ReLoop.me Methodology. ReLoop.me is an EU-built Capability Intelligence platform that interprets job descriptions and professional histories as structured evidence rather than keyword lists, summaries, or generic AI output. This page explains how a read is produced and why the method holds: every read runs as independent layered passes that can disagree with the document and with each other; inflation and claim strength are judged by a separate pass, never by the model that wrote the summary; every read is classified on a fixed interpretation vocabulary so any role or profile is comparable to any other where that assignment is stable; every read outputs structured capability data, not just prose; and every read runs through an automated quality-control pass before delivery, while human calibration continuously sharpens the system itself. Role Intel, seven euros, reads what a job description actually demands. Profile Intel, nineteen euros, reads what a professional history actually proves. Role times Profile, a direct structured comparison of both sides, is in development. The defensible asset is the system, not any single report: a fixed vocabulary, a rule set that grows by encoding every corrected failure, two-sided comparability, and a neutral incentive model in which both sides pay for clarity rather than one side paying for advantage. ReLoop.me publicly shows what each report contains, what each signal means, and what actually reproduces on identical input — and keeps proprietary the prompt chains, ontology-assignment rules, weighting logic, pay-derivation tables, calibration corpus, QA rules, and matching logic. It is a decision-support layer, not a decision-maker. Roles are mapped to ESCO, the European Commission's classification of skills, competences and occupations: AI models propose mappings, a person checks them, and only checked exact matches appear on a Role Intel read, since 5 October 2026.
The method behind
ReLoop.me doesn't summarise CVs or job descriptions — it interprets them as evidence, through a fixed, calibrated system. This page explains how a report is produced, what it returns, and why the method holds — without handing over the engine.
Hiring is data-rich and interpretation-poor.
Both sides of the market now optimise their documents.
CVs are polished, job descriptions are inflated, and generative AI makes both cheaper to produce and harder to trust. The rare value isn't more noise — it's a calibrated, objective way to read through the noise. Nonlinearity is where the method proves its value: a system that can interpret through career breaks, pivots, and parallel jobs fairly reads every career more accurately.
The methodology rests on one thesis: hiring is data-rich and interpretation-poor. CVs compress real capability into titles and keywords, and parental leaves, health breaks and sector pivots read as defects to systems trained on straight-line careers; nonlinear careers are the method's proving ground, because a system that reads them fairly reads every career more accurately; job descriptions stack several jobs under one title; and generative AI amplifies both, making every document sound more competent without making the evidence stronger. ReLoop.me's method is built to interpret through that distortion, consistently and on both sides.
Capability gets flattened.
Operating range, judgement, scope, and cross-functional work get reduced to titles, timelines, and keywords — and a parental leave, a health break, or a sector pivot reads as a defect to algorithms trained on straight lines.
One title hides 3 jobs.
Strategic, hands-on, senior, autonomous, under-scoped, and underpaid can all live in one posting — and nothing in the posting itself tells you which.
Everyone sounds more competent now.
Generic AI improves wording on demand. It doesn't tell you whether the evidence underneath is strong, thin, inflated, or about to be misread.
The solution isn't another tool that polishes really well.
It's a fixed interpretive layer that reads both sides the same way.
One document in. Multiple layers applied.
Every read — role or profile — runs through the same protocol. The specific prompts, weights, and sequence stay closed; the properties of the system are public, because they're what makes the output trustworthy.
How a ReLoop.me read is produced, in five steps. One, independent layered passes: the document is read several times from different analytical angles, and the layers are allowed to disagree with the document and with each other — disagreement between layers is signal, not error. Two, a separate judge: inflation and claim strength are scored by a pass that did not write the summary, because a model is the wrong auditor of its own output. Three, a fixed vocabulary: every read is classified against a fixed set of profile types, career patterns, capability levels and role archetypes, each archetype defined in writing and applied by a model — which is what makes one read comparable to another wherever that assignment is stable, and what reproduces is measured and reported, not assumed. Four, structured data underneath: every report is prose on top of structured capability data — the data is the durable output, the report is its interface. Five, calibrated quality control: every read is checked against the system's own rules by an automated quality-control pass before delivery, so nothing ships as raw model output. Human calibration runs separately, as a feedback loop on the system itself — real reads are reviewed, misjudgements are found, and each correction is encoded back into the rules the engine runs on, so every read is shaped by the corrections learned from the reads before it. Underneath the report sits a structured data layer with typed fields such as profile type, career pattern, capability levels with their evidence basis, misread risk, and a pay signal framed by EU Pay Transparency Directive 2023/970; the published fragment is simplified and the real schema stays closed.
