# Changelog (the admission record)

- 2026-11-09: replay_commercial_price_proof withdrawn (replacement compare_hospital_rates)
- 2026-11-09: prove_commercial_price withdrawn (replacement compare_hospital_rates)
- 2026-11-09: plan_commercial_price_proof withdrawn (replacement compare_hospital_rates)
- 2026-11-09: lookup_rates_for_tin withdrawn (replacement compare_hospital_rates)
- 2026-11-09: lookup_rates_for_npi withdrawn (replacement compare_hospital_rates)
- 2026-11-09: get_negotiated_rate withdrawn (replacement compare_hospital_rates)
- 2026-11-06: get_network_groups admitted (Groups count Medicare-enrolled clinicians only (reassignment); contract level, so a contract's HMO and PPO networks read as one; ACOs are not rolled up yet (CMS publishes ACO participants by legal name only).)
- 2026-11-05: get_plan_hospitals admitted (Most directories list clinicians, not facilities: strong evidence exists for 113,461 plan-hospital pairs in 2027, the rest is inferred from CMS's clinician affiliation (2026-05-29); absence is never 'out of network'.)
- 2026-11-05: get_network_system_change admitted (406 contracts compare like for like; contracts whose 2027 list is a stand-in or whose 2026 list came only through the API carry-back have no rows; 91 exit and 76 join signals, of which two are publicly confirmed; a signal is a directory signal to verify, never proof.)
- 2026-11-05: get_network_breadth admitted (Breadth is a share of what MA plans list in the county, not of every practising clinician; name-matched lists under-count (matched_share on the row); 2% of listings carry a ZIP with no county; coverage follows the directory build (stage 25: 99.17% of 2026 members, 96.42% of 2027 from 2027 sources).)
- 2026-11-02: is_provider_in_plan admitted ("Not listed" is a claim about the covered contracts only (99.1% / 96.1% of members); name-only listings (grade C/D, about 11.3M rows) carry no NPI and cannot answer for the provider they name.)
- 2026-11-02: get_plan_network admitted (Coverage is 99.1% of 2026 MA members and 96.1% of 2027 members from 2027 sources (stage 21): the open-gap contracts (HCSC, Geisinger 2027, Highmark Wholecare, Elderplan, Mercy Care, Elevate, CarePartners CT, six thin Elevance contracts) answer no_coverage; 42 contracts' 2027 rows are their 2026 directory through the crosswalk and are labelled; API-served listings carry no printed name or location.)
- 2026-11-02: get_network_change admitted (A unit whose 2026 directory came through the API only as a thin snapshot reads a very high joined rate (H5959: 92 members in 2026, 50,505 in 2027, joined 3,301 in the shared ZIP3 areas); the answer flags any joined rate over 20% as unverified, and a directory signal is never evidence that a contract ended.)
- 2026-11-02: get_medicaid_rate admitted (Coverage is state by state: 47 states hold rows and 147 of 255 state x service-line cells are served (measured 2026-11-02), so lookup answers not-found for an unbuilt state or line and says so; compare leaves out every state whose key carries more than one rate in the edition (about 11,800 ambiguous keys, NY, IA, NC most) and names how many, and converts time units to an hour but never another basis.)
- 2026-11-01: get_medicaid_mco_enrollment admitted (The guard is a label rule, not the publisher's own classification: a sole plan in a program is flagged equals_publisher_total_suspected (Florida: 11 rows), a total whose label carries no total word is not caught, and a total filed under its own program (Arizona's two combined-display rows, 1,404,364 and 75,580) is set aside but not paired with its plans, so Arizona's cells read source_total_without_plan_rows. Washington's 1,168 of 1,170 checked cells fall short of their total because masked 0s hide 1 to 10 members each. The ledger view version that classifies each publisher's totals is the Data Tracker's pending fix (measured 2026-10-05).)
- 2026-10-29: resolve_entity admitted (Layer one only: facilities, doctors, networks and PBMs are later layers (resolve_provider for a CCN or NPI). Exact matches first; a contains-match is offered only as a candidate. Walked in-process from this Mac before the bridge rolls; the release checks are its first live grade.)
- 2026-10-29: describe_entity admitted (A date written inside a sentence (an event label, an evidence field note) is removed by the customer scrubber; link dates are served bare with their reading beside them (from_is, to_is), so ownership windows survive. The date the caller asks for (on) is not echoed in answer.scope.filters on the customer wire (the door drops a bare date from the scope for every tool). scope=group on the largest parents reads many rows and is the slowest call.)
