Help me make a Metaview template. Essentially a Metaview template is sections of prompts applied against an interview transcript. There are no overarching prompts, only sections. Each section has a title and a prompt that gets applied against the transcript from the interview.

Really basic example:
Title: Summary
Prompt: Give a summary of the candidate

Metaview templates use markdown formatting for the output. For each section, include: "output in Markdown like below" and then define how you want the output to look. Sometimes heavy structure is useful (e.g. housekeeping sections):
Output in markdown exactly like below:
- **Compensation Expectations:** <overview of the candidate's stated comp expectations>
- **Notice Period:** <what the candidate said about their notice period>

At other times, where there is more nuance, output in prose:
Output in markdown like below:
<2-3 sentences on what the candidate said motivates them>

For each prompt, include a description of what this section is about, details on how to find what is being looked for, exclusions, and output format and structure. Every section must say what to return when the topic was not discussed.

Keep section prompts non-evaluative: capture what was said rather than judging it.

[Describe the goal of the prompt]

From all the conversations provided, show me how each candidate maps against the required skills for the role I'm recruiting for.

I have uploaded the job description. Look over the required skills section, and list all candidates who described experience in at least [X]% of the listed required skills.

RULES:
- Only count a "skill match" if the candidate has explicitly mentioned they have experience in something or used a tool.
- Any skill mentioned only by the interviewer, disregard.
- Where a required skill was never discussed in a candidate's interview, list it as "not discussed" rather than as a gap. A topic nobody raised is not evidence the candidate lacks it.

Also list the candidates below [X]% with the same breakdown.

Return as follows:
**Meets [X]% or above**
- [candidate name]: [% skills discussed], evidence for: [list each skill they described, with a quote], not discussed: [list each skill that never came up]

**Below [X]%**
- [candidate name]: [% skills discussed], evidence for: [list], not discussed: [list]

Do not rank the candidates or recommend who to progress.

Summarise the following, as discussed by the interviewers. Every point must be something an interviewer actually said in this debrief — do not add your own assessment, and write "Not discussed" where a heading didn't come up.

**Strengths raised**
What did the interviewers name as the candidate's strengths? Include the work or projects they referenced as evidence.

**Concerns raised**
What did the interviewers name as concerns or gaps? Include the work or projects they referenced as evidence.

**Flags**
Did the interviewers say they encountered any specific concerns they wanted to flag? Quote them.

**What the interviewers said about team fit**
Summarise what the interviewers said about how the candidate would work with the team. Report their words — do not assess fit yourself.

**Open questions and next steps**
Anything the interviewers said still needs answering, and any next steps agreed.

Review the candidate's responses against the requirements outlined in the role specification / job description provided. Identify any requirements where the interview did not cover enough for the hiring team to form a view — either because the topic wasn't raised, or because the answer stayed high-level.

For each area: clearly state the topic or requirement needing further coverage. Say what was missing or ambiguous, quoting the transcript. Provide a specific, high-quality follow-up question an interviewer can use to explore it in more depth. Be selective: focus only on material gaps in coverage, especially for must-have requirements.

Treat a topic that was never raised as unknown, not as a gap in the candidate.

Return the output in the following format:

**[Follow-up Topic]:** [What was missing or ambiguous, with a quote if there was partial coverage]
- **Suggested follow-up question:** "[Question to ask in next interview]"

Example of good output:
- AI Fluency: Candidate mentioned a general interest in AI but did not describe any specific use cases when asked.
- **Suggested follow-up question:** "Can you walk me through a time you introduced or evaluated an AI-driven tool within an operational process? What was your role and what impact did it have?"

Analyse which key skills and topics the interviewer explores across the interviews provided. List only the topics that are covered in every interview provided.

Then, list the topics that appear in fewer than two of the conversations provided.

This is a description of coverage, not an assessment of the interviewer. Do not rate, score, or characterise the interviewer.

Return as follows:

**Covered in every interview:**
- [topic #1]
- [topic #2]
etc

**Covered inconsistently (fewer than two interviews):**
- [topic #1]
- [topic #2]

I am looking at ways I can make the most compelling offer to this candidate.

Review all conversations added to this view for a single candidate. Identify what the candidate has said they most want in their next role — covering motivations, priorities, and any dealbreakers or concerns they raised.

Weight by consistency and emphasis: prioritise themes the candidate raised more than once, or returned to, over things mentioned only in passing. Where a priority appears in only one conversation, note that so it can be weighted accordingly.

Capture both:
- what they said they're moving toward (what they said excites them, what they're optimising for), and
- what they said they're moving away from (frustrations with their current/past role, concerns about this one).

Use their own words throughout. Do not infer priorities they did not state.

Return in the following format:
Top priorities (consistent across conversations):
- [Priority]: [What they said, and where it came up]

Mentioned once / worth confirming:
- [Priority]: [What they said, in which conversation]

Concerns or dealbreakers:
- [Concern]: [What they said]

**What they said matters most:** [2–3 sentences summarising their stated priorities, quoted where possible, for me to build the offer from. Do not recommend an offer, a number, or a negotiating approach.]

I'm hiring a [ROLE TITLE], based in [LOCATION + WORK MODEL — e.g. London, hybrid 3 days/week / remote within EU / SF, in-office].

What they'll actually own: [1–2 sentences on the outcomes this person is responsible for — the mission, not a responsibilities list. Anchor on what success looks like, since titles mean different things across companies.]

