Versions

Version Name	Internal Release Date	API Release Date	Latest Payer MRF Month	Production Schema	Release Notes
2026_08	10/13/2026	TBD	2026_08	tq_production.clear_rates_2026_08	Release Notes
2026_07	09/17/2026	09/17/2026	2026_07	tq_production.clear_rates_2026_07	Release Notes
2026_06	08/07/2026	08/17/2026	2026_06	tq_production.clear_rates_2026_06	Release Notes
3.0.1	06/09/2026	06/15/2026	2026_04	tq_production.clear_rates_2026_04
3.0.0	05/10/2026	06/01/2026	2026_03	tq_production.clear_rates_2026_03
2.4.3	04/10/2026	05/01/2026	2026_02	tq_production.clear_rates_2026_02
2.4.2	03/12/2026	04/01/2026	2026_01	tq_production.clear_rates_2026_01
2.4.1	02/19/2026	02/25/2026	2025_12	tq_production.clear_rates_2025_12	Slack Release Notes
2.4.0	02/05/2026	02/11/2026	2025_11	tq_production.clear_rates_greg_2025_11	Slack Release Notes
2.3.2	01/22/2026	01/28/2026	2025_11	tq_production.clear_rates_calcifer_2025_11	Slack Release Notes
2.3.1	01/08/2026	01/14/2026	2025_11	tq_production.clear_rates_2025_11	Slack Release Notes
2.3.0	12/4/2025	12/10/2025	2025_10	tq_production.clear_rates_2025_10	Slack Release Notes
2.2.2	11/14/2025	11/19/2025	2025_09		Slack Release Notes
2.2.1	10/30/2025	11/5/2025	2025_08		Slack Release Notes
2.2.0	10/15/2025	10/21/2025	2025_08		Slack Release Notes

v2026_08 Release Notes

Internal notes for tq_production.clear_rates_2026_08. This is the first release built on the v2 selection engine (clear_rates_sub_dag + clear_rates_combiner), replacing core_licensable_data_orchestrator. Full QA, with every number traceable to a query: findings register and open decisions. An SE and CSM guide to the changes: narrative.

At a glance

	2026_07	2026_08
Rates (score above 1)	3.54B	3.67B (+3.6%)
Payers	103	105
Networks	342	352
Providers with rates	79,778	79,592
Billing codes with rates	12,027	13,898

The engine now decides, per hospital contract, which machine-readable file to trust and fills from it first; picks one gross-charge method per provider; breaks ties by distance to a validated or claims-based reference instead of taking the larger rate; and demotes rates with no Medicare reference that fall outside national bounds. Scoring rules are mostly unchanged.

For the same provider, payer, network and code in both months: 85% of rates are exactly equal, 12% changed by more than half a percent (three quarters of those are the same rate type moving in value from refreshed files), and 90% of scores held. Where a different kind of rate won, the old rate was still a candidate about 90% of the time and lost a same-score tie-break.

What changed

• Rate selection
  • One source per hospital contract (mrf_selected_source, is_consistent). 62% of hospital rates come from the file chosen for their contract, 20% from the other file where the chosen one had no rate, 18% on contracts with no preference.
  • One gross-charge method per provider (primary_gross_charge_methodology).
  • Ties broken by distance to a reference (canonical_median_validated_rate, _method), with posted allowed amounts preferred over charge-based figures.
  • National-bounds override for rates with no Medicare reference: 321,410 rows demoted to outlier, 88% hospital inpatient.
  • Crosswalked rates (MS-DRG to APR-DRG and back, RC to HCPCS) now score 2 as experimental. 38M inpatient rates went 4 to 2 on the same value. Outpatient surgical grouper rates also 4 to 2, pending validation.
  • Contract provisions (MS-DRG base rates, obstetrics) score 5 and report the file they came from; last month they scored 4 as estimates. Hospital inpatient estimate share fell from 32% to 8%. Hospital validated rows rose 28.1M to 51.4M; 74% of that is this reclassification, true payer-hospital pairs went 28.1M to 33.4M.
• Payers, networks, codes
  • New payers AvMed (382), Sharp (809), Optum (820). Horizon remapped (OMNIA to Omnia, Managed, State), Centivo renamed, Highmark PremierBlue new.
  • Aetna's individual-market HMO and EPO plans aged out: last published 2026_04, outside this release's 2026_05 to 2026_08 window. 53M rates on eleven networks. Aetna's national networks grew 5 to 6%.
  • Laboratory code list doubled by design (CLD-208, CLD-210): 939 to 1,899 codes, 18.6M rates, 0.8x to 1.1x Medicare. Laboratory rates +84%.
  • Physician group plausibility repointed to the 2026_10_05 table; net PG space +0.2%.

