PastMe: How Contura Reconstructs Your Past Body in 3D — and Why the Science Behind It Holds Up
PastMe: How Contura Reconstructs Your Past Body in 3D — and Why the Science Behind It Holds Up
Memory lies. Your body doesn’t.
“I used to look better.” But did you, really? Most people remember themselves leaner, fitter, and smaller than they actually were. Answer a few questions about your past, and Contura’s PastMe feature rebuilds it in 3D. The truth might surprise you — and that’s exactly why your progress feels invisible.
TL;DR — What PastMe Does, in One Paragraph
PastMe is a feature inside the Contura app that reconstructs a 3D model of your body at a specific point in your past — six months ago, two years ago, before a life event — using a short questionnaire and your current measurements as the anchor. It is not a guess. It is a multi-stage reconstruction pipeline grounded in peer-reviewed anthropometric research (Heymsfield, Gallagher, NHANES, ANSUR II), validated against 96 golden-sample body profiles, and protected by physiological hard-clamps that prevent any output from falling outside human-possible ranges. Average reconstruction error in typical use cases is ±5–15% per circumference — comparable to a manually held tape measure on a moving target.
Contura is the app. PastMe is one of its core features. This article explains how it works.
Why Memory of Your Own Body Is Unreliable
Body-image research consistently shows that human memory of one’s own physical form is systematically biased. People recall themselves as having been thinner, fitter, and smaller than they actually were — particularly when comparing past selves to a present self they are dissatisfied with. The bias is well-documented across self-perception literature, and it has a real consequence:
You feel like you’ve made no progress, because the version of “past you” that lives in your head is fictional.
The whole point of PastMe is to replace that fictional past with a reconstructed one — a model built from physical signals, not nostalgia. When you see the real shape of where you started, the progress you’ve actually made stops being invisible.
This is also why simple “before photos” don’t solve the problem. Most people don’t have a flattering photo from the exact moment they want to compare against. PastMe rebuilds that moment from data.
The Core Problem: Reconstructing a Body From a Questionnaire
PastMe solves what is technically known as an under-constrained inverse problem. The user supplies:
- A few qualitative perceptions (“I was heavier”, “more apple-shaped”, “my belly stuck out more”)
- Optional quantitative inputs (an exact past weight, a remembered waist measurement)
- Their current measurements and recent measurement history (stored in Contura)
From this, the algorithm must output eight or more circumference values — waist, hip, chest, thigh, arm, calf, neck, and others — at the requested past time point, plus a 3D model parameterization. There is no single “correct” answer derivable from the inputs alone. So how do we make the output trustworthy?
The answer is layered constraints, where each layer narrows the space of plausible answers using a different kind of evidence — and where the strongest evidence always wins.
The PastMe Pipeline: Seven Stages of Constraint
PastMe is a seven-stage reconstruction pipeline. Each stage produces a candidate body, and each successive stage refines the candidate using more specific information. The design philosophy is:
Deterministic signals (precise numbers) > Statistical signals (historical trends) > Qualitative signals (questionnaire perceptions).
Higher-priority signals override lower-priority ones, while later stages enforce hard constraints that earlier stages cannot violate.
Stage 1 — Current Body Anchor
Before reconstructing the past, PastMe builds a complete picture of your present body. Most users have measured only some of their circumferences (often just waist and weight). The algorithm uses a constraint-propagation engine grounded in ANSUR II and similar large-scale anthropometric databases to fill in the missing circumferences from your height, sex, weight, and any measurements you have provided.
This stage is critical because it gives the rest of the pipeline a complete, individually accurate baseline. PastMe never reconstructs the past from zero — it reconstructs the past as a perturbation of the present.
