Candidate Self-Audit
Anthropic AI-Collaboration Compliance
Self-assessment by Dr. Joshua T. Fraser against Anthropic's Candidate AI Guidance (updated July 10, 2025). Prepared as a transparency artifact and a pre-submission discipline.
The policy, in one sentence: Anthropic encourages AI collaboration for research, polish, and prep, and disallows it for first drafts, take-home assessments, and live interviews. Core expectations: use Claude thoughtfully, be yourself, be transparent.
What I have followed
FollowedTransparency about AI collaboration
The Writing DNA practice explicitly documents human-AI collaboration as a discipline. The Heartbeat protocol is public. My consulting work with FlowState Intelligence surfaces Claude use as a feature, not a hidden assist.
Evidence: flowstate.iaotgroup.co/admin/decisions; Heartbeat integration in the learning log.
FollowedReal experience in portfolio
The shipped products are live production systems with real users. Verifiable URLs across FlowState, Christina's Child Care, BHMC Live, the SIP Webapp, and the 369 Room. Claude cannot fabricate the commit history, the user sessions, or the operational deployments.
Evidence: six live, linkable production surfaces (FlowState, Christina's, SIP, BHMC Live, Intelligence Admin, the 369 Room) across four sectors of client delivery.
FollowedThoughtful, structured Claude use
My collaboration with Claude is not ad-hoc. It runs on documented protocols: the Heartbeat operating system, voice rules, a methodology-tracking engine that flags when a pattern repeats across three contexts, reflexive rubric-scoring for voice compliance, and framework embodiment checks before any new product build. These are methodologies, not prompts.
Evidence: ~/.claude/skills/heartbeat, ~/.claude/skills/spiral-research, reflexive rubric-scoring methodology candidate confirmed across three contexts.
FollowedClaude as collaborator, not replacement
Iron Rule #11 in the Heartbeat says it plainly: the human does the thinking, the system does the remembering. Capture mode asks questions and I answer. The insights in my learning log are mine; Claude writes the record.
Evidence: Heartbeat SKILL.md Iron Rules section, 40+ captured learning entries.
What I need to adjust before submitting to Anthropic
AdjustDraft first, refine with Claude second
This resume was assembled through a Claude-assisted synthesis of source materials (an older resume, the learning log, the framework registry). Technically that is refinement of existing work rather than AI writing a resume from scratch, but it is refinement-heavy. For an Anthropic application specifically, the intent of the policy asks for more of my own hand.
Action: Before submitting to Anthropic, handwrite a 2-paragraph why-Anthropic cover letter from scratch. Use the resume as factual scaffolding, but revise the summary paragraph and positioning statement until they sound like me.
AdjustTake-home assessments: Claude off unless permitted
My normal development environment is Claude Code. Building without it is slower than my natural shipping pace. If Anthropic issues a take-home coding challenge, the policy says complete it without Claude unless explicitly allowed.
Action: Read any take-home instructions carefully. If AI use is not explicitly permitted, complete the assessment without Claude. Budget 3-4x my normal shipping time. Do not negotiate the rule; follow it.
AdjustLive interviews: no Claude in the loop
My natural thinking mode involves talking through problems with Claude as a sounding board. During live interviews at Anthropic, that is not allowed.
Action: Practice verbal problem-solving alone for 30 minutes before each interview. Interview prep with Claude is explicitly allowed (practice answers, prepare questions) so rehearse extensively before the day. During the interview itself, no Claude.
AdjustAuthenticity read on the resume copy
The resume copy follows consulting-register house voice (zero em-dashes, zero filler, direct and specific). Anthropic reads thousands of Claude-assisted resumes. Authenticity signals cut through.
Action: Before the Anthropic submission, read the resume aloud once. Rewrite any phrase that does not sound like my natural speaking voice. Add one detail that only I would include: a specific story about a shipped product, a lesson from the field, a decision I made that was wrong.
Pre-submission checklist
- Read the resume aloud; rewrite any phrase that feels AI-assembled or that I would not say
- Handwrite a 2-paragraph why-Anthropic cover letter, unassisted, from scratch
- Use Claude only for polish on the cover letter, not for generation
- Confirm all portfolio URLs are live and reachable
- Check application instructions for take-home assessment; plan AI-off work session if present
- For interviews, practice verbal problem-solving unaided for 30 minutes before each session
- Add a sentence in the cover letter acknowledging use of Claude for refinement (transparency)
- Attach portfolio case study package as supporting artifact
Strategic positioning
Lead with Claude fluency as the strength
Anthropic states explicitly that they are looking for candidates who excel at collaborating with AI. My entire operating practice is structured Claude-human collaboration, and the artifacts prove it. The Heartbeat protocol, the methodology-tracking engine, and the reflexive rubric-scoring pattern are not prompts or outputs. They are reusable methodologies for doing this work well at scale.
The cover letter should lead with that frame. Two years of production systems built with Claude as primary collaborator. Documented protocols. Shipped products across four sectors. Then link the portfolio. The application does the opposite of most Anthropic applications: instead of claiming AI fluency, it shows the artifacts of sustained AI collaboration over time.