Changes no determination: the screening-out suite proves the answer cannot move immediate action, HUD category, chronicity, support need, or any discussion-set disposition.

Alternatives considered: Ask how many animals and what kind, as structured items (rejected): Closed-option determinism and the no-storage design make the free-text worker note the right carrier for kind and count; a count would invite treating pets as a burden score. · Leave pets to the worker to raise informally (rejected): It is raised inconsistently and usually too late, after a unit is identified, which is when the letter delay bites. · "Do you have a pet or animal that stays with you?" (v0.1.0) (replaced): 2026-08-10, practitioner flag 198 (Elisa): the singular made people with three cats and a dog hesitate over a question that has no wrong answer, and an answer of yes then said nothing about how many. The wording is now plural-inclusive and the help says one or more, any kind. The other half of the flag, actually collecting the number and the species, is new data collection and went to the public register. Item version 0.1.0 to 0.2.0. · "One or more, any kind. Pets never affect your place in line or any result here." (v0.2.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). JUDGMENT, with a gating consequence. This item is the entry condition for the pets module, which holds the assistance-animal letter question, and it triggers SS-R19. Two truthful answers were being lost. A person with a service animal or an emotional support animal may firmly not call that animal a pet, and the question asks about "pets or other animals", which a careful reader can still resolve as no; answering no closes the module whose entire purpose is the letter that animal's housing rights depend on. And a person whose dog is being kept by a friend while they are in a shelter answers no about an animal they intend to get back, which is the case people refuse placements over. The help now says a service animal or an emotional support animal counts here, and that an animal staying with someone else for now counts too. Never any determination, in either direction. The structural half, that a gating question should probably not depend on whether a person accepts the word pet, is filed as IR-34. Item version 0.2.0 to 0.3.0.

Known risks: Disclosure could bias a worker against the household. Mitigated by determination-inertness (tested) and offer-only framing; the help text says plainly that a pet never affects anyone's place in line or any result.

What this answer changes: SS-R19 (supports worth offering); plus the pet_household flag. No determination reads this answer.

Review status: pending external review · last updated 2026-08-28

pet_paperwork (v0.5.0 · module: About your pets · sensitivity low)

Question: Do you have a letter for your animals?

Evidence: Fair Housing Act, 42 U.S.C. 3604(f)(3)(B), and Section 504 of the Rehabilitation Act for federally assisted housing: a housing provider has to consider an assistance animal as a reasonable accommodation, and a letter is the usual way that request is supported. HUD FHEO Notice 2020-01, which set out how to assess such a request, was withdrawn by HUD on 2025-09-17 (Notice of the Withdrawal of FHEO Guidance Documents, hud.gov, withdrawing FHEO 2013-01 and FHEO 2020-01 among others) and no replacement has been issued, so what documentation a provider asks for is now less settled than the duty behind it; Field practitioner report (Elisa, 2026-07-28): the paperwork, not the animal, is the delay that loses units

Reason kept: Separates the household that already has a letter from the one that needs help getting one, so the offer is made only where it is useful. Asked only when a pet was reported. Changes no determination (tested). The withdrawal of FHEO Notice 2020-01 on 2025-09-17 makes the question more useful, not less: the statutory accommodation duty is untouched, but with no federal guidance on how a request is assessed, practice varies by provider, so a household that already has a letter is measurably further along than one that does not. HEAT states that position rather than implying a settled federal standard, and what the question asks is unchanged.