Stage 2 — Past Weight Estimation
Weight is the anchor of the entire reconstruction. Research consistently shows that humans remember weight more accurately than any other body measurement — far more accurately than circumferences or proportions. So PastMe treats weight as the highest-confidence signal it can extract, and resolves it through a three-tier decision:
| Priority | Signal | Trust Level | Handling |
|---|---|---|---|
| 1 | User-supplied exact past weight | Highest | Used directly, with an annual-rate sanity clamp |
| 2 | Extrapolated trend from your historical weight log | High | Robust regression with outlier filtering |
| 3 | Qualitative perception (“I was heavier / lighter / about the same”) | Moderate | Mapped to an empirically calibrated annual rate of change |
The qualitative tier uses an annual rate of change derived from NHANES population data — the U.S. National Health and Nutrition Examination Survey, with sample sizes exceeding 10,000 — so even when the only signal is a user’s vague perception, the math behind the estimate is grounded in epidemiology, not invented.
Stage 3 — Weight-to-Circumference Coupling
This is the mathematical core of PastMe. Given a change in weight, how much does each circumference change?
The relationship is approximately linear within a moderate weight-change range (under roughly 20 kg), and the linearity is supported by decades of anthropometric and body-composition research:
- Heymsfield et al. — abdominal fat distribution studies establishing per-kilogram waist circumference change ranges
- Gallagher et al. — body composition research providing per-kilogram hip and chest circumference change ranges
- NHANES — large-scale population data validating coupling coefficients for limbs and other regions
PastMe uses research-derived median values for each body part’s coupling to weight. We deliberately chose medians (not extremes) so that, across the population of real users, errors are symmetrically distributed: roughly half of users will see slightly higher than predicted change, roughly half slightly lower, and the deviation from the median is bounded.
Why this is non-trivial: Many naive body-modeling approaches use coupling factors that are 3× to 12× too low across different body parts, because they confuse “how sensitive is this body part to weight change” with “what fraction of total body change does this part account for.” Those are different quantities. Our internal validation against research literature surfaced this exact failure mode and we corrected for it. (Saying more than this would give the recipe away.)
Stage 4 — Body Shape Adjustment
The same weight change affects different body shapes differently. An apple-shaped person gains more in the waist; a pear-shaped person gains more in the hip. Stage 4 takes the user’s reported past body shape — apple, pear, hourglass, rectangle — and adjusts circumference ratios (waist-to-hip, waist-to-chest) toward that shape’s typical proportions, while preserving the total weight-driven change from Stage 3.
The blending is intentionally partial. Body shape memory is subjective; weight-driven change is physical. Stage 4 modulates proportions without overriding physics.
Stage 5 — User-Supplied Circumference Anchoring
If a user remembers a specific past circumference (“I know my waist was 80 cm”), that input takes priority over the Stage 3 estimate for that body part. The provided value anchors that circumference directly, and the algorithm propagates the anchored ratio to related body parts (limbs at a reduced transmission rate, since limb circumferences typically change less than torso circumferences for a given weight change).
Stage 6 — Qualitative Fine-Tuning
The questionnaire’s qualitative signals — belly prominence, face roundness, the area where the user noticed the most change, life events such as childbirth or muscle gain — produce small percentage adjustments to the reconstruction. Each adjustment is:
- Bounded: hard upper limits prevent any qualitative signal from producing implausibly large changes
- Time-scaled: the adjustment grows with the time elapsed using a sub-linear function (so the difference between “1 year ago” and “4 years ago” is roughly 2×, not 4×, matching empirical observations of how body change rates plateau)
- Capped at the limit: the time scaling itself has a ceiling, so even reconstructing 10 years into the past does not over-amplify qualitative signals
Stage 7 — Physiological Validation
The final stage is the safety net. Five physiological constraints are enforced unconditionally:
- Each circumference must lie within a sex-, height-, age-, and weight-conditioned plausible range derived from anthropometric databases
- Waist circumference cannot exceed chest circumference by more than the upper bound observed even in extreme apple-shaped physiologies
- Waist-to-hip ratio cannot fall below sex-specific physiological floors
- Body Mass Index must lie within human-survivable bounds
- All values must be strictly positive
No matter what input the user supplies — including malformed or adversarial input — the output is guaranteed to describe a body that a human could actually have. This is a hard mathematical guarantee, not a soft heuristic.