Data samples

• Hospital source selection. Cigna National PPO, CPT 80048: candidates were the hospital's fee schedule $18.13 (chosen source), the payer's current fee schedule$15.90, and a $5.50 payer row from a three-month-old file at a higher score. Last month shipped$5.50. This month ships $18.13, within 14% of the payer's current posting. Clear Lookup
• Provision rate. MS-DRG base rates read from contract language now score 5 and carry base_rate, drg_weight and fee_schedule_year in the external table; base_rate * drg_weight reproduces the rate. 16.7M rows.
• Coverage gain. Claritev MultiPlan and PHCS hospital outpatient contracts went from a few dozen rated codes to about 7,400, derived from the payer file the engine selected for them.
• Rate type demotion. Reeves Regional, BCBS TX, APR-DRG 0661-2: every candidate is crosswalked from MS-DRG 809 and 813. The $5,374 average of the posted case rates scored 2 and was demoted by national bounds; a$15,626 state gross-charge estimate won. A known issue, below.

Schema and type changes to prod_combined_abridged

Read these before querying. Every one is a silent change for a downstream filter or join.

• 26 new columns, appended after payer_id (positions 182 to 207). Consumer-facing: mrf_selected_source, is_consistent, inconsistent_fill, payer_mrf_selection_tier, hospital_mrf_selection_tier, canonical_median_validated_rate, _method, _percent_of_medicare, claims_validation_source, claims_validation_method, claims_benchmark_p25, _p75, _median, is_national_bounds_outlier_override. Gross-charge plumbing: gross_charge_methodology, gross_charge_count_floor, gross_charge_selection_scope, primary_gross_charge_methodology, primary_gross_charge_rank, primary_gross_charge_non_outlier_rate_count, is_primary_gross_charge_methodology, roid_has_primary_gross_charge_match. Internal, do not project: chunk_id, experimental_priority, raw_priority, experimental_geography_priority. Dictionary: R1 §8.
• canonical_rate_type values lost their prefixes. Every 2026_07 value began with raw: , transform: , impute: or provision: . No 2026_08 value does. best_payer_rate_type and best_hospital_rate_type same. Filters on the old strings return nothing.
• canonical_gross_charge_type vocabulary changed to Provider, State, CBSA from six mrf_ / komodo_gross_charge_* values.
• Type changes: clear_market_utilization_commercial/medicare/medicaid decimal(38,11) to double; asc_avg_medicare_rate decimal(21,2) to (20,2); lab_avg_medicare_rate decimal(20,2) to (19,2); device_percentage decimal(38,9) to (38,10). payer_location_data_source_name and canonical_rate_gross_charge_original_billing_codes remain array(varchar) and are 100% NULL.
• No canonical_rate_score = 0 rows. Unrated rate objects are not materialized (18,964,707 last month). Compare at score above 0.
• best_payer_rate_score and best_hospital_rate_score are NULL where 2026_07 had 0. ClickHouse renders them as -1.
• canonical_rate_percent_of_* are no longer rounded to two decimals.
• payer_last_updated_on is the first of the data month on every payer row, not the payer's file date. External rate_last_updated_on inherits this.
• Columns 2026_07 populated that are empty in 2026_08: canonical_rate_validation_method, best_payer_rate_validation_score, best_hospital_rate_validation_score, anesthesia_base_units, payer_file_hash, payer_location_data_source_name, clear_market_utilization_* (ClickHouse defaults these to 1.0). Queued for 2026_09.
• canonical_rate_subversion values are 2026_05 to 2026_08.

Downstream table changes

External clear_rates: three script edits and six new columns

• Collapse prod_rollup_codes to one row per code before joining (BOOL_AND(external_release_include) ... GROUP BY 1,2,3,4). The 2026_08 rollup table has a row per modifier for 1,318 codes; the current join adds 1.95B duplicate rows.
• Make the gross_charge_type CASE case-insensitive; the source values are Provider, State, CBSA and the current patterns match none.
• Append the six columns announced in September: case_rate, case_period, base_rate, drg_weight, fee_schedule_year, provision_name, read from canonical_method_params and gated on the formula. Populated on 159,311 obstetrics and 16,745,780 MS-DRG rows. No other new columns on the external table.
• Consumers also see: rate_type prefixes gone, no rate_score = 0, percent-of-Medicare unrounded, rate_last_updated_on as a month stamp, 12.4M SSP-only rows excluded (4.3M last month).