Alternatives considered: Ask everyone with a pet to obtain a letter (rejected): A letter is never a requirement of housing and asking that way would make it read like one. · Fold the paperwork question into the pet question (rejected): Two different facts with two different responses; folding them would ask about paperwork of everyone, including the majority with no animal. · Help text "...Tell the assessor what kind of animal and how many..." in every mode (v0.1.0) (replaced): 2026-08-10, practitioner flags 199 and 200 (Elisa). Flag 199: the help told the person to tell the assessor even when they were answering alone with no assessor anywhere, so the item now carries a self-administered variant. Flag 200: people confuse a service animal with an emotional support animal and guess wrong about who can write the letter, so the help now states the difference and names who can write the letter (any licensed health professional who treats the person), and both terms are also glossary entries carrying the fuller definitions. The glossary is deliberately not applied to question screens (question wording is never visually altered), so the fact people actually need had to be in the help text itself; the tooltips reach the same terms on the Questions page, the Evidence Register and check-your-answers. The help promises no outcome. The wording also matches the plural change to pets_any. Item version 0.1.0 to 0.2.0. · Purpose and evidence cited HUD FHEO-2020-01 as current authority for the accommodation duty (v0.2.0) (replaced): 2026-08-10, citation audit. HUD withdrew FHEO Notice 2020-01 (Assessing a Person's Request to Have an Animal as a Reasonable Accommodation Under the Fair Housing Act, issued 2020-01-28) together with FHEO Notice 2013-01 on 2025-09-17, in the Notice of the Withdrawal of FHEO Guidance Documents, and the documents were removed from hud.gov. Citing withdrawn guidance as though it were in force is exactly the failure this register exists to prevent, so the reason now rests on the statute the guidance only ever explained: the Fair Housing Act reasonable-accommodation duty at 42 U.S.C. 3604(f)(3)(B), plus Section 504 for federally assisted housing, with the status of the guidance stated plainly. The question is unchanged, no help text changed, and no determination moved: this is a correction to the published reason, not to what anyone is asked. Item version 0.2.0 to 0.3.0; register version 1.8.0 to 1.9.0. · “Do you have paperwork for them, like a service animal letter or an emotional support animal letter?” (v0.3.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). The worst string in the instrument at grade 11.9 in the second person and 13.1 in the third: seventeen words carrying two four-word program names. The stem is one short question now, “Do you have a letter for your animals?”, and the two program names move to the first line of the help text directly under it, kept in full because they are what a housing provider will ask for: “It can be a letter for a service animal, or one for an emotional support animal.” That is the ordinary hint-text pattern, and it is also what stops a six-line question heading on a phone. The help text keeps both definitions, the ADA phrase “trained to do a specific task” and the fact that any licensed health professional who treats the person can write the letter; it is five short sentences instead of two long ones (5.0). Item version 0.3.0 to 0.4.0. · "It can be a letter for a service animal, or one for an emotional support animal. If not, help getting a letter can be offered." in all three voices (v0.4.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). JUDGMENT. "Do you have a letter for your animals?" is a stem that needs its help text to mean anything, and the help never said what a letter is not. Vaccination records, a city license and a vet's invoice are all letters about an animal, and a yes on any of them withholds SS-R19, the offer of help getting the letter a housing provider will actually want. The help now says vet records or a license are not the same thing. "If not" also became "If there is no letter", because a bare "if not" after a two-item list is ambiguous about which not. Offer-only in every direction; the statutory reasoning about the Fair Housing Act recorded in this item's purpose is unchanged. Item version 0.4.0 to 0.5.0.

Known risks: The word paperwork can read as a hurdle. Mitigated by help text saying help getting a letter can be offered, that it is never required, and that there is no wrong answer.

What this answer changes: SS-R19 (supports worth offering). No determination reads this answer.

Review status: pending external review · last updated 2026-08-28

military_service (v0.5.0 · module: Adult context · sensitivity low)

Question: Have you, or anyone in your household, ever been in the United States military? Going to boot camp or basic training counts too.

Evidence: Veteran-dedicated resource system (SSVF, HUD-VASH, HCHV)

Reason kept: Gateway to a parallel resource system; asked of every adult, at household level. Not asked of anyone under 18: from instrument 0.15.0-pilot the item sits in the conditional module now titled "Adult context" ("Service history" at the time), entered by every age band except under 18. HUD's Core Elements of Coordinated Entry uses this exact skip as its worked example of assessment skip logic. A skipped answer is missing, not a no: no veteran flag is raised, and none is denied.