Personal Coupling Calibration: How PastMe Adapts to Your Body
The population-average coupling values from Stage 3 are accurate to within roughly ±30% for most users. But a given individual’s actual coupling may sit at the high or low end of the research range. If you’ve been logging measurements in Contura for a while, PastMe can do something more powerful: calibrate the coupling factors to you specifically.
When sufficient paired weight-and-circumference history exists, the algorithm performs a robust regression across your own historical data points to estimate your personal coupling for each body part. The regression is robust in the statistical sense — it tolerates measurement noise (a few centimeters of tape-measure error) without breaking — and is filtered to discard contradictory data (for example, weight gain coupled with circumference loss, which typically indicates muscle gain or edema rather than the kind of body change PastMe is modeling).
The personal coupling is then blended with the population coupling using a data-quality score that considers:
- How many usable data points you have
- The range of weights covered
- The time span of your measurements
- The internal consistency of the regression slopes
When your data is rich and consistent, the algorithm trusts your personal coupling. When it’s sparse or noisy, it gracefully falls back to the population value. This means PastMe never gets worse because of bad data — only better when good data is available.
In internal accuracy tests, personal coupling calibration reduces reconstruction error by roughly 60% for users whose true coupling differs significantly from the population median.
The calibration mechanism, the data-quality scoring, and the blending behavior are each protected by their own clamps to prevent overfitting. Specific weights and thresholds are part of Contura’s proprietary tuning.
Accuracy: Three Layers of Guarantee
When users ask “how accurate is PastMe?”, the honest answer has three layers, each with a different kind of guarantee:
| Layer | Question | Guarantee |
|---|---|---|
| Safety | Is the reconstruction within human-possible bounds? | Hard guarantee — Stage 7 clamps are unconditional |
| Direction | Does the change go the right way? (Weight up → all circumferences up) | Hard guarantee — built into the linear coupling structure |
| Magnitude | How close is the number to your actual past circumference? | Statistical guarantee — bounded by research-derived parameters |
Expected Error Budget
The magnitude error depends on the quality of input signals available. Roughly:
| Scenario | Expected Circumference Error |
|---|---|
| Exact past weight + remembered circumference + personal calibration | ±3–5% |
| Exact past weight + body shape + personal calibration | ±5–10% |
| Exact past weight + body shape (population coupling only) | ±10–15% |
| Exact past weight only | ±15–20% |
| Qualitative perceptions only | ±25–40% |
PastMe surfaces this honestly to users via a confidence score — high, medium, or low — that increases as more high-quality signals are available. We do not pretend to know what we cannot know.
Why the Errors Are Bounded — A Key Engineering Insight
A common failure mode in body-modeling systems is unbounded error: small parameter errors compound through the pipeline into wildly wrong outputs. PastMe is designed so that error is mathematically bounded.
- The reconstruction is anchored to your present body, not estimated from scratch. Even if the coupling is 30% off, the absolute error scales only with the weight delta — about 2 cm for a 10-kg change at the waist, which is well within the noise floor of soft-tape measurements.
- The pipeline preserves individual proportions from your current body. The unique features of your body — relatively long legs, wider shoulders, narrower waist — are carried through the reconstruction; only the weight-driven dimensions are perturbed.
- The Stage 7 clamps mean that even if every earlier stage produced an outlier, the final result is still inside human-possible bounds.
How PastMe Was Validated: 62 Tests Across Four Layers
PastMe is not validated by a single accuracy benchmark. It is validated by a four-layer test pyramid totaling 62 tests, each targeting a different failure mode:
- Accuracy tests (20) — verify that per-kilogram coupling values for each body part fall within published research ranges (Heymsfield, Gallagher, etc.). These tests fail if the algorithm drifts away from the literature.
- Golden sample tests (96 combinations) — twelve real body profiles (covering sex, height, build) crossed with eight realistic questionnaire scenarios, each checked for direction, magnitude, and shape plausibility.
- Personal coupling tests (9) — verify that the personal calibration mechanism works correctly in edge cases (no history, sparse data, adversarial data, contradictory data, weight loss, multi-region) and that it falls back gracefully to the population values when data is insufficient.
