Assessment Builder Research

Part of the Anaya Care Handbook. This page records the research and the plan behind the Assessment Builder (rules ABLD-1 – ABLD-33). The rules live on that page; this one explains why they are what they are, and the order to build them in.

Superseded in part, 2026-10-01. The owner chose the One platform design below over the additive one it recommended: every assessment becomes a template, the six care assessments move into the Anaya library as locked instruments under licences the owner settled (the Meal Assessment without the MNA-SF or IDDSI), the initial assessment and the care proposal each run on a template the agency chooses, value-added services give way to each agency's service add-ons, and the Simplified Routine Task Inventory's score no longer sets prices. This is template-driven intake, live since 2026-10-06; its rules are ABLD-34 – ABLD-39 and the rewritten ABLD-1, ABLD-13, ABLD-15 and ABLD-19 – ABLD-21 on the Assessment Builder page. The research below is kept as it was written; notes dated 2026-10-01 mark where the decision moved past it.

Research done 2026-09-29. Ten research threads (four through Anaya's own product, six across the web), three competing designs, three independent reviews of them, and a separate check that tried to disprove every claim below that the plan leans on. Claims that couldn't be confirmed from a primary source are marked. Nothing here is legal advice: where a licence is described, the page says what the rights holder's own terms say.

The short answer

(2026-10-01: superseded — see the note at the top of this page.) Build it — beside the six care assessments, not instead of them. Agencies need their own forms (state forms, franchise forms, payer forms, supervisory visits), and every serious competitor offers some way to make them. But Anaya's six care assessments and the intake do real work that a generic form engine would put at risk: the Simplified Routine Task Inventory sets proposal prices, the Meal Assessment's swallowing answer keeps food away from clients who take nothing by mouth, and the care plan is shaped around the inventory's sections. So agency assessments are a new, separate kind of record, filled from versioned templates, whose completed answers reach Anaya's plans as labelled context — and nothing that works today changes shape.

The experience is designed first: a structured card canvas with an outline and a settings panel for building, one long sectioned page for filling on a tablet, and Anaya as a co-author whose every contribution is visible, sourced and undoable. The experience was agreed on a clickable prototype before it was built; the Assessment Builder page now shows the real screens.

Checking the product along the way turned up a few things that need their own fixes before the builder; they are recorded as known gaps.

What the handbook said, and what changed

Until now the handbook called an agency-configurable Assessment Builder "a possible future direction" and left it as an open decision (AS-26, AMODE-7). This research settles the part that could be settled: agency assessments are built, on the client record; the intake stays the fixed standard form, with reopening it kept as a decision for later. (2026-10-01: reopened and decided — the intake runs on a template the agency chooses.)

What agencies are required to record

No two rule sets ask for the same form, which is the strongest reason the builder exists. Timing, required contents, signers and review schedules differ by state, payer and franchise.

Sources: the state administrative codes (Texas 26 TAC 558.404, Washington WAC 246-335-440, New Jersey N.J.A.C. 13:45B-14.9, Georgia 111-8-65-.11, Minnesota 144A.4791, Illinois 77 IAC 245.210, New York 10 NYCRR 766.3), 42 CFR 441.301 and 441.725, and 26 USC 7702B, read on 2026-09-29. Arizona, Ohio, the Joint Commission, several franchise sites and Australia's Support at Home pages could not be read and are not included.

What most of them share: functional limits per daily activity, the services with their frequency and times, equipment, nutrition and allergies, mental or cognitive status, home safety risks and what's being done about them, goals, the family's part, an emergency or contingency plan, and signatures. That list is the backbone of the requirement packs planned for the last phase.

One tension to design around: regulators want the client or family to sign and receive the plan, but CA-6 keeps assessments staff-only. The shared, signed document is therefore the service plan or care proposal, never the raw assessment (ABLD-29).

How other platforms do it

Almost every home care platform reviewed lets an agency make its own forms, but they split in two: self-serve builders (ShiftCare, CareSmartz360, AxisCare, HHAeXchange's clinical product, AlayaCare, Access Care Planning, CareLineLive, PointClickCare) and vendors that build the agency's forms for it, often for a fee or after a wait (Alora, older HHAeXchange forms, and reviewers of KanTime and Axxess). The strongest template libraries are in the UK (Birdie ships more than 25 validated assessments co-designed with agencies).

What they agree on: drag-and-drop fields, conditional questions, required fields, e-signature, pre-filling from the client's profile, and a published version that never changes, with earlier forms kept on the version they were filled on.

