For public benefits programmes
No eligibility determination, no benefit calculation, no denial, no termination, no sanction, and no adverse action of any kind. A system that made one of those would be wrong on the day it was right, because a household would have no person to appeal to. What can be carried is the document work that stands between a caseworker and the decision they are qualified to make.
The standard is a number of days, set in regulation, and it does not adjust for a hiring freeze, a policy change, an economic shock or a new verification requirement. When any of those arrives, the standard stays and the queue lengthens.
Most of a caseworker’s day is not deciding. It is establishing whether an application is complete, identifying which verification is missing, contacting a household that changed phone numbers, re-requesting a pay stub that arrived unreadable, rescheduling an interview somebody could not attend because they could not leave work, and re-entering information that already exists in another programme’s file.
The households are the least equipped population to navigate a document process. Somebody applying for food assistance is often also moving, also working an irregular shift, also without a printer, and also managing a health situation. A request for one more document is not a small ask, and the failure to answer it is usually not refusal.
Churn does the rest of the damage. A household loses benefits at redetermination for a paperwork reason, reapplies within weeks, and the full intake burden runs again — for the agency and for them. Nobody chose that outcome and everybody pays for it.
And the consequence of the queue is measured in the wrong unit. Reported as processing days, it sounds administrative. Experienced by a household, it is a month without rent assistance, and the appeal that follows is another proceeding for an agency that is already behind.
A caseworker cannot decide until a file is complete, and completing a file is a chase across several parties on several clocks — so the timeliness standard is consumed almost entirely by work that requires no professional judgement at all.
That imbalance is why hiring more caseworkers helps less than expected. It adds decision capacity to a queue whose delay is upstream of deciding, so the new capacity waits along with the old.
The first act, therefore, is to measure the file rather than the caseload. For a sample of applications: how many days from receipt to complete, how many verification requests were issued, how many were issued sequentially rather than together, how many households became unreachable, and how many redeterminations failed for a document reason. Most agencies have the caseload number and not one of these.
What can then be operated is the completeness band. Checking an application against the programme’s own written requirement list at receipt. Naming every missing item at once instead of discovering them one at a time. Reminding a household before a deadline in the channel that actually reaches them. Confirming that a document received is legible and is the thing requested. Answering the status question so it does not consume a caseworker.
What never moves, at any tier, for any efficiency: eligibility determination, benefit calculation, denial, termination, sanction, overpayment finding, fraud referral, and any exercise of the discretion a caseworker holds. Those belong to a person the household can name and appeal to. That is not a constraint we are working within — it is the reason the arrangement is legitimate at all.
Days from application receipt to a complete file — measured by elapsed days to completeness, separated from days to determination, against your own baseline.
Verification requests issued sequentially rather than together — measured by count of separate requests per application, before and after.
Documents returned as illegible or as the wrong item — measured by count of re-requests caused by a document problem rather than by a household not responding.
Households that become unreachable during processing — measured by count reaching a terminal unreachable state, and at which step it happened.
Redeterminations lost for a paperwork reason — measured by count of closures for procedural rather than substantive reasons, and reapplications within ninety days.
Caseworker hours spent on completeness rather than determination — measured by time-on-task sampling across both bands, taken the same way before and after.
Status enquiries reaching a caseworker — measured by count of contacts asking only where an application stands, which is answerable without a caseworker.
any eligibility determination, benefit calculation, denial, termination, sanction, overpayment finding, fraud referral, or exercise of caseworker discretion. Nothing here scores, ranks, prioritises or triages a household by any predicted characteristic, and nothing assesses what a document proves. Those are the job, they belong to a named public employee a household can appeal to, and no efficiency justifies moving them.
The requirement lists come from your own policy — the programme manual, the notice text, the verification matrix your workers already use. Nothing here writes a rule. Where two of your own sources disagree about what is required, that disagreement is reported as a policy question rather than resolved, because resolving it would be setting policy.
Everything is written back into the eligibility system of record. No second record of a household is created. A second record is how a fair-hearing officer ends up with two versions of what an agency did, and there is no efficiency worth that.
Household contact is through the channels you already operate and in the languages your programme already supports. Language access is scoped explicitly rather than left to who is on shift, and every household-facing surface is built to be usable by somebody with a screen reader on an old phone.
