Conformance up 22%.
The biggest fix was two fields on a form.
- Organization
- Solutions 2 Recovery
- Where
- Valley Village, Los Angeles
- Levels of care
- Detox, residential, PHP, IOP, outpatient
- Accreditation
- Joint Commission, DHCS licensed
- EHR
- Kipu Health, direct integration
- Patients in this story
- 18, under active treatment
- Checks run on their records
- 3,433
Solutions 2 Recovery takes people at the point where the other options have run out. The programme sits in Valley Village, north of central Los Angeles, and it runs the whole distance: medical detox, residential, partial hospitalisation, intensive outpatient and outpatient. A patient can move through four levels of care without ever leaving the organisation, and every one of those moves has to be justified in writing to somebody.
That is a well-run place by the usual measures. Joint Commission accredited, DHCS licensed, a quality process that predates us. What they did not have, and what almost nobody in behavioral health has, was a way to know on any given Tuesday whether the record in front of a payer would actually hold up.
Nine rulebooks, one chart
The difficulty with a full continuum is that the requirements stack rather than merge. A chart written for a residential patient at ASAM 3.5 sits under California's certification standards for the programme, under the ASAM criteria as whichever payer holds the contract chooses to interpret them, and under 42 CFR Part 2 because it is a substance use record. Move the same patient down to partial hospitalisation and the applicable rules change again, while the earlier documentation still has to stand behind the days already billed.
- Treatment planning and progress (SUD)
- California DHCS AOD certification
- ASAM criteria as payers apply them
- 42 CFR Part 2 readiness
- Consents and agreements
- Admission and assessment
- Medication and MOUD
- Discharge and transitions
- Incidents and safety
Fifty-eight controls in total, applied according to the level of care each patient was in rather than uniformly across the census.
The standard answer to all of this is a quality review that samples. Somebody pulls one chart in ten, works down a checklist and writes up what turns up. It is honest work, and it finds real problems, but it finds them one at a time. A reviewer working from a sample can tell you that a note is missing something. Only a reviewer who has seen every note can tell you that the same thing is missing from all of them, which is the difference between a correction and a fix.
What the full sweep turned up
In September we put the active census under continuous review, reading directly from Kipu Health, the EHR the programme runs on. Eighteen patients, 3,433 checks, every document measured against the controls that applied to that patient's level of care on that day. Roughly half came back clean, which is an unremarkable result for a first full sweep anywhere that has never had one.
The shape of the other half was the interesting part. Findings spread evenly across a clinical team look like a training problem. These were not spread at all. Five of the fifty-eight controls carried half of everything found, and the two largest were not clinical judgement in any sense. One was service modality, meaning whether the session happened in person or over video. The other was client location for synchronous services. Both are required under DHCS certification, and neither is a decision anybody makes at the bedside. They are fields that either exist on the progress note form or do not exist on it. They did not, so every note that went out failed both checks, every day, for as long as the form had been in use.
The gap between those two bars is not clinical performance, it is a form.
Which distinction matters a great deal to whoever has to act on the report. A conformance figure of 37% reads as an instruction to retrain the staff, and a month of everyone's time disappears into that. The same records at 81%, with two absent fields explaining the rest, point somewhere much cheaper. The clinical work was being done all along. The form was simply not asking for the proof.
We were confident in the clinical work. What we did not have was a way to see, every day, whether the record proved it. The useful surprise was how concentrated the gaps turned out to be, and that two of the biggest were fields missing from a template. Fixing that once takes a whole category of problem off the table permanently.
Three weeks later
Conformance across the reviewed records rose by 22% over the following three weeks, and none of that came from a policy rollout or a training day. Findings were arriving while patients were still in treatment, each one carrying the control it failed and the correction it needed, so a supervisor could hand the work to a named person that afternoon. The window in which a record can still be completed honestly is a short one and it closes at discharge, which is the whole reason retrospective audit finds so much that can no longer be repaired.
The problems that stop coming back
A finding you correct is worth one chart. A finding class you retire is worth every chart that comes after it. Sorted by how each group actually gets solved rather than by how often it shows up, the work at Solutions 2 Recovery falls into three piles of very different value.
- 01
Gone for good, roughly a third of findings
Put the two fields on the progress note template. The controls carry on running, but the condition that generated the finding no longer exists anywhere, so this class stops being produced rather than being corrected chart by chart.
- 02
Prevented while the note is being written, roughly another third
Plan of care mechanics account for this group: goals carrying measurable targets and dates, interventions with a frequency and a named owner, each session note tied back to the goal it serves. Enforced at the point of writing rather than audited afterwards, the gap closes before the document is ever saved.