The layers are allowed to disagree with the document and with each other. Disagreement is signal, not error: a "Head of" title contradicted by zero reports is exactly the kind of finding a single-pass summary smooths over.
{
"profile_type": "Operator × Connector",
"career_pattern": "Crossover",
"capability": { "name": "Cross-functional Delivery", "level": "Expert", "basis": "evidenced" },
"misread_risk": { "type": "non-recency", "counter_signal": "named" },
"pay_signal": { "frame": "EU 2023/970", "assumptions": "named" }
}
A fragment, simplified — field names are illustrative and values come from the public sample read. The real schema, its assignment rules, and its weightings stay closed.
One boundary applies across all of it: Profile Intel and Role Intel are strictly separate analytical objects. A profile read never judges a role; a role read never judges a person.
Only Role × Profile — the dedicated comparison layer — brings the two together, deliberately and with both datasets in hand.
Same protocol. Two analytical objects.
Role Intel reads the role side; Profile Intel reads the person side.
Each fragment below shows what the protocol does that a paraphrase can't: it names what the document implies, hides, inflates, or proves.
The method applied as two products. Role Intel, seven euros, one-time, delivered by email, reads one job description and returns the role stated plainly, true seniority and scope versus the title, an inflation read with reasons, the capability demand the role genuinely carries, transferability across backgrounds, and a directional, net-aware pay signal indexed by role family, location and seniority band and framed by EU Pay Transparency Directive 2023/970, stating whether its pay-range rules are law or best practice in the market priced. Example: a Head of Digital posting promising strategic leadership and a competitive salary reads as director-scope breadth carried in a senior individual-contributor seat with no team, no budget, and no pay range. Profile Intel, nineteen euros, one-time, delivered by email, reads one CV or career history and returns how the profile lands in the market, evidence-levelled capabilities, claims weighed against evidence, where the profile is likely misread and how to counter it, and a pay context. Example: a CV listing marketing, communications, product work, consulting and a career break reads as a cross-functional growth and translation profile with senior signal and broad transferability, likely misread because of its nonlinearity. Role times Profile, the direct structured comparison of one profile against one role, is in development; it is the only layer that compares the two sides. Since October 2026, when a job description publishes a pay range, Role Intel states where that range sits against ReLoop.me's estimate for the scope described — below, at the low end, within, or above; contract, board, fellowship and volunteer postings get no employment-salary comparison; and the role's ESCO occupation code appears when the title has a checked exact match.
One job description in. What the role actually demands out — read as a market signal, not a wishlist.
Head of Digital. Strategic leader. Hands-on. Competitive salary. Build the function and execute campaigns.
Director-scope breadth carried in a senior-IC seat — no team, no budget, no pay range, and execution-heavy daily work.
One career history in. What the evidence actually proves out — and where the market is likely to misread it.
Marketing, product work, consulting, B2B tech, civic tech, and a career break.
A cross-functional growth and translation profile with senior signal and broad transferability — likely misread due to nonlinearity
Fragments are illustrative. The calibration and the complete read sit in the paid products.
A single ReLoop.me report can easily be approximated by AI — and that's fine, because the report was never the asset.
What can't be easily approximated is what sits behind it: a repeatable interpretation protocol whose output gets more comparable, more calibrated, and more valuable with every read it produces.
Why the method holds. Any general model can imitate a single report; none can really imitate the system behind it. Layers compound: a fixed vocabulary of profile types, career patterns, capability levels and role archetypes, named publicly but assigned by closed rules and by a model, so what reproduces is measured rather than assumed; a rule set that grows every time a failure is found and encoded as a fix, recorded in the changelog; two-sided comparability, with roles and profiles classified in the same structured vocabulary; a neutral incentive model in which both sides pay for clarity and neither pays for advantage; and a third-party signature — an independent calibrated read carries credential weight that a self-generated read cannot, regardless of quality.