Full script and QA: builds/clear_rates_v2. Dev copy: tq_dev.internal_dev_npattison.cr08_clear_rates_v2 (all payers).

ClickHouse prod_clickhouse_abridged: one script edit and 26 schema entries

• Same prod_rollup_codes collapse (BOOL_AND(analyze_include)); the current join adds 634M duplicate rows.
• 26 new entries in ABRIDGED_COLUMNS after payer_id, all nullable; exclude chunk_id and the three *_priority columns from production. Declare hq_latitude and hq_longitude as Nullable(Float64) (currently Nullable(String)).
• Analyze consumers also see: canonical_rate_type prefixes gone, canonical_gross_charge_type new vocabulary, best scores -1 where 0, best_*_contract_methodology reads No Payer Rate / No Hospital Rate where there is no best rate, utilization 1.0 on every row, six columns empty or defaulted, a new Unknown Provider Type bucket (58,584 rows).

Full script, renderer and dictionary additions: builds/clickhouse_abridged_v2. Dev copy: tq_dev.internal_dev_npattison.cr08_clickhouse_abridged_v2 (payers 643, 7, 76 as a test sample).

Known issues, queued for 2026_09

Shipped as is on 10/13 by decision. Each has a ticket draft in the findings register.

• The projection step empties seven columns 2026_07 populated (above).
• Extreme rates ship at rated scores: rated rows above $10M rose 1,553 to 5,086; 9,999,999,999 placeholders at score 1. 85 Medicare Advantage outpatient contracts use a "100% of gross charge" transform (median 1.4x to 6.5x Medicare).
• Medicare references missing on 8.2% of rated rows (5.5% last month): the ASC and outpatient MPFS fallback was removed, and the APR-DRG Medicare reference snapshot omits 443 IPPS hospitals and 33 codes.
• Three pipeline losses: Ambetter HMO (20.2M rated rows, tuple removed), AmeriHealth (0.2M, stale label), APR-DRG imputations not ported (11.4M).
• Lab network-consistency test fails whole contracts over the doubled code list (BCBS IL 381 to 37 passing contracts).
• Cigna physician group fee-schedule rates 4 to 2 via the network-consistency flag (12.6M rows, same rate).
• Crosswalk-only hospital contracts get no source selection (61M rows with NULL is_consistent); the raw payer de-dup collapses sibling ROIDs that differ by facility or modifier.
• Claims-only validation carries the payer_hospital label and the whole-claim band validates some outpatient estimates.

v2026_07 Release Notes

Internal notes for tq_production.clear_rates_2026_07. Customer-facing notes are staged as a Zendesk draft (preview/edit); the permanent article link goes live when it is published closer to the API release. Full ticket detail on the CLD board. Full growth decomposition: Clear Rates 2026_07: why the table grew 72%.

At a glance

This is the largest Clear Rates release to date. Rated rate objects grew from 2.05B to 3.54B (+72.4%), and every added row is attributed to a cause below. Roughly three quarters of the growth comes from refreshing the physician group rate object space, which about doubled the set of billing codes each physician group is priced for. Most of the rest is new networks (Aetna Select is the big one) and repairs to network label mappings that recovered rates which had been silently failing to attach.

	2026_06	2026_07
Rated rate objects	2.05B	3.54B (+72.4%)
Payers	97	103
Networks	302	342
Providers	76,807	79,778
Billing codes	11,651	12,027

Two changes to know before querying the new table:

1. One new column, provision_traceability. Nothing else about the schema changed.
2. The external table now excludes a small set of codes. 185 Hospital Outpatient codes that exist only to support Standard Service Packages (SSPs) are filtered out of the external clear_rates table and Analyze, about 4.3M rated rows (0.12%).