- 2026-10-26: get_volatility_index admitted (Every year is plans as filed against the prior September's members (what members face, not outcomes); the benefit part is a floor (OTC and giveback not counted) and starts in 2021. Walked in-process against the live table from this Mac before the bridge rolls; the release checks are its first live grade.)
- 2026-10-18: get_service_line_prices admitted (Hospital geography is the state partition, a CCN list or CMS's wage-index CBSA (metro divisions, e.g. Miami 33124), not city. Facility types are not separated (a long-term acute hospital can rank beside an acute one). Walked in-process from this Mac before the bridge rolls; the release checks are its first live grade.)
- 2026-10-18: get_contract_profile admitted (The record carries the 40 largest plans and 40 largest counties of a contract (plan_count and county_count are the full counts), and a segmented plan's benefits are one segment's filing. Walked in-process against the live build from this Mac before the bridge rolls; the release checks are its first live grade.)
- 2026-10-13: get_star_risk admitted (Walked offline over rows recorded live from the _v4 tables, not from the live customer seat (the bridge carrying the tool is not rolled at admission); the post-deploy walk is its first live grade. The Overall chance to drop and below 4.0 near the line are the locked blend's, tested held out 2024-2027; its odds sit somewhat too close to average (slope 1.33; 5.0 contracts under-predicted, 46% vs 65%), disclosed in every answer; Upside and Down / Same / Up are tested; the simulation's own odds are shown for comparison only; Part C / D are simulated shares; the expected star is simulated and runs low; the answer says each.)
- 2026-10-12: get_star_benchmark admitted (Walked in-process against the live tables from this Mac, not from the live customer seat; CAHPS cut points are approximate and family averages use each year's own measure set (the count is in every row).)
- 2026-10-09: get_stars_rollup admitted (Walked through the tool in-process against the live tables from this Mac, not from the live customer seat, because the bridge carrying it is not rolled at admission; the post-deploy walk is its first live grade.)
- 2026-10-09: get_star_decomposition admitted (Walked in-process against the live tables from this Mac, not from the live customer seat; survey (CAHPS) cut point distances are approximate, and the answer says so.)
- 2026-10-09: get_quality_bonus admitted (Walked in-process against the live table from this Mac, not from the live customer seat; PY2025-2028 are modeled and labeled, and very small and new-parent contracts are a labeled residual.)
- 2026-10-09: get_plan_value admitted (Walked through dispatch in-process against the live views from this Mac, not from the live customer seat; the rank mode reads a shelf table that is not yet a catalog card, so a seat cannot reproduce it by raw SQL.)
- 2026-10-09: get_benefit_change admitted (Walked through dispatch in-process against the live tables from this Mac, not from the live customer seat, because the bridge carrying it is not rolled at admission; the post-deploy walk is its first live grade.)
- 2026-10-06: get_contract_service_area admitted (Walked through dispatch in-process against the live table from this Mac, not from the live customer seat, because the bridge carrying it is not rolled at admission; the nightly walk is its first live grade. Contracts serving the whole country are left out of a state or county filter unless the caller asks for national service areas.)
- 2026-10-01: state_expert admitted
- 2026-10-01: get_stars_cutpoints admitted
- 2026-10-01: get_stars admitted
- 2026-10-01: get_landscape admitted (Rows are one per plan segment; the Plan and Premium family (the Part C / Part D premium split before CY2025) is stored one year early for CY2017 to 2023 (the Tracker's LAND-3, a loader fix) and is not read by this tool, so part_c_premium and part_d_premium are null before 2025.)
- 2026-10-01: get_enrollment admitted
- 2026-10-01: get_county_view admitted (benefit_change is STATE grain under the code COUNTY_GRAIN_NOT_HELD on every county (the county-grain benefits table is not published); measured 2026-09-30 on Cook, Maricopa and DuPage, status partial with sections_unavailable naming benefit_change_county on each.)
- 2026-09-30: get_landscape_exits admitted (The county grain (1.1.0) was walked through finalize_answer against the live table from this Mac, not from the live customer seat, because the bridge carrying it was not rolled at admission; the nightly walk is its first live grade. County names match CMS spelling only ('Cook', 'Maricopa'; 'St. Louis' as CMS writes it), and a misspelling is an honest not-found, not a suggestion.)