Alternatives considered: Discharge-status detail (rejected): Eligibility detail belongs to the VA conversation, not a universal screen. · "Have you ever served in the United States military?" (v0.1.0) (replaced): 2026-07-28, field practitioner feedback (Elisa): people who did not complete service, or who do not consider themselves veterans, answered no and lost the whole veteran resource system. v0.2.0 adds boot camp or basic training, which is the concrete fact rather than the identity. · "Have you ever been in the United States military, or been to boot camp or basic training?" (v0.2.0) (replaced): 2026-07-28, field practitioner feedback (flag #101): SSVF serves veteran families, so a non-veteran respondent with a veteran partner answered no and the household lost veteran-specific resources it qualified for. v0.3.0 asks the question at household level, and the flag reads "Veteran in household". · “Have you, or anyone in your household, ever been in the United States military, or been to boot camp or basic training?” (v0.3.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). Grade 9.5. The boot-camp clause, which exists because people who washed out of basic training often do not call themselves veterans, is now its own sentence rather than a third branch: “...ever been in the United States military? Going to boot camp or basic training counts too.” (5.3). Household scope unchanged. Item version 0.3.0 to 0.4.0. · No help text at all (v0.4.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). JUDGMENT, on a single gating item with a very large consequence: a yes opens SSVF, HUD-VASH and HCHV, which in most communities is a parallel and faster resource system. Two well-documented reasons to answer no wrongly had nothing on the screen to prevent them. Guard and Reserve service is the standard ambiguity in this question and people who served in either routinely say they were not in the military; and people discharged other than honourably often answer no because they have learned to expect the follow-up. Neither is what this item asks. The help now says the National Guard and the Reserves count, and that any length of service counts and so does any kind of discharge. The item continues to record a household-level fact and a skipped answer continues to be missing rather than a no. Item version 0.4.0 to 0.5.0.

Known risks: A veteran adult in the household of an under-18 respondent is not captured while the under-18 skip is in force; the youth-pathway routing that under 18 already triggers is where that household's veteran resources have to be raised instead.

What this answer changes: ID-R8 (options to discuss); SS-R3 (supports worth offering); plus the veteran flag.

Review status: pending external review · last updated 2026-08-28

barrier_active_eviction (v0.2.0 · module: Adult context · sensitivity low)

Question: Is there an eviction case against you in court right now?

Evidence: Legal-aid outcome literature: representation changes eviction outcomes

Reason kept: Triggers legal-aid offer; never gates anything (evictions are a banned screening factor).

Alternatives considered: Also ask about eviction history over past years, with a records-check resource (practitioner feedback, Josh Johnson, 2026-07-28) (undefined): Deferred to pilot data. An active court case drives a time-critical legal-aid offer; historical evictions drive no determination and their remedy (record mitigation during the housing search) is case-management work after a match. A history question would cost an item and risks reading as a burden score. A records-check resource cannot be promised uniformly across communities. · Ask minors about eviction cases (2026-07-30 under-18 audit) (undefined): Minors generally cannot hold leases, making an eviction case against them nearly unanswerable; the item now sits in the age-gated adult module, the same gate as military service. Owner decision. · No help text at all (v0.1.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION at both ends of a legal process. "Is there an eviction case against you in court right now?" raises the active_eviction flag and drives a time-critical legal-aid offer. People call a landlord's notice an eviction, which produces a yes with no case behind it; and people whose case has already been decided say no because it feels finished, which is the moment before a writ and the moment legal aid is most worth having. The help now says a case already filed in court is what is meant, that a letter or notice from a landlord is not a court case, and that a case a judge has already ruled on still counts. Offer-only, never a determination, and still not a criminal-history question. NR-11 records why historical evictions are not asked and is unaffected. Item version 0.1.0 to 0.2.0.

Known risks: None; screening-out suite guards.

What this answer changes: SS-R2 (supports worth offering); plus the active_eviction flag. No determination reads this answer.

Review status: pending external review · last updated 2026-08-28

disabling_condition (v0.2.0 · module: Chronicity documentation · sensitivity high)

Question: Do you have a condition that is expected to be long-lasting and makes it harder to live on your own? You do not need to say what it is.

Evidence: 24 CFR 578.3 chronicity requires a disabling condition

Reason kept: Statutory chronicity element; asked only when chronicity is live (module gate), yes/no only, never which condition (II.B.12.f).