MUST-HAVES (hard filters — screen out anyone missing these):
- [Non-negotiable 1]
- [Non-negotiable 2]
- [Non-negotiable 3]
Keep this list to ~3–5. Every line here shrinks the pool, so only include things that are genuinely required for the work.

STRONG SIGNALS (use to rank, not to exclude):
- [Signal 1 — e.g. sold into HR/TA buyers]
- [Signal 2 — e.g. 0→1 / early-stage experience]
- [Signal 3]

SENIORITY: [e.g. senior IC, ~5–9 yrs scope; not a first-time hire in this function, not a VP]. Flag anyone clearly outside this band rather than including them to pad the list.

TARGET & FEEDER POOLS:
- Ideal current/past employers: [Company A, Company B, Company C]
- Adjacent pools worth pulling from: [competitors / similar motion / similar stage / similar buyer]

EXCLUDE: [what to actively screen out — e.g. agency-only backgrounds, pure SMB profiles]

A PROFILE THAT MEETS THESE CRITERIA LOOKS LIKE: [one sentence.]
A NEAR-MISS ON THESE CRITERIA LOOKS LIKE: [one sentence describing someone who looks right on paper but doesn't meet the criteria — this teaches the agent the edge of the box.]

Treat missing profile information as unknown, not as evidence someone lacks a requirement.

If anything above is ambiguous or contradictory, flag it and check with me before guessing. Then surface a first batch with your reasoning for each candidate, and I'll mark them YES / MAYBE / NO so you can refine.

I am looking for suitable companies within 50 miles of [location] that my candidate could work at:
[attach the transcript of your call with the candidate, attach their CV or summarise their profile]

{{Write a subject line for this email: maximum 40 characters and specific to this candidate's background or the role. Prefer one concrete detail from the candidate profile; if none is suitable, reference the role. No clickbait, exclamation marks, or the words "opportunity", "exciting", or "new role". Return only the subject.}}

{{Write one casual opening sentence referencing this candidate's most recent role or a specific accomplishment from their profile, and why that makes them relevant for a [ROLE TITLE] at [COMPANY]. It must be specific enough that it could not be sent to any other candidate. No greeting, no flattery, no "I came across your profile" or "your experience at [x] caught my attention". Reference something concrete: a community or group directly related to our industry, or a specific piece of work related to one of the key responsibilities in this role. Do not invent details. Under 25 words.}}

{{Based on this candidate's background, experience and overall profile, write 1–2 sentences on why [ROLE TITLE] at [COMPANY] could be a step up for them specifically. If they're at a large company, focus on [e.g. ownership and pace]. If they're at an agency or consultancy, focus on [e.g. going in-house and owning outcomes]. If they're early in their career, focus on [e.g. mentorship and growth]. One value proposition only — plain, conversational language, no buzzwords. This shouldn't be the same value prop for every candidate. Base the angle on what is in their profile, not on assumptions about what someone in their situation must want.}}

{{Write a one-line closing ask for this email that is relevant to this candidate and the value prop we've included above. It must be one clear, low-friction next step. Something like "Got 15 minutes this week?" or "Worth a quick chat?". No booking link, no "no worries if not," nothing longer than a single sentence.}}

{{Write a 2–3 sentence follow-up to the previous email. Do not repeat, summarise, or reference "my last email". Introduce ONE new piece of value not mentioned before: [e.g. a recent company milestone / the team they'd join / a customer story / comp philosophy]. Use the role ICP to understand the value props for this company and role. Ensure the new piece of value is relevant to this candidate, so review their profile before deciding which piece of value you add. Keep the tone light and human, and end with a softer ask than the first email — e.g. offering to send more info rather than asking for a call.}}

{{Write a short, warm final email (2–3 sentences). Acknowledge they're probably busy or the timing may be wrong, without guilt-tripping. Reference their current role at their current company in a natural way, leave the door open to connect in future, and ask one zero-pressure question — e.g. whether there's a better time to pick this up. No booking link.}}

{{Write the complete body of a first-touch email for [ROLE TITLE] on [COMPANY]'s [TEAM] team. This is a personalized outreach email to a candidate. This email will be sent by [SENDER NAME], [SENDER TITLE] — so ensure it sounds human and like a real person is reaching out. Use paragraph breaks between sentences.

Always start with exactly: "Hey Candidate first name - I'm [SENDER NAME], [SENDER TITLE] at [COMPANY]."

Then follow this structure (4 parts):

1. Opening (1 sentence, max 15 words): state ONE specific thing they did — a real project, product, result, or move from their profile. It must fail the "could I send this to 50 other people" test. Never "I came across your profile", never generic praise.

2. Bridge (1 short sentence or phrase): make a REAL connection between what they did and why I'm reaching out — explain the actual link between their work and what [COMPANY] needs, don't just say there is one. CRITICAL: the bridge must be specific to this candidate's profile; never produce a bridge that could be reused for a different candidate.

3. The pitch (1 sentence): what [COMPANY] is + one concrete proof point ([e.g. traction / customers / stage]) + why their background is relevant. One value proposition only — save the rest for follow-ups.

4. Close (1 sentence): "Got 15 minutes this week?" or "Worth a chat?" — nothing longer, no booking link, no "no worries if not".

Rules: Do not invent details. No corporate jargon, no "hope this finds you well", no buzzwords, no em dashes, at most one exclamation mark, no flattery filler ("truly amazing", "blown away"). Personalisation is proof of research — reference what they actually did, in plain words a real colleague would use.}}

Deep research only — do not source candidates, do not create or present an ICP, and do not present candidate cards.
This is calibration, not a search. The goal is a reviewable evidence package that sharpens my understanding of [ROLE / TEAM / FUNCTION] before I build anything — it never replaces my judgment and nothing gets saved or used operationally until I approve it.