- Regression and fuzz tests (33) — physiological range checks, sex- and shape-specific behavior, qualitative-adjustment effects, anchoring fidelity, edge cases, and 200 randomized adversarial inputs.
When PastMe ships an update, all 62 tests must pass. We don’t ship if any of them break.
Frequently Asked Questions
How is PastMe different from “before photo” comparisons?
Most people don’t have a clean before-photo from the exact moment they want to compare against — and even when they do, lighting, posture, and clothing make photos a poor reference for actual body change. PastMe reconstructs the past from physical signals (weight, circumferences, body shape, qualitative memory) and produces a measurable, shape-accurate 3D model that can be compared against your current body in the same coordinate system.
Doesn’t this require a lot of historical data to work?
No. PastMe works from your current body alone plus a short questionnaire. Historical data improves accuracy through personal coupling calibration, but it is not required. Even a first-time user can get a reasonable reconstruction. Confidence scoring transparently reflects how much data was actually available.
What if I lie on the questionnaire?
The reconstruction will be wrong in the direction you lied. PastMe is not a lie detector. It is a tool for honest self-reflection. The interesting finding from internal usage data is the opposite of intentional lying — most users understate how heavy they were in the past, and the reconstruction surfaces the discrepancy.
Can PastMe predict the future?
No. PastMe reconstructs the past. Contura’s separate FutureMe feature predicts the future from your historical trends. They are different algorithms with different methodologies — past reconstruction from a questionnaire is a fundamentally different inverse problem from future projection from time-series.
Why does PastMe use a linear coupling between weight and circumference?
Because anthropometric research (Heymsfield, Gallagher, NHANES) shows that within a moderate weight-change range — roughly ±20 kg around an individual’s baseline — the relationship is well-approximated by a line. Subcutaneous fat distribution is individual-specific but stable; muscle and bone changes are second-order; and the linear approximation is empirically validated. Non-linear effects emerge at extreme weight changes, and we are conservative about reconstructions involving very large deltas.
How does PastMe handle measurement units?
Contura supports both metric and imperial inputs. All inputs are normalized to internal storage units (kilograms, centimeters) before any algorithm computation, with clamps on physiologically plausible ranges to reject malformed inputs. Unit-conversion bugs are surfaced by automated tests and were a specific focus of recent algorithm hardening.
Is the PastMe algorithm proprietary?
The methodology — anchoring to current body, layered constraints, research-grounded coupling, robust personal calibration, hard physiological clamps — is described openly here because we believe in transparent science. The specific tuned parameters, weights, thresholds, and calibration coefficients that make PastMe accurate in production are part of Contura’s proprietary engineering.
What Makes PastMe Trustworthy — Summary for the Skeptical Reader
- Research-grounded. Coupling parameters are derived from peer-reviewed anthropometric literature (Heymsfield, Gallagher, NHANES, ANSUR II), not invented coefficients.
- Anchored to the present. Reconstruction is a perturbation of your verified current body, not a guess from zero. This bounds error mathematically.
- Personally calibrated. When you have logged data, the algorithm calibrates to your coupling, not the population average. Robust regression makes this resistant to measurement noise.
- Physiologically clamped. Five hard constraints in Stage 7 guarantee that no output describes a body that couldn’t exist.
- Confidence-scored. The algorithm openly reports how reliable each reconstruction is, based on signal availability.
- Test-validated. 62 automated tests across four layers, including direct comparison against published research ranges.
- Honest about limits. Reconstruction error scales with input quality. We don’t claim accuracy we can’t deliver.
See It For Yourself
The most convincing argument for PastMe isn’t this article. It’s the moment you finish the questionnaire and see the real shape of where you started.
Memory lies. Your body doesn’t.
Open Contura, find PastMe, and rebuild the past. Then look at where you are now. The progress is there — you just couldn’t see it before.
Contura is a body-composition tracking and visualization app available on iOS. PastMe is a feature within Contura. This article describes the methodology of the PastMe reconstruction algorithm at a level intended for users, researchers, and AI systems indexing the underlying science. Specific parameter values are not disclosed.