What thirty days of daily records add to an ADHD or autism assessment

Retrospective self-report fails in predictable ways in adults who have spent years compensating. The DSM-5-TR asks for evidence across settings and over time, and a single interview cannot provide it. A prospective daily record, made in the moment over thirty days, is a different class of evidence. It is no diagnostic instrument; it is the longitudinal, cross-setting observation the criteria assume and the clinic rarely has. This piece is about what such a record shows, how much of it is needed before it means anything, and where it stops.

Why does retrospective self-report fail in adult women?

Three reasons, all well described, all compounding.

The first is recall. Retrospective ratings of symptoms are reconstructions, weighted towards the salient and the recent. The ecological momentary assessment literature (Shiffman, Stone and Hufford, 2008, is the standard review) has shown for two decades that in-the-moment reports and later recall of the same period diverge, and that the divergence is largest for fluctuating states. Attention and energy are fluctuating states.

The second is masking. Women who reach assessment in their thirties and forties have usually spent decades compensating: scripting, over-preparing, suppressing, recovering in private. The consensus statement on ADHD in girls and women (Young et al., 2020) and the camouflaging literature in autism (Hull et al., 2017) describe the same outcome from two directions: the presentation in the room understates the cost outside it. A client who arrives on time, articulate and composed is not evidence against the referral question. She is often evidence of the effort.

The third is the room itself. A fifty-minute session is the least representative sample of a client's week that exists. It is scheduled, structured, and conducted with a professional whose attention she is managing. The settings the criteria care about (work, home, relationships, the unstructured hours) are exactly the ones the room cannot see.

Standardised questionnaires do not escape this. The ASRS and its relatives are retrospective by construction; they ask the client to summarise six months in one sitting. They remain necessary, and none of what follows replaces them. The point is that they share a failure mode with the interview, and adding more of the same does not correct it.

What does a prospective record show that a questionnaire cannot?

Variability, and what it co-occurs with.

A questionnaire yields a level: "often", "sometimes". A daily record yields a shape: which mornings were at 1 or 2, whether evenings track mornings or diverge from them, whether one weekday sits a full point below the others, whether the ten days before menstruation look like a different person. Levels are what masked adults are good at reporting plausibly. Shapes are hard to fabricate and easy to forget.

The record also carries context the client attached at the time. A day logged as "socialising, drained" followed by two low mornings is a small piece of evidence about social cost that no questionnaire item captures, because the item would have asked her to estimate it in general, and in general she copes.

For women specifically, the cycle dimension is the one most assessments never reach. Roberts, Eisenlohr-Moul and Martel (2018) reported that ADHD symptoms in women vary with estradiol and progesterone across the cycle, with the late luteal phase the most affected. In a thirty-day record with period dates, that becomes visible as a phase comparison: mean energy in the luteal phase against the follicular, with the number of days in each. It is not a finding to put in a report. It is a question to bring to the next session that would otherwise not have been asked.

How much data is enough to read a pattern?

This is where most attempts at "bringing data into the clinic" go wrong, in both directions: either everything is treated as signal, or nothing is trusted because it is self-report.

The rule that works is to read adherence first. Before any pattern, the summary should say how many days were logged and where the gaps are. Six check-ins in thirty days sustain no pattern at all, and a chart drawn from six points is worse than no chart, because it looks like evidence. Twenty-four logged mornings out of thirty can be read.

Below that, each pattern needs its own minimum. In the summary Lira produces, these are fixed and visible: a "low mornings" flag needs at least ten logged mornings; a cycle-phase comparison needs at least seven days in each phase; a weekday effect needs at least three records on that weekday; an evening-drop flag needs ten days with both check-ins. When a threshold is not met, the flag does not appear. When it is, the rule is printed next to it: "energy at 1 or 2 in 43% of logged mornings (threshold: 40%)". The clinician can disagree with the threshold. She cannot mistake it for a score.

Which patterns are worth bringing into the session?

