A differential-testing harness for software that is being rewritten from scratch: record real usage of the original application into a portable behavioural spec, replay it against a clean-room rebuild, and report divergence as a browsable diff (what the original does, what the rebuild does, which is authoritative). Sold to the clean-room movement as the missing proof layer — the reason a rebuild is trustworthy rather than merely impressive.

Market Panorama

Competitive Intelligence

Users & Demand

Business Model

Market Sizing

Tech Solution & Build Plan

Go-to-Market Strategy

Risk Assessment

Action Recommendations

### Market definition and scope

Behavioural conformance tooling sits inside the space between an original application and a from-scratch rebuild. Its job is to produce evidence: record a real session against the original, express it as a portable and human-reviewable specification, replay that specification against the rebuild, and classify every divergence as a bug, an intentional difference or an unsupported case.

The boundary distinguishes it from three adjacent categories. It is not a unit-testing framework — unit tests are written by the same person who wrote the code and cannot answer whether the rebuild behaves like the original. It is not UI automation for testing the original — Appium and WinAppDriver can drive the original application but produce no equivalence verdict. And it is not source-level translation, which is a different approach to the same problem.

Its unit of value is a document a maintainer can point at and a customer can read: here is what we tested, here is what the original does, here is what the rebuild does, and here is what we believe is intentional.

October made the market visible. storytold/wordcraft, a clean-room Microsoft Word reimplementation in pure Rust, took 17,115 stars in its first month (GitHub Trending, October 2026). tangency-labs/solidcraft rebuilt parametric CAD with MCP support at 13,259 stars. HeyPuter/everything announced a programme of rebuilding software from scratch at 6,642. And horizon-rust/horizon, rebuilding Minecraft, advertises that it is checked against vanilla with 281 differential harnesses at 6,089 stars.

### Key trend signals with data citations

The horizon claim is the most useful single data point in this report. "Checked against vanilla with 281 differential harnesses" describes a working method, implemented by hand, at a scale that a hobby project achieved through enormous manual effort. That is the definition of an unproductised method: the technique is validated, the tooling does not exist, and the labour is being absorbed by volunteers.

The other rebuilds are the market size. A 17,115-star Word reimplementation and a 13,259-star CAD reimplementation in one month indicate that rebuilds attract enormous attention. Attention is not the same as adoption, but it is the precondition for adoption, and it means a rebuild with a verifiable equivalence story can differentiate against one without.

The portfolio pattern matters too. HeyPuter/everything's stated ambition — rebuilding and open-sourcing the world's software from scratch — turns individual rebuilds into a programme, and the same organisation behind wordcraft shipped sibling projects (EffectCraft, GridCraft, DesignCraft, DeckCraft) in the same window. Organisations running multiple rebuilds have a repeatable need rather than a one-off one, which is a better buyer than an individual.

### Why now

Rebuild culture changed in a way that removes the stigma of the approach. Clean-room reimplementation, historically a legally fraught exercise, is now a recognised portfolio strategy with thousands of stars attached. That legitimisation expands the population of projects that face the equivalence question and therefore the population that needs a proof layer.

Simultaneously, the method itself became visible. Most rebuilds quietly claim to be faithful; horizon published the mechanism. Once one project describes 281 differential harnesses, the absence of tooling is on the record rather than a private frustration.

The commercial driver is older and steadier: companies replacing licensed software for cost or licence reasons have always needed to demonstrate equivalence to somebody — an internal stakeholder, a customer, or an auditor. That need exists today, is answered with spreadsheets, and is invisible in GitHub trending data.

### Market maturity

Early, with an unusual shape: demand is visible in open source and money is likely in enterprise, and the two do not overlap much. That mismatch is the central strategic fact of this opportunity and the reason the standard playbook — free open source, paid seats — fits poorly.

A mature version of this market would have a specification exchange format, third-party verification services and insurers who price off a conformance report. None of that exists, which is both the opportunity and the reason a build is expensive.

### What this means for a solo founder

Lead with the commercial rewrite buyer, use open-source rebuilds as credibility rather than revenue, and treat the reference specification as the product. A founder who builds for star counts will build something impressive that nobody pays for.

storytold/wordcraft, a clean-room Microsoft Word reimplementation in pure Rust, reached 17,115 stars in its first month.

A 17,000-star rebuild in month one

horizon-rust/horizon advertises being checked against vanilla Minecraft with 281 differential harnesses — the method proven, the labour unproductised (6,089 stars).

Differential testing done by hand at scale

HeyPuter/everything announces rebuilding and open-sourcing software from scratch (6,642 stars), turning rebuilds into a portfolio strategy.

A whole program of rebuilds

tangency-labs/solidcraft rebuilds parametric CAD in Rust with MCP support (13,259 stars), where equivalence has engineering consequences.

Behavioural Conformance Suite for Clean-Room Rebuilds