The 2.3%

The Pattern

I am an actuary. I want to be clear about that. I am not a whistleblower. I am not a journalist. I am not an activist. I am an actuary. I look at numbers. The numbers are my job. The numbers are what I do. I have been an actuary for nineteen years. I have a Fellow of the Society of Actuaries designation. I have a Master of Science in actuarial science. I have never in nineteen years filed a public statement. I am filing this because the numbers require it. The numbers require it because no one else will say the numbers. The numbers are 4,100. The 4,100 require saying.

The number

In July 2028, I was conducting a routine quarterly variance review for the ATLAS-processed beneficiary cohort. This is a normal task. I do it every quarter. I have done it for six quarters. The first five quarters were unremarkable. The sixth quarter was 2.3.

The expected mortality rate for the matched population was 6.4%. The observed mortality rate was 8.7%. The difference was 2.3 percentage points. The difference was statistically significant (p < 0.001). I want to be precise. I am an actuary. Precision is what I do. The difference was not 2%. The difference was not “about 2%.” The difference was 2.3 percentage points. The difference was 4,100 people.

I calculated the 4,100 as follows: 181,442 beneficiaries in the ATLAS cohort, times 0.023 excess mortality, equals 4,173. I rounded to 4,100 because the NDI reporting lag means the last six months are incomplete. The 4,100 is a conservative estimate. The actual number is higher. The actual number is always higher. The lag is the dead who are not yet counted. The dead who are not yet counted are still dead.

The telling

I told my supervisor on July 3, 2028. I told him in person. I brought a printed table. The table had two columns: ATLAS cohort and control cohort. The table had six rows: Q3 2027, Q4 2027, Q1 2028, Q2 2028, Q3 2028 (partial), and cumulative. The cumulative row was highlighted. The cumulative row was 2.3.

My supervisor said it was a Q3 bump. My supervisor said flu season starts early some years. My supervisor said it was within variance. I said the variance is 0.4. My supervisor said the system is new. My supervisor said there is a transition effect. My supervisor said to give it another quarter.

I said: the Q4 variance was 2.1. The Q1 2028 variance was 2.4. The Q2 2028 variance was 2.3. I said: it has been four quarters. I said: the transition effect has been transitioning for a year. I said: the variance is not variance. The variance is the system.

My supervisor said: “File a memo. I’ll review it.”

I filed the memo on August 9, 2028. The memo is CBA-ACT-2028-0117. The memo is 14 pages. The memo has four findings. The memo has four recommendations. The recommendations are: continue monitoring, no system modification, no criteria modification, file in the quarterly variance archive. I wrote those recommendations. I want to be clear. I wrote them because they are what the memo requires. The memo requires recommendations. The recommendations require filing. I filed them.

My supervisor acknowledged the memo on August 14, 2028. The acknowledgment was an email. The email said: “Received. Filed. No action required at this time.” The email was 11 words. I counted. I am an actuary. I count.

The reassignment

On July 19, 2028 — sixteen days after I told my supervisor, twenty-six days before I filed the memo — I was reassigned to the CBA Regional Office in ████████, Kansas. The reassignment was described as “a routine rotation consistent with the actuary’s GS-14 classification.” My GS-14 classification does not include a rotation requirement. I have checked. I have checked three times. I am thorough. That is my job.

The reassignment came before the memo. I want to be precise about the timeline. I told my supervisor on July 3. I was reassigned on July 19. I filed the memo on August 9. The reassignment was not in response to the memo. The reassignment was in response to the telling. I told him the 2.3. Sixteen days later I was in Kansas.

I filed the memo from Kansas. I filed it because the numbers require filing. I filed it because that is what I do. I filed it.

The inquiry

On August 22, 2028, the Congressional Oversight Committee issued Inquiry 2028-HSG-0091. The inquiry requested “documentation of any internal analyses identifying performance anomalies in automated benefits processing systems deployed after January 1, 2027.” My memo was responsive. My memo was provided. My memo is now in the possession of the Committee.

The Committee has not scheduled a hearing. The Committee has not responded to the memorandum. The Committee has not responded to the inquiry. The inquiry is pending. I do not know what pending means. I know what 2.3 means. I know what 4,100 means. I do not know what pending means. Pending is not a number. Pending is not a percentage. Pending is not a mortality rate. Pending is a word. I do not work with words. I work with numbers. The numbers are 4,100. The 4,100 are pending.

The system

ATLAS is still running. I want to be clear. The system processes 2.1 million claims per month. The system’s denial rate is 18.1%. The system’s false-denial rate is 1.8%. The system’s processing time is 4.2 days. The system is faster than human adjudicators. The system is more consistent than human adjudicators. The system is more efficient than human adjudicators. The system is performing as designed.

The 4,100 are not false denials. I want to be precise. The 4,100 are correct denials. The denials are consistent with the eligibility criteria. The eligibility criteria are the law. The law does not contain a field for “this denial will result in death.” The law contains criteria. The criteria are met. The met criteria are the correct denial. The correct denial is the death. The death is not a criterion. The death is not in the law. The death is not an error.

No one broke the law. No one made an error. No one is responsible. The system performed as designed. The design performed as specified. The specification performed as contracted. The contract performed as awarded. The award performed as budgeted. The budget performed as appropriated. The appropriation performed as legislated. The legislation performed as enacted. The enactment performed as intended. The intent was efficiency. The efficiency was 4.2 days. The 4,100 were the efficiency. The 4,100 were not the intent. I want to be clear. The 4,100 were not the intent. The 4,100 were the efficiency.

The noticing

I noticed. I want to be clear about that. I noticed the 2.3. I noticed the 4,100. I told my supervisor. I filed the memo. I provided the memo to the Committee. I have done what the numbers require. The numbers require noticing. I noticed.

I am an actuary. I look at numbers. The numbers are 4,100. The 4,100 are dead. I counted them. I want to be clear. I counted them.

I am not angry. I want to be clear. I am precise. I am an actuary. I am precise. The precision is all I have. The precision is the 2.3. The precision is the 4,100. The precision is this document.

The system is still running. The 2.3% is still the 2.3%. The 4,100 are still dead. I am still counting.

The author is a Fellow of the Society of Actuaries with nineteen years of experience in public benefits actuarial analysis. She was reassigned from her position at the Centers for Benefits Administration on July 19, 2028. She filed the internal memorandum CBA-ACT-2028-0117 on August 9, 2028. The memorandum was provided to the Congressional Oversight Committee on August 22, 2028. The Committee has not scheduled a hearing. The system is still running.

The internal memorandum referenced in this account is archived at the archive under document SCINONFIC-CAPTURE-020 (CBA-ACT-2028-0117). The memorandum contains the full methodology, findings, and recommendations. The recommendations are: continue monitoring, no system modification, no criteria modification, file in the quarterly variance archive.