What changed

• Payers & Networks
  • Six new payers landed: Wellcare (706), AmeriHealth Caritas (959), Northwell Direct, Independent Health, Preferred Health Professionals, and Capital Health Plan (+12.7M rated combined). Wellcare and AmeriHealth Caritas are the 2026_06 spillover finally shipping, and BCBS Rhode Island Medicare Advantage (397, 51.6k rated) also landed, closing the full "did not ship" list from the 2026_06 notes. Only three of the six payers were newly configured this cycle; the others were set up in earlier releases and their data arrived now [CLD-183]
  • Aetna Select was added as its own network (carried as an HMO), 211.2M rated rates, 93% of the entire new-network bucket. The other 29 new networks are much smaller: Molina landed 23 state-level Medicare Advantage, Exchange, and HMO networks (+7.4M total), Baylor Scott & White added two Texas HMOs (+4.0M), and Regence's UT PPO was renamed and split into Utah Participating and Utah Preferred BlueOption [CLD-183]
  • Plan label repairs recovered rates on 26 existing networks (+104.0M rated directly attributable). Payers publish MRFs whose plan labels we map to our named networks; when a payer renames a label and our mapping lacks the new variant, real rates silently fail to attach. This release repaired labels on 26 networks, which grew 95.2% on average vs 58.7% for the other 275. Largest recoveries: UHC Navigate (label "NAVIGATE" added) 100.3M → 169.6M, Cigna NY HMO 0.8M → 11.8M, Cigna New England HMO 0.5M → 6.1M, Anthem VA HMO 0.3M → 5.2M, Healthfirst NY Essential Plan 0.3M → 5.0M, BCBS Wyoming PPO 15.6x. Several of these are recoveries of networks flagged as "lost their payer MRF" in the 2026_06 notes [CLD-183]
• Rate object space
  • The physician group plausibility table was refreshed for the first time since March. Plausibility decides which billing codes each physician group can be priced for. A group qualifies for a code only if payers have actually posted a rate for that pair, and only if the group's specialty plausibly performs the service. The pin moved from physician_group_plausibility_2026_03_09 to physician_group_plausibility_2026_08_31, and the allowed set grew from 394.6M group and code pairs to 1,476.6M [CLD-196]
    • Two upstream changes landed between the two builds: the claims-evidence thresholds a code must clear were lowered in April (DA-359), and Core Rates moved to taxonomy plausibility v2 in May, which is stricter and pulls the other way
    • Effect on coverage: physician group rated rows 1.42B → 2.83B. The average group now carries 3,488 rated codes, up from 1,759. 569 codes are rated for physician groups for the first time
  • The per-provider-type code lists changed modestly: Hospital Outpatient +93/-3 codes, Physician Group +33/-25, ASC +15/-42, Lab +2, Imaging Center -4. The 74 removed combinations carried 3.62M rated rows and read as deliberate tightening, removing procedures implausible for the setting (amputations and trauma surgery at ambulatory surgery centers, emergency department visit codes at imaging centers)
  • 56 outpatient facility codes needed by SSP v2.0 were added to hospital coverage. They generated ~668k rate objects, but none received a price that survived selection, so their rated contribution this month is zero [CLD-180]
  • Customer-requested codes were added to the ASC, hospital outpatient, and physician group code sets, plus a Urology service line fix [CLD-196]
• Rate selection & scoring
  • Rate score tiers were restructured to describe how each rate was derived: 5 validated or provision-sourced, 4 raw posted rate, 3 derived through a transformation (crosswalk, charge-based calculation), 2 experimental derivation, 1 outlier, 0 no rate (PR 6717, no CLD ticket). The boundaries are unchanged: > 1 still means rated and > 0 still means a rate exists, so headline counts remain comparable month over month. Individual tier counts are not comparable to 2026_06 (old tier 3 meant partially validated, new tier 3 means transformation-derived), and any downstream logic keying on specific tiers should be reviewed
  • Laboratory rate selection now prefers rates posted for POS 81, the independent laboratory place of service. Payers often post two rates for the same lab, code, and network that differ only by place of service, and the old tiebreak (highest rate wins) picked the facility-only rate an independent lab can never bill. Measured against paid claims, the fix corrects up to 1,073 currently-wrong cells while risking 415 currently-correct ones, about 2.6:1 favorable [CLD-191]
  • Institutional lab rates are now brought in for Cigna x Quest. Some payers post large portions of their lab rates under an institutional billing class rather than professional, which previously excluded them entirely: Cigna posts ~62% of its lab rates institutional, including 100% of its Quest rates. The override is built to extend to more payers or payer-provider combinations as research firms up [CLD-192]
  • Nomi Health (1216) ASC facility rates were recovered. Nomi posts ASC facility rates under a professional billing class, so they were dropped before reaching Clear Rates; Nomi is now on the professional-to-Outpatient bypass list. All 5 Michigan ASCs in the customer's provider list previously had zero coverage [CLD-186]. Coverage verification and remaining limitations (percentage-based rates dominate, no gross charge source for ASCs, several psych hospitals absent) are documented in [CLD-187] and [CLD-177]; both were investigation/verification only, no further fix this release
• Provisions & Traceability
