How Plutarch Builds a Brief
Plutarch is hiring feasibility intelligence for the energy industry: evidence on whether a role can be filled, at what price, and who else could do it. You paste a job description; you get a brief. This page describes how that brief is made.
1. The job description is read once
A model reads the job description a single time — the role, its location, whether it depends on a safety or technical certificate, and how far its skills carry across sectors — and every later step uses that one reading, so no two parts of the brief disagree about what the role is. Text that is not a job description is refused, and the credit is returned.
2. Evidence is gathered
Plutarch keeps a knowledge base of individual facts — a salary, a demand signal, a certificate requirement — extracted from official wage statistics, published salary guides, employer careers sites and industry reports. Each fact keeps the source it came from. For pay, Plutarch also searches the web live for your exact role on every brief. That live research sets the pay figure; the knowledge base adds its sources only where it agrees. Each research step runs on a frontier AI model, with a model from a different provider behind it if the first fails.
3. The verdict is computed, not written
The feasibility verdict comes from fixed rules applied to that evidence — not from a model’s opinion. A model writes the explanation around it; it does not choose the verdict. Pay figures are taken from the evidence, never made up, with one labelled exception below.
4. Every pay figure says what it rests on
- Official source
- From a named official/government source.
- Licensed data
- From a licensed third-party data vendor.
- Researched live
- Researched from public market sources for this exact role while the brief ran — cited, directional, and being corroborated into the knowledge base.
- Derived estimate
- Synthesized from public market sources we collected earlier, not one named source — treat as directional, not exact.
- AI guess
- No cited pay evidence exists for this role anywhere we looked — the knowledge base or live web search. This is a plain AI guess from general knowledge, not sourced or verified data. Treat it as a rough starting point only.
Each figure also carries a confidence level. An AI guess is always low confidence, and a gap that ends in one is queued so that the role gets researched properly.
5. You can see the reasoning
Every brief opens to its trace: what each step read, what it chose to research or skip and why, and the evidence behind each number.
6. Three labels for AI-written content
- AI guess
- A pay figure in a brief that no source backs. Shown only when every search found nothing, always low confidence.
- AI-written summary
- The short summary on a role page. A model writes it from the counted figures beside it, and it contains no numbers of its own.
- Hypothesis
- On the training page, one possible reading of what training bodies have published. It is shown with the statements behind it, and it is never a measurement.
What Plutarch does not do
- It holds no candidate data — no CVs, no profiles, no individual people.
- Your job descriptions and briefs are never visible to another customer, unless you share a brief’s link yourself.
The public role pages — AI, digital & trading roles and technical roles — have their own counting rules. Figures published on this site may be reused under these terms. To try it, paste a job description — your first 2 briefs are free.