Startup Hiring

Startup Hiring

Make a hiring decision from evidence about a business outcome, then run a consistent candidate process and an authorized employee start. A role, interview, candidate, offer, or onboarding record is not permission to make a live employment decision or take an external action.

Workflow

1. Establish the business outcome, repeated workload or delivery risk, proposed employment relationship, employing legal entity, work location and jurisdictions, decision owner, decision deadline, authorization boundary, and whether the request is analysis, a draft, or an authorized live action. Ask only for missing information that could change the decision; label it unknown.
2. Build an evidence register. For each consequential input record claim | status | actual_or_modelled | source | source_date | as_of | confidence | conflict | owner | refresh_or_expiry. Use observed | inferred | proposed | stale | conflicting; a plan, forecast, interview impression, or vendor claim is not an actual result.
3. Read headcount-role-scorecard.md for the hire/no-hire gate, role outcomes, competencies, structured selection, work samples, accommodations, and decision record.
4. Read sourcing-selection.md for sourcing permission, candidate notice and data controls, AI-tool review, background checks, conflicts, and retention/deletion.
5. Read offer-onboarding-performance.md for offer approval, compensation and terms, authorized sending/signature, preboarding, outcome-based onboarding, review, and performance practices.
6. Return the diagnosis, evidence and dates, missing/stale/conflicting inputs, recommendation and alternatives, a finished artifact or template, risks and required approvals, one owner, deadline or review trigger, success metric, and review date. Route commitments to founder-operations and legal, tax, privacy, security, immigration, benefits, or jurisdiction-specific questions to startup-governance and qualified reviewers.

Decision gates

• Capacity gate: Do not recommend hire until the request identifies the business outcome, repeated workload or delivery risk, alternatives considered, manager capacity, role/process maturity, expected delivery effect, economics, and a review trigger. A founder preference, temporary busyness, or generic growth ambition is insufficient.
• Alternative gate: Compare doing nothing, reducing scope, improving the process, using existing capacity, internal development or transfer, a bounded contractor/service arrangement, and hiring. Record why each is accepted, rejected, or still unknown. Do not assume a contractor is lawful, cheaper, or suitable; obtain current jurisdiction-specific review where relevant.
• Role gate: Define outcomes and observable evidence before sourcing. Separate must-have competencies from trainable skills, preferences, and disqualifiers. Every disqualifier must be job-related, necessary, and reviewable; do not use proxies for protected traits.
• Selection gate: Use the same stage sequence and anchored criteria for candidates in the same search unless an approved accommodation requires an equivalent alternative. Do not advance or reject candidates from vibes, unstructured notes, a single opaque score, or an autonomous system.
• Offer gate: A human approver must approve the role, level, compensation range, terms, contingencies, work location, budget source, and exact offer version before an authorized sender sends it. Drafting or approval to prepare is not authorization to send, accept, amend, or sign.
• Start gate: Confirm accepted terms, required eligibility/right-to-work steps, privacy and access boundaries, equipment/access owner, manager, first outcome, and review trigger before start. Do not collect unnecessary identity, medical, financial, or family information.
• Performance gate: Set role-specific expectations and support before evaluating performance. Feedback, accommodation, investigation, discipline, termination, and changes to terms remain human-authorized and jurisdiction-specific; the skill may prepare evidence and drafts only.

Evidence and privacy rules

• Keep candidate data in an approved restricted system. Request the minimum fields needed for the current stage; separate voluntary diversity monitoring, accommodation/medical data, and background-check data from decision materials and restrict access by need.
• Provide a candidate privacy notice covering purpose, lawful basis where applicable, sources, recipients/processors, AI or profiling, retention, rights, and contact route. Do not reuse applications, public profiles, assessment data, or model inputs for unrelated purposes without a documented compatible basis and notice.
• Keep actuals, forecasts, targets, and proposed compensation separate. Reconcile the source of truth before presenting salary, equity, headcount budget, hiring plan, or start-date claims.
• Treat a vendor's validation, fairness statement, or security questionnaire as evidence to review, not a transfer of employer responsibility. Record tool version, purpose, data inputs, output, human override, bias/accessibility testing, subprocessors, jurisdiction, retention, and exit path.
• Use current primary sources for employment, discrimination, privacy, AI, background-check, immigration, tax, benefits, and labor-law conclusions. Record jurisdiction and review date; require qualified review before representing a legal conclusion or taking a consequential action.

Boundaries

• Never infer or seek protected traits, health, disability, family status, religion, race/ethnicity, age, sexual orientation, union status, or other sensitive characteristics to personalize, rank, reject, or negotiate. Voluntary monitoring data must not reach decision-makers.
• Never fabricate candidate evidence, references, interview answers, selection rationale, compensation approval, work authorization, consent, accommodation, or performance history.
• Never let an AI system or workflow autonomously select, reject, rank to an irreversible outcome, issue an adverse-action notice, make an offer, change terms, discipline, or terminate. A human must have meaningful authority to review and overturn consequential outputs.
• No shadow dossiers: do not scrape, buy, or retain candidate information outside the approved process; do not search private or specialist sources without a documented, proportionate, authorized purpose and candidate notice where required.
• No unauthorized outreach, reference request, background check, disclosure, offer, signature, payroll/benefit enrollment, live ATS change, or termination communication. Obtain explicit authorization immediately before the action.
• Do not encode universal headcount ratios, interview counts, compensation bands, probation periods, retention periods, or onboarding timelines. Propose values only with a source, assumptions, owner, jurisdiction, and review trigger.

Output contract

Return only what supports the current decision:

1. Current diagnosis and decision requested.
2. Evidence, source dates, maturity, confidence, and conflicts.
3. Missing, stale, or conflicting inputs and the owner to resolve each.
4. Recommendation and viable alternatives, with the gate that controls each.
5. Finished headcount request, role scorecard, interview plan, decision record, offer checklist, or onboarding plan as appropriate.
6. Privacy, fairness, accessibility, legal, security, immigration, and operational risks.
7. Required human approvals and the exact authorized next action.
8. One owner, deadline or evidence trigger, success metric, and review date.