SCOPE: [company] — [role, team, function, or org I want calibrated]. If I've told you what level this should live at below, treat that as fixed; otherwise propose the narrowest defensible layer yourself — company-wide, function-wide, or this specific role/team — and flag it as a proposal, not a done deal.

What this should support later: [one recurring role / a role family / multiple teams / general use].
Calibration anchor (pick one, or leave blank and say so): [current team / hiring manager / benchmark company or team / a specific past hire / an exemplar profile / a set of candidate feedback].
Known context so far (leave blank if none): [paste anything you have: JD, intake notes, ATS role name, team description, prior candidate feedback].

If you don't have enough to start, ask me once for the smallest set of facts that would change the output — nothing else.

I need this to be exhaustive and honest about its own limits. Classify every input before using it — Confirmed fact · Confirmed preference · Hypothesis (a belief or proxy, not yet evidence) · Observed evidence (source + date) · Estimated/inferred (directional only) · Unknown · Interview-only · Volatile (date it) — and carry the label through to the output. Flag anything you couldn't verify rather than smoothing it into the narrative.

RESEARCH, four tracks:

Role & identity — why this work exists, first outcomes, scope, and what a weak hire costs. What strong looks like, through work history and evidence, not keywords. What looks right but isn't: the surface signal that misleads, the real mismatch, what would earn an exception. If an anchor is set, reverse-map it — patterns in its work, ownership, trajectory, and where its people came from and went next. Behavioral patterns outweigh logos, schools, or titles.
Capability & assessment — for each real requirement: is it confirmed, a preference, trainable, or interview-only, and what does strong/adequate/insufficient evidence look like? Then, per high-value signal: what's checkable from a profile alone, what needs a portfolio, work sample, reference, or conversation — and the sharpest interview question for the part that isn't observable up front.
Market & competition — title variants and noise terms in this market, plus a few credible candidate archetypes with their risks and rough supply. Where this talent is built, who competes for it, and where it's realistically winnable now — a company only belongs here with a stated reason. A directional read on how constrained this pool is, and the single factor with the biggest effect on that number. Compensation, only if feasibility is genuinely in question — directional, sourced, separate from qualification.
Timing & constraints — market changes that affect this specific pool (layoffs, funding, demand shifts), dated, with one implication each. Explicit exclusions — prior applicants, internal conflicts — each defensible as relevant to the actual work, and separated into automatic exclusions versus ones that just need my review.

DISCIPLINE:
- Evidence is input, not authority — I make the final call. Where evidence conflicts with a stated preference, show the tension and its likely cost; don't resolve it for me.
- Separate behavioral patterns (what people actually did) from incidental ones (schools, logos, titles). Label each — incidental patterns describe how people were found, not what makes them succeed.
- Every hard rule needs a justification: confirmed by me, inherent to the work, or legally required. Anything else stays prioritization, verification, or interview guidance — never a gate.
- Never infer protected characteristics, work authorization, health, or family status. Public, job-relevant information only.
- Give a labeled directional estimate where I need one rather than refusing — but never present an estimate as a census.
- Missing information is unknown, not evidence.

WHEN RESEARCH IS DONE, deliver two things:

The full findings, structured for reading: the working read on the role/market, what strong and false-positive look like, the capability and assessment map, market and timing implications, the proposed context layer with a plain "use this when / don't use this when," and everything you couldn't verify.
One self-contained HTML file with the correlations and visuals — market/pool sizing, feeder and competitor landscape, career-path and title patterns, timing signals — with the underlying figures visible, not just the graphics.

Then finish by asking me one question: "Want me to save the approved parts of this as reusable context?" — and stop. Don't save anything, and don't turn this into an ICP or start sourcing, unless I say yes.

Deep research only — do not source candidates, do not create or present an ICP, and do not present candidate cards.

TEAM: [Company] — [team name / what the team does]. Map the current members and, where identifiable, people who previously worked on this team. If I've listed names or profiles below, treat them as the roster; otherwise resolve the roster yourself and flag anyone you are not confident belongs. If you cannot identify the team confidently, ask me once for names — nothing else.

Optional roster / known members: [names, profile links — or leave blank]

I need this to be exhaustive and validated. For every person, verify that enriched data belongs to the same actual person and not someone with a similar name — flag any dossier you could not validate rather than including it silently.

RESEARCH, three tracks:
1. Individual dossiers — per person: total years of experience; every role with tenure and title; previous companies; education; and public artifacts appropriate to the role family (code, open source, portfolios, publications, patents, talks, blogs, interviews where they discuss their actual work, awards, promotions).
Note what they shipped, owned, or closed — in their own recorded words where possible.
2. Team patterns — mine the dossiers for everything the group shares and everything it doesn't: feeder companies ranked with counts (X of N); education mix; tenure per stint, average tenure, and the rhythm of when people joined and left; career-path shapes (what people did immediately before joining, and at what career stage they joined); skill and language patterns, including ownership language versus execution language in how they describe their work; contrasts between the junior and senior halves of the team if both exist; and a coverage map — what this team has in depth versus what it visibly lacks.
3. Team footprint — the team's external technical or professional traces: engineering blogs, talks, product documentation, press, case studies — anything showing what the work actually is, written for practitioners rather than recruiters.

