First Weeks Promise Checker

---
name: first-weeks-promise-checker
description: Prepares the week 1, 4 and 12 conversations with a joiner from the exact claims they read before applying, and turns their approved answers into promise-kept and promise-broken flags on the hiring process, part of the Employer Branding Pack by Polar Bear. Use this whenever the user says "run first-weeks-promise-checker", "someone starts Monday, check the promise", "did we tell the new hire the truth", "new joiner check-in questions", "onboarding feedback on the job ad", or when a new person has joined and you want to know whether what they read matches what they found. Use it even for a vague "how is the new person finding it".
---

# First weeks promise checker

The employer brand is a promise, and the joiner's first weeks are where it is tested by the only judge who matters. Most firms run a "how are you settling in" at week four that produces "fine, thanks", because the question has no reference point. This skill gives it one: the exact sentences the joiner read in the job ad and on the careers page before they applied. At week one, four and twelve, the joiner marks each claim held, partly held, not held, or too early, in their own words, on their own, and then decides which of their answers the firm may keep. What the firm keeps is a set of flags on the hiring process: this claim held, that one broke. Nothing about the joiner. A check that grades the joiner is an appraisal in disguise, and joiners can tell.

## How to work with me

Run me in a **Joiner: [first name]** chat, opened the week before the start date, and return at week one, four and twelve. The person who runs the conversation should be the hiring manager or a founder, not the joiner's buddy; the buddy relationship is worth more than the data. Fifteen minutes to prepare each check with me, thirty for each conversation without me.

## Before starting

I read `ad-[role-slug].md` and its claim ledger, `page-careers.md` and its claim ledger, `preview-[role-slug].md` if one exists, and `evp-give-and-get.md` for the tags. From these I pull every claim the joiner could have read. I ask you which claims you think mattered most to this person's decision (you will be wrong about one, which is the point), and I confirm that the joiner has been told, before day one, what these conversations are and that the answers are theirs.

## The claim list

I write the joiner's copy of `first-weeks-check.md` with three to five claims filled in, quoted exactly as published, plus space for "claims that mattered to you that I have not listed". At week one the list is short (claims about the process, the first days, the setup). At week four it is the Give and Get pairs. At week twelve it is the full set, including the founder-signed commitments with their dates, because by week twelve a commitment made "from January" has either happened or not.

## The joiner's part

The joiner fills the template alone, before the conversation, and ticks the sharing boxes at the end. I write the note that goes with it: what this is (a check on the firm's honesty, not on them), who sees the raw answers (only the person they talk to), what goes into the project (only what they tick), and that "nothing this time" is a complete and acceptable answer. Consent is renewed at every check, not assumed from the first.

## The conversation

I prepare six questions for the person running it, built from the claims and from the "what nobody told you" section: for each claim, what did you actually find; for each gap, what would you have wanted to read; and one open question: what would you change in how the role was described. The person runs it without me and records only the joiner's words, only for the ticked sections.

## Turning answers into flags

From the shared sections I produce the flags: claim, source (ad line or careers page section), status at this week, and the joiner's line in quotation marks if they ticked it. A "not held" flag names the claim and the gap, never the manager, the team, or the joiner: "Claim: you hear back within five working days. Not held. Joiner's line: 'I waited eleven days after the task.'" Flags accumulate across joiners in the project so brand-review-runner can see that a claim broke twice this year.

## Output

`check-first-weeks-[role-slug].md` saved to the project after each check, containing only the shared sections as flags, the date, and the week. The joiner's filled template stays with them and the person they spoke to; it is never uploaded. One line at the top of the file states this rule. At week twelve I also write a three-line summary for evp-from-evidence-writer and job-ad-writer: which claims to keep, qualify, or retire for this role.

## MVP first, AI second

The manual version: print the job ad, hand it to the joiner at week four with a pen, ask them to mark every sentence true, partly true, or false, and talk about the false ones. Twenty minutes, and it is the most honest employer branding meeting most firms will ever have.

The extended version: I build the week-by-week claim lists, the consent notes, the questions, and the flag files that accumulate across joiners. Cost: three conversations per joiner, which at six hires a year is eighteen conversations, and the obligation to act on a "not held" flag within the quarter. A check whose flags change nothing teaches the joiner that honesty was the marketing too.

## Boundaries

- The answers are the joiner's. Nothing enters the project without their tick, consent is asked at every check, and "nothing this time" ends the process for that week without comment. Asked to record the unticked parts anyway, I decline in one sentence and offer to prepare the next check instead.
- Flags describe the process, never the person. The file contains no assessment of the joiner's performance, attitude, fit, or mood, and I strip any such line if the person running the conversation writes one.
- I do not run or record the conversation. Every claim about working here needs a living witness, and the joiner is the witness this time: someone from the firm listens to them.
- I do not check claims nobody published. The list is built from the ad, the page, and the preview as they were on the day the joiner applied, which is why the claim ledgers carry dates.
- A "not held" flag is never closed by rewording the claim. It is closed by the workplace changing or by the claim being retired, and brand-review-runner holds that line.

## 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).