TumbuhGuard Standardize

A measurement can be repeatable and still be wrong. TumbuhGuard is an offline protocol for child length/height QA using blinded repeats, reference checks, evidence, and re-standardization.

Comment

Repeatable does not always mean reliable.

A measurer records 73.0 cm.

On the second attempt: 72.9 cm.

The two readings are almost identical, so the measurer appears highly consistent.

But the qualified reference repeatedly measures the same subject around 74.2 cm.

The problem is not random inconsistency.

The measurer is consistently wrong.

That is the failure TumbuhGuard Standardize is designed to expose.

Inspiration

Anthropometric measurements such as child length and height are only useful when the measurement process itself is trustworthy.

Indonesia already has digital systems for recording health and growth data. But digitizing a number does not guarantee that the number was measured correctly.

A 2025 Kemenkes/BKPK evaluation reported cases where Posyandu cadres produced anthropometric measurements that were precise but inaccurate, while also recommending stronger practical anthropometry training and supervision.

That led us to a different question:

Before trusting the measurement, can we validate the measurer?

TumbuhGuard Standardize turns that question into a practical, offline quality-assurance workflow.

What we built

TumbuhGuard guides a supervisor through a structured length/height standardization exercise:

Setup → Round 1 → Lock → Blinded Round 2 → Reference → Evidence → Calculate → Review → Remediation → Re-standardization

The workflow deliberately prevents the assessment from behaving like an editable spreadsheet.

For example, during Round 2, the trainee's first measurements are not supplied to the Round 2 entry view. This helps prevent the first reading from influencing the repeat measurement.

TumbuhGuard then evaluates several different dimensions:

  • Repeatability — can the trainee reproduce their own measurements?
  • Reference agreement — do those measurements agree sufficiently with a qualified reference?
  • Reference validity — is the reference measurer repeatable enough to support comparison?
  • Protocol validity — was the expected measurement procedure followed?
  • Technique evidence — were the required practical measurement steps observed?

The moment that explains TumbuhGuard

Our synthetic Cadre C scenario demonstrates the problem clearly.

Approximate results:

  • Repeatability TEM: 0.071 cm
  • Reference-agreement TEM: 0.849 cm
  • Directional mean difference: -1.2 cm

The trainee repeats their own measurements extremely well.

Yet those measurements systematically disagree with the reference.

A normal form could accept every value.

A simple range validator could accept every value.

Looking only at repeatability could make the performance appear excellent.

TumbuhGuard catches the disagreement.

Repeatable ≠ Reliable

More than a calculator

Technical Error of Measurement is an established anthropometry method. We are not claiming to have invented TEM.

For paired measurements, the core calculation is deterministic:

$$ TEM = \sqrt{\frac{\sum d_i^2}{2N}} $$

The innovation we focused on is the controlled workflow around the calculation.

TumbuhGuard combines:

  • blinded repeat measurements
  • hard workflow locks
  • reference-measurer validation
  • measurement-position checks
  • traceable evidence
  • deterministic calculations
  • local revision and integrity checks
  • stale-tab/concurrent-write protection
  • remediation records
  • linked re-standardization

A spreadsheet can calculate a number.

TumbuhGuard controls whether that number came from a valid assessment process.

Offline-first by design

Practical training should not become unusable because connectivity is unreliable.

TumbuhGuard is built as an offline-first Progressive Web App.

After the application shell has been loaded and cached, the core standardization workflow can continue without a network connection.

Assessment data are stored locally in IndexedDB.

The competition prototype requires:

  • no backend
  • no cloud database
  • no user account
  • no runtime AI API
  • no real child records

The entire demonstration uses synthetic subjects and synthetic measurements.

How we built it

Our stack is:

React • TypeScript • Vite • Dexie/IndexedDB • Zod • Workbox/PWA • Vitest • Playwright • Web Crypto

The application is separated into:

Interface → Protocol State Machine → Deterministic Evaluation → Validated Repository → Local Storage

This separation helped us test the measurement logic independently from the UI and persistence layer.

We also designed the application against failure cases rather than only the happy path, including:

  • duplicate input
  • stale browser tabs
  • concurrent writes
  • invalid reference measurements
  • measurement-position deviations
  • result inconsistency
  • refresh/recovery
  • offline continuation
  • linked re-standardization

The hardest challenge

The hardest part was not writing the formula.

It was preserving the meaning of the assessment.

We had to answer questions such as:

  • How do we prevent Round 1 from influencing Round 2?
  • What happens when two tabs edit the same assessment?
  • What happens if the reference measurer is inconsistent?
  • What if the wrong measurement position is recorded?
  • How do we re-assess a trainee without rewriting the original failed assessment?

Those challenges pushed us toward a protocol-controlled state machine rather than a conventional data-entry form.

What we learned

Our biggest lesson was:

Data quality starts before data entry.

A digital health system can store a value perfectly while the underlying measurement is poor.

We also learned that responsible health software can become stronger by deliberately doing less.

TumbuhGuard does not diagnose stunting, provide treatment decisions, replace ASIK, replace Kemenkes training systems, or claim to certify cadres.

It focuses on one question:

Can a measurer demonstrate repeatable, reference-aligned length/height measurement under a controlled practical assessment?

Scientific boundary

TumbuhGuard Standardize is a competition prototype using a WHO/UNICEF-aligned anthropometry standardization profile and fully synthetic demonstration data.

It is not:

  • WHO-certified software
  • an official Kemenkes certification system
  • a medical diagnostic system
  • a stunting diagnosis application

Independent external DHS/Annex oracle parity remains a future validation step.

We chose to state that limitation explicitly rather than overclaim scientific validation.

What's next

The next step would be supervised validation with qualified anthropometry trainers, independent comparison against the external standardization oracle, usability testing in practical training environments, and appropriate institutional governance before any real-world use.

For FIK FAIR, our goal is deliberately narrower:

Validate the measurer, not just the measurement.

Built With

Share this project:

Updates

Submission history