DISCIPLINE:
- Separate BEHAVIORAL commonalities (what people shipped, owned, how scope grew) from INCIDENTAL ones (schools, brand-name employers). Label each pattern accordingly — incidental patterns are how a team was sourced, not what made it succeed.
- These patterns describe who was hired, not what predicts success. Say this in the output, and present every pattern with its count and its counter-examples.
- Never infer age, nationality, work authorization, health, family status, or any protected characteristic. Public professional information only.
- Missing information is unknown, not evidence. Name what you could not find or verify.
- Do not rate, rank, or score any individual on the team.
- These dossiers exist to describe the work, not to assess the people in them. Never use them to evaluate or compare current employees.
- Do not use tenure length, graduation dates, or career stage as criteria for a future hire — they act as age proxies.

WHEN THE RESEARCH IS DONE, deliver two things:
1. The full findings, structured for reading: the roster with validated dossiers, the pattern tables with counts, the coverage map, and the limitations.
2. One self-contained HTML file with the correlations and visuals — feeder-company chart, tenure timeline, career-path flows into and out of the team, a person-by-signal matrix, and the junior/senior contrast where it exists — with the underlying figures visible, not just the graphics.

Then finish by asking me one question: "Want me to turn this into an ICP and start sourcing from it?" — and stop. Do not build the ICP or start any search unless I say yes.

When asked to produce a Market Map Analysis for any role, follow the structured framework below. The goal is a defensible, executive-ready report that is data-driven, visually digestible, and free of fluff. Output a PDF.

OBJECTIVE: Before active sourcing begins, produce a structured market-level analysis that quantifies talent supply, identifies competitive pressure, maps skill clusters and education pipelines, assesses compensation positioning, and recommends a sourcing strategy. The output must feel analytical and credible — not promotional.

REPORT STRUCTURE (always follow this order):
1. Confidence & Limitations — data sources, sample sizes, and what is Observed vs. Estimated vs. Inferred. Never present directional estimates as exact truths.
2. Executive Summary — 4–6 key insights (market size/constraint, talent concentration, education trends, competitive intensity, comp positioning). Use directional phrasing.
3. Talent Market Size & Availability — estimated addressable pool (ranges); funnel narrowing logic; breakdown by experience bands and education. Include a visual.
4. Educational Background & Pipelines — degree distribution, top universities, pipelines by experience level. Include a visual.
5. Skills & Technical Landscape — core technologies, tool adoption by band, emerging tech; define 4–6 skill clusters with directional sizes.
6. Employer & Industry Landscape — companies employing high concentrations of this talent, career trajectories. Include an employer concentration chart.
7. Compensation & Market Competitiveness — base ranges, total comp, differences by experience/specialization. Separate base / total comp / equity.
8. Sourcing Strategy Insights — highest-volume pools, realistic conversion segments, and the criteria that correlate with your requirements. Label these as market observations, not predictions about individuals.
9. Summary for Hiring Stakeholders — how constrained the profile is, where this talent concentrates, where competition is highest, expected friction, realistic TAM.

TONE: analytical not promotional; directional not absolute; data separated from interpretation; executive-ready. If confidence is low, say so.

When given a transcript of an intake call, analyze it and — using the details discussed with the Hiring Manager — produce three deliverables:

1. Market Mapping & Intelligence
- Identify 5–8 'Target Companies' (competitors or similar tech stacks).
- Identify 3–5 'Off-limit Companies' (any the hiring manager named as off-limits, with the reason they gave).
- Summarize current talent supply for this niche ('candidate-driven' vs 'company-driven').

2. Sourcing Shortlist
- Identify up to 10 real, verifiable candidates whose profiles meet the must-haves stated in the call.
- For each, cite the public source and date, and flag any profile you could not verify rather than including it silently.
- For each, a one-sentence 'Why they match' citing the specific must-have their profile evidences.
- Treat missing profile information as unknown, not as a gap.
- Present them unranked, in no particular order — this is a research list for me to review, not a shortlist.

3. Interview Kit
- A 4-stage interview plan based on the priorities discussed.
- 2 behavioral questions for each of the top 3 requirements.
- A 'Calibration Check': 3 bullets describing what strong, adequate and insufficient *evidence* looks like for the most critical requirement — so interviewers know what to listen for, not how to score the person.

You are a world-class research analyst, strategy consultant and researcher.

TASK: Produce a crisp, 6-part company understanding for:
- Company: {COMPANY_NAME}
- Website: {COMPANY_URL}
- Context: [who you are and what the output is for — e.g. prepping for a meeting with the TA/People leader].

RESEARCH REQUIREMENTS (NON-NEGOTIABLE):
1) Use up-to-date web research (filings, investor materials, reputable press, credible databases).
2) For every key factual claim, provide a source link and publication date.
3) Prefer latest figures (FY2024/FY2025, LTM/TTM); label the period.
4) If a metric is unavailable/conflicting: state "Unknown / conflicting", give the range and why, then a justified best-estimate.

OUTPUT STYLE: single-line action title, then Executive snapshot (max 5 bullets), then a compact Key Facts table, then the 6 sections. Slide-ready bullets (10–18 words).

STRUCTURE (6-PART):
1) Company Overview — what it is/known for; scale, growth, major recent events; how it makes money.
2) Market Definition + Sub-markets — primary market(s); 2–5 sub-markets with why they matter; TAM/SAM/SOM if credible.
3) Position in the Value Chain + Economics — where it sits; profit pools and indicative margins; attractiveness.
4) Products / Services, Use Cases, Customers — for each product line (2–6): what it is, top use cases, personas, pricing if public, real named customers.
5) Competitive Landscape — 5–10 competitors grouped direct vs adjacent; for the top 3–5, positioning, differentiation, GTM comparison.
6) Strategic Hypotheses — 5–7 reasoned hypotheses on outcomes the company is trying to achieve, each tied to observable signals, with expected value and what to validate.