- 03
Caught the same day, the remainder
Co-signatures, timeliness and the genuine clinical judgement calls stay human, and should. What changes is how long they stay invisible. These now surface within a day of the note being written, while the record can still be completed properly.
Two thirds of everything found sits in the first two piles, and neither responds to asking people to be more careful. What that changes, in practice, divides cleanly between the two things a provider of this size actually worries about.
Money that was quietly at risk.
Telehealth lines that could not be defended
Service modality and client location are not paperwork for its own sake. They are the two facts that make a remote session billable as a remote session and defensible if anybody asks later. With neither field on the form, every telehealth line carried that exposure quietly, and nobody would have discovered it until a payer or a contractor went looking. Once the fields exist, the exposure closes for every session that follows.
Program days that were never billable
Intensive outpatient pays by the program day, and the day only stands up if attendance and duration are on the record for each patient individually. A group that ran without its per-patient notes is not a denial waiting to happen, it is simply revenue that was never claimable in the first place. Surfacing that the same week means the note can still be written properly, which is the difference between recovering the day and losing it.
Level of care downgrades at concurrent review
Payers do not usually refuse a residential stay outright. They move the patient down a level, and they do it when the justification in the chart has gone stale relative to the day of the review. The plan of care findings in the second pile are precisely what keeps that justification current, which shifts the conversation with the utilization reviewer from argument to evidence.
Denials that no appeal can win
Medical necessity is the one denial category where the appeal frequently cannot help, because the missing evidence is clinical documentation and a record cannot legitimately be added to after the fact. Every finding closed before submission is a denial that never gets written, which is worth considerably more than a strong appeal after the fact.
Findings a surveyor can no longer raise.
The guaranteed DHCS finding
Certification surveys work from a pulled sample. When a required field is absent from the template, it is absent from every chart in that sample, so the finding is guaranteed before the surveyor has read a word of clinical content. Putting the fields on the form removes an entire line of questioning from the next survey, and it removes it for every chart rather than for the ones somebody remembered to fix.
The extrapolation multiplier
This is the item worth the most money and the one least often discussed. Medicaid and UPIC contractors extrapolate: they take the error rate found in a statistically valid sample and apply it across the whole population of paid claims. A scattered human error produces a modest rate. A systematic error present in every record produces a rate approaching total, and the recoupment that follows is sized against years of billing rather than against the handful of charts actually reviewed. Removing a systematic error does not just reduce the finding, it removes the multiplier that turns a small finding into a six-figure demand.
Part 2 and accreditation evidence
Consent records, disclosure tracking and the segmentation of substance use records are checked the same way and fail the same way, on evidence rather than on intent. So is the golden thread that Joint Commission surveyors trace from assessment through to outcome. Both were already in the review set here, and both improve through the same mechanism as everything else: findings arriving while somebody can still act on them.
The numbers behind this story
Adentris platform data, September 2026, covering 18 patients under active treatment. Aggregate only: no patient-level information appears here or in any Adentris material.
| Patients covered | 18 |
|---|---|
| Documentation checks run | 3,433 |
| Controls applied | 58 |
| Frameworks covered | 9 |
| Share of findings from the top five controls | 50% |
| Share from the two template fields | 32% |
| Share from plan of care mechanics | 36% |
| DHCS conformance, all controls | 37% |
| DHCS conformance, excluding the two template fields | 81% |
| Change in conformance, first three weeks | +22% |
Questions
01 How many patients and records does this story cover?
Eighteen patients under active treatment during September 2026. Adentris ran 3,433 checks across their records, applying 58 controls drawn from nine frameworks to whichever level of care each patient was in at the time.
02 What improved in the first three weeks?
Documentation conformance rose 22%. The clinical team was acting on findings the same week they appeared, rather than learning about them at a quarterly audit or in a payer's denial letter.
03 Were the gaps spread across the clinical team?
No, and that turned out to be the useful part. Two controls accounted for a third of everything found, and both were fields missing from a progress note template rather than clinical judgement. Set those two aside and conformance against the California DHCS framework was already 81%.
04 Which frameworks were applied?
Treatment planning and progress standards for SUD, California DHCS AOD certification, ASAM criteria as payers apply them in utilization review, 42 CFR Part 2 readiness, consents and agreements, admission and assessment, medication and MOUD, discharge and transitions, and incidents and safety.
05 How does Adentris connect to the EHR here?
Solutions 2 Recovery runs Kipu Health, and this deployment uses a direct integration with Kipu rather than the AI Web Agent layer. Where a direct integration exists it is the cleaner route, because records arrive structured and the review can run continuously against the live census. For systems without one, the AI Web Agent reads the EHR through its interface instead, which is what keeps Adentris workable on any EHR.