Live pages mean corrections land everywhere at once. This log records the ones that matter: value corrections, rule fixes, and engine upgrades, dated and in the open. Errors caught by readers arrive through the report a miscode link on every tool.
The E/M Assistant now has a See an example button: a clearly fictional diabetes and hypertension note runs through the real deterministic engine on your own device, producing the full two column result, the 99214 plus G2211 coding, the rationale, and every live control, with no AI call and no free analysis consumed. The Op Note Assistant points new users to its fifteen published worked examples the same way. The homepage now has one primary action, coding an office note, with the operative tool second and a quiet start here link, per the external review. Tool results announce themselves to screen readers and receive keyboard focus, and errors are announced as alerts. A plain language Terms of Use page now covers acceptable use, AI limitations, user responsibility for coding and de-identification, early access billing with a 30 day full refund promise, and intellectual property, linked from every footer and written for counsel review before paid access scales.
An independent product review drove this release. The de-identification instruction was rewritten: removing the name, date of birth, and record number alone may not be sufficient, HIPAA Safe Harbor covers 18 identifier categories, your organization's policy governs, and you remain responsible; the warning now says so on every tool. Privacy language now matches the implementation exactly: the raw note is never stored, analysis results live briefly in temporary job storage, are deleted on pickup, and expire automatically within one hour, with a cleanup process enforcing it. Paid access now uses one shared key across every tool, with automatic migration for anyone holding the old one. The Five Note Audit computes every dollar figure from one user-entered rate instead of mixing a hardcoded Medicare figure with your rate. The pricing page now labels the purchase path honestly as early access by email until the self-service checkout ships. The self-test system was rebuilt so a reset is instant and every kick automatically retries failed cases, and the homepage slogan finally has readable contrast. A set of small editorial defects flagged by the review was cleared in the same pass.
Every page now lives at exactly one address: the extensionless URL is canonical, every .html address redirects permanently to it, canonical tags and the sitemap agree, and all internal links across 73 pages were rewritten and machine verified to resolve. Six new pages each answer one search question directly in the first paragraph: 99213 vs 99214, G2211 documentation requirements, modifier 25 with a joint injection, modifier 57 for fracture care, and the two shoulder and knee arthroscopy code pair questions (29826 with 29827, 29877 with 29881), all linked from the guide library. The two assistant pages gained substantial crawlable descriptions covering what each tool does, who it is for, what it returns, how it was validated, and common questions. And every page now carries an author, review date, and sources block above the footer.
The self-test battery caught the operative extractor occasionally omitting posterior segmental instrumentation from a multi-procedure lumbar fusion when template noise surrounded it. A deterministic backstop now guarantees it: when the note states segmental instrumentation with an explicit segment count alongside a fusion, the add-on is restored with the stated count and prices as the correct 22842 to 22844 family. The version shown in the page footer also now reads the deployed build directly, so what the footer says is always what is actually live.
A second, independent implementation of the AMA E/M framework and CMS payment rules was written from the rulebooks and 6,000 generated scenarios combining common and complex rules (procedures with modifier 25, global periods with real date math, AWV combinations, prolonged time, consults, counseling add-ons, hospital visits) were scored by both and compared field by field. The sweep caught a genuine overcoding defect: when a procedure and a separate problem were managed on the same day, the visit level could still rest partly on the condition treated by the procedure, even while the rationale claimed otherwise; a knee injection plus stable hypertension could print level 4 when the separately addressed work supports level 3. The problem filter now runs before the level is computed. The sweep also caught the annual wellness combination pinning its modifier to the wrong code in rare pairings, now scoped precisely. Separately, the extractor gained a deterministic guard so a same-day procedure can never erase G2211 eligibility, which the live battery had flagged, and one new live case pins the procedure-related flagging. All 6,000 scenarios now agree exactly.
Twenty two new scenarios, deliberately different from every prior test, were pushed through the engine and every output was reviewed line by line. One real defect surfaced: when high risk came from drug therapy requiring intensive monitoring, such as warfarin with INR management, the risk sentence could print a moderate element next to the words "reaches high," a mismatch any auditor would question. The rationale now names the intensive monitoring itself. The problems sentence also gained its missing reasons for stable acute illnesses and self-limited problems, so every tier now states why it was reached. Boundary behavior was verified across the set: the 40 minute level 5 line, the 75 minute prolonged threshold, day 90 inside versus day 91 outside the global period, day zero, and dates crossing a year boundary. Separately, the reader now waits at most one minute for an analysis; a note that cannot finish in that time gets a clear timeout message naming the likely cause, template boilerplate, and suggesting a trimmed or different note.
