About This Book
For most of its life, software that scores job candidates worked like a vending machine with the buttons taped over. A résumé went in, a number came out, and nobody — not the recruiter, not the candidate, not even the regulator who's eventually going to come knocking — could see how the machine decided.
This book is about the thing we built to fix that, which we call the Rubric Foundation. You don't need to write code, you don't need a statistics degree, and you definitely don't need to read every page. Skip around. Read the cartoons. We won't be offended.
One promise up front: where something isn't finished, this book says so, right on the page. No "coming soon" fairy dust. If a number is a best guess, you'll see the words best guess.
Icons Used in This Book
Like every book in this series, the margins are littered with little pictures. Here's what they mean, so you can read the ones you want and ignore the rest:
The whole thing, in one breath
Here's the deal. The old scoring software kept its math inside the program, the way a chef keeps a secret recipe inside their head. It worked. People bought it. But the moment anyone asked "why did this candidate get a 62?", the honest answer was a shrug.
The Rubric Foundation throws out the secret recipe and writes it on a card. Every number now points to a real, published authority. Every number records why it was picked. And — this is the part that makes everyone relax — a trained person can change it by editing a document, without a programmer rewriting anything.
The old way wasn't wrong. It just couldn't do the one thing the next few years of hiring are going to demand: hand someone a score they can read, challenge, and check.
What it actually is
Strip away the jargon and a rubric is just a written document — one for each kind of job, such as "Senior Project Manager." It lists the scoring buckets, says how much each bucket is worth, names the method that reads each bucket, and keeps a shelf of real-world authorities that every number points back to.
Every single number in that document — every weight, every cutoff, every points value — is wrapped in a little tag carrying its source, how confident we are, who set it, when, and the reason. At scoring time the engine reads the rubric, hands each bucket to its named method, takes a photo of the whole thing next to the evidence and the citations, and files it. Three ideas are hiding in there. Here they are inside one real number.
value: 25.0 # the number itself calibration_status: CALIBRATING_SP3_LIVE calibration_authority: intelletto_curator_team calibration_effective: 2026-06-10 framework_anchors: [ SH-1998, ONET-13-1082.00 ] numeric_rationale: > Hard-skills weight set at 25% from the Schmidt & Hunter 1998 meta-analysis: structured assessment of job knowledge + skill explains ~26% of supervisor-rated performance for professional roles. SP3's calibration loop is now live and this number is being checked against our own outcome data; its badge flips to "calibrated" once enough decisions clear the bar.
What every number points at
The list of authorities isn't decoration — it's a bouncer. A curator cannot publish a number whose source doesn't resolve to a recognized framework. If a citation is left dangling, the document refuses to publish, full stop. Here's the shelf a rubric can pull from today, so the next time someone asks "where did this score come from?" the answer is one piece of paper instead of a week of detective work:
- O*NET-SOC — the occupation map that defines role families, seniority brackets, and the go-to hard-skill list for each job.
- SIOP 2023 — work-psychology principles, used to set soft-skill weights and interview anchors.
- CEFR — the A1-through-C2 language scale, locked at the framework level so language scoring isn't a matter of opinion.
- EEOC four-fifths rule — the 80% line that scoring has to stay above for every protected group, or you have a problem.
- Schmidt & Hunter (1998) — the meta-analysis used as the starting point for how much each bucket should be worth.
- APA Standards · PMI BoK · UK ONS · ENISA · ISO · HHS Safe Harbor · Lightcast — validity rules, PM competencies, UK role bindings, security skills, certification tiers, de-identification methods, and skill-taxonomy cross-walks.
Count them up and you've got the twelve external frameworks every number leans on. None of them is ours. That's the point — we don't get to grade our own homework.
The part that surprises people: the honesty label
The single feature most likely to make a reader sit up is that every number tells you how sure we are of it. There are exactly three levels of sureness, and each one makes a different promise. The old method had nothing like this — a hardcoded weight came with no little flag saying whether it was nailed down by science or just an engineer's Tuesday-afternoon guess.
"A score you can't question is a score you have to take on faith. The whole point of the foundation is to get rid of the faith."
Three pillars, and why we built all three at once
The Rubric Foundation isn't a bet on one clever idea. It's a bet that three different pressures all show up in the same eighteen-to-thirty-six-month window — and that paying once for a citation-backed scoring base answers all three. Each pillar on its own is arguable. Stacked together, they're the right shape for a company that plans to keep selling in regulated markets.
Seven sub-projects, one falling row of dominoes
The foundation was the first of seven sub-projects, named SP1 through SP7. They're not seven separate features you can pick off a menu — they're a chain, where each link unlocked the next as it landed. That chain is now built end to end. Understand the chain and you understand the whole thing, including what each link does and which still leans on data, not code, to finish maturing.