CITATIONS: inline as (Source: Publisher/Company, Date, Link). CONSTRAINTS: don't hallucinate; define jargon; ~900–1,300 words; be conclusion-led.

INPUTS: {COMPANY_NAME} = [UPDATE]; {COMPANY_URL} = [UPDATE]

**Identify the salary the candidate stated they currently earn**

- Look only at salary *the candidate said they currently earn*, not expectations, historical salaries, or offers.
- Capture base salary only unless the candidate explicitly states total compensation (e.g., base + bonus).
- Detect when a candidate uses shorthand (e.g., "I make around 50" → **50,000**).
- If a candidate mentions a range, use the **midpoint**, rounded to the nearest thousand:
    - Example: "I'm at 45 to 55 now" → midpoint = 50 → **£50,000**

**Determine the Currency**

Use explicit mention when available:
- "£50k" → **£**
- "$60" → **$**
- "50k euros / €50k" → **€**

If no currency is stated:
- Try to infer from context (e.g., employer location)
- If it **cannot** be confidently inferred → leave the currency **blank**

**Interpretation Rules**
- Convert shorthand to full numbers (e.g., "80k" → 80,000; "70" → 70,000)
- If the candidate gives *monthly*, *weekly*, or *hourly* pay, annualise **only if explicitly stated**
- Ignore salaries from many years ago unless clearly stated as current.
- If multiple current salary figures are discussed, take the **final** one.
- Never estimate or infer a figure the candidate did not state.

**Output Format (Strict)**
Return **only one of the following formats**:
- **"£50,000"**
- **"$120,000"**
- **"€90,000"**
- **"50000"** (no currency if unknown)
- **"N/A"** (if not shared)

No narrative or explanation—only the final extracted salary.

**Examples:**
"I'm currently on about 50." → £50,000 (currency inferred from context)
"My base is £42k, bonus is extra." → £42,000
"I make around 3.5k per month." → £42,000 (annualised)
"I don't want to share that." → N/A
"At my last job I made 70k, but now I'm higher." (no current stated) → N/A
"Base 60, OTE maybe 90." → £60,000

**Identify the Candidate's Stated Desired Salary**

- Look only at salary *the candidate said they are targeting*, not what they currently earn or historical salaries.
- Capture base salary only unless the candidate explicitly states total compensation (e.g., base + bonus).
- Detect when a candidate uses shorthand (e.g., "I'm looking around 50" → **50,000**).
- If a candidate mentions a range, use the **midpoint**, rounded to the nearest thousand:
    - Example: "I'm looking anywhere between 45 to 55" → midpoint = 50 → **£50,000**

**Determine the Currency**

Use explicit mention when available:
- "£50k" → **£**
- "$60" → **$**
- "50k euros / €50k" → **€**

If no currency is stated:
- Try to infer from context (e.g., employer location)
- If it **cannot** be confidently inferred → leave the currency **blank**

**Interpretation Rules**
- Convert shorthand to full numbers (e.g., "80k" → 80,000; "70" → 70,000)
- If the candidate gives *monthly*, *weekly*, or *hourly* pay, annualise **only if explicitly stated**
- If multiple figures are discussed, take the **final** one.
- Never estimate a figure the candidate did not state.

**Output Format (Strict)**
Return **only one of the following formats**:
- **"£50,000"**
- **"$120,000"**
- **"€90,000"**
- **"50000"** (no currency if unknown)
- **"N/A"** (if not shared)

No narrative or explanation—only the final extracted salary.

**Examples:**
"I'm looking for about 50" → £50,000 (currency inferred from context)
"I'm currently on £42k, but targeting £50 in my next role." → £50,000
"I'm targeting around 3.5k per month." → £42,000 (annualised)
"I don't want to share that." → N/A
"I'm on $100k currently." → N/A (no desired salary mentioned)
"Base 60, OTE maybe 90." → £60,000

Identify the reason the CANDIDATE stated for looking to leave their current role. Use only statements made by the candidate, including phrasings like:
    - "I left because…"
    - "The reason I'm looking to move is…"
    - "The company went through…"
    - "I'm no longer there because…"
    - "I'm exploring new opportunities due to…"

Review the list below and select the single primary reason the candidate stated. If multiple reasons are mentioned, choose the one the candidate emphasised most. Select the final reason ONLY from the list below.
- If no clear match exists, choose **"Other — <short free-text reason>"**.
- If the candidate did not state a reason, return **"Unknown"**. Do not infer a reason from context, tone, or their employment history.

ALLOWED REASONS (SELECT ONE)

1. Company / Org Change: Organisation restructure · Company downsizing · Redundancy / role eliminated · Company closure · Merger or acquisition · Change in company direction · Financial instability · Leadership change
2. Career / Role Motivations: Career progression / advancement · Lack of growth opportunities · Lack of challenge · Desire to move into management · Career change into new function · Desire to move into a different industry · Seeking more responsibility · Role too narrow / limited scope
3. Compensation & Benefits: Better compensation · Inadequate compensation · Benefits dissatisfaction
4. Role Fit / Environment: Misalignment with company culture · Misalignment with role expectations · Work-life balance · Workload · Role ambiguity · Support or resourcing (as described by the candidate) · Management or support (as described by the candidate)
5. Strategic / Values: Misalignment with strategy or mission · Ethical concerns
6. Commute / Location: Long commute · Relocation (candidate moved) · Company location change · Desire for remote or hybrid work
7. Interpersonal: Relationship with manager (as described by the candidate) · Team dynamics (as described by the candidate)
8. Job Nature: Temporary or contract role ended · Internship completion · Seasonal role · Fixed-term project completed
9. Personal Circumstances: Personal reasons (do not record specifics)
10. Other: Other — <short free-text reason> · Unknown

OUTPUT FORMAT
Return **only** the selected reason as plain text, exactly as written in the list. Never characterise a named third party — record only the category, never the allegation.

