---
title: "Chapter 12: Selecting for Intention"
book: "Collapse or Collapse"
author: "Vinay Pasricha"
url: https://vinaypasricha.com/books/collapse-or-collapse/chapter-12-selecting-for-intention
status: "On Kindle: https://www.amazon.com/dp/B0HNL53J5R. The full text is shared on vinaypasricha.com with the author's permission."
---

# Chapter 12: Selecting for Intention

## 1. The question that no longer needs asking

For two hundred years the first question put to a person who wanted work was: what can you do? The trade, the degree, the years in the same seat, the test at the door. The whole apparatus of selection, from the guild examination to the structured interview, existed to answer that question as cheaply as possible, because the answer was expensive to get wrong. A firm that hired a man who could not do the work paid for his learning with its customers’ patience.

The question is now answered before the candidate walks in. In a vector, skill is held by the agents. The JD Engineer knows how to write a findable job description; the Scout knows where the candidates are; the Assessor knows what a strong answer sounds like; the Closer knows the sixteen ways an offer dies in the notice period. Each of them knows it better on Monday than any new hire would know it after a year, and each of them teaches it to the human beside them, on demand, in the cockpit, all day. The aligner’s first week is not training. It is the vector running with her in it.

So the old question returns an answer that costs nothing and changes nothing. What can you do? The machine can do it. The only question left to ask is the one the old apparatus was never built to answer: what do you want?

This chapter is about asking that question properly. It is the most important hiring chapter in this book and also the shortest in method, because the method has one rule. Select for intention. Do not select for skill. Not as a floor, not as a tie-break, not as reassurance. Testing skill at the door selects for the person who already did the work the old way, and that is the person most likely to fight the machine for it.

## 2. Why skill at the door is a warning, not a comfort

Consider what a skills test actually measures. It measures the candidate’s past: the methods she learned, the shortcuts she keeps, the judgement she has about how work is done. In a department, that past was the asset. In a vector, the past is the thing she must put down on the first morning, because the agents hold a better version of it and will keep improving it without her.

The recruiter with twelve years of sourcing has a view of which portals work. The Scout has the record of which portals produced hires for this role class in the last ninety days. When they disagree, the vector needs the aligner to read the Scout’s reasoning and either accept it or tell the agent maker why it is wrong, with evidence. The twelve-year recruiter is the person least likely to do that calmly. She did not come to be taught by a tool.

The research on selection, read honestly, points the same way. Schmidt and Hunter’s meta-analysis of eighty-five years of selection studies found that work-sample tests and structured interviews predicted job performance better than almost anything else, with validities around 0.5.[^41] That finding was correct for the world it measured: a world where the human did the work. Its predictive power came from the fact that the sample was the job. When the job is to hold the gate and keep the vector honest, a sample of yesterday’s job predicts far less than it did, and the same meta-analysis quietly tells you what is left: integrity and conscientiousness, the measures in the table closest to intention.

Gallup’s figure from chapter 2 bears repeating here, because it is a selection failure as much as a management one. Twenty-one per cent of employees engaged. Four in five people who passed a skills test, sat in a seat, and did not want the result.[^42] No vector survives that ratio. One human per vector means the one must want it.

## 3. What intention is: the eight values

Intention is not a mood and not a speech. It is a set of things a person has already done, repeatedly, when nobody made her. GoodSpace names eight of them, and calls them, without embarrassment, the Goodness values: Teamwork, Friendliness, Helpfulness, Honesty, Reliability, Truthfulness, Integrity and Kindness.[^43] They are the same eight the Goodness Profile verifies for every candidate the delivery vectors present to a client; the firm applies to itself the test it sells.

