Privacy clocks
Right-of-access requests, breach clocks, and review replies.
A fair practice, run by the rules you set. Compliance, specialty billing, care management, home care, and money recovery for practices, clinics, and agencies, and sixty checks in all, now with pharmacies and hospitals. It does not bill, file, or send.
Right-of-access requests, breach clocks, and review replies.
Work RVUs per provider, tiered pay, and the draw true-up.
Exclusions, licenses, staff health, associates, risk analysis, training, physician contracts.
Chronic care, remote monitoring, and primary care management time.
PT, OT, and speech units, ABA authorizations, behavioral health.
Election notices and the cap, home health periods.
Attribution and capitation against your roster.
EVV against claims, caregiver hours, and plan-of-care hours.
Surprise-billing disputes, secondary claims, recoupments, code updates, wound products, liens.
The packages rolled out for a billing company or MSO and its client practices.
Support after setup: $300 a month per package, or $750 a month for three or more. Monthly exclusion screening runs are included for the compliance pack. A billing company partner rollout continues at $2,000 a month.
Column names were written from common practice exports and have not yet been checked against a live export; the first setup confirms them with you. Payer and state rules are settings you confirm, not advice.
The compliance pack reads staff and vendor files only. Every other package reads billing files with patient details: they stay on your machine, setup is with you over a shared screen, and a business associate agreement is signed before we ever hold a file. Masking is a design choice, not a HIPAA determination, and nothing here is billing, coding, legal, or clinical advice.
Everyone you pay against the exclusion files you download.
Licenses and certificates expired, expiring, or missing for the role.
TB tests, vaccines, fit tests, and screens due. HR records, not care.
The 30-day clock, extensions, and fees for patient access requests.
Risk assessments not done and notice deadlines coming due.
Vendors that touch patient data with no agreement on file.
The annual evidence checklist and high risks with no owner.
Required courses by role and state, and refreshers overdue.
Drafted replies that would confirm someone is a patient. It does not post.
Payments to referral sources against signed, current agreements.
Units delivered against authorized hours and supervision.
The 8-minute rule, the KX threshold, and discipline modifiers.
Election notices, benefit periods, and the aggregate cap.
Late admission notices, low-use periods, and overlaps.
Session minutes against codes, add-ons, and yearly limits.
Units against product size used and wasted, and invoices.
Open balances by attorney and case, and settled cases not paid.
Minutes per patient per month against each code.
Days of readings against each device code.
Monthly codes against tiers and consent.
Patients the payer lists that you never saw, and the reverse.
Per-member payments against the roster, month by month.
Claim lines with no verified visit, or more units than verified.
Overtime, travel time, split shifts, and missed breaks.
Hours delivered against hours authorized, week by week.
Negotiation and dispute windows, and offers below the QPA.
Balances left after the primary paid, with no secondary sent.
Takebacks matched to the original claim, and appeal windows.
Deleted and changed codes still in your fee schedule and open claims.
Work RVUs from charges, and the draw true-up. It does not run payroll.
Each one does one job, on your computer, from a file you export. There is no login to your EHR, a clearinghouse, a payer portal, or a state system.
Wave 1: Measured in the final pass, 19 Nov 2026
On fictional test data, 19 Nov 2026. Every agent measured: each problem planted in its test files found and nothing else reported, no finding on the same files with the problems repaired (2,297 rows, most of them the provider pay charges and remote monitoring readings), and each one-fact change caught. The test files and changes were written by the same author as the checks, so this shows the checks do what was intended, not how they do on a real practice's files. The clean set found one wrong finding, fixed in this pass (caregiver travel time listed although payroll paid it). The test files and the checks were written by the same team, so these numbers show the rules do what they say. How often they are right on your own files is measured at your setup.
git clone https://github.com/fpavia111-creator/fairpractice-agents
cd fairpractice-agents
npm install
node packages/fairpractice-exclusion-screening/bin/fairpractice-exclusion-screening.js demo
Open source, MIT licence, on GitHub: fpavia111-creator/fairpractice-agents. Needs Node.js 20 or later. The demo runs on made-up files and needs no network and no setup.