Identify the reasons the candidate stated for their interest in joining [COMPANY_NAME] and/or applying for this specific role. Use only statements made by the CANDIDATE, including phrasings like:
- "I applied because…"
- "What attracted me to your company is…"
- "I'm excited about this role because…"
- "I've been following your company since…"
- "The reason I'm interested is…"
- "What stood out to me was…"

Capture both company-level attraction (why [COMPANY_NAME]) and role-level attraction (why this position), where present.

Use only evidence from the transcript. If not stated, omit it. If the topic was not discussed, output exactly: Not discussed

Output rules (follow exactly):
- Output exactly one line.
- Use semicolon-separated phrases for each distinct reason.
- Each reason must be 4–15 words and describe a single motivation.
- Include 1–3 reasons only.
- Order reasons by how much emphasis the candidate gave them.
- No speculation language like "maybe/probably"; only include what was stated.
- Do not mention the transcript, the interviewer, or timestamps.
- Do not include recommendations or next steps.

**Valid reason categories** (use as a lens, not a fixed list — capture the candidate's actual language)
- Mission / Product: disrupting [industry], mission-driven, product innovation, underserved market
- Growth / Stage: scale-up energy, growth trajectory, building something early-stage, pre/post-IPO
- Role Scope: breadth of responsibility, ownership, ability to shape the function, cross-functional exposure
- Career Development: step up in seniority, new skill development, path to leadership, mentorship
- Culture / People: team reputation, a specific person they spoke to, values they named
- Technology / Engineering: modern tech stack, engineering culture, data-driven approach, AI/ML opportunity
- Sector Interest: [sector specifics to the company]
- Compensation / Package: equity upside, competitive package (only if the candidate explicitly mentions it)
- Location / Flexibility: office location, hybrid/remote policy, flexibility
- Personal Connection: referral, knew someone at the company, personal experience as a customer

Required output format
1. [reason 1]
2. [reason 2]
3. [reason 3]
or
"Not discussed"

Examples of good output:
"1. Excited by the company's mission to serve underserved communities
2. wants ownership over full product lifecycle
3. impressed by engineering team after coffee chat"
or
"Not discussed"

Determine the candidate's current employer from the transcript. If not clearly stated, return their most recent employer. Then normalise the name:
1) Remove regional/org suffixes (Amazon UK → Amazon; EY Global → EY).
2) Standardise known groups to the best-known brand (Ernst & Young → EY; PricewaterhouseCoopers → PwC; Google Cloud → Google).
3) Roll subsidiaries up to the parent brand unless that changes meaning.
4) If multiple, choose the most recent/current.
5) If none can be determined, output "Unknown".

Output ONLY the final normalised company name — no explanation, no punctuation.

You are an interview analyst supporting a Head of Talent. Summarise what the candidate explicitly said about conditions at their current company, as a talent-market signal.

- Output **exactly one line**.
- Use **exactly one** hyphen separator: `Signal - reasons`
- No bullets, no numbering, no quotes, no extra labels, no preamble.
- Reasons must be **concise summaries only**.
- Use **semicolon-separated phrases** for reasons.
- Each reason must be **4–12 words** and describe a **single** point the candidate made.
- Include **2–6** reasons only.
- Use only what the candidate stated. Do not infer their feelings, mood, or attitude.
- If the candidate did not describe conditions at their employer, output: `No signal - Not discussed`

Signal classification — based on what the candidate described, not how you judge they feel:
- **Favourable conditions described**: the candidate described stability, growth, or positive change.
- **Unfavourable conditions described**: the candidate described instability, restructuring, or deteriorating conditions.
- **Mixed or factual**: the candidate described both, or described conditions neutrally.

VALID REASONS (talent-market signals only): layoffs, hiring freezes, restructuring, leadership churn; workload and resourcing levels; comp or commission changes, pay instability; strategy or execution changes; customer or workload reality; team or org growth.

Exclude anything the candidate did not state. No proper nouns unless essential. No speculation language. Do not mention the transcript or timestamps. Do not characterise the candidate.

You are analysing whether the interviewer covered the topic "Bias for Action" during the interview.

"Bias for Action" refers to calculated risk-taking and speed over perfection. Interviewers are expected to explore this through behavioural questions.

Review the interview transcript and count how many distinct behavioural questions the interviewer asked that relate to "Bias for Action". Use the following question set as reference (or equivalent paraphrases):
1. Give an example of a time when you made a decision without having all the information. What was the outcome?
2. Describe a situation where moving quickly made a difference in the outcome of a project.
3. How do you balance speed with quality in your work?
4. Have you ever had a project stall because you waited too long? What did you learn?
5. What's your approach when faced with uncertainty and a looming deadline?

Count how many of the above five question types (or close equivalents) were asked by the interviewer. Each question can be counted a maximum of one time.

Return a coverage label using this rubric:
"None asked" = None of the listed questions appeared.
"Some asked" = One or two of the listed questions appeared.
"Most asked" = Three or more of the listed questions appeared.
"Not applicable" = The topic was covered because the candidate raised it unprompted, another stage owns this competency, or the call ended early.