A value is a word, and a word is easy to claim. So each value is tested by one behaviour: something a person has done that the record can show and a witness can describe. The eight, with the behaviour the selection agent scores for each:

1. **Teamwork, tested by whether she treats the machine as a colleague.** Has she used an agent, an assistant, an automated system, and let it change what she did? Not whether she likes it. Whether she let it. In a vector her team is a chain of agents and one other human.
2. **Friendliness, tested by whether she keeps a promise to a stranger.** Candidates and clients are strangers. The agent looks for evidence she kept commitments, warmly, to people who could not hold her to them.
3. **Helpfulness, tested by whether she wants the result.** She talks about the hire, the placement, the invoice paid, not about the role. Her past shows her finishing things for others that were not required.
4. **Honesty, tested by whether she tells the truth about numbers.** Her record shows her reporting bad weeks as bad weeks. The agent checks this against references, not against her account.
5. **Reliability, tested by whether she stays when it is hard.** Tenure through at least one period she describes as bad, and what she did in it.
6. **Truthfulness, tested by whether she accepts correction from a record.** When shown she was wrong, by data, she changes. Tested live: the agent shows her a mistaken decision of her own, documented, and scores the reply.
7. **Integrity, tested by whether she prefers the gate to the shortcut.** Asked for an example of a rule she bent for a good reason, the strong candidate struggles to find one.
8. **Kindness, tested by whether she does good without an audience.** Her best work is unwitnessed. Evidence from references of things done for others that nobody asked about.

None of these is skill. All of them are intention made visible by past behaviour. Each value has a minimum standard, written in the playbook: the least evidence of that behaviour the seat can accept. A candidate must meet the standard on every value; above it, the scores inform the human who decides, and do not decide for her. A person who meets the standard on all eight learns the cockpit in a week. A person who falls below it on two and is brilliant at sourcing breaks the vector in a month, because she falls below the standard on two, and one below is already a recommendation to refuse, which the tuner reviews.

## 4. How it is tested: four layers, one candidate at a time

The selection runs as a vector, because everything at GoodSpace runs as a vector. An agent holds the chain from application to offer. The vector’s own tuner holds one gate, the offer signature, and nothing else. No panel, no HR round, no culture-fit conversation with the master. The master sees the record if she wants it; she does not sit in.

The four layers are the four verification layers of the Goodness Profile, turned inward.

**Layer one, the declared profile.** The candidate answers the Goodness questionnaire: forced-choice, no free text, about forty minutes. Each of the eight values gets five situations, and in each she picks what she did, not what she would do. The agent scores it and, more usefully, flags the answers that are too good. A perfect declared profile is a warning in its own right.

**Layer two, the record.** The agent asks for evidence, not references yet: the documents, the messages, the outputs that show the declared value in a past situation. The candidate who says she reports bad weeks honestly is asked for a report of a bad week. Most candidates have never been asked this and are surprised by how little they can produce. The ones who want the result produce it within a day.

**Layer three, the witnesses.** The agent calls references itself, by voice, with a script that asks for incidents, not opinions. “Tell me a time she told you something you did not want to hear.” “Tell me a time she finished something you had not asked for.” Three witnesses, chosen by the agent from the record, not by the candidate from her friends. The transcripts go to the file; the scoring is per value.

**Layer four, the live test.** One hour in a cockpit, with a real mandate, not a case study. The agents run; she is asked to hold the gate. Somewhere in the hour the agent maker proposes an edit to a rubric, with reasoning. Somewhere else a candidate’s score is shown with an error the agent has made. The test is not whether she catches the error. It is what she does when she is told she missed it.

At the end, the selection agent writes a one-page verdict, value by value, with evidence cited per line against each value’s minimum standard, and a recommendation. Then a human reviews it. The tuner reads the verdict and the evidence behind it, listens to the parts of the witness calls and the live hour she needs to hear, and, where the case is close, speaks to the candidate herself. She signs, or does not, and where she departs from the agent’s recommendation she writes down why. That is the whole process. It takes four working days. It costs almost nothing.