  • Obstetrics contract provisions now price inpatient deliveries. Case rate and per diem terms extracted from contract language are applied as imputation rates for vaginal, cesarean, and bundled delivery episodes, using a full-episode formula (case rate plus per diem for days beyond the case period, with lengths of stay derived from Komodo claims). A provision hierarchy ensures the most specific plausible provision wins per billing code [CLD-188]
  • A new provision_traceability column (varchar) documents the provision behind a rate, so provision-derived rates can be traced back to their contract term. This is the only schema change this release; the other 180 columns are unchanged in name, type, and order [CLD-188] [CLD-189]
  • Base-rate imputation frequency fields now report real numbers from the provisions pipeline instead of placeholder 9999 sentinels, and an always-true gate left over from the old logic was removed [CLD-189]
  • Obstetrics rates now report their true origin (payer or hospital) in canonical_rate_source instead of a generic imputation label [CLD-198]
• Codes
  • When a hospital reports both a specific custom code and a generic bucket for the same family (for example emergency room visit levels), the specific mapping now wins, so rates land on the exact code instead of being spread across the whole family. Implemented via a new hierarchy column in the CSTM crosswalk (EMR1 → 99281 and EMR5 → 99285 beat the generic EMR → all-ER-codes fallback) [CLD-190]
  • Florida Blue publishes rates under plan-specific local codes. Those are now crosswalked through the outpatient surgical grouper to standard HCPCS codes, so the underlying rates surface under codes customers can actually use [CLD-194]
• Bug fixes
  • CAR-T imputation suppression was keyed to the wrong code: Q2043 (not a CAR-T code) instead of Q2053. Fixed, so Q2053 no longer carries imputed rates that are unreliable at CAR-T price points [CLD-195]
  • Physician group names are no longer overridden by infusion center names when a provider appears in both spines. This is a general fix, not an allowlist: it affects ~637 provider_ids, and roughly two-thirds of them now surface a distinctly different (correct) organization name [CLD-175]
• Schema & data release changes
  • Column changes: one addition, provision_traceability (varchar); the other 180 columns of prod_combined_abridged / clear_rates are unchanged in name, type, and order (see Provisions & Traceability). rate_score tier meanings shifted (see Rate selection & scoring); the > 1 rated and > 0 any-rate boundaries are stable
  • A new prod_rollup_codes table says, per provider type and billing code, whether the code belongs in Analyze and in the external release, via analyze_include / external_release_include flags (PR 7026). The flags are false for 185 Hospital Outpatient codes whose only reason for inclusion is supporting SSPs (facility-fee line codes such as 99213, 99202, 99211, plus the CLD-180 catch-up list); everything else is true. In 2026_07 the flagged codes carry 4.28M rated rows, 0.12% of the table [SE-14670]
  • The external clear_rates table is no longer built by core_licensable_data_orchestrator: creation moved to the LS team's clear_rates_api_dev_tables DAG (PR 7204, merged 9/14), which applies the external_release_include filter (rows without a flag entry default to included). prod_clickhouse_abridged (Analyze) applies analyze_include the same way and its subtables inherit the filter. Practical consequence: the external clear_rates_2026_07 and Analyze will run ~4.3M rated rows below the counts in these notes, which are measured on prod_combined_abridged. Partitioning unchanged (state, payer_id) [SE-14670]
  • Coming: real partitioning for the tq_production clear rates tables [SE-14694] (selected for development; payer_id + network_id is the leading candidate). This is a large query-performance lever for downstream users; partition keys should be announced to DS and SE once landed
• Data shifts worth knowing
  • Rated rate objects 2.05B → 3.54B (+72.4%, +1,485.9M). Payers 97 → 103, networks 302 → 342, providers 76,807 → 79,778, billing codes 11,651 → 12,027 (rated grain, matching the published-notes convention). Physical size 411 GB → 552 GB
  • The growth is fully decomposed, each added row attributed to exactly one cause, reconciling to the total: plausibility refresh +1,122.3M (75.5%), new networks +226.1M (Aetna Select +211.2M), network label repairs +104.0M, densification of existing provider-code pairs +53.2M, new payers +12.7M, drops -32.4M. See the whiteboard writeup for method and itemized plausibility changes
  • Provider-type totals: Physician Group +1,410.6M (+99.5%), Hospital Outpatient +41.1M (+10.4%), Hospital Inpatient +13.8M (+13.3%), DME +9.8M (+23.5%), ASC +6.9M (+10.7%), Laboratory +1.6M
  • Outlier-scored rows grew 378.1M → 590.0M, roughly in line with the space expansion
  • QA watch item: the refreshed physician group plausibility grants broad code lists to generalist groups (internal medicine and wound care practices now carry rated rows for advanced imaging and radiation therapy at ~4,500-5,000 codes each). This is consistent with how the table is built (the clinical filter operates at taxonomy grain, and claims-evidence thresholds were loosened), but it is the kind of breadth customers may ask about
  • Three movers exceed the uniform space-refresh band without an established cause: Anthem NV Choice PPO 967k → 2.20M (+127%), Kaiser CO 163k → 401k (+146%), Health New England PPO 71k → 108k (+53%). All small in absolute terms; not traced to a 2026_07 ticket