| Source | What ran and what it found | Licence |
|---|---|---|
| OIG List of Excluded Individuals/Entities (LEIE) 19 Nov 2026 | The full file (84,224 entries) through the exclusion screening agent: 500 of 500 listed individuals found; 1,000 made-up staff with very common names gave 6,357 findings before a fix this run led to (near names with a different birth date are now counted, not listed) and 529 after | Public HHS-OIG data (US government work) |
| CMS physician fee schedule RVU file, January 2026 19 Nov 2026 | The CSV as published (header on row 10, a STATUS "CODE" column, two "RVU" columns) through the provider pay agent: every code came back with no work RVU, then malpractice RVUs were read as work RVUs; both fixed, and the fixture charges now price with the real 2026 work RVUs | Public CMS data (US government work) |
| Nevada Medicaid sample 835 file 19 Nov 2026 | The shared core's 835 reader keeps the 1,812.27 check made only of provider-level adjustments; no wave-1 agent posts ERAs | Public sample from Nevada Medicaid (state government), no real patients |
Public, lawful files only; no private data was used. The files are not copied into the repository: a script there downloads them again.
| Package | What a person reviews and decides |
|---|---|
| Privacy clocks | The privacy officer decides every breach determination and sends every notice, record copy and review reply; the agents only list the clocks and flag risky words. |
| Provider pay | The practice administrator checks the RVU file and the comp plan and decides every true-up; the agent does not run payroll. |
| Compliance pack | The compliance officer confirms every exclusion, licence and health finding against the source and decides what to do; nobody is suspended or terminated by an agent. |
| Care management | A person checks the time and consent records and decides what to bill; the agents never submit claims or read clinical readings. |
| Therapy and ABA | The billing lead and the clinicians decide units, modifiers and plan-of-care signatures; the agents never submit claims or read notes. |
| Hospice and home health | The billing office files every notice and decides on the cap; the agents only list dates and amounts. |
| Value-based contracts | A person raises roster and payment disputes with the payer after checking the lists; nothing is sent by the agents. |
| Home care | The agency decides payroll corrections and claim fixes after reading the lists; the agents never submit payroll or claims. |
| Money recovery | The billing team decides every dispute, appeal, secondary claim, code change and lien reduction; the agents never file, accept or send. |
| Billing company partner | Each client practice reviews its own lists and settings; every claim, payment, notice and dispute stays with a named person there. |
| System | Report or export | Status |
|---|---|---|
| HHS-OIG | LEIE downloadable file (UPDATED.csv) | Tested on the real published file |
| CMS | Physician fee schedule RVU file (PPRRVU, CSV) | Tested on the real published file |
| Any payer or clearinghouse | 835 ERA files read by the shared core | Tested on a public sample |
| State Medicaid agencies | State exclusion lists | Not yet tested on a real export |
| Practice-management and EHR systems | Claims, visits, sessions, time logs, plans of care, open claims, chargemaster and fee schedule lists | Not yet tested on a real export |
| Hospice and home health systems | Census with election, NOE, NOTR and face-to-face dates; 30-day periods and visits | Not yet tested on a real export |
| EVV vendors and payroll systems | Verified visit exports, shifts, payroll registers | Not yet tested on a real export |
| Payers | Attribution rosters, capitation statements, recoupment notices | Not yet tested on a real export |
| HR, credentialing and learning systems | Staff rosters, licences and certificates, health records, course completions | Not yet tested on a real export |
| Accounts payable and spreadsheets | Vendor inventories, payments, physician agreements and payments, incident logs, risk registers, access request logs, lien case lists, dispute logs, review exports | Not yet tested on a real export |
No layout has yet been tested on a customer's own export. Every setup starts by confirming the layout on your files, and any column can be mapped by hand.