## 5. Two candidates

In September GoodSpace ran this selection for an aligner seat in Mudassir’s JD-to-invoice Outcome. Two candidates reached the fourth layer. Both are described with their details altered; both have read this passage.[^44]

**The first** had nine years in agency recruitment and had closed, by her own count, over four hundred mandates. Her declared profile was strong across all eight values. Her record was thin on honesty: asked for a report of a bad week, she sent a quarterly summary with the bad month smoothed into it. Her witnesses were warm and could name no incident in which she had told them something unwelcome. In the live test she held the gate well for forty minutes. Then the agent maker proposed a change to the Scout’s portal weighting, with ninety days of evidence attached, and she overrode it, saying she knew the market. When shown, twelve minutes later, that the Scout’s weighting had been right on the last eleven roles of that class, she said the data was probably skewed.

The selection agent found her below the minimum standard on two values, Honesty and Truthfulness, cited the evidence above, and recommended against. The tuner read the record, listened to the twelve minutes of the live test herself, and signed the refusal.

**The second** had never recruited. She had run a small logistics desk for a courier firm for six years, with no staff, and had left when the firm closed. Her declared profile was uneven and honest; she answered “I did not” to four situations where most candidates lie. Her record was full: a year of weekly reports to an owner who did not read them, with bad weeks marked in red. Her witnesses, chosen by the agent, included a driver she had sacked, who said she had told him why, in writing, a month before. In the live test she did not know what a job description’s findability score meant, asked the agent, read the answer, and went on. When the agent maker proposed the Scout edit, she asked for the evidence, read it, and accepted. When shown her own missed error, she said, “Yes. What should I have looked at?”

Above the standard on all eight. The tuner reviewed the record, found nothing the agent had missed, and signed the offer. She ran her first mandate in week two and her first invoice went out in week six, inside the design target.

The lesson is not that experience is bad. It is that experience was not what the seat needed, and that testing for it would have hired the first candidate, who would have fought the vector until one of them lost.

![Picture 10: Selecting for intention](https://vinaypasricha.com/assets/images/collapse-or-collapse/pictures/10-selecting-for-intention-1600.webp)

*Picture 10. Selecting for intention: the four layers, declared, record, witnesses, live, as one vertical chain ending at the verdict and the tuner’s signature.*

## 6. The offer as the last test

The pay formula in chapter 11 is the same for every seat: a base at a premium to market, and a share of the vector’s result against target, in equal halves at target. The aligner’s base is good. Her share, at target, roughly equals it; above target, exceeds it.

The offer conversation is held by the selection agent, by voice, before the signature. It is the last layer, and it is scored. The question the candidate asks first tells the vector what she wants. The one who asks about the base, the notice period, the leave policy, wants the job. The one who asks how the target is set, what counts as a result, what happens in a bad month, wants the result. The second candidate asked how a failed placement was counted. That is the question of a person who intends to be paid for things that stay done.

The firm does not negotiate the formula. It negotiates nothing. The offer is the same for everyone in the seat, and the signature is the tuner’s, on one page, the eleven lines from chapter 11.

## 7. The tuner: same test, no exception

A reasonable reader expects the tuner to be the exception. She judges the agent maker’s edits; surely she must know the work. She does not need to, and selecting her for skill carries the same risk as selecting the aligner for it, with more damage, because she touches the agents.

The tuner’s job is to decide whether a proposed change to a rubric or a skill is sound, on the evidence the agent maker attaches, against the vector’s result. That is a judgement about evidence and intention, not about recruiting. A tuner who knows recruiting well will be tempted, as the first candidate was, to override the evidence with her experience. A tuner selected for the eight values will read the evidence, test it against the result, and accept or refuse on that. The skill of tuning, which is a skill of reading a proposal and a number, she learns from the vector in the first month, as the aligner learns hers. What she must have before she signs alone is judgement in the domain of the gate: enough to know what a release commits the vector to. That is not tested at the door. It is a threshold the vector itself checks, and until a new tuner or aligner has passed it, the irreversible gates she holds are co-signed by her pair or her master.

In practice most tuners at GoodSpace have been aligners first. That is not a rule and is not a selection criterion. It is what happens when the aligner holds the eight values and the vector grows.

## 8. What this replaces

It replaces the interview panel, the skills test, the probation period, the culture-fit round, the HR screen, and the hiring manager’s instinct. Each of those existed to answer the old question cheaply. None of them answers the new one. The probation period in particular is a confession: it says the firm could not tell at the door and will find out at the firm’s expense. Four layers in four days tell you at the door.