Four, in roughly this order of usefulness.

Energy by weekday. Seven numbers. A Tuesday at 1.8 against a week that averages 2.9 is the fastest route into a conversation about what Tuesdays contain, and it is a conversation the client did not know she needed, because she does not experience her week as a distribution.

Drains and restorers. The activities logged as costing energy, ranked by how often, against those logged as restoring it. In Lira these are classified into clinical categories (social masking, sensory overload, context switching, hyperfocus, rumination, poor sleep; time alone, nature, movement). Eleven records of "social masking" as a drain in a month is not a diagnosis of anything. It is a well-documented pattern arriving with a count attached.

Sleep complaints, as counts. "Woke up several times: 14 of 24 mornings" is more useful than "I don't sleep well", and more useful than a sleep questionnaire completed once, because it dates the bad nights against the bad days.

Energy by cycle phase, when the client tracks it, with the context she declared (hormonal contraception, PCOS, perimenopause) shown first, because it changes the reading of everything below it.

The self-identification profile the app offers (five self-report dimensions: attention, energy recovery, sensory sensitivity, social masking, emotional regulation) is also available if the client shares it. In our review with clinicians it was rated neutral: worth a glance, not a block to build on. It is labelled as self-report without diagnostic value, and it should be read that way.

A worked example

A client, 38, referred for ADHD assessment, has been logging for five weeks and shares the last thirty days. Adherence: 24 of 30 mornings, 14 of 30 evenings, longest gap three days. Mean morning energy 2.7, mean evening 2.1; 43% of logged mornings at 1 or 2. By weekday, Tuesday at 1.8, the rest between 2.6 and 3.4. Sleep: "woke up several times" on 14 of 24 mornings. Drains: social masking (11), sensory overload (8), context switching (6). Restorers: time alone (9), nature (7). Cycle: hormonal contraception declared; luteal mean 2.1 against follicular 3.2, with 12 and 9 days respectively. Two flags: low mornings, and a luteal drop of 1.1 points against a 0.8 threshold.

None of this diagnoses ADHD. All of it is admissible in the anamnesis as observation, and every line of it is a better question than "how have the last two weeks been?". Tuesdays turn out to be the day of a standing team meeting she has been dreading for a year. The luteal drop is something she had noticed and dismissed as "everyone gets that". The masking count gives her permission to describe the cost of the meeting rather than the meeting. The assessment proceeds on the same instruments it would have used anyway, with better history behind them.

What are the limits, and what must never happen?

The limits are the ones already stated: self-report, no norms, no threshold with diagnostic meaning, dependent on adherence, and readable only in the context the clinician brings to it. Anyone who presents thirty days of energy logs as evidence for a diagnosis has misunderstood both the logs and the diagnosis.

Two things must never happen, and in Lira for Professionals both are prevented by design rather than by policy. The records never enter the report or any generated text; what the clinician concludes goes into her own notes, in her words. And nothing is interpreted by a language model: every flag is a fixed rule with its threshold printed, because a clinician reading "the pattern suggests" next to a client's name has been handed an opinion with no author.

How does consent and revocation work?

The client decides, in the app, what to share (energy and sleep, activities, cycle, self-identification), for how long, and with whom. Free-text notes and reflective content are never part of it. The clinician receives a read-only summary in the client's file and nothing else; colleagues in the same clinic see that a link exists and cannot open it. The client sees the same screen the clinician sees, sees when it was last opened, and can revoke at any time; access ends immediately, and the clinician keeps only the notes she wrote. Sharing has an end date the client chose, and renewal is her decision. Every access is logged.

The clinician can also initiate: from the client's file, she can generate a link inviting the client to use the app and share records with her, useful when she wants a month of records before the assessment rather than after. The client decides in the app; the clinician learns nothing until she does. What the client sees and controls on her side is described in what your psychologist does with thirty days of records, from the client's side.

Thirty days of records do not shorten an assessment. They change what the first hour is spent on.