When an AI assistant calls a Fairpractice tool, the agent reads the files on the user's computer and returns a result. Before it leaves the agent, patient, client and member names and numbers, birth dates (cut to the year), phones and e-mails, and the names of providers, therapists, caregivers, employees and reviewers are replaced with stable pseudonyms or placeholders. NPIs, claim and case numbers, codes, dates, amounts and business names such as payers, vendors and law firms are sent as they are; free text such as review texts, reasons and risk plans is sent with known names and phone, e-mail and SSN patterns replaced, and risk owners in the security risk analysis are sent as written. A check of 3,645 masked values in the test files found none in a result. Results go to whatever model the user's assistant uses; for patient data use a model covered by a business associate agreement, or a local model. The reports written on the user's computer keep the real values.
Every threshold is a named setting in a settings file the customer keeps: 116 settings in 28 agents, 24 of them following a dated rule such as the 60-day HIPAA breach notice, the 30-day right of access, the No Surprises Act 30 and 4 business days, the 5-day hospice and home health notices, the KX threshold and the hospice cap, which CMS sets each year. Each dated default, its source and the date it was checked are listed in docs/settings.md. CPT and CMS time and day rules (care management minutes, remote monitoring days, the 8-minute rule, APCM tiers) are in the agents' code; Mythos Machines changes them, and any default, only in a release with a changelog line, and checking them is part of the support plan.
Thirty more checks for the back office: credentialing, prior authorization, coding integrity, pharmacy money, front desk, patient financial aid and hospital finance, by the rules you set; they never bill, pay, file or send.
NPI and taxonomy records, NP and PA supervision, locum billing and hospital privileges.
Which scheduled services need an authorization, and the open requests by age and payer.
Record request and audit deadlines, and Advantage plans against traditional Medicare.
Intake forms, text consent, insured estimates, referrals and Medicaid renewals.
PBM payments below cost, PBM audits and controlled substance counts.
340B duplicate discount risks and buy-and-bill drugs bought against billed.
Financial assistance, sliding fees and collection agency placements.
Chargemaster, unsigned encounters, coverage lists, the payment window, late charges, unbilled accounts and bad debt.
Agency staffing invoices and supply and implant prices against contracts.
All thirty, mapped once per export layout and rolled out for a health system, MSO or billing company.
Support after setup: $300 a month per package, or $750 a month for three or more. Hospital revenue integrity is $600 a month on its own, and the partner rollout continues at $2,500 a month.
These packages add to the first Fairpractice packages; a practice can mix both.
Column names were written from common exports; the CMS NPPES and NADAC files were tested as published, the other layouts have not yet been checked against a live export, and the first setup confirms them with you.
Payer, Medicare, Medicaid, 340B, state and policy rules are settings you confirm, not advice.
Estimates and totals are arithmetic on your files, not guarantees.
Patient files stay on your machine, names are masked before anything reaches the model, and a business associate agreement is signed before we ever hold a file. Nothing here diagnoses, suggests codes, reads clinical notes, or is billing, coding, legal, pharmacy or clinical advice.
Your roster against the NPPES file: invalid, deactivated or mismatched records.
Agreements missing or ending, supervisors who left, ratios and chart reviews.
Claims past the 60-day limit and Q5 or Q6 modifiers missing or wrong.
Reappointments due, procedures with no privilege, temporary privileges too long.
Scheduled services on a payer list with no authorization, urgent ones first.
Requests past the payer decision days, peer-to-peer dates and appeal windows.
Denials, pay per unit and days to pay against traditional Medicare.
Record requests due, appeal windows and money taken back during an appeal.
Renewal dates past or near, coverage ended and plans that changed.
Deleted codes, revenue code problems, prices below cost and duplicates.
Open encounters holding claims, by provider. It never reads a note.
Diagnoses already on a claim against the coverage lists, and ABNs. No code suggestions.
Outpatient services in the days before an admission that belong on the stay.
Charges posted after the claim went out, and lag by department.
340B drugs billed to Medicaid against your carve settings and the MEF.
Claims paid below invoice cost or the benchmark, with appeal windows.
Audit responses due, documents missing and early recoupments.
Counts against purchases and dispensing. Numbers only.
Drug units bought against units billed, and claims paid below cost.
Referrals not scheduled, not called back, not seen, and sent outside.
Consents, cards, ID and forms missing or out of date before the visit.