Where they have gone with AI in 2025–2026: Birdie SmartPlans drafts answers into the agency's assessment with each answer cited to the moment in the conversation, gives no AI answer on high-risk questions, and saves nothing until the assessor confirms. AlayaCare's form assistant maps dictation to fields for review, and its care plan agent needs a person's final yes. WellSky reports cutting about three hours of intake to one with ambient documentation. ShiftCare drafts forms from a prompt or a photo of a paper form. PointClickCare turns assessment answers into suggested care plan items — but only once the assessment is locked.

The gap Anaya can fill: every AI filling tool reviewed fills the vendor's templates. None fills an agency's own template from a conversation. Anaya already has a grounded, durable interview; running it against any agency template, with each answer cited, and feeding completed answers into Anaya's care plans, is a real difference — as is turning a photographed paper form into a checked template in minutes without a services fee.

What makes a builder easy

For the person building. Agency owners build templates rarely, so every way in must be visible: a + between questions, with / as the shortcut for people who build often (Tally, Airtable, SurveyJS, Paperform). Questions are edited in place and look exactly as the care manager will see them (research from the UK Ministry of Justice's form builder supports this). Settings that would clutter the canvas — points, "Not applicable", show-when rules, what an answer means to Anaya — live in a panel for the selected question. Show-when rules are written as sentences, with one list of every rule; a node-graph flow map is avoided because it breaks down on long forms (Typeform's own community reports this at 100 questions). Publishing follows proven patterns: a draft with an "unpublished changes" badge and a history where restoring doesn't publish (Tally), filled records kept on the version they started on (ODK), and a review that flags changes that would break answers already given (REDCap's draft mode).

We considered a document-style editor, where the whole template is one rich-text page, using the editor Anaya already has. We chose the card canvas: the editor's known stability issues, its text-level undo, and drag-and-drop that can't be done by keyboard would all have had to be worked around, and a free-flowing document invites prose that Anaya can't use as structured answers. The canvas keeps what makes a document pleasant — typing in place, Enter for the next question, / to insert.

For the person filling. Care managers are expert, repeat users: one question per screen would slow them down and hide the overview (Nielsen Norman Group on wizards; GOV.UK allows grouping questions for staff). So the form is one long page with a sections rail and each section's state; answers save as they go; "required" is checked only when marking the assessment completed, with an error summary that links to each gap (GOV.UK); "Not applicable" and "Not evaluated" are real answers (REDCap's missing-data codes); answers carried forward show where they came from and ask to be confirmed (WCAG 2.2, 3.3.7); and a check-your-answers page ends it. On a tablet, targets are at least 44 px, nothing depends on hover, and a lost connection keeps changes on the device with an honest message.

AI that helps without taking over

Three different jobs, kept separate:

1. Anaya builds the template — from a description, a Word file, a PDF or a photo of a paper form. Every major form builder now does this; the difference is control. The best designs let the author keep parts of the draft (Google Forms' tick-to-insert suggestions, Microsoft Forms' "keep all or some"), while builders that apply AI edits directly have a documented case of an agency's form being wrecked and needing revision history to recover. Clinical documentation tools add useful ideas: a plain "what this captures" hint and synonyms per field (Microsoft's Dragon Copilot), and testing a template against a sample conversation before rollout ("Tune accuracy"). Imports are best-effort — vendors say tables and scales break — so every imported question shows its source and needs keeping.
2. Anaya fills an assessment — from a recorded or written conversation. Independent 2025–2026 studies of AI scribes reported errors in about a quarter of the key facts they documented, with omissions making up three-quarters of those errors, and a verified failure in about a third of notes from three deployed tools. So suggestions carry their quote, what wasn't said stays Not discussed rather than guessed, gaps get a short guided follow-up, high-risk questions get quotes only, and completion waits until every suggestion is reviewed. No vendor reviewed shows a per-field confidence percentage; they use citations and blanks instead, and so does Anaya.
3. Anaya uses the answers — in its plans; today the care plan only (ABLD-19). Axxess and AlayaCare both keep AI-generated plans as drafts a person must approve; Anaya already does the same.

What the library can include

Many well-known assessment tools belong to someone, and permission to use one on paper in practice is not permission to build it into commercial software. The table records what each owner's own terms say.