v2026_06 Release Notes

Internal notes for tq_production.clear_rates_2026_06. Customer-facing notes are published separately to Zendesk. Full ticket detail on the CLD board.

• Payers & Networks
  • Claritev / MultiPlan (1109) wired up: PHCS (33.7M rates) and MultiPlan (31.9M rates) national PPO networks. Needed a payer-scoped exception (ppo_exclusion_payer_overrides) because MULTIPLAN and PHCS are global ppo_exclusions keywords in variables_ppo.sql and were zeroing out 1109's own plan-bridge rows [CLD-165]
  • UnitedHealthcare Neighborhood Health HMO Open Access (FL) added as its own network, payer data only (5.0M rates) [CLD-165]
  • Did not ship despite being in scope: Wellcare (706), AmeriHealth Caritas (959), and BCBS Rhode Island Medicare Advantage (397) are absent from prod_combined_abridged — 706 and 959 have zero rows, 397 has only PPO and Exchange. Needs follow-up [CLD-165]
• Rate selection & scoring
  • Prefer ffs over bundle rates. negotiation_arrangement was absent from the rates_ordered ORDER BY in raw/payer_mrf/build.sql, so the negotiated_rate DESC tiebreaker picked the flat bundle amount for every code in a family [CLD-152]
  • CMS MS-DRG weights migrated off the stale cld_utils.msdrg_weights_by_year (FY2025) to reference_internal.ipps_msdrg (FY2026). 761 of 765 shared codes had a materially different weight [CLD-169]
  • Base-rate imputations now join each contract's own canonical_base_rate_weight_year rather than always the latest fiscal year — ~30% of scored provision rows use a non-current weight year [CLD-158]
  • Removed msdrg_mrf_base_rate_mult_cms_weight end-to-end, plus its 5 supporting columns and dead upstream CTEs. MRF-inferred base rates are not trustworthy enough to publish when a provider has no provision-sourced base rate [CLD-168]
  • Base rates no longer get a flat provision_score of 8 (which mapped to canonical_rate_score 5). They now score 7 when validated against a non-outlier counterpart, else 6.999 → floor 6 → canonical_rate_score 4 [CLD-173]
  • provision_* columns pinned to the current release's own latest subversion. The provision_latest CTE previously searched backward across all subversions, pairing current rates with provision detail from runs up to 8 months stale [CLD-166]
• Codes
  • APR-DRG severity-of-illness suffix stripped before the spines_services_clinical_categories join in code.sql. Before the fix 0 of 1,340 APR-DRG codes matched a clinical category; after, 1,312 [CLD-157]
  • APR-DRG billing codes padded in hospital MRF build.sql and build_pg.sql, a follow-up to v2.4's hosp_prepare_base.sql fix that left the raw ingestion path unpadded [CLD-171]
  • EAPG crosswalk join normalized on both sides ahead of the Aug 21 hospital_rates code-format change. APC, RC, and APR-DRG paths already used LPAD / SPLIT_PART and were unaffected [CLD-159]
  • Verified every mapped_from_apc = true row has an APC sibling with byte-identical rates across all 1,318,965 groups. No change required [CLD-170]
• Traceability
  • Anesthesia conversion-factor formulas carry the payer's time_unit as an explicit parameter instead of a hardcoded literal, flowing through to canonical_method_params [CLD-164]
• Bug fixes
  • Plan-bridge network_name not matching the network spine silently dropped every hospital-posted rate for Health Net (504), HealthPartners (121), AultCare (155), University of Utah Health Plans (353), and BCBS Kansas City (47). Health Net PPO went 8,924 → 1,046,040 rates [CLD-162]
  • EPO removed from exchange_exclusions in variables_exchange.sql, which was dropping legitimate Individual EPO marketplace plans already self-tagged Exchange by the hospital (Oscar 567, Cigna Connect EPO Exchange) [CLD-172]
  • best_hospital_rate_score / best_payer_rate_score non-zero while the rate is NULL — traced to the best_payer_hospital_cols CASE block in merge_combined_abridged_joined_chunks.sql. Diagnosed only, no fix in this release [CLD-167]
• Data shifts worth knowing
  • Rated rate objects 2.16B → 2.05B (-4.90%). Payers 96 → 97, networks 299 → 302, providers 74,670 → 76,807
  • Physician Group -8.24% (-127.4M), which is two independent effects: the PG rate object space contracted 11.8% for existing payers (1,683.2M → 1,484.0M objects, provider count flat), while Claritev added 193.1M objects of which only 22.8M rated. Rated share improved 92% → 95%. Not traced to any 2026_06 ticket — needs engineering sign-off
  • Rate score 1 +50.33% is almost entirely Claritev: 163.5M of its rows land outside outlier bounds, at a median 850% of Medicare. Its 11.9% rated share is well below the 50-60% typical for PG, so its rates may not be landing as negotiated PPO amounts
  • Eight networks lost their payer MRF entirely (payer-sourced rates → 0, hospital-posted retained): Cigna NY / New England / TN HMO, BCBS Minnesota Aware PPO and HMO, Healthfirst NY Essential Plan, WPS PPO, Providence PEBB Choice

