Corrections & Retractions
We keep a public record of findings we have withdrawn or revised, with dates. Nothing on this site is meant to read as a permanent claim: model outputs reflect a specific build on a specific date, and when we find that an earlier result was wrong — or was an artifact of a bug — we say so here rather than quietly changing the numbers.
If you are citing a figure from this site, cite it as of the build date noted on the page, and check this log for any later revision.
July 2026 — five findings retracted following an engine audit
A systematic audit of the calculation engine in July 2026 found four defects. After fixing them and re-validating against the SSA's published benchmarks, the following earlier headline findings did not survive and are formally withdrawn:
| Earlier claim (pre-audit) | Status as of July 2026 |
|---|---|
| "The DC alternative beat Social Security in 43 of 43 scenarios (100%); minimum 1.66×." | Retracted. On the corrected engine, all-fixed-income allocations lose to Social Security at nearly every income level. The universal-win claim was false. |
| "The advantage generally rises for younger generations." | Retracted. This was an artifact of the frozen-earnings and missing-wage-indexing defects. The corrected benefit formula is cohort-invariant; remaining cross-cohort differences reflect each cohort's actual market-history window, not formula bias. |
| "Education was tested and found inert — identical to the cent." | Retracted. An artifact of a preset field overriding demographic inputs. Demographics move benefit levels substantially; they move the DC-vs-SS ratio only about ±10%. The August 2026 audit additionally validated the size of the education effect: it corresponds to a back-inferred college/HS career premium of ~1.41×, within the observed U.S. range. |
| "Population-weighted: DC beats SS for 100% of workers under every investment band." | Retracted. The robustness argument required every cell to win, which is false on the corrected engine. A rebuilt population-weighted analysis reports results separately by investment behavior instead. |
| Retirement-age "tier-dependent reversal." | Not reproduced / withdrawn. On the corrected engine, later retirement widens the DC advantage in both fixed-income and equity tiers, with no reversal. |
Why we published this rather than burying it
Any calculator can show numbers. Very few will show where their numbers used to be wrong. We found these defects in our own model — several of which had been inflating our own thesis — fixed them, and published the retraction. We consider that the most valuable credibility statement the project can make, and we would rather do it than present figures we cannot defend.
July 2026 — near-retirement cohort SS benefit overstatement corrected
A cross-validation against the SSA Quick Calculator found that the model overestimated Social Security benefits for near-retirement cohorts (workers past their earnings peak) by roughly 2×. The career-scaling anchor inverted the age-earnings profile at the worker's current age; at deep post-peak ages the profile multiplier is floored at the age-61 value (an artifact of Census unconditional means averaging in retired workers), so anchoring there deflated the denominator and inflated the entire career. The anchor is now floored at 0.97 of peak, so a still-employed worker's current income is treated as approximately peak earning power rather than the retirement-attrition-deflated late-career value.
| Test case (male, retire 67) | Before (real 2026) | After (real 2026) | SSA Quick Calculator |
|---|---|---|---|
| Born 1960, $48K current income | ~$3,100/mo | ~$1,730/mo | $1,558/mo |
| Born 1970, $55K current income | ~$2,100/mo | ~$1,785/mo | $1,943/mo |
| Born 1980, $65K current income | ~$2,150/mo | ~$2,150/mo (unchanged) | $2,343/mo |
| Born 1980, $120K current income | ~$3,175/mo | ~$3,175/mo (unchanged) | $3,445/mo |
All four cases now fall within ±15% of the SSA Quick Calculator. Younger cohorts (at or near their earnings peak) are unaffected because their current-age profile multiplier is already ~1.0; the correction only raises the anchor for post-peak workers whose late-career multiplier had been deflated by the retirement-attrition floor.
Side effect (honest, not a bug): because the Social Security leg of the comparison falls for older cohorts while the defined-contribution leg is unchanged, DC/SS ratios rise for near-retirement cohorts under this correction. That is the fix working as intended — the prior ratios were inflated on the SS side. The headline grid and replacement-rate validation figures published elsewhere on this site were produced on the pre-correction engine and are being re-checked; figures tied to near-retirement cohorts may shift in later builds.
August 2026 — Audit v2 corrections
A second independent audit in August 2026 (documented on the Methods Audit page) targeted the app around the engine. The following earlier behaviors are corrected or retracted as of that audit:
| Earlier claim / behavior (pre-August 2026) | Status as of August 2026 |
|---|---|
| The app's spousal benefit at claiming age 62 equaled the full 50% top-up, and appeared to grow past FRA. | Corrected. SSA reduces spousal benefits claimed before FRA (to ~65% of the top-up at 62) and pays no delayed credits after FRA. The engine now applies the reduction; the apparent post-FRA growth was a display deflation error, also fixed. Verified claiming curve (real 2026, single-earner couple): ~$2.8K / ~$3.2K / ~$3.2K at spouse ages 62 / 67 / 70 (was ~$3.2K / ~$3.2K / ~$3.4K). |
| PDF reports stated "Trust Fund Shortfall: Disabled" while showing benefit-cut figures. | Corrected. The label is now bound to the live scenario and verified in both modes. Reports generated before August 2026 may carry the wrong label; the figures themselves were correct for the benefit-cut setting. |
| "Ask the Model" quoted benefit amounts "from the app's modeling engine." | Retracted. The assistant had no connection to the engine and the quoted figures did not match it. It now declines to quote benefit numbers for described scenarios and routes to the simulator. If the assistant gave you a specific dollar figure before August 2026, disregard it and run your details through the simulator. |
| Survivor benefits under the 2033 benefit-cut scenario. | Corrected. Survivor benefits were paid at the full scheduled rate while all other post-2033 benefits were reduced — overstating Social Security in survivor scenarios by roughly the cut percentage. The reduction now applies uniformly. |
Spousal claiming-age correction (August 2026). The spousal top-up is capped at 50% of the primary earner's PIA, reduced for early claiming, with no delayed credits above FRA. The engine was keyed to the wrong filing age for the reduction, and a calendar-year deflation error made post-FRA benefits appear to grow. The corrected curve is flat above FRA: a single-earner couple sees ~$3.2K at FRA-67, ~$2.8K at 62 (the 0.65 floor), and ~$3.2K at 70 — no step-up. The pre-correction curve showed ~$3.4K at 70, which was the top-up inflated by undeflated COLA adjustments, not delayed credits.
Standing note on reliability
- All figures on this site are model outputs as of a stated build date, not predictions or guarantees, and may change in later versions.
- Past performance does not guarantee future results. Historical return series used by the model describe the past; the future may differ.
- Where a page shows a deterministic figure and a Monte Carlo figure that disagree, treat the Monte Carlo figure as the more reliable one.
- This is an educational tool, not financial advice.
Have you found something you believe is an error? We welcome methodology challenges — see the Contact page. A model built to be checked needs checkers.
Educational purposes only — not financial, tax, or investment advice. Past performance does not guarantee future results.