CORNERSTONE

Roughly 95% of people in American prisons come home. The question was never whether they return to our communities. It is what version of them arrives.

CORNERSTONE turns the time between sentencing and release into the thing it should always have been: a period of structured training that produces a person with a credential, a job offer, treated clinical need, and a verifiable record — instead of a person with a release date and nothing else.

Everything downstream assumes a building where a person can breathe. That assumption does not currently hold. A rehabilitation framework that opens with college programs while people are getting sick in the same building will be read as unserious, and it will deserve to be.

Two structural failures sit underneath all of it: in a number of states, correctional facilities are wholly or partly exempt from the routine public health inspection that governs restaurants, schools, and nursing homes — and roughly half of states have no corrections ombudsman or independent oversight body with real access.

Download the framework document (Markdown, v1.0 — July 2026)

Each one runs entirely in the browser as a single self-contained file. No login, no server, no install. A family member on a phone in a parking lot outside a facility should not need an account, and a public defender's office should not need IT approval to open a web page.

FOOTING App 5 · The floor, made operable

Conditions reporting for families

A family member enters what is happening — mold, water, medication not arriving, assault, retaliation — and gets back every office with authority over it, how fast each one moves, and which ones do not require naming the person inside. The safety brief comes before the form. 51 jurisdictions, agency names only, no unverified contact ever rendered.

KEYSTONE App 1

Campus blueprint builder

Enter a facility's population, length of stay, and profile; get tier allocation, staffing at defined ratios, a cost and return model on RAND's figures, a funding stack matched tier by tier, and a 36-month build sequence. Includes a hard viability gate that refuses to produce a blueprint where length of stay cannot support one.

BEARING App 2

Three-axis classification engine

Enter a case profile; the three axes resolve independently and visibly. Surfaces the counterintuitive evidence when a profile would have triggered an offense-label exclusion, and carries the constitutional floor on screen. This is the app that makes the separation framework concrete instead of theoretical.

PARCHMENT App 3 · Build first

Portable credential ledger

Hash-sealed, transfer-proof credential record — each entry's seal includes the one before it, so an altered or removed entry breaks every seal that follows. The employer verification view has offense, sentence, facility, risk score, and clinical record structurally absent from the payload rather than permission-gated. There is no field to leak.

CHALKLINE App 4 · Research pass 1

50-state readiness report card

Fifty states and DC graded on six indicators: PEP approvals, Workforce Pell adoption, registered apprenticeship in corrections, occupational licensing barriers, Medicaid reentry waiver status, and earned-time structure. Every cell says whether it is verified, likely, or unverified; no jurisdiction is graded until at least four of six indicators are verified. It grades 1 of 51 and says so in large type instead of filling the grid with plausible guesses.

Everyone quotes four to five dollars back per dollar spent. That figure is real, it is RAND's, and it is narrow — it counts reincarceration cost avoided over three years, to taxpayers, and nothing else. No earnings. No taxes paid. No crime that never happened to anybody.

Count the whole life and the number moves a long way.

Sources: RAND IG-113 / RR-266 / RR-564, correctional education meta-analysis. Washington State Institute for Public Policy benefit-cost results, adult criminal justice — figures in 2023 dollars, benefit-cost methods updated December 2024, literature review May 2024; "confident" is WSIPP's own modelled probability that benefits exceed costs. Both are VERIFIED against the publishing body.

Every one of these is a narrated walkthrough of the real app — recorded off the live page, not a mockup — with captions. If a video shows a screen the app no longer has, the video is wrong and we want to know.

Every factual cell across the suite says one of three things about itself: verified (checked against a primary source), likely (follows a known pattern or a secondary source — confirm before relying on it), or unverified (recorded but not yet checked). Only verified cells count toward a grade. Gaps are displayed rather than filled in. The coverage bar is a feature: it is what makes the grid trustworthy the first time someone checks a cell.