The Glass Box Principle

The opposite of a black box is not an open-source repository. It is a set of artifacts that resolve when the reader clicks.

Black box has become the shorthand for any model whose innards are hidden. The natural opposite would be white box, or open source, or fully explainable AI, or one of a dozen other phrases the industry has tried on. None of them describe what a financial product actually needs. A user asking where does this number come from is not asking for source code. They are asking for four things: the formula, the version, the constants, and the sources. Those are not code. They are artifacts. The question is whether the artifacts open when the reader clicks them.

Glass box is our name for the discipline of making every claim on a product surface openable. It is a specific working constraint on what ships, not a marketing posture. Applied consistently, it produces a product in which every number the user sees is one hop away from the evidence that produced it. Applied inconsistently, it produces a product with a lot of badges.

What the discipline requires

If the product says based on current IRS limits, the word current is a link that resolves to a dated row in a registry. If the product says assuming a four-percent withdrawal rate, the number four is a link that resolves to the spec's assumptions section. If the product says verified, the word verified is a link that resolves to the changelog entry for the current spec version. If any of those links resolves to a 404, the claim comes down until the link is fixed. There is no third option, no working-on-it version, no we-cited-it-in-the-annual-report substitute.

• ✓Every named constant → link to registry row with source URL and revision date.
• ✓Every methodology claim → link to spec section.
• ✓Every version reference → link to changelog entry.
• ✓Every verified badge → link to verification cases and pass/fail history.
• ✓Every embedded number a user sees → recomputable from the stored envelope.

Why the discipline is different from transparency

Transparency is a value. Glass box is a working constraint. A transparent product might document its methodology in a whitepaper that the marketing team wrote three years ago. A glass-box product cannot make a claim on the marketing site that does not resolve to a live artifact. The whitepaper is a document; the glass-box artifact is a hyperlink to a document that is currently true.

The distinction matters in practice because transparency is easy to claim and hard to verify. Glass box is easy to verify — click the link — and hard to claim without doing the work. Every glass-box claim on the Worthune site is a link. If a claim on the site is not a link, either it does not warrant one (a stylistic sentence, a rhetorical framing) or it has not been substantiated yet, in which case it comes down until it is.

The audit surface glass box creates

A glass-box product exposes a different audit surface than a transparent-but-closed product. When a customer's compliance team asks how the mortgage-tax-benefit figure is computed, the answer is a spec URL. When a regulator asks whether the product uses current tax figures, the answer is a facts registry URL with revision dates. When a journalist asks where a claimed verification rate comes from, the answer is a changelog URL with the release notes for the referenced spec version.

None of these audit conversations require an engineer's time. The artifacts are self-service. The compliance team, the regulator, and the journalist can each answer their own question in the same way a curious user can — by clicking the link. The engineering cost of the glass-box discipline is up front, in the work of publishing the artifacts. The engineering cost of the alternative — answering the same audit question over and over with a hand-written explanation — recurs forever.

“A black box is opaque. A glass box is not transparent — it is auditable. The difference matters when a customer, a journalist, or a regulator asks the same question.”

How the discipline shows up on the Worthune site

The homepage claim that one hundred percent of models publish a spec resolves to the catalog. The claim of two hundred fifty verification cases per model resolves to the downloadable eval datasets, one per model. The claim of TY2026 IRS and SSA figures verified at the primary source resolves to the facts registry, with the primary-source URL and the revision date visible on each row. The claim of zero silent changes resolves to the full changelog. If any of those resolved to a 404, the claim would come down.

The same discipline governs this content library. Every article that names a constant links to the registry row for the constant. Every article that names a regulation links to the primary regulatory source. Every article that references another article links to that article. If an artifact does not exist yet, the article that would reference it does not ship. This is not a stylistic choice. It is the working constraint the manifesto committed the product to, applied to the surface most content libraries treat as separate from the product.

Where the discipline breaks

Glass box is easier to describe than to sustain. Two patterns break it most often. The first is the incrementally-added marketing claim — a phrase added to a landing page by someone who did not check whether the referenced artifact exists. The remedy is a lint step that fails the build when a landing-page phrase matches a known-glass-box pattern (verified, sourced, current, up to date) without an accompanying link. The second pattern is the drifted link — an artifact that moved without a redirect. The remedy is a scheduled crawl that reports 404s on the site's own claims, weekly.

Both remedies are small pieces of engineering that pay for themselves within a quarter. The alternative is a product that describes itself as verified in twenty places and can prove it in three. That is the product the glass-box discipline exists to prevent.

What the discipline changes for the reader

A reader landing on a glass-box product for the first time does not consciously notice the discipline. What they notice is that questions get answered when they click. Curiosity becomes evidence rather than frustration. A reader who wonders where a number came from finds a spec; a reader who wonders whether a limit is current finds a dated registry row; a reader who wonders whether the vendor has changed the math finds a changelog entry. The cumulative effect is a product that rewards attention, which is a different and rarer property than a product that survives inattention.

The commercial argument for glass box tracks the technical one. Products that reward attention accumulate the specific kind of user who reads carefully and shares generously — developer-adjacent, professional, procurement-influencing. Products that only survive inattention accumulate users who churn quietly the first time a number is questioned. The two user bases compound in opposite directions.