Messages after an opt-out, with no consent, or in quiet hours.
What insured patients may owe, against the quotes you logged.
Discounts below policy and collection actions too early or without notice.
Discounts against your tiers, and income proof missing or old.
Accounts placed with an open claim, pending aid or too early.
Which write-offs meet your collection-effort settings, and why others do not.
Agency hours against timekeeping, rates, overtime and guarantees.
Invoice prices against contracts and POs, bill-only implants and rebates.
Discharged not final billed, interim bills due and holds by reason.
Each one does one job, on your computer, from a file you export. There is no login to your EHR, pharmacy system, PBM or payer portal, clearinghouse, state system or bank, and people make every decision.
Wave 2: Measured in the final pass, 19 Nov 2026
On fictional test data, 19 Nov 2026. Every agent measured: each problem planted in its test files found and nothing else reported, no finding on the same files with the problems repaired (407 rows), and each one-fact change caught. The test files, the clean set and the changes were written by the same author as the checks, so this shows the checks do what was intended, not how they do on a real practice's files. The clean set found one real rule error, fixed in this pass (Medicare bad debt for documented indigent patients). The test files and the checks were written by the same team, so these numbers show the rules do what they say. How often they are right on your own files is measured at your setup.
git clone https://github.com/fpavia111-creator/fairpractice2-agents
cd fairpractice2-agents
npm install
node packages/fairpractice-pbm-reimbursement/bin/fairpractice-pbm-reimbursement.js demo
Open source, MIT licence, on GitHub: fpavia111-creator/fairpractice2-agents. Needs Node.js 20 or later. The demo runs on made-up files and needs no network and no setup.
| Source | What ran and what it found | Licence |
|---|---|---|
| CMS NPPES weekly NPI file (9 to 15 Nov 2026) 19 Nov 2026 | 29,288 real NPI records through the NPI and taxonomy agent, with a made-up roster of 3,000 providers that agrees with NPPES plus the file's 675 deactivated NPIs: 989 wrong findings before the fixes this run led to (primary taxonomy in any of 15 slots, names with middle names and credentials, foreign postal codes), 0 after; 675 of 675 deactivated NPIs found; 0 check-digit failures | Public CMS data (US government work) |
| NADAC 2026 (data.medicaid.gov) 19 Nov 2026 | The 18 Nov 2026 week (30,211 NDCs) as the PBM agent's benchmark file, as published, with 1,000 made-up claims on real NDCs: 100 of 100 claims paid under NADAC found, no other findings | Public CMS data (US government work) |
| FDA National Drug Code Directory 19 Nov 2026 | All 219,432 listed package NDCs (4-4-2, 5-3-2 and 5-4-1) turned into 11-digit keys with 0 unread and 0 collisions; 27,220 of the 30,211 NADAC NDCs matched an FDA listing (the rest are no longer listed) | Public domain (US government work) |
| HHS poverty guidelines (ASPE API), 2025 and 2026 19 Nov 2026 | For household sizes 1 to 10, the two numbers the financial aid and sliding fee agents take (one person, each added person) reproduce every published amount exactly; the test files use the 2025 numbers and each customer sets its own year | Public domain (US government work) |
Public, lawful files only; no private data was used. The files are not copied into the repository: a script there downloads them again.