v3.0.0 Release Notes

• Accuracy
  • Tightened counterparty validation to a flat 5% tolerance across all provider types (was 20%/10% with a $15k threshold)
  • Medical validation now requires 5% agreement in both directions with both sides non-outlier — prevents outlier-to-outlier cross-validation
• Codeset / Data
  • Added 26 Drug SSP/Episode J-codes (24 J-codes + 2 biosimilars) to PG codeset
  • Combined PG + infusion center NPIs in provider.sql — fixes NPI drops for dual-type providers
  • Coalesced HPT v3 median_amount with estimated_allowed_amount for hospital MRF rates
  • Excluded venipuncture codes (36415, 36416) and HCPCS Category II F-codes from is_surg_code heuristic
• Plan Bridge / Network Mappings
  • Cigna (payer 76): added plan_name ranking hierarchy
  • BCBS Kansas City (payer 47): prefer Preferredcare over Preferredcare Blue
  • UHC (payer 643): added Options PPO network
  • Medical Mutual of Ohio (payer 272): prefer SUPERMED COMMERCIAL plan_name
  • Added HMO/PPO exclusion exceptions (affects Priority Health, payer 299)
• Traceability
  • Fixed payer_hospital traceability join that was dropping opposite-side metadata (hospital or payer columns NULL depending on which side won)
  • Fixed per_diem traceability: OP rates now show {rate}; IP rates keep {per_diem} × {glos} decomposition
• Bug Fixes
  • Fixed combined_brit imputation filter dropping non-first payers
  • Fixed ambiguous service_code column in build_payer_mrf

v2.4.1 Release Notes

Updates:

• Professional Rates now have Modifiers 26 and TC. These new ROIDs will not be loaded to the API yet, but they will be included in the underlying data release
• Adds hospital subtype: Long Term Acute Care Hospitals
• Adds networks: Cigna National PPO, Aetna National PPO, UHC Select EPO
• For Exchange, start pulling payer data as well (we'd only been using hospital data for Exchange)
• Added Nomi rates for Consumer
• Cleans up canonical_gross_charge so that canonical_gross_charge is only populated if greater than canonical_rate
• Include service code 23 for professional fees
• Use provider_parent_system as the health_system_name for labs
• SelectHealth network naming and product_network_label selection updates

v2.4.0 Release Notes

• Provisions
  • Introduced limited-scope, data-only provision-based features to support UX ideation and early data feedback
  • New Provisions Table including:
    • Implant carveouts
    • Drug carveouts
    • Stoploss
    • Inpatient base rate / percentage / per diem
    • Outpatient surgical groupers
    • Draft table: tq_dev.internal_dev_mmalhotra.provisions_final
  • Implant carveouts:
    • Uses inferred provisions and DevPR to generate bundled rates for device-dependent procedures
    • Logic currently applies to a limited set of providers
    • New fields added to prod_combined_abridged
• Taxonomy updates
  • Added taxonomy_classification
  • Added taxonomy_specialization
  • Fields added to the API table
• Provider updates
  • provider_subtype now includes all physician group specialties
  • Physical Therapy Physician Groups added