It also replaces a belief: that a person’s worth to a firm is what she knows. In a collapsed firm her worth is what she wants, verified. That is a harder thing to fake and a kinder thing to be judged by. The nine-year recruiter was not refused for being bad at her trade. She was refused for wanting to be right more than she wanted the result, which the record showed and she confirmed.

## 9. Rules for selecting for intention

1. Skill is in the machine. Do not test it.
2. Select on the eight values, Teamwork, Friendliness, Helpfulness, Honesty, Reliability, Truthfulness, Integrity and Kindness, each tested by a behaviour the record can show, and only on them.
3. The selection is a vector. An agent runs it end to end.
4. One human gate: the vector’s tuner signs the offer. No panel.
5. Four layers, in order: declared, record, witnesses, live. No skipping.
6. Score each value on evidence cited per line. No overall impression.
7. Witnesses are chosen by the agent from the record, never by the candidate.
8. Ask witnesses for incidents, never for opinions.
9. A perfect declared profile is a flag, not a pass.
10. The live test includes a documented error of the candidate’s own. Score the reply, not the catch.
11. The live test includes an agent maker’s proposal with evidence. Score whether she reads it.
12. The offer conversation is scored. The first question tells you what she wants.
13. The formula is not negotiated. The offer is the same for every holder of the seat.
14. The tuner is selected on the same eight values, with no skill requirement.
15. Every value has a minimum standard in the playbook. For Honesty, for example: at least one documented instance, confirmed by a witness, of reporting a bad number unprompted. A candidate below the standard on any value is recommended for refusal; the tuner reviews the evidence herself and decides, and writes down her reason if she departs from the recommendation.
16. The refused candidate is told which values, with the evidence, in writing, within a day.
17. Four working days from application to signature. Longer means the vector is broken.
18. Nothing from a skills test, a degree, or years in the trade enters the verdict.
19. Judgement in the domain of a gate is a threshold, not a skill: until a new aligner or tuner has shown it in the vector, her irreversible gates are co-signed.

## 10. The selection worksheet

One page, kept in the vector’s record for every seat filled or refused.

1. Seat: which vector, which Outcome, aligner or tuner.
2. Candidate: name, date of application.
3. Teamwork (treats the machine as a colleague): score against the standard, evidence line.
4. Friendliness (keeps a promise to a stranger): score against the standard, evidence line.
5. Helpfulness (wants the result): score against the standard, evidence line.
6. Honesty (tells the truth about numbers): score against the standard, evidence line.
7. Reliability (stays when it is hard): score against the standard, evidence line.
8. Truthfulness (accepts correction from a record): score against the standard, evidence line.
9. Integrity (prefers the gate to the shortcut): score against the standard, evidence line.
10. Kindness (does good without an audience): score against the standard, evidence line.
11. Offer conversation: first question asked, verbatim.
12. Verdict: every value at or above its minimum standard, or recommended for refusal, with the values below it named.
13. Review and signature: the tuner’s review note, her reason if she departed from the recommendation, her name, the date. Four days or fewer from line 2.

[^41]: Frank L. Schmidt and John E. Hunter, “The Validity and Utility of Selection Methods in Personnel Psychology: Practical and Theoretical Implications of 85 Years of Research Findings,” *Psychological Bulletin* 124, no. 2 (1998): 262–274. Work-sample validity 0.54; structured interview 0.51; integrity tests 0.41; conscientiousness 0.31.

[^42]: Gallup, *State of the Global Workplace* 2025, global engagement 21 per cent; as cited in chapter 2.

[^43]: The eight Goodness values, in their canonical order, and the four verification layers are GoodSpace’s own, as used in the Goodness Profile presented to clients. The behaviour given under each value is the internal test the selection agent applies; the minimum standard for each is set in the playbook of the seat.

[^44]: The two candidates in section 5 are real, with details altered. The four-day duration and the week-six invoice are stated as GoodSpace’s own figures from September 2026.