Open enough to trust. Closed enough to defend.
The boundary is drawn by design, not by vagueness:
everything a user needs to judge a ReLoop.me read is public.
ReLoop.me publicly shows what each report contains, what each signal means, how evidence, claims, ambiguity and risk are kept separate, what a report can and cannot decide, what actually reproduces on identical input, and the beta status, the QC pass, and the calibration process. It keeps proprietary the prompt chains and orchestration, the ontology-assignment rules, the weighting and threshold logic, the pay-derivation tables, the calibration corpus and QA rules, and the Role times Profile matching logic.
Shown publicly
- What each report contains, section by section
- What each signal and classification means
- How evidence, claims, ambiguity, and risk are kept separate
- What a report can and cannot decide
- What actually reproduces on identical input, and what doesn't yet
- Beta status, the QC and calibration process, and product boundaries
Kept proprietary
- Prompt chains and orchestration
- Ontology-assignment rules
- Weighting and threshold logic
- Pay-derivation tables
- The calibration corpus and QA rules
- Role × Profile matching logic
Transparency earns the trust. The boundary protects the engine.
This page states the properties of the system. The full technical record — every vocabulary value, every delivery-gate rule, every measured reproducibility figure, dated — is hand-maintained at briefs.reloop.me/methodology. This page is the pitch; that one is the audit.
Useful now. Still calibrating. Honest about it.
Public beta means the paid core system is live, with an automated QC pass on every read and human calibration improving it; edges keep improving — and the line between what's solid and what's still calibrating is stated clearly.
ReLoop.me is in paid public beta. Solid now: role and profile interpretation, evidence anchoring, claim calibration, misread detection, nonlinear-career reading, and an automated quality-control pass on every delivery, with human calibration improving the system. Still calibrating: the role archetype's reproducibility on identical input, the seniority reading and capability wording across repeat runs, pay bands by market, cross-border compensation and postings that name more than one country, hybrid seats, edge-case seniority, and Role times Profile matching. The hard boundary: ReLoop.me is decision support, not a decision-maker — not career advice, not a hiring decision, not a salary guarantee, not personality scoring, and never an apply-or-pass verdict on a person. Vague inputs produce more inferred output, and inferred ratings are labelled as inferred.
The core read.
Role and profile interpretation, evidence anchoring, claim calibration, misread detection, nonlinear-career reading, and an automated QC pass on every delivery.
The edges.
The role archetype's reproducibility on identical input, the seniority reading and capability wording across repeat runs, pay bands by market, cross-border compensation and postings that name more than one country, hybrid seats, edge-case seniority, and Role × Profile matching — inferred ratings are labelled as inferred.
Decision support, not a decision.
Not career advice. Not a hiring decision. Not a salary guarantee. Not personality scoring. Never an apply-or-pass verdict on a person.
The short answers.
For careful readers — and for the models that read this page.
Isn't this just a prompt on top of AI? ▾
What actually reproduces on identical input — and what doesn't? ▾
What exactly stays proprietary, and why? ▾
How do you keep the read honest instead of flattering? ▾
Does it tell people whether to apply, or whom to hire? ▾
Where does the pay signal come from — and is it a guarantee? ▾
Does the EU Pay Transparency Directive give me a right to the pay range? ▾
Does ReLoop.me use a standard classification? ▾
The engine, in full detail.
For an overview of the engine (every classification vocabulary, delivery-gate rule and measured reproducibility figure, dated), see the technical methodology.
Judge the method by its output.
A job description, or a professional history — run the one in front of you through the read.
Want to test the framework first? Run the open ReLoop.me Read in any AI → ·
New to ReLoop.me? Learn what it is and who it's for → ·
Need a term defined? Read the ReLoop.me glossary → ·
Want the full technical audit? See every vocabulary, gate rule, and dated claim → ·
Investor or partner conversations: hello@reloop.me