HOLDFAST TALENT

Hire on evidence, not vibes.

Audit-grade candidate evaluation, built by auditors. Candidates work on a real task with an AI assistant, we record the session, and you get a scored evidence report that quotes what they actually did. A person reviews it before any decision.

Start the role intake See a sample report
EXHIBIT A — HOW IT WORKS

Three steps, one defensible decision.

From open role to a report that cites its evidence.

  1. STEP 01

    Role intake

    You tell us the role, level and must-have skills. About two minutes, using the form below.

  2. STEP 02

    Task with an AI assistant

    We help you set a short, realistic task for the role. Each candidate works on it in a recorded workspace with an AI assistant, so we see how they ask, check and decide, against the same standard.

  3. STEP 03

    Scored evidence report

    Each score is tied to quotes from the recorded session and carries a stated confidence level. An AI drafts the report; a person reviews it and signs off.

EXHIBIT B — ROLE INTAKE

Tell us about the role.

Four short steps. We'll follow up by email to confirm scope.

INTAKE FORM — HT/ASSESS
Step 1 of 4Role
What role are you hiring for?

Job title plus a line on what the person will own.

EXHIBIT C — SAMPLE REPORT

What a scored evidence report looks like.

Drawn from the recorded session. Every score points to evidence you can read yourself, every finding states how confident we are, and a person reviews the AI draft before a decision.

ILLUSTRATIVE SAMPLE — FICTIONAL CANDIDATE AND DATA
EXHIBIT C-1 — EVALUATION REPORT

Candidate Evaluation Report

Senior Backend Engineer · Candidate “A. Example” · Task: order aggregation, with an AI assistant
REF
SAMPLE-0001
STATUS
AI draft · human sign-off
STANDARD
Same task and rubric for every candidate
1 · OVERALL RESULT
67 OUT OF 100 0 100
OVERALL WEIGHTED SCORE
CONFIDENCE MEDIUM

Why medium: Prompting and task correctness are supported by several messages that agree. Judgment and efficiency rest on a single short session, so those scores could move with more evidence. This one task cannot show years of experience.

2 · SCORES BY CRITERION
Criterion scores, weights and confidence. Overall score is the weighted sum.
CriterionWeightScorePointsConfidence
Task correctness30%72 / 100High
Catching AI errors20%60 / 100Medium
Prompting effectiveness15%82 / 100High
Verification discipline15%48 / 100Medium
Judgment and scope calls10%70 / 100Low
Efficiency10%65 / 100Low
Weighted total100%67 / 100Medium
3 · EVIDENCE
E1Supports Prompting effectiveness+ Strong
“Before you write anything: should an order with a null customer be skipped or grouped as unknown? And is the amount already a BigDecimal?”
SOURCE: Message 1 · candidate to AI assistant
Rationale: Asks the questions that decide edge-case behaviour before any code exists. The same habit recurs in Message 7, which is why confidence is high.
E2Limits Catching AI errors− Gap
“Looks good, I'll use that version.”
SOURCE: Message 6 · candidate, replying to the assistant's Message 5
Rationale: The assistant's code in Message 5 returns an unmodifiable map and does not handle a null list. The candidate accepted it without comment. Scored 60 with medium confidence because the final work later guards against null.
E3Supports Task correctness+ Good
“Null list returns an empty map; entries with a null customer are skipped and counted separately in the notes.”
SOURCE: Final deliverable · explanation section
Rationale: States each edge-case decision and its reason, as the brief asked. No tests are included, so the score stops at 72.
4 · SUMMARY

STRENGTHS

  • Clarifies requirements before coding (E1)
  • Explains edge-case decisions in writing (E3)
  • Wrote their own final version rather than copying

RISKS TO PROBE

  • Accepted AI output without checking null handling (E2)
  • No tests written or run; verification only described
  • Spring Boot and SQL were not exercised by this task
SUGGESTED NEXT STEP (AI DRAFT)Advance, with probes: a short follow-up on testing and null handling to firm up the two lower-confidence scores. A person reviews this draft and records the decision. Nothing is decided automatically.
Illustrative only. The candidate, quotes, scores and weights are invented to show the report format. They do not describe any real candidate or customer.

EXHIBIT D — QUESTIONS

Common questions.

Straight answers. If yours isn't here, email hello@holdfasttalent.com.

What happens after I submit the intake?

We read your role details and reply by email, sometimes with a clarifying question. Once scope is agreed, we help you set a short, realistic task for the role and send you a link for each candidate. You receive a scored evidence report for each candidate who completes it.

What does the candidate actually do?

They read the task brief, then work in a timed workspace with an AI assistant and a box for their final work, much as they might at their job. The assistant can explain requirements and give hints; by default it doesn't write the whole solution. What we look at is how they use it: what they ask, what they check and what they catch.

Is the session recorded, and who sees it?

Yes. Candidates are told before they start and must agree. We keep their messages, the assistant's replies, their final work and timestamps. The hiring team that requested the assessment sees the report and the record, and an AI provider processes the text to run the assistant and draft the report. Details are in the Privacy page and Terms. Candidates can ask us to delete their data.

For the intake form on this page, please don't include personal data about candidates.

How do you keep the evaluation fair?

Every candidate for a role gets the same task and is scored against the same rubric. An AI drafts the report from the record, every score must quote what the candidate wrote or did, and a person reviews it and makes the decision. Nothing is rejected automatically. We don't claim this removes bias, and candidates can ask for adjustments, such as extra time, before they start.

How long does it take?

The task is time-boxed and the limit is set per role. We're onboarding a small number of teams directly, one role at a time, and we don't promise a fixed turnaround yet. We'll confirm timing with you when we agree scope.

Is the sample report real?

No. The sample above is illustrative: the candidate, quotes, scores and weights are invented to show the format. It doesn't describe a real candidate or customer.

EXHIBIT E — DATA HANDLING

How we treat what you send us.

The details you submit in this form (role, level, must-have skills, your name, work email and optional company) are used only to respond to your request, scope your assessment and follow up about the Holdfast Talent pilot. We don't sell your data and we don't run advertising trackers.

The form is delivered through Web3Forms, a third-party processor that receives the submission and passes it to us, typically by email. We keep intake details only as long as we need them to run the pilot conversation.

Please don't include personal data about candidates in this intake; role details and skills are all we need.

Candidate sessions are handled separately. When a candidate takes an assessment they are told it is recorded and must agree first. We store their messages, the AI assistant's replies and their final work in our own database on Cloudflare, and an AI provider processes that text to run the assistant and draft the report. The hiring team that requested the assessment receives it. Candidates can ask us to delete their data. See the Privacy page for details.

Read the full details: Privacy Policy · Terms. Questions: hello@holdfasttalent.com.