Real clinic notes carry a lot of template text, and the longest ones were being cut off by the hosting platform's ten second limit on synchronous requests, surfacing as a generic "Something went wrong." The E/M reader now runs on a background architecture with a fifteen minute budget: the page submits the note, polls for the finished analysis, and reports progress while it reads, so note length can no longer kill a run. The extraction brain lives in one shared module used by both the new path and the old one, which remains as an automatic fallback, and the live self-test battery gained a case that exercises the new pipeline end to end on every sweep. Error messages are also specific now, saying what happened and what to do instead of asking you to guess.
When a postoperative note carries the surgery date and the service date, the assistant now computes the interval itself: inside the 90 day global period it says which day you are on, and outside it says so plainly, drops any postoperative modifier, and codes the visit as separately payable. The global period question arrives pre-answered from the dates with a one line explanation, and you can still override for 10 or 0 day procedures. A routine in-global visit now codes 99024 directly, with no visit level and no more contradictory advice about raising the level of a visit that is part of the surgical package; if an unrelated problem was also managed, one click switches to the paid visit with modifier 24. Counseling the note already documents no longer asks whether it happened, only for the minutes. The tool pages also got a redesign: a larger paste box, one collapsed information section holding privacy, pricing, and workplace guidance, and on wide screens a two column results view with the questions on the left and the running codes, total, rationale, and coaching pinned on the right, updating live as you answer.
Two rule fixes from live clinic use. First, a test the clinician orders at the visit now counts exactly once, as an ordered test; it can no longer also be claimed as an independent interpretation, and the assumption checkbox now names the specific test and states the condition that makes an interpretation hold up: performed or ordered elsewhere, personally read and documented, and not separately billed. Second, a drug at over the counter strength, such as topical diclofenac 1 percent gel, now scores as over the counter treatment (low risk) rather than prescription drug management (moderate risk), even when the note says prescribe. Both fixes carry new permanent cases in the live self-test battery.
Both AI tools now run two channels on every note. The deterministic engine produces the verified codes, values, and rationale as always, and the model independently proposes the codes it would report; agreements are shown, and model only suggestions appear in a clearly labeled unverified panel with library values, never in the total. This build also folds in a day of live clinical feedback: the G2211 longitudinal sentence now appears exactly once, Modifier 25 explains itself in a single unified paragraph, social determinants name their specific factors, the problems line states why it reaches its tier, level advice is honest about how many categories must rise, follow up ranges like 4 to 6 weeks count for G2211, a surgical discussion no longer counts as a surgical decision, fracture visits point to the fracture pathway, and when a note names a CPT the catalog missed, the tool now looks it up and says so.
A new Search page finds any phrase across every guide, and typing a code like G2211 or 99214 shows its current work RVU with a one-click jump into the calculator alongside every page that discusses it. Behind the scenes, a key-protected self-test endpoint now feeds twenty two canonical notes, including deliberately bloated ones with planted decoys, through the live AI reader and reports exactly which extracted facts passed, so reading accuracy is verified end to end after every release.
The complete PPRRVU2026 April relative value file was applied to the library: 59 codes that displayed verify are now valued, and 232 existing values were corrected to the April revision, including the wound repair family, the 2023 hernia family (49593 rose to 10.00), the debridement set, and two arthroplasty companions whose older values were materially low (27165 is 19.78 and 27495 is 16.13 in the current file). Rankings that depend on value, and therefore modifier 51 placement, updated automatically and are re-pinned. The library now carries 540 codes, all on one authoritative file.
The tumor catalog grew from ten thigh and femur entries to fifty two spanning shoulder, upper arm, forearm, hand, pelvis and hip, thigh, leg and ankle, and foot soft tissue (with each site's own size thresholds, including the 3 cm lines for forearm, hand, and foot radicals and the reversed pelvis subcutaneous pair), plus bone families for the humerus, scapula, radius and ulna, tibia, and fibula, the femoral fixation add-on 27358, osteoarticular allograft reconstruction (20932), and antibiotic drug-delivery placement (20700) for infection cases. Every code and band assignment was verified against published CPT references this build. One value is loaded from the January 2026 CMS file (27059 at 28.62); the remaining tumor values display verify until the fee schedule file is applied, which is the next planned load.
