What to expect

Live today

• The full 14-dimension profile on any file, read against the profile you declare or the dimensions you rank, with auto-inference as the fallback
• A polished Pro-style report: profile fit, capability and violation poles, recommendations, sharing, and PDF export
• Concrete fixes on report recommendations for Pro subscribers, with a worked example on the no-sign-in sample report
• A free essential security review, Security (D6) only, for every signed-in account, plus a quick deterministic scan that needs no account
• Evidence-cited measurement: the judge cites lines internally for every call; harmful-pattern findings and top recommendations surface those citations in the report
• All fourteen dimensions currently clear Facet's four-bar reliability gate (agreement, minority-recall, test-retest) before they ship
• Severity-weighted, fail-closed security scoring, externally grounded and adversarially reviewed
• Your source code is never stored - only secret-scrubbed measurements, scoped to your account

What we are finishing for the Pro MVP

Facet is a measurement instrument, and the paid product has to make the measurement useful at the moment a developer reads it. The Pro MVP is therefore report-first: the HTML report, shared report, and saved PDF should all be credible user-facing artefacts; recommendations should explain the profile-relative gap; and concrete fixes should upgrade those recommendations without exposing the proprietary instrument.

Free users see the same report structure and the same recommendations, because that is the proof of the product. Pro adds the workflow on top: concrete fixes generated against your own code, more capacity, private report handling, and the next private-repo/editor paths.

On the roadmap

The single-file profile is the instrument. The roadmap turns it into something that lives inside how you already work, so the question is answered continuously, not once:

• Private repositories and an editor integration for Pro
• Pull-request profile comments and profile-drift CI gates
• The external-weighting treatment extended beyond security, robustness and concurrency first
• A validation-status audit across all fourteen dimensions: confirming which are fully validated end-to-end, which still need more work, and where per-indicator severity weighting (the D6 security treatment) should extend next - most dimensions plausibly have at least one indicator that matters more than the rest, by some objective or subjective standard, and D6 is unlikely to be the only one that needs it
• Team pooled credits, author and section segmentation, codebase coherence, and an expected-profile conformance manifest
• A metered API for tool builders and CI
• Enterprise private deployments, SSO, custom profiles, conformance dashboards, and audit exports

Where this is heading

Nothing below is dated or promised. It is the direction, and it is deliberately ambitious: optimisation targets as a first-class concept for the whole practice of coding.

• Codebases, not just files. Per-section targets, coherence across a repo, drift per commit. Then mixed-target files: parts of one file often should be optimised for different things, and that should be measurable and reviewable, feature by feature, against targets a team declares per path. Tests made functional correctness enforceable in CI; reliable measurement does the same for every quality tests cannot see.
• Engineers, teams, organisations. The same problem can be solved several ways, and engineers genuinely differ in which ways they are best at. Opt-in measurement of your own code shows what you already optimise well, what you could do differently, and how to become versatile; teams see whether they hold the right mix for what they build; organisations learn which dimensions their product actually needs. Developer-first and opt-in, always: a self-understanding instrument, never a surveillance one. For seasoned engineers, this is expertise finally demonstrated by measurement rather than asserted.
• AI-assisted coding. Published research on how reliably LLM judges measure code, and on what AI models optimise for when nobody tells them. Then the loop closes: an interview helps you work out what your build should optimise for, and the resulting profile travels with the project as a declared target that coding agents honour and Facet verifies. When generation is cheap and measurement is reliable, the declared profile becomes the source of truth and code becomes build output.
• More languages and better judges, permanently. Each new language qualifies and is badge-gated before its reads are called reliable, and every judge re-sits its exam as the model landscape moves, so the instrument keeps improving underneath you.
• Educators are not a market we are entering yet, but the instrument is built so it can serve them when the time comes.

If you share the mission, making code quality measurable for novices reliant on AI and for seasoned developers alike, we want to hear from you.

How the Pro preview works

The preview is designed to let people inspect the actual product shape before paying:

• The sample report shows the same report structure Pro users get.
• Recommendations are visible on every report, free or paid. Concrete fixes are Pro; a full worked example sits on the sample report so you can judge one before paying.
• The essential security review stays free with an account, and the quick scan needs no account, so the security read is never the thing behind the paywall.
• Team and Enterprise remain greyed out until pooled/team and private-deployment workflows exist.

Founding members

People who help shape Facet during the beta become founding members. That means our best price, locked in for as long as you stay; a direct line into what we build next; and early access to the paid features above before general release. We will set the exact founding price when the paid tiers go live and the measurement work is done - and founding members get it first and lowest. Join the list below to be one.