v2.3.2 Release Notes

• RC to HCPCS imputations
• Best payer and best hospital rate columns to kick off API work on filtering templates to payer- or hospital-data only
• Add remaining OPPS codes that were previously excluded. We should now have all OPPS codes
• Bug fix Exchange plans that were previously including many "Commercial" payer_class_name rates from hospital MRF

v2.3.1 Release Notes

• For Market Solutions:
  • New Networks and Payers:
    • HealthPartners
    • PEHP
    • Capital Health
    • HealthLink
    • SelectHealth
    • The Alliance
    • HealthSmart
    • Medica
  • Add ASP Dosage Amount and Unit to prod_combined_abridged
  • ML to MG Conversion Logic
  • Add PLA Codes
• Cigna Anesthesia transformation update
• Ensure only single affiliated PGs get hospital MRF rates
• ASC and PG Provider Subtypes

v2.3.0 Release Notes

• Provider Types:
  • Infusion Center Spines [DATA-243]
  • Lab Spines (select labs w/ national coverage) [DATA-242] | [DATA-235]
• Add CAR-T DRGs + Plausibility [DATA-232]
• Accuracy Methdology Updates see accuracy-scores page for details
  • PG-specific accuracy methodology [DATA-238]
  • ASP/Medicare-based accuracy methodology for drugs
  • Anesthesia-specific outlier bounds
• Outpatient Outlier Lower Bound increased to 50% of Medicare [DATA-237]
• OPG imputation identification uses stricter parameters [DATA-240]
• Exchange plans

v2.2.2 Release Notes

• Labs
  • Guardant rates are now available for all payers
  • Implemented lab-specific accuracy scoring methodology
• More Hospital MRF Rates:
  • Include rates for Providers Posting HCPCS as Inpatient. This adds 260 more providers with OP hospital rates.
  • Include rates for Providers ALWAYS Posting DRGs w/ Rev Codes. Previously, we only included rates with NULL Rev Codes. Affects just a handful of providers, but important ones!
• New Payers:
  • Added SummaCare
• Misc:
  • Always populate medicare_rate in prod tables
  • Remove "capitation" negotiation_arranagment from payer MRF data pulls (only happens 0.02% of the time, but on average these rates are 7x higher than ffs rates)
  • Validated Rates tiebreaker fix: in v2.3.0 we started selecting the highest rate. This introduced problems where a hospital PoC and payer PoC validated each other, because then the highest GC would be used (even if it were a state-average when a hospital-reported GC is available). Now, in this case, we rely on the hierarchy of GC sources to break ties.

v2.2.1 Release Notes

• Tables are in tq_production!
• Accuracy :
  • Validated rates: select the higher rate as the canonical rate
  • Anesthesia rates: always prioritize normalized over raw (@Stefano DiSano)
• Life Sciences:
  • Add NDC to Drug Code Crosswalk for Hospital rates with an NDC but NULL billing code (ty @Monika Malhotra for PRs!)
• Employer:
  • A Network Check version of Clear Rates enhanced w/ extra imputation tiers
• Medicare:
  • Update to Fixed Medicare Reference Data Tables (we had been using static ones)
  • Lab Medicare Fee Schedule for Physician Groups (@Sam Jones)
• Plan Mappings:
  • Fix Premera Blue Cross mapping so we have AK rates for Alaska Heritage Network (@Hunter Brown | thread)
  • Fix Wellmark Bluecard to use PPO plan and rename from IA HMO to IA PPO (@Sam Jones | thread)
  • Implement Plan Bridge Hierarchies for MO Blue (@Stefano DiSano)
  • IBC Hospital Data added for Main Line Health (@Stefano DiSano)
• Misc:
  • Round all rate features to 2 decimals

v2.2.0 Release Notes

• Highlights:
  • Analyze: features for next week's lab template demos
  • Adds is_lab_code indicator
  • Adds medicare reference pricing for the new lab codes
• Spines:
  • We built Physician Group Spines and Clear Rates is now using it (s/o @Nick Pattison for the design and architecture, it's quite a technical marvel he cooked up!). This will help to boost coverage and improve rates for Employer Network Check reports and add more PGs for Analyze.
• Market Solutions:
  • Two New Drug Dosage Standardization Methods (s/o @Monika Malhotra for the research and PRs to add these net new transformations to the Clear Rates codebase!)
  • Uses new unit/quantity fields in Hospital MRF data to dosage standardize
  • Uses NDC field from Hospital MRF data + strength/strength_unit from Medispan to dosage standardize