Alternatives considered: Condition type detail (rejected): Forced diagnosis disclosure is prohibited and unnecessary. · Asked at the end of the assessment, with the other module questions (placement through instrument 0.20.0-pilot) (replaced): 2026-08-11, public register item 14 (Elisa): the long-lasting-condition question should sit next to the question about needing regular help, because that is how the conversation runs. It now does: when the housing history opens it, it is asked directly before the daily-help question, condition first. The gate, the wording and the yes/no form are unchanged, and the order-invariance proof in guards.test.ts shows the move touches no determination. The risk the reviewer named in the same breath, a disability question followed by a checklist of support needs, is why the support-kinds follow-up below it is unscored, is never a disability assessment, and says so on the page. · No help text at all (v0.1.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION and JUDGMENT, on the single most determination-ending answer in the instrument, asked with nothing under it. CH-R2 reads a no and returns "does not currently appear to meet" for chronic homelessness immediately, regardless of every duration answer given before it. The stem asks about "a condition that is expected to be long-lasting" that "makes it harder to live on your own", and three misreads of that were reachable. A person does not think of a substance use disorder, or of a mental health condition, as a "condition"; a person whose condition comes and goes does not read it as long-lasting; and a person who is in fact living on their own, which is most people answering, hears the second clause as false about them and says no, when the definition at 24 CFR 578.3 asks whether the condition substantially impedes the ability to live independently, not whether the person is currently failing. The help now names the kinds of condition without asking for any of them, says it counts if it comes and goes, and says it counts even for someone managing right now. The promise that no diagnosis is disclosed is preserved and restated in the help, per CPD-17-01. No option and no gate changed. That a determination-ending binary has no second look is filed as IR-35. Item version 0.1.0 to 0.2.0.

Known risks: None with yes/no form.

What this answer changes: CH-R2, CH-R3, CH-R6 (chronic homelessness); ID-R1 (options to discuss).

Review status: pending external review · last updated 2026-08-28

inst_stay_length (v0.2.0 · module: About the facility stay · sensitivity moderate)

Question: How long have you been in the facility so far?

Evidence: 80 FR 75791: institutional stays under 90 days with qualifying entry continue the episode; 90+ breaks it

Reason kept: Resolves chronicity and Category 1 through institutional stays.

Alternatives considered: No help text at all (v0.1.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION and RESPONSE MAPPING at a determination boundary. CH-R9b reads "90 days or more" and returns "does not currently appear to meet"; under 90 days with a qualifying situation before entry keeps the episode running at CH-R9a. "How long have you been in the facility so far?" does not say which stay, and a person who has been in and out of the same jail or the same hospital this year can answer with the total, which crosses 90 days on a stay that has not. And nobody carries their admission date in days: 90 is a number a respondent has to convert. The help now says this stay only, counted from the day of going in this time, and that 90 days is about three months. Both options are unchanged and so is the regulation behind them. Item version 0.1.0 to 0.2.0.

Known risks: None identified.

What this answer changes: HC-R2, HC-R2a, HC-R2b (HUD homeless category); CH-R9, CH-R9a, CH-R9b (chronic homelessness).

Review status: pending external review · last updated 2026-08-28

inst_prior_situation (v0.5.0 · module: About the facility stay · sensitivity moderate)

Question: Just before you went in, were you staying outside or in an emergency shelter? A place not meant for people to live in counts too.

Evidence: Same Federal Register rule: what came immediately before the facility decides everything

Reason kept: Companion fact to stay length.

Alternatives considered: "somewhere not meant for living" (replaced): 2026-07-28, field practitioner feedback (Daniel Gore): same phrase change as the chronicity items. · No help text (v0.2.0) (replaced): 2026-08-10, practitioner flag 189 (Elisa): this item carries the same phrase as the chronicity items and was the only one with no help text at all, so it got the examples sentence the others now carry. Item version 0.2.0 to 0.3.0. · “Just before entering, were you staying on the streets, in an emergency shelter, or a place not meant for people to live in?”, with the one-sentence place definition as help text (v0.3.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). The stem scored 9.8 and the help text 12.9. “Just before entering” has no object; it is “Just before you went in” now, the third branch moves to its own sentence, and “on the streets” becomes “outside” with the rest of the instrument: “Just before you went in, were you staying outside or in an emergency shelter? A place not meant for people to live in counts too.” (4.9). The chronic-homelessness institutional-stay rule this item feeds is unchanged. Item version 0.3.0 to 0.4.0. · "A place not meant for people to live in means a car, a van, a shed, or a garage. It also means a storage unit, an abandoned building, a park, or a bus or train station." as the whole help (v0.4.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION of the time anchor. 24 CFR 578.3(2) requires a qualifying situation IMMEDIATELY before entry, and CH-R9c reads a no here as ending the pathway. "Just before you went in" carries no unit: a person who spent months in a shelter and the last two nights at a sister's house can answer either way and be sure they are right, and the two answers produce different determinations. The help now opens by saying it means the night before going in. The definition sentences that were the whole of the help are unchanged and remain the shared wording used in every item carrying that phrase. Item version 0.4.0 to 0.5.0.

