ConstantForce whole-life insurance ignores the benefit amount

GitHub · Hugging Face · LinkedIn · Maintainer issue

Frozen evidence archive (0.65 MB) · Reproduction guide · Candidate patch

Research result verified 15 September 2026. Evidence edition prepared for publication on 15 September 2026.

ConstantForce.whole_life_insurance(..., discrete=False, moment=1 or 2) in actuarialmath returns the moment for a unit benefit even when the caller passes a different benefit b. The same class's term_insurance(t=WHOLE, ...) includes the benefit and returns the expected result.

The recorded current-source pin is 7d18f11ad304898f177b7922b3c53f70e4c2b4f4. The published PyPI wheel is version 1.1.0, SHA-256 b19990e4378aaa19fe6bc1182b4269faec6617cb62b0677fea1e624fbbb3ff6f. Its constantforce.py is byte-identical to the pinned current source. Both were executed, and their 144 main observation rows match exactly. The current source's project metadata still says 1.0.1; the source pin and wheel identity distinguish the two distributions.

Minimal example

from actuarialmath import ConstantForce

life = ConstantForce(mu=0.02).set_interest(delta=0.03)
print(life.whole_life_insurance(35, b=100000, discrete=False))
print(life.term_insurance(35, t=life.WHOLE, b=100000, discrete=False))

Recorded outputs are approximately 0.4 and 40,000. These calls describe the same continuous whole-life benefit under the class's constant mortality assumption. Changing age from 35 to 70 leaves the results unchanged, as this memoryless lifetime model requires.

For the second moment at the same parameters and benefit, the whole-life call returns approximately 0.25, while the expected answer is 2,500,000,000 in squared monetary units. A zero benefit also incorrectly returns a positive unit-benefit moment.

Independent oracle and candidate

With a lifetime T ~ Exponential(mu) and discounted benefit Z = b * exp(-delta*T), direct integration gives

E[Z**m] = b**m * mu / (mu + m*delta).

The main oracle uses exact rational arithmetic for the declared decimal inputs. The source's shortcut omits b**moment. The candidate adds this single factor in this branch. It does not change the generic variance branch or mortality assumptions.

Executed checks

• 144 observations: three positive mortality forces, four nonnegative interest forces, six benefit amounts and the first/second moments.
• Oracle failures: 120 original → 0 candidate → 120 original-formula mutation. The released wheel reproduces the same 120 failures.
• All 144 term_insurance(t=WHOLE) controls satisfy the independent oracle.
• 288 finite-term controls, at terms 1 and 10, pass their analytical oracle and remain byte-for-byte identical in all four result sets.
• All 24 unit-benefit main observations remain unchanged. The age check passes throughout. These controls overlap in purpose; they are not separate customer trials.
• Fixed tolerances: relative 2e-12, absolute 2e-14; unchanged between runs.
• All 91 original source files retain their recorded hashes. Only src/actuarialmath/constantforce.py differs in the candidate. The candidate was restored after mutation replay.

Runtime: Python 3.12.14; NumPy 2.3.5, SciPy 1.16.3, pandas 2.3.3, matplotlib 3.10.6. The source also imports IPython, which was installed in this isolated environment; the full resolved dependency set is in environment-requirements.txt. Numerical library thread counts were set to one. The full upstream test suite was not run. A separate portable runner verified the bundled source/wheel hashes, extracted fresh copies and reproduced every recorded main and finite-term numeric row exactly. The original evidence files were preserved.

Duplicate review and limits

Five saved GitHub issue/PR searches and the canonical 95-publication GERO catalog were checked on 15 September 2026. The related existing report, issue #3, concerns the generic whole-life variance formula squaring the first moment twice. This finding concerns the positive-moment shortcut in ConstantForce and a missing benefit factor; it is not a new version of that variance finding. No exact duplicate was found within the recorded search scope. This is not a guarantee of global novelty.

These are synthetic actuarial calculations on a public educational library. No insurer deployment, premiums charged, reserves booked, customer losses or production exposure were measured. Variance and other shortcut candidates are outside this correction's scope.

Primary contract references: source, official guide, PyPI release.

Research: Xamit Kadirbekov / GERO. AI-assisted preparation, with executed public Python code and an independent mathematical oracle. Upstream MIT license is preserved in the source archive.

Archive integrity

SHA-256: f9c41cdf7e788b0aab2f3c1c4c52b26a6b63959a59cbda0d68893fb202bf3d1b

Portable package verification · Component licenses