The Op Note Assistant now carries the arthroplasty tool's full playbook: the infection staging map for hip and knee (single stage with modifier 22, head and liner or poly exchange, explant with static or articulating spacer, and replant), navigation with the correct split between imageless 20985 and image-based 0055T (and the rule that pre-operative patient-specific guides alone have no CPT), the true intertrochanteric or subtrochanteric osteotomy (27165) versus the bundled extended trochanteric osteotomy, and prophylactic femoral cerclage (27495). It also gained negative pressure wound therapy by size and disposability, a full excisional debridement system (depth, summed area, open fracture family, size add-ons) with on-page detail buttons when the note leaves them open, and a starter orthopaedic tumor set for the thigh and femur. Debridement and open fracture site values await fee schedule verification and are marked accordingly.
The E/M Assistant now codes psychotherapy-only sessions (90832, 90834, 90837 by documented minutes) and psychiatric diagnostic evaluations (90791, or 90792 when medical services such as prescribing are documented), completing the behavioral health core alongside the existing E/M-with-psychotherapy pairing.
Critical care (99291 with 99292 units) now codes by documented time, home and residence visits (99341 to 99350) joined the engine, and psychotherapy documented alongside medication-management E/M now pairs the correct add-on (90833, 90836, or 90838) automatically. The independent cross-validation reached 26,550 generated encounters in full agreement. The library grew by twenty high-use entries; critical care carries values, while home, psychotherapy, and nursing facility discharge values display as verify pending fee schedule confirmation.
Prolonged services in the hospital and nursing facility now code 99418 at the published thresholds, and the nursing facility family arrived end to end: initial care (99304 to 99306), subsequent care (99307 to 99310, four levels), and discharge (99315 and 99316 by documented time). The independent cross-validation now agrees with the engine on 24,371 generated encounters across every evaluation setting a working physician bills. Discharge values (99315, 99316) await library verification and display as verify.
The E/M engine now codes hospital consultations (99252 to 99255, with their own time table) and same-date admission and discharge (99234 to 99236). The independent cross-validation grew to 21,838 generated encounters across every visit family, in full agreement.
An automated audit found thirteen engine values that disagreed with the site's fee schedule library, including total knee and total hip arthroplasty and both meniscectomy codes. Every engine value now reads from the library as the single source of truth, and duplicate cholecystectomy entries were consolidated.
Initial hospital or observation care (99221 to 99223), subsequent care (99231 to 99233), and discharge day management (99238 and 99239, selected by documented time) now code end to end, with G2211 and wellness codes correctly barred outside the office. The independent cross-validation grew from 9,559 to 16,798 generated encounters covering office, consult, emergency department, hospital, and discharge visits, in full agreement.
When documented tobacco cessation minutes appear in a note, the E/M Assistant now reports 99406 or 99407 directly with modifier 25 on the visit, instead of only coaching about it.
The Op Note Assistant now selects 22842, 22843, or 22844 automatically when three or more fixation segments are documented, reserving 22840 for non segmental constructs, and reports multi level add-ons as units (a three level ACDF reports 22552 x2).
The E/M engine now produces the Medicare AWV combination directly: G0438 or G0439, the problem visit with modifier 25, and G2211 under the CMS preventive exception.
The laparoscopic hysterectomy family (58570 through 58573) and hysteroscopy with biopsy (58558) now carry verified 2026 values.
A prose sentence stated 99204 at 3.00 wRVU; the correct 2026 value is 2.60. Caught by external review; every value on the site resolves from one shared library, and this sentence now does too.
Removal of a loose body from a different compartment now bundles into knee arthroscopy with the Medicare G0289 guidance shown in the output; some commercial payers accept 29874 with modifier 59.
CMS applied a 2.5 percent efficiency adjustment to most surgical wRVUs for 2026; the library reflects the post adjustment file and the companion says so.
Anterior abdominal hernia repairs now resolve to the exact code among 49591 through 49596 and 49613 through 49618 from documented size, reducibility, and recurrence; deleted legacy codes were removed.
29877 with meniscectomy now bundles per current NCCI in the same or a separate compartment, and the public validation example demonstrates it. An earlier version kept it with modifier 59; that was wrong and is documented here on purpose.
Two sources conflicted on 33533, so the value was removed rather than guessed; the code shows "verify" everywhere until an official figure is confirmed.