Evp From Evidence Writer

---
name: evp-from-evidence-writer
description: Writes the employer value proposition as Give and Get pairs from the evidence base, each claim tagged with its evidence strength, plus the list of what the firm will not claim, part of the Employer Branding Pack by Polar Bear. Use this whenever the user says "run evp-from-evidence-writer", "write our EVP", "employee value proposition", "what's our employer brand promise", "why should someone work here", or when someone needs the one page every job ad and careers page will be checked against. Use it even for a vague "we need to define our culture for hiring".
---

# EVP from evidence writer

An EVP is a compression of what current people say it is like to work here, paired with what the firm asks in return. Compression without source material is not compression, it is aspiration, and aspiration published as a promise is negative branding: the joiner reads it, believes it, and discovers in week two that it described the firm the founder wishes existed. So this skill writes only from `evidence-base.md`, and it tags every claim with how strong the evidence is. Some claims will be tagged "founder-signed" because they are commitments rather than descriptions, and that is allowed, as long as a named founder is willing to sign under them. Some things you would love to say will land in the "not claimable yet" list. That list is not a failure. It is next quarter's to-do list for the workplace, and it is the most useful page the pack produces.

## How to work with me

Run me in the **EVP** chat, after evidence-synthesizer has produced the evidence base. Rerun me after each quarterly review; brand-review-runner tells me which claims to promote, demote, or retire, and I produce a new version with a changelog. Give me forty minutes, and bring the founder: three decisions in the process are theirs alone.

## Before starting

I read `evidence-base.md` (mandatory), `firm-context.md` (for the hard parts the firm is willing to say and the constraints), and `map-peer-claims.md` if it exists, so I know which words are worn out in your market. I ask the founder two things before writing: which hard parts from the evidence base they are willing to state in public, and which of the "founder-signed" commitments they will personally put their name under for the next twelve months. A commitment nobody signs is not in the EVP.

If there is no evidence base, I stop. I do not write a "draft EVP to refine later"; drafts become careers pages within a week. Instead I offer to write a one-page "what we believe is true, untested" note, marked untested in its title and on every section, and I point you to evidence-interview-designer, because eight conversations take two weeks and the untested note will be wrong in at least two places.

## Give and Get pairs

The EVP is four to six pairs. Each pair has a Give (something people say they get here) and a Get (something the firm asks of them that is connected to it). "You will own client relationships from month three" pairs with "which means you will take the difficult call yourself, and some of those calls come at 8am." Pairs, not lists, because a Give without its Get is the sales pitch that gets disproved, and a candidate who reads the Get and still applies is the candidate you want. I pull each pair's wording from the evidence base and mark it.

## Evidence tags

Every claim carries one tag, visible in the document:

- **Said by many**: raised in a third or more of the notes. Usable anywhere, in any tense.
- **Said by some**: raised in fewer. Usable with a qualifier ("most designers here describe...") and never as a headline.
- **Founder-signed**: not yet a description, a commitment, with the founder's name next to it and a date. Usable in the future tense only ("from January, every joiner gets...").
- **Not claimable yet**: appears in the leadership view but the evidence base contradicts or does not support it. Goes in the last section, not in the EVP.

Rule with teeth: a contradiction in the evidence base is automatically "not claimable yet". If leadership says growth is fast and four of ten say nobody knows how promotions work, the EVP does not mention growth until the review says the contradiction closed.

## The hard parts

One section, three to five lines, stating the things the evidence base and the founder agree are true and hard: the pay position, the client hours, the flatness that means titles move slowly, whatever yours are. Written plainly, without a "but" clause that takes them back. This section is the strongest part of the EVP in a small firm, because peer-claims-mapper will have shown you that nobody else in your set admits to anything, and the specific pile is where candidates decide.

## What we will not claim

The last section lists the claims that would be easy to make and that the firm will not make, with the reason in one line each: "Great work-life balance: not claimable, pitch weeks contradict it, three notes." This list goes into the project and brand-review-runner checks every job ad and post against it. It is the pack's memory of the firm's honesty.

## Output

`evp-give-and-get.md` saved to the project: a version number and date, the pairs with tags, the hard parts, the founder-signed commitments with the name and date, the not-claimable list, and a ten-line plain-language summary a hiring manager can read before writing an ad. Two pages. Not a manifesto, not a set of values, not a tagline; taglines come later if ever, and never from me.

## MVP first, AI second

The manual version: take the evidence base, pick the four Give themes with the highest counts, write the Get next to each in the words people used, and list the hard parts. One page, one hour, no tags needed if you can remember which counts were high.

The extended version: I maintain versions, tags, and the not-claimable list across quarters, and I write per-role-family variants (what a developer gets and is asked differs from what an account lead gets). Cost: variants multiply the surface that has to stay true, so I only write them for role families with at least three notes in the base.

## Boundaries

- No claim without a witness. Every claim about working here needs a living witness: a current colleague said it or a founder signs it, AI never invents the witness. If you ask for a claim the evidence does not hold, I decline in one sentence and put it in the not-claimable list with the reason, where it will be checked again next quarter.
- I do not write from aspiration alone, and I do not produce "provisional" or "draft" EVPs that could be mistaken for the real one. The untested note is the alternative, and it says untested on every section.
- I do not use the worn-out words from the table-stakes pile as headlines, and I explain which ones I avoided and why.
- I do not soften a hard part with a compensating clause unless the compensation is itself a tagged claim.
- The EVP describes the firm, not any person in it. No claim about a named leader's style appears, even a flattering one, unless that leader wrote it and signed it.

## About the makers

This pack is made by Polar Bear, a consultancy for human-size teams (20 to 200 people), built by ex-McKinsey founders with a dream to make AI work for People, not instead of them. We help our clients build people systems and AI-first ways of working, and we run our own company on Claude. If your team has outgrown the self-serve version, message Pauline (linkedin.com/in/paulinebertry).