Known risks: None identified.

What this answer changes: HC-R2, HC-R2a, HC-R2b (HUD homeless category); CH-R9, CH-R9a, CH-R9c (chronic homelessness).

Review status: pending external review · last updated 2026-08-28

inst_discharge_soon (v0.1.0 · module: About the facility stay · sensitivity moderate)

Question: Do you expect to leave this facility within the next few days?

Evidence: CMS 42 CFR 482.43 discharge planning; round-3 external review

Reason kept: Plan-before-discharge urgency trigger (RA-R8). Deliberately does NOT change the band while inside: exposure model integrity.

Alternatives considered: Raise band before discharge (rejected): Mortality inside a facility is a different construct; the urgency flag prompts action without corrupting the model.

Known risks: None identified.

What this answer changes: RA-R8 (reassessment guidance). No determination reads this answer.

Review status: pending external review · last updated 2026-08-28

safety_leave_help (v0.3.0 · module: Support preferences · sensitivity high)

Question: Do you want help leaving the unsafe situation?

Evidence: 24 CFR 578.3 Category 4: "fleeing, or ATTEMPTING TO FLEE" domestic violence and other dangerous or life-threatening conditions; HUD CE and Victim Service Providers FAQs; trauma-informed practice: survivor choice, no required action

Reason kept: Splits the attempting-to-flee half of Category 4 out of the violence question so neither answer can hide the other. Wanting help to leave, or being unsure, is treated as attempting to flee; an explicit "not right now" is respected and removes Category 4 without removing any safety routing, support, or housing option (HC-R3).

Alternatives considered: Keep it inside the fleeing-violence question ("leaving, or trying to leave") (replaced): 2026-07-28: a compound question let a person who is being harmed but not yet leaving answer no, losing both the Category 4 test and the routing. · Yes/no only (rejected): Ambivalence is the normal state when leaving is dangerous. "I am not sure" is a real answer and is treated as attempting to flee, not as a no. · Infer intent to leave from the housing answers (rejected): Inference about a survivor's intentions is exactly the judgment HUD reserves to the person; self-certification governs. · Answer label "I am not sure" rendered in the first person under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.1.0 to 0.2.0. · "Do you want help leaving this situation?" and "Do they want help leaving this situation?" (v0.2.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). COMPREHENSION: an ambiguous referent on a sensitive screen. This item is reached only after a yes to the violence question, and "this situation" was meant to point back at that; but "situation" is also the ordinary word for homelessness, and it is the word the rest of the assessment uses for it, so a person can answer yes meaning they want help getting housed and be routed toward leaving a home they were not talking about. The stem now says "the unsafe situation", which is the exact phrase the next question in the same module already used, so the module now names one thing one way. The construct is unchanged: this is the attempting-to-flee half of Category 4 asked in the survivor's own terms, it gates nothing, and "Not right now" is still an answer that is respected. Item version 0.2.0 to 0.3.0.

Known risks: Asking someone to state an intention to leave can be dangerous if overheard. Mitigations: the assessor is told to ask this privately (high-sensitivity ask-privately hint), the question is skippable with no penalty, the answer is never printed, the resulting Category 4 is marked sensitive and stays off the printable card, and the inline quick-exit affordance sits directly under the question.

What this answer changes: HC-R3 (HUD homeless category).

Review status: pending external review · last updated 2026-08-28

dv_provider_pref (v0.3.0 · module: Support preferences · sensitivity high)

Question: Would you rather work with an advocate? An advocate helps people who are leaving an unsafe situation.