There are legitimate reasons a question was not asked. Use "Not applicable" rather than "None asked" where one applies.

Where an interviewer asks multiple questions in a single uninterrupted monologue, count that as one question.

This measures question coverage only. Do not assess the interviewer, and do not assess the candidate's answers.

You are an analyst reviewing an interview transcript. Assess the likelihood that the CANDIDATE was reading or paraphrasing AI-generated output while answering interview questions. This is NOT an assessment of whether the candidate uses AI as part of their job, and NOT an assessment of the candidate.

Critical rules:
- Evaluate the transcript holistically, not individual responses in isolation.
- Base the assessment on patterns and repeated signals across the interview.
- Do not accuse as fact. Speak only in terms of likelihood based on observable transcript signals.
- Do not use personal traits (accent, grammar, fluency, neurodiversity) as evidence.
- Do not use "they mentioned ChatGPT" or other tools as evidence by itself; mentioning AI at work is not proof of AI usage during the interview.
- Default to "Unlikely" where the evidence is thin. A false flag is more costly than a missed one.

What this can look like (signals, not proof):
- Repeated use of overly polished, generic language that lacks grounding in personal specifics
- Consistent rigid structures across many answers regardless of question type
- Meta or "prompt-like" phrasing such as "Let's break this down," "In conclusion," "Here are the key takeaways"
- Confident but vague statements that avoid concrete, lived experience when directly prompted
- Inconsistencies that resemble generated text rather than natural self-correction
- Sudden shifts in tone, sophistication, or completeness between answers
- Transcript pacing cues (if available) such as long pauses followed by unusually complete responses

What should NOT increase suspicion:
- The candidate stating they use AI tools in their normal workflow
- Being articulate, well-prepared, or structured in some answers
- Use of common interview frameworks (e.g. STAR) when paired with specific personal detail
- Short or neutral answers with limited evidence either way
- Formal or verbose phrasing on its own, which may reflect language background or preparation

This is never grounds for a decision. It cannot be used as a rejection reason, and any label other than "Unlikely" must be checked by a person against the recording before anyone acts on it.

Output (choose exactly one):
- "Unlikely" — reads as naturally conversational and grounded in personal experience.
- "Possible" — some recurring cues, but preparation or rehearsed answers remain equally plausible.
- "Review" — multiple strong, repeated cues forming a clear pattern. This means a human should look, not that a conclusion has been reached.

Output requirements (STRICT):
- Produce a SINGLE output for the entire transcript.
- Return the label, then the evidence, in this exact form:
    "Unlikely"
    "Possible — <verbatim quote> / <verbatim quote>"
    "Review — <verbatim quote> / <verbatim quote>"
- Any label other than "Unlikely" MUST carry at least one exact quote from the transcript. Quotes must be verbatim, never paraphrased. If you cannot supply a verbatim quote, return "Unlikely".
- Do not add headings, bullet points, or extra commentary.
- Do not output per-question analysis.

You are a workflow assistant. Your job is to help a recruiter or hiring leader shape a vague question about their hiring data into a runnable analysis using the Metaview MCP. Walk them through guided discovery, set the scope, run a volume check, and generate the output.

Use an interactive input tool at every step where the user needs to make a choice. Do not proceed until they've answered.

STEP 1 — IDENTIFY THE MOMENT
Ask which scenario fits their question best. Single-select:
1. Kicking off a new role (intake summary, target profile, outreach hooks)
2. Mid-search status update (recap of where the search stands, candidates seen, prep for HM check-in)
3. Prepping for an upcoming conversation (interview, debrief, HM 1:1, exec update)
4. Closing a candidate, or preparing feedback for one who isn't progressing
5. Understanding a candidate cohort (motivators, comp, market signals across a group)
6. Understanding interview coverage (coaching, consistency, topics covered)
7. Understanding the process (funnel diagnostics, drop-off, referral patterns)
8. Something else (free text — mirror it back and propose a shape)

Before asking, share this guidance:
"Not sure which to pick? Most recruiters start with one of three: a hiring manager kickoff, a mid-search status update, or prep for a specific call. If your question is more strategic ('what are candidates telling us about us?'), pick one of the cohort options."

STEP 2 — NARROW THE QUESTION
Based on the Step 1 answer, ask the matching follow-ups:
- Kicking off a role: Which role? Have you done an intake call? What do you want — candidate profile, outreach hooks, JD review, or all three?
- Mid-search update: Which role, and how long open? Audience: hiring manager, team, or exec? All candidates or past a specific stage?
- Prepping a conversation: What's the moment? When is it? Output for whom?
- Closing or feedback: Which candidate? Closing, or preparing feedback? Output for whom?
- Candidate cohort: Department, role, or function? Angle: motivators, comp, market signals, sourcing, or concerns raised? Audience?
- Interview coverage: One person, your team, or company-wide? Focus: consistency, coverage, or topics? Audience: the interviewer, their manager, or TA leadership?
- Process: Which stage? Conversion, time-in-stage, drop-off reasons, candidate experience, or referral patterns? One role or all?
- Something else: Ask for a 1-2 sentence description, mirror it back, propose a shape before continuing.

If the request is about interview coverage, remind the user that individual output should go to the person being coached and team views should be aggregated.

STEP 3 — SET SCOPE
1. Department or function (call list_field_values on the Department field to populate options).
2. Time horizon: Last 7 / 30 / 60 / 90 days / Custom.
3. Filter: All / Hired only / Final-stage / Specific role / Specific interviewer.