| Package | What a person reviews and decides |
|---|---|
| Provider credentialing | Medical staff services confirms each NPPES, supervision, locum and privilege finding and decides what to correct; the agents never change NPPES, grant privileges or bill. |
| Prior authorization desk | A person checks each payer's current list and submits and follows up every authorization; the agents never submit, appeal or call a payer. |
| Payer audits and Medicare Advantage | A person sends every record request and appeal and decides what to raise with a plan; due dates come from the letters and your settings. |
| Front desk and estimates | Front desk staff collect forms, confirm consents and quotes with the patient and decide on every message; the agents never text, call or take payment. |
| Independent pharmacy | The pharmacist decides which claims to appeal and files each appeal and audit response; inventory changes and DEA reports are made by a person. |
| 340B and clinic drugs | The 340B program manager checks each flagged claim against the state's rules and the HRSA registration and decides on any correction or rebill. |
| Patient financial aid | Financial counsellors decide applications and discounts, and a person decides whether to recall or hold an account at an agency. |
| Hospital revenue integrity | Revenue integrity and HIM staff decide every chargemaster change, charge move, rebill, hold release and bad debt listing. |
| Hospital spend control | Accounts payable decides what to short-pay, dispute or approve; the agents never pay or hold an invoice. |
| Health system partner rollout | Each site's staff review their own lists and settings; every payment, filing, claim and message stays with a named person at the site. |
| System | Report or export | Status |
|---|---|---|
| CMS NPPES | NPI downloadable file (monthly or weekly npidata_pfile) | Tested on the real published file |
| CMS (data.medicaid.gov) | NADAC weekly file | Tested on the real published file |
| Any pharmacy or billing system | NDC columns, 10-digit dashed or 11-digit | Tested on a public sample |
| Practice-management and EHR systems | Appointment, encounter, charge, claim, referral, prior-authorization and document lists | Not yet tested on a real export |
| Hospital patient accounting systems | Unbilled (DNFB), late charge, outpatient and inpatient account, bad debt and collection placement lists; chargemaster | Not yet tested on a real export |
| Pharmacy management systems | Claim payment lists, controlled-substance purchases, dispensing and counts, PBM audit logs | Not yet tested on a real export |
| Wholesalers and specialty distributors | Invoice lines with NDC, HCPCS and billing units | Not yet tested on a real export |
| HRSA 340B OPAIS | Medicaid Exclusion File entries (NPI, state, quarter) | Not yet tested on a real export |
| Medical staff and credentialing systems | Rosters, privileges, reappointment dates, supervision agreements, locum assignments | Not yet tested on a real export |
| Staffing agencies and timekeeping systems | Agency invoice lines, timesheets, contract rates | Not yet tested on a real export |
| Supply chain and accounts payable systems | Invoice lines, contract prices, purchase order lines | Not yet tested on a real export |
| Patient texting platforms | Message logs and consent lists | Not yet tested on a real export |
No layout has yet been tested on a customer's own export. Every setup starts by confirming the layout on your files, and any column can be mapped by hand.
When an AI assistant calls a Fairpractice tool, the agent reads the files on the user's computer and returns a result. Before it leaves the agent, patient names, patient and account numbers, member numbers, phone numbers, and provider and staff names (surgeons, NPs and PAs, locums, agency nurses) are replaced with stable pseudonyms or placeholders. NPIs, claim and invoice numbers, CPT, HCPCS, NDC and revenue codes, dates, amounts and business names such as payers, PBMs and vendors are sent as they are; free text such as descriptions and hold reasons is sent with known names and phone, e-mail and SSN patterns replaced. A check of 6,639 masked values in the test files found none in a result. Results go to whatever model the user's assistant uses; for patient data use a model covered by a business associate agreement, or a local model. The reports written on the user's computer keep the real values.
Every number is a named setting in a settings file the customer keeps: 97 settings in 30 agents, 31 of them following a dated rule such as the Medicare 3-day payment window, the 120-day bad debt collection effort, the 501(r) 120-day and 30-day notices, the 60-day locum limit, the 7-day prior-authorization decision time from 2026 and the 10-business-day text opt-out from 2025. Each dated default, its source and the date it was checked are listed in docs/settings.md; poverty guidelines are required settings the customer updates each January. Mythos Machines changes a default only in a new release with a changelog line, and checking the dated defaults is part of the support plan.
We send the list of reports by name. You export them from the systems you already use; nothing is connected.
Over a shared screen: the checks are installed, your columns are matched, and your own rules and dates are set.
The checks run on your files. We go through the worklist with you, and every flagged row is a person's call.
You run them whenever you like. Support keeps the dated rules and the export layouts current.
A fixed scope and a fixed price, agreed before any work starts. Your files stay on your computer the whole time.
Written by the check itself on made-up files. Every person, business and number is invented.
Exclusion screening All samples