Where an interface into the eligibility system does not exist, that is a limitation and the step stays with your staff. Automating a screen against an eligibility system is a fragile process operating on the most consequential record an agency keeps, and it should be refused rather than engineered.
This is among the most sensitive data any government agency holds: income, household composition, health circumstance, immigration-adjacent information, and the fact of applying at all. It stays inside your tenancy, on your retention schedule, exportable by you, and it is not used to train anything serving another organisation. That boundary is architectural rather than contractual language.
The fair-hearing record is the artefact that matters. Every action carries what was done, under which written grant, by which actor, at what time, against which application, with what result — produced during the work. When a household appeals, the agency needs a record that shows what was requested and when, not a reconstruction.
Programme-specific obligations are agreed in writing before anything runs. Federal benefit programmes carry data-sharing and safeguarding requirements that differ by programme, and where a requirement cannot be met, the honest answer is that the scope closes rather than that the requirement is interpreted generously.
And the boundary is enumerated rather than described: an explicit list of administrative acts, with everything else — including anything unclassifiable — routing to a caseworker by default.
Programme integrity’s question is whether anything can affect a determination, directly or by shaping what a caseworker sees. The answer is that the administrative band is enumerated and contains no assessment of any kind — a document is checked for legibility and identity, never for what it establishes.
Counsel and the hearing officer should review the record format before anything runs, against a real appeal. If the record would not let a hearing officer establish what was requested and when, that is a reason to stop rather than a detail to improve later.
Civil rights review matters here more than in any other page of this architecture. Anything that changes how quickly different households move through a process is a civil rights question, so nothing here prioritises, and the measurements are reported in a form that lets your own analysts test for disparate effect rather than take an assurance.
And accessibility and language access are delivery requirements. A household-facing surface that a person cannot use has not been delivered, whatever it does.
One programme’s intake, sampled over a defined period, measured from receipt to file completeness — with no authority to contact a household or touch a record.
The observation phase separates two numbers that agencies usually report as one: days to a complete file, and days from a complete file to a determination. The split almost always shows that the standard is being consumed upstream of any professional judgement, and that finding belongs to the agency whatever happens next.
Stopping there is a legitimate outcome. An agency that learns most of its timeliness gap is completeness rather than decision capacity has a different hiring conversation, a different training conversation, and possibly no vendor conversation at all.
If it continues, the first grant is one act — completeness checking at receipt, with every missing item named at once — for one programme, with the determination boundary written and reviewed by counsel before a single application is touched.
So have we, and the documented failures share a shape: a system was permitted to make or shape a determination, and the household had no person to appeal to. That is why the boundary here is a refusal rather than a safeguard — no determination, no calculation, no adverse action, no prioritisation, and no assessment of what a document proves. If a proposal ever includes one of those, it is the wrong proposal and it should be refused rather than fenced.
It is the distinction the whole design rests on, so it is worth being precise. Checking completeness asks whether the items on your written list are present and legible. It does not ask whether a pay stub establishes income, whether a household composition qualifies, or whether a circumstance is verified — those are assessments and they belong to a caseworker. If your requirement list cannot be applied without judgement, then it is not a completeness list and that programme is out of scope until it is.
That is a real risk and it is why unreachable is a reported state rather than a silent lapse. The design change is not more contact, it is fewer rounds: every missing item named at once instead of discovered one at a time, and a reminder before a deadline rather than a notice after it. If the measurement shows contact volume up and completeness time flat, the arrangement is not working and that should end it.
Then that is established before an engagement and may close it, which is the correct outcome. Federal benefit programmes carry safeguarding requirements that differ by programme, and the honest position is that some scopes we should not take. What we will not do is interpret a requirement generously to keep a scope alive. Determine the obligation first; if it excludes us, it excludes us.
Correct, and it is why nothing here prioritises or triages, and why the measurements are defined so your own analysts can test for disparate effect rather than accept an assurance from us. The observation phase produces those measurements before anything is delegated, so the baseline exists independently. If the analysis after a bounded first scope shows differential movement that cannot be explained by the written rule, that is a reason to stop.