Sources: phqscreeners.com form PDFs, web.stanford.edu/~yesavage/GDS.html, cdc.gov agency materials, iddsi.org guidelines for use, the British Pain Society reprint, mna-elderly.com (forms and FAQ), the Allen Cognitive Network RTI-E manual, mocacognition.com/permission, parinc.com, hign.org reprints, eprovide.mapi-trust.org, bradenscale.com and catalog.interrai.org, read on 2026-09-29. The Norton scale, the UCLA loneliness scale, SLUMS and the Cornell scale rest on secondary sources only.

What this means for the builder: the library records each template's licence and refuses a template whose licence isn't confirmed (ABLD-27); credit lines are shown with the template on screen, and will be on every printout once agency assessments can be printed; Anaya refuses to reproduce a licensed check's questions when asked to "make an MMSE" (ABLD-23); and the outside result type captures scores Anaya can't embed (ABLD-28).

What the design has to protect inside Anaya

Some of Anaya's features read particular answers from the six care assessments and act on them directly. These are why the six stay as they are:

• Prices. The Simplified Routine Task Inventory's answers add up to points that raise a proposal's hourly rate (up to 50%) and set its severity band. Any change to the inventory's answers or shape would silently change prices. (2026-10-01: the uplift and the severity band are removed with template-driven intake; stored prices are kept as they are, and only service add-ons carry a price on top of the base rate — ABLD-13, ABLD-21.)
• Safety. The Meal Assessment's swallowing answer keeps meals and food activities away from a client who takes nothing by mouth, in five places, and its allergies and texture levels screen activity sheets. A template answer must never be mistaken for these (ABLD-15). (2026-10-01: these become named questions of the Meal Assessment instrument, the only answers the checks read.)
• The dementia check. Whether the inventory says dementia is confirmed decides whether the Allen Cognitive Level shapes activity recommendations. (2026-10-01: read from the Simplified Routine Task Inventory instrument's Confirmed dementia question; unanswered means not known, never No.)
• The care plan's shape. The care plan is organised around the inventory's daily-living sections, so agency answers can inform its wording but not add sections (ABLD-21).
• The intake chain. The intake's inventory is copied without loss into the proposal and then the client; a second shape would have to travel all three steps. (2026-10-01: replaced by the intake and proposal templates, with a lossless draft copy or Anaya's pre-fill at convert and a copy of the proposal's answers on the new client — ABLD-35 – ABLD-37.)
• Who can read what. Agency answers themselves are never shown to care providers, representatives or medical professionals (ABLD-20). They reach only the wording of the care plan (ABLD-19, ABLD-21); client memory, which those roles can read, does not take them in.

Two smaller findings shape the plan. Anaya's AI tools have a size limit on what they can reliably ask the model to fill in one go, so Anaya fills a template one section at a time rather than all at once. And the save pattern the dashboard uses today sends the whole form and keeps a local copy only after a successful save — so the builder needs a save queue that keeps unsent answers on the device before sending, or a home visit with a poor connection could lose exactly the answers that weren't sent (ABLD-17).

Options considered

Three designs were drawn up in full and scored by three independent reviewers — one for the experience, one for engineering risk, one for product, compliance and scope.

The recommendation combines them: the additive design as the foundation, the experience-first design's builder and filling screens on top, and the one-platform design's safety checks and wording rules. Moving the six onto the engine is not scheduled; it can be revisited once there's evidence from agencies using the builder. The reviewers' main corrections are already in the rules: agency answers shape only the wording of the care plan, and the answers themselves are never shown to care providers, representatives or medical professionals (ABLD-20); a save queue that never loses unsent answers (ABLD-17); AI never writes into a draft the author is editing (ABLD-22); a completed assessment opens read-only (ABLD-18); and the safety-overlap warning is visible in the builder, not just stored (ABLD-15).

Rollout

Each phase ships on its own and ends with people using it, not only with code.

Where it stands (2026-09-30): phase 1a is built, with completed answers reaching the care plan only: client memory was left out on 2026-09-30, because linked care providers, representatives and medical professionals can read it (ABLD-19). Of phase 1b, Describe it and the licence guard are built; the meal plan and care provider tasks are not yet opened to agency answers, and wait for the check against clients who take nothing by mouth. Phase 2 is built, apart from testing a template against a sample conversation and the locked CDC checks. The builder is released to customers now, so the "Done when" measures can be taken as owners and care managers use it; none has been yet. Of phase 3, only care topics on questions exist so far, and they label a question without yet sending its answer to a part of the plan.