Evidence: CPD-17-01 II.B.10; HUD CE and Victim Service Providers FAQs: survivor chooses the process

Reason kept: Consent-based routing; routing is offered, never automatic.

Alternatives considered: Automatic VSP routing on any disclosure (rejected): Removes survivor choice; HUD guidance requires the person to choose. · "Either" as third option (replaced): HUD guidance frames it as both-together; 'either' read as no-preference. · No third-person variant of the help text; the second-person text rendered under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.1.0 to 0.2.0. · “Would you feel more comfortable working with an advocate who helps people leaving unsafe situations?” (v0.2.0) (replaced): 2026-08-27, owner readability standard: “especially for the client side of the assessment, even the sample card things should be easy enough that a 12 year old could fill it out.” A 12-year-old reads at roughly grade 6 to 7, so every client-facing string is now gated at Flesch-Kincaid grade 8.0 or below, with an average sentence of 20 words or fewer (packages/test-scenarios/src/readability.test.ts). Grade 10.7: fifteen words, seven of them three syllables or more. “Would you rather work with an advocate? An advocate helps people who are leaving an unsafe situation.” (6.5). The choice itself is untouched, and so is the help text that says it closes no doors. Item version 0.2.0 to 0.3.0.

Known risks: None beyond disclosure context.

What this answer changes: Context and cross-checks only; changes no determination.

Review status: pending external review · last updated 2026-08-28

housing_worked_factor (v0.3.0 · module: What worked before · sensitivity low)

Question: Looking back, what made it possible for you to stay housed?

Evidence: Strengths-based planning: the ingredients of past stability are the best predictor-free blueprint for the next plan

Reason kept: Names the single biggest ingredient of past stable housing; printed on the card so the next plan starts from it. Display only.

Alternatives considered: Multi-select of all factors (deferred): Single strongest factor is more actionable; multi-select UI is not yet built. · Free-text 'what made it work' (rejected): Free text cannot be deterministic or translated; the choice list covers the common ingredients with an Other. · Allow selecting multiple factors (practitioner feedback, Josh Johnson, 2026-07-28) (undefined): Kept single-select deliberately: asking for the single biggest factor sharpens the strength that leads the plan, and the interface has no multi-select control by design. Whether workers need a second factor is a pilot question. · Answer label "A person who supported me" rendered in the first person under a third-person question in proxy mode (replaced): 2026-08-10, practitioner flags 188, 190, 192 and 195 (Elisa): in proxy mode the question stem read in the third person and everything under it stayed in the second person, so one screen asked about "them" and explained it to "you" and a reviewer could not tell who the assessment was for. A third-person variant of every string the item renders. The second-person text is unchanged in every other mode. Item version 0.2.0 to 0.3.0.

Known risks: None identified.

What this answer changes: Context and cross-checks only; changes no determination.

Review status: pending external review · last updated 2026-08-28

housing_worked_when (v0.2.0 · module: What worked before · sensitivity low)

Question: About when was that stable place, and were you an adult then?

Evidence: Strengths-based planning: the recency of a protective factor changes what can be built on it, without changing whether it counts

Reason kept: Unscored recency context for the protective factor, adopted from public register item 1 as the alternative to narrowing the "ever" window. Prints in the what-worked planning text and in the action plan; no rule reads it and no determination moves.

Alternatives considered: Bound the parent question to the past three years instead (rejected): It would erase the strength for the people who have been homeless longest. Recency is worth showing and is not worth screening on, so it is shown. · Ask for a year (rejected): A year is a memory test on a question whose whole purpose is to find something that worked. Three bands answer the planning question (last few years, longer ago, childhood) and the third one answers the practitioner's actual worry directly. · Ask it of everyone rather than only after a yes (rejected): There is nothing to date for someone who has never had stable housing, and asking would make the absence conspicuous. The gate is the parent question's yes. · "An estimate is fine. It is asked so the next plan can build on what actually worked, and no answer here changes any result." (v0.1.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). RESPONSE MAPPING: two of the three answers were both true at once for a real case. "In the past 3 years" and "When I was a child or teenager" both fit a nineteen-year-old whose stable place was at sixteen, and the item exists to let a worker tell last year from childhood, which is exactly the distinction the overlap destroys. IR-26 records the item as double-barrelled and stays open, because the real fix is the option set. What is fixed here is the arbitrary half: the help now says that if it was in childhood, choose that answer even if it was recent, so two respondents in the same situation now give the same answer. No rule reads this item and no determination moves either way. Item version 0.1.0 to 0.2.0.