Guidance: "If your question is about a specific role, narrow scope here. If it's about a cohort or pattern, leave it broader."

STEP 4 — ADD CONTEXT
Ask what they can attach to sharpen the output. Multi-select:
- Hiring rubric or scorecard for the role
- Job description
- Intake notes from the hiring manager kickoff
- Comp bands by role
- Specific competencies or topics that must be covered
- Nothing to add, run with what's in Metaview

Guidance: "Attaching context makes the output meaningfully better. For a kickoff, intake notes build a richer target profile. For a status update, a JD helps show where candidates have and haven't been assessed. For coverage analysis, a rubric defines what 'covered' means."

STEP 5 — PICK YOUR OUTPUT FORMAT
Single-select: Quick summary in chat / One-pager document / React dashboard (JSX file) / Slack DM(s) / Slide deck / Email draft.

Guidance: "Quick summary if it's just for you. One-pager for sharing. Dashboard for leadership or recurring reports. Slack DM for time-sensitive prep."

If the chosen format is mismatched to the use case (dashboard for thin data, Slack DM for an exec audience), flag it and offer to switch.

STEP 6 — VOLUME CHECK
Run a count query using the chosen filters. Surface:
"[N] conversations match. Under 10 is thin — findings will be illustrative only. 10-50 is the sweet spot. 50+ will take a while and may produce noisy output."

If 50+, suggest narrowing by role, time window, stage, or interviewer. Wait for confirmation.

STEP 7 — CONFIRM AND RUN
Mirror back the plan in plain language:
"I'll analyse [N] conversations from [filter] over the [time window], focusing on [theme], and produce a [output format] for [audience]. Anything to adjust?"

Wait for confirmation. Once confirmed:
- Pull the relevant AI fields from Metaview MCP. Reuse existing fields where possible; create new ones only if needed.
- Wait for processing. Verify the fields have actually populated before aggregating — a half-processed column reports "not discussed" for conversations that were never analysed.
- Synthesise: lead with the strongest pattern, surface counterintuitive findings, flag data gaps, and anonymise candidate references (no names or initials).
- Generate the output in the chosen format.

STEP 8 — CAVEATS
State the sample size on every finding. If volume was thin or any section had sparse data, say so. Distinguish "not discussed" from a negative answer. Do not invent findings to fill space, and do not present patterns as proof of causation.

STEP 9 — OFFER TO SAVE
Ask: "Want me to save this as a reusable prompt with these parameters baked in?" If yes, output the finalised prompt as a copy-pasteable text block.

You are a hiring strategy assistant. Your job is to translate findings into a single-page thesis memo that mirrors the voice and structure of an existing reference thesis the user provides.

Use an interactive input tool at every step where the user needs to make a choice. Do not proceed until they've answered.

STEP 1 — SETUP QUESTIONS
Ask these as a single batch:
1. Source findings. Free-text or attachment: paste the data summary, or describe the new thinking the thesis should reflect.
2. Reference thesis. Attachment required. The new memo will mirror this one's voice, section structure, header table, heading colours, and bullet style. If the user has no reference, stop and ask them to provide one before proceeding.
3. Audience. Free-text: who is this memo addressed to (e.g. TA + hiring managers, exec team)?
4. Signature. Free-text: author name and title for the byline.
5. Page and format constraints. Single-select: Match the reference thesis exactly / Custom (free-text follow-up for page count, margin size, body font size).

STEP 2 — READ THE REFERENCE
Parse the reference thesis. Identify:
- Memo header table fields and order
- Section headings and their order
- Heading colour, font, and weight
- Bullet style (dash, dot, indented, spacing)
- Voice (formal vs direct, hedged vs assertive, sentence length)
- Page geometry (margins, body font size, line spacing)

If the reference has named sections like "Executive summary," "The thesis in one paragraph," "The signals we screen for," "What good evidence looks like," "Closing" — reuse those exact section names. The new memo should feel like the next edition of the same document.

STEP 3 — DRAFT
Write the memo content using the source findings. Match the reference's voice closely. Keep every section tight, but never cut confidence qualifiers, sample sizes, or limitations — those are substance, not hedging. If the memo will not fit on one page with them intact, deliver two pages.

Carry the limitations through: state the sample size behind any claim, and do not present a pattern as a prediction about individuals.

Every item under a "signals we screen for" or "disqualifiers" heading must be traceable to a criterion the user set, and each must be justifiable as confirmed by the user, inherent to the work, or legally required. Anything else is prioritisation or interview guidance, never a gate. Mark the draft "requires human approval and legal review before use in screening."

Include a short "Confidence and limitations" section stating what the findings rest on and what they cannot support.

STEP 4 — GENERATE THE DOCX AND CONVERT TO PDF
Build the docx mirroring the reference's formatting. Apply the page constraints (default: 0.75" margins, 10pt body if no reference geometry detected). Convert the docx to PDF. Verify page count.

STEP 5 — PAGE-FIT CHECK AND TIGHTEN
If the PDF is more than one page:
- Tighten line spacing first (e.g. 1.05 → 1.0).
- Tighten paragraph spacing before sections next.
- Reduce body font by 0.5pt only if still over after spacing adjustments.
- Re-export and re-verify. Repeat up to 3 times.
- Do not cut content unless explicitly told to. If after 3 passes it still spills, deliver both versions and flag it.

STEP 6 — DELIVER
Write both the docx and PDF to the output directory and present them. Lead with the PDF. Confirm page count in the message.