Each link leans on the one before it. Read top to bottom, the chain shows what each stage does now that all seven are built — and the one thing that still finishes on data, not on a deploy:
- SP1 · Rubric Foundation — the citation-backed base: rubrics, the eight typed methods, the resolver, the audit-packet snapshot. Live in production.
- SP2 · Outcome Capture — built. Writes down every recruiter decision next to the exact rubric version that produced the score, so outcomes can be measured against the math that drove them.
- SP3 · Empirical Calibration — built. Turns "pending" best guesses into actually-calibrated numbers, checked against real outcomes with confidence intervals and a fairness pass. The loop is now live; it flips the third honesty badge on per number as enough decisions accumulate behind that number. This is the one stage that still finishes on data volume, not on a build.
- SP4 · Bias monitor — built. Watches outcomes against the EEOC four-fifths rule nonstop, with framing for the US, EU, UK, Singapore, and the Philippines.
- SP5 · Candidate Portal — built. Lets a candidate see their own scorecard, trace the evidence, and challenge a specific line.
- SP6 · Candidate AI Assistant — built. A chat layer over the portal that explains a result in plain language.
- SP7 · Recruiter Explanation — built. Shows the recruiter the rubric breakdown, evidence, citations, and bias context, retiring the old staff-only debug page.
No "coming soon." Here's the real status board.
Every Dummies book has the table you actually flip to. This is ours. Green means it's built and running. Yellow means it's wired but still maturing on data. White means designed but not yet built — and after this update, nothing sits there anymore.
| Piece | Status | What it actually is |
|---|---|---|
| Rubric & citation schema | Live | In the production database; every scorecard can point at a rubric and its citation chain. |
| 8 typed scoring methods | Live | Skills, education, language, domain terms, scope, tenure, culture-fit, compliance. |
| Extraction normalizer | Live | Turns real résumé data into the shape the methods expect. |
| Two seeded rubrics | Live | A Senior Project Manager rubric (O*NET-SOC 13-1082.00) and a general fallback exist as real production rows. |
| Job role-family tagging | Live | Backfilled tenant-wide on 13 June; the pilot tenant now resolves most job descriptions to their specific rubric, with the general fallback as the safety net. |
| Calibrated numbers | Calibrating | SP2 capture and SP3 calibration are now built and running; numbers move from pending to calibrated as outcomes clear the bar. None are stamped calibrated until that bar is met. |
| Curator team | Headcount 0 | The CTO maintains the rubrics today, so the maintainability payoff hasn't started yet. (See the cartoon above. That's not entirely a joke.) |
| SP2–SP7 builds | Built | All six are built — Outcome Capture, Empirical Calibration, Bias monitor, Candidate Portal, Candidate AI Assistant, and Recruiter Explanation. What's left now finishes on data, not on a deploy: SP3 needs outcome volume to calibrate. |
What actually changes on your desk
The neat part of the design: recruiters, candidates, and operators all look at the same explanation — just from their own side of the glass.
"One investment, three loads, one shared explanation — put together honestly, with the gaps printed on the page instead of hidden behind the score."
The Part of Tens
No book in this series is finished without it. Ten things to remember about the Rubric Foundation, in case you skipped everything else and flipped straight to the back (we see you):
- The recipe is on a card now. The scoring math is a readable document, not a secret in the code.
- Every number cites a real source. No anchor, no publish — twelve external frameworks back the math.
- Eight typed methods do the scoring, and on production they're failing 0 times out of 416.
- Every number wears an honesty label. Framework-fixed, pending, or calibrated — and it says which.
- Calibration is now running. SP3 is live; numbers move from pending to calibrated as real outcomes clear the bar, and none claim "proven" before their badge says so.
- Three pillars, one base. Regulators, candidate trust, and maintainability — paid for once.
- A skeptic has to be right three times to make the bet wrong; a believer needs one payout to break even.
- SP1 through SP7 is a chain — and it's now built end to end. Each link landed in order; what's left finishes on data, not on a deploy.
- The dev tenant is live on the rubric path, running about ten points below legacy (median −11.32) while it calibrates.
- The curator team's headcount is zero. Until that changes, the CTO is the curator, and the maintainability payoff hasn't started.
White paper · Rubric Foundation · derived from RUBRIC-FOUNDATION-RATIONALE. Presented in the plain-English reference style as an affectionate homage to the classic how-to book — original cartoons drawn for this document. Not affiliated with, endorsed by, or produced by the …For Dummies® series or its publisher; "For Dummies" is a registered trademark of its respective owner.