Known risks: A childhood answer could be read as making the strength count for less. It does not: nothing scores it, and the pages that print it treat it as planning context. Watch for it in cognitive testing.

What this answer changes: Context and cross-checks only; changes no determination.

Review status: pending external review · last updated 2026-08-28

imminent_housing_loss (v0.3.0 · module: About the place you are staying · sensitivity low)

Question: Are you about to lose the place you are staying within the next two weeks?

Evidence: 24 CFR 576.2 Category 2 (14-day imminent loss)

Reason kept: Category 2 gateway; same-day reassessment trigger; prevention/diversion discussion.

Alternatives considered: Ask it of everyone, in core (replaced): 2026-07-28, practitioner report: asking a person sleeping outside whether they are about to lose their housing is nonsense, and the answers were driving prevention and diversion recommendations for unsheltered people. The item moved into the housing_risk module, gated on living situations where the person still has a place to lose (staying with family or friends, own housing, somewhere else). Unsheltered, emergency shelter, and institution never see it, and an unasked item is missing, never a no. · No help text at all (v0.2.0) (replaced): 2026-08-28, misinterpretation audit (owner directive: "go through each question and think of ways that someone might misinterpret or answer in a way that isn't intended"). JUDGMENT, on an item that decides a HUD category. HC-R4 reads a yes and returns "possible category 2: imminent risk". The module it sits in is entered by people staying with family or friends, in their own housing, or somewhere else, and for that population the loss is usually a conversation rather than a document: a host saying "you need to find somewhere". With no help text, "about to lose" reads as requiring something official, and the people the category exists for are the people least likely to have anything official. The help now says being asked to leave counts even if nothing is in writing, restates the two-week window that is the category's own boundary, and says a place that can be stayed in for longer than that is not what is being asked, so the sentence does not push in only one direction. Item version 0.2.0 to 0.3.0.

Known risks: None identified.

What this answer changes: HC-R4 (HUD homeless category); ID-R6 (options to discuss); RA-R6 (reassessment guidance); plus the imminent_housing_loss flag.

Review status: pending external review · last updated 2026-08-28

support_types_needed (v0.1.0 · module: What kind of help · sensitivity moderate)

Question: What kinds of help would make it easier to keep housing? Choose as many as fit.

Evidence: Person-centred support planning: naming the kinds of help wanted is what makes a plan a plan; supportive-housing practice treats the list as a conversation, never as a profile

Reason kept: Unscored planning detail, adopted from public register item 2 as the alternative to replacing the assessed yes/no with a menu. Asked only after a yes, printed on the case manager's planning card as a starting list. No rule reads it, no band moves, and it is never a disability assessment.

Alternatives considered: Replace the yes/no support question with this list (rejected): The yes/no is what the support-planning rules read. A scored menu of support types is a disability profile sitting in a housing record, and it would let the number of boxes ticked move a band, which is not a thing HEAT will ever do. · Score the list, for example one point per kind of help (rejected): It would punish honesty and reward under-reporting on the most personal question in the instrument. Proven inert instead: every combination of this answer produces a byte-identical canonical payload. · Free text instead of a list (rejected): Free text cannot be translated or reviewed, and the results page already carries a free-text message to the next worker for anything the list does not hold. "Something else" is the seventh option here for the same reason. · Print it on the client's page as well (deferred): It is planning material for the conversation that follows, and the client's page already says what the person wants said, in their own words. Whether people want their own list back on their copy is a pilot question.

Known risks: Read carelessly, a list of support needs beside a disabling-condition question looks like a disability assessment. Mitigations: nothing is scored, the card says so in print, and the list never appears on the printable client card. If a pilot reviewer reports it being used as an eligibility screen, the surface comes off immediately.

What this answer changes: Context and cross-checks only; changes no determination.

Review status: pending external review · last updated 2026-08-28