- 2026-09-30: get_landscape_benefit_change admitted (1.1.0 (county grain) and 1.2.0 (same-members change) were walked through the door in-process against the live tables from this Mac, not from the live customer seat (bridge not rolled at admission); the nightly walk is their first live grade and the fixture's routed case reads red on the live surface until the bridge serves 1.2.0. Y+1 stars stay empty for 2026 to 2027 until CMS assigns them, at every grain.)
- 2026-09-30: find_views admitted (Walked in-process through finalize_answer on this Mac (the three fixtures and the three free-trial starter questions), not from the live customer seat: the bridge that serves it and migration 129 that entitles it were not rolled at admission, so the nightly walk is the first live grade and the seat answers entitlement_denied until the migration is applied.)
- 2026-09-19: submit_query admitted
- 2026-09-19: resolve_provider admitted
- 2026-09-19: replay_hospital_price admitted
- 2026-09-19: recall_spend_window admitted (the tool's own DESCRIPTION still quotes a Medicaid-leg reference figure ($63,724,880.54) that the live answer does not reconcile to ($30,181,579.43 — the difference is a national roll-up double count fixed in #1801), with no restatement caveat. The live number is the right one; the description is the stale one)
- 2026-09-19: query_substrate admitted
- 2026-09-19: query admitted
- 2026-09-19: prove_ma_enrollment_trend admitted (Its own demo call (Humana Inc., 2024, FL) took 41-46s on the first machine walk (run 35516039355, 2026-09-20) — past the 40s synchronous dispatch budget — so a customer asking the documented question gets a SCAN_LIMIT_EXCEEDED refusal rather than the number, not on every call but whenever the estate is cold. Since 2026-09-20 that refusal is a real envelope with a working route (the tool is now in ASYNC_HEAVY_TOOLS, so it names submit_query with the same arguments); before that it was a bare error object with no status at all. Making the sync path finish is the actual fix and it is not done.)
- 2026-09-19: prove_hospital_price admitted
- 2026-09-19: poll_query admitted
- 2026-09-19: plan_hospital_price admitted
- 2026-09-19: pbm_compensation_rollup admitted
- 2026-09-19: medicare_allowed admitted (The `medicare_year` pin is still not on the SERVED inputSchema. Measured from the customer seat 2026-09-20: tools/list publishes {billing_code, care_class, setting, limit}, so the bridge drops `medicare_year` and the tool answers for its default year rather than the pinned one — silently, because an argument that never reaches the door cannot be refused by it. The handler validates the pin correctly; the published schema is what is stale, which makes this a catalog- publish defect and not a tool one.)
- 2026-09-19: how_to_join admitted
- 2026-09-19: get_pbp_benefit admitted
- 2026-09-19: get_market_share admitted
- 2026-09-19: get_hospital_quality admitted
- 2026-09-19: get_hos admitted
- 2026-09-19: get_exchange_plan_profile admitted
- 2026-09-19: formulary_churn admitted
- 2026-09-19: find_data_sources admitted
- 2026-09-19: fetch_query admitted
- 2026-09-19: excluded_prescriber_payment_screen admitted
- 2026-09-19: describe_schema admitted
- 2026-09-19: describe_data_source admitted
- 2026-09-19: compare_hospital_rates admitted
- 2026-09-19: certified_measure admitted
- 2026-09-19: asp_pricing_lookup admitted (The reference call (J0129) took 31s and then returned 502 from the edge on the first machine walk (run 35516039355, 2026-09-20). 31s is INSIDE the 40s synchronous dispatch budget, so the door never timed out and has nothing to restate: the answer was built and the connection was dropped above it. The tool is not in ASYNC_HEAVY_TOOLS, so a caller who hits this has no background lane either. Owner: the serving lane (the bridge deadline / ALB idle timeout), not the door.)
- 2026-09-18: search_tools withdrawn (replacement find_data_sources)
- 2026-09-18: ira_mfp_exposure withdrawn (replacement find_data_sources)
- 2026-09-18: ira_drug_label_risk_screen withdrawn (replacement find_data_sources)
- 2026-09-18: get_drug_source_inventory withdrawn (replacement find_data_sources)
- 2026-09-18: coverage_status withdrawn (replacement state_expert)