Instructions for any model preparing or executing this plan

Read the current week first, then this file, DECISIONS, GATES and the prior week's review/evidence. Inspect actual project repositories before assuming anything exists. Follow applicable workspace instructions.

The v5 master controls thesis, sequence, 10-hour capacity and gates. Its Parts 0, 3, 4, 8, 9, 11 and 12 provide operating rules, curriculum, prompts and evidence standards. Use v4 for retained offers, company map, culture filter and templates. Do not revive Vara, clinical-data projects, a fourth instrument, a new SaaS or full-course completion.

This folder is prepared planning material, not completed scientific work. Do not treat existence of a specification as evidence that Peter has the skill, produced the artifact or contacted anyone.

When asked to prepare Week N, complete its local preparation pack:

1. Inspect prerequisite artifacts and record what is present, missing, stale or inaccessible. State the smallest action that resolves each essential gap.
2. Fill the week commitment with proposed blocks and deliverable checklist; leave Peter's commitment/date unconfirmed until supplied. Do not invent a calendar start.
3. Create study-notes.md as a lesson pack: selected free source units, specific reading objective, plain-language explanation, worked example, exercise with separate solution, five quiz questions and grading rubric. Leave Peter's answers/completion blank.
4. Create the three week-specific artifacts named in its acceptance table at the requested maturity: specification, draft, runbook or measured result. During preparation, a result file contains schema and execution instructions with explicit pending status; never invented measurements.
5. Create founder-paper.md: one verified author paper, question/method/result digest, data workflow, three intelligent questions and one honest connection sentence. If the text cannot be accessed, mark the missing section rather than infer methods from the title.
6. Create branding-pack.md: Wednesday note (optional, 80–150 words), Sunday build log (150–250 words), concise X adaptation, figure/caption/alt-text brief, and that week's special publication asset. Drafts describe verified results or clearly marked slots for results not yet earned.
7. Create opportunity-actions.md: current qualified targets, reason for fit, source/date, ladder stage, message drafts or interview preparation, intended next action and a dated follow-up plan. Record messages as drafted until sent.
8. Update evidence.md with file paths, state and verification method. Finish with the next physical action and unresolved dependencies. Do not jump into later weeks.

Preparation quality requires complete local drafts, not an offer to prepare them later. When asked to execute a week, implement its scoped work and verify it; distinguish the learner's required practice/answers from model-produced assistance. The model must not mark Peter as having studied material on his behalf.

Weeks 1–20: Mon build 90 min; Tue biology 90; Wed build 90 including the last 10 for a note; Thu AI 60 plus founder paper 30; Fri build 90; Sat buffer 90; Sun review/publishing 60. Total 600 minutes. Setup Week 00 has no mandatory study syllabus.

Weeks 21–40: the more specific v5 Part 7 lightens formal study to Tuesday. Thursday remains a 90-minute block for applied scientific review, collaboration, hiring or selling, with a founder paper when relevant and feasible. On selling/submission/decision weeks with 'none' as study, the block serves that work. The weekly total does not increase.

Every block includes a two-minute start and three-minute close. Tuesday's indicative internal allocation is 10 recall + 35 selected source + 30 exercise + 10 teach-back + 5 log. Thursday's first hour is 5 recall + 20 selected source + 25 application + 5 teach-back + 5 log; the remaining 30 is the paper digest. These are ceilings, not extra assignments.

Long chapter/video lists in the source are menus within these limits. The preparing model must name the exact subsection/timestamps that fit, and place unselected material in the parking lot. Reading, coding exercises, content, calls and setup all consume the same budget. API/compute jobs may run unattended after a bounded smoke test, with cost/time limits; no implied extra human work.

Saturday first absorbs slipped critical tasks and any longer content draft. It becomes sandbox time only when committed work is complete. Sunday is 20 min build log, 5 checklist/evidence, 20 gap review, 15 next-week physical actions. Draft long articles during a designated build/buffer block, not a fictional extra Sunday session.

Peter owns learning, personal claims, career preferences, commitment, final scientific responsibility and external communications. Models prepare, implement and review within the requested scope. Independent humans perform independent annotation and scientific review. Partner labs own wet-lab feasibility/execution. Co-authors approve their contribution and final manuscript.

The current request authorizes local planning artifacts. It does not itself send messages, apply for jobs, publish content, register accounts, submit papers, sign agreements, spend money or issue invoices. For future execution, use the authorization actually given in that session; do not repeatedly ask for permission already granted. Keep external action drafts concrete and ready if authorization is still absent.

• For every dataset: source URL, access date, version/checksum, license/permissions, inclusion/exclusion rules, exact N, unit and label definitions.
• For every experiment: question, hypothesis or exploratory label, baseline, training/calibration/test split, seed/config/code version, command, output and limitations.
• Split by independent biological/experimental unit: batches, papers, or protein families/clusters as appropriate. Fit transforms and thresholds without held-out leakage; early predictions use only information available then.
• Report denominators, uncertainty method, failed runs and negative results. Never require the model to win or manufacture the master's example headline.
• Separate synthetic, simulated, public experimental and partner data. Keep original and normalized values; keep provenance to the exact supporting location.
• An LLM cannot act as the second independent human annotator. Use categorical agreement where appropriate and numeric agreement for continuous fields. Do not tune on the final test set.
• Public manuscripts, dataset DOIs, endorsement, journal review and real adoption are separate achievements. No fabricated citation, DOI, reviewer, co-author, testimonial, result, outreach receipt or payment.
• Defer clinical/patient data and production biological interventions. Document labdocs as a research demonstration; a hash chain or local deployment does not establish regulated suitability.
• Refer to SOURCES and verify current technical documentation, venue policies, vacancies and prices when needed. Source tables in weekly specs are inherited planning guidance, not fresh verification of each unit/link.

Use the career asset requirements. Translate executive experience into concrete scientific delivery strengths without claiming a research degree, lab skill or publication absent from evidence. Keep applied-scientist and scientific-AI-lead narratives distinct enough to match actual roles.

GridX weakness columns are private hypotheses. Verify companies and people before outreach; write about public archetypes, not named unverified deficiencies. Founder conversations, recruiter screens, applications and clients are separate funnels in OPPORTUNITIES.csv. Count sent, answered, booked and held separately.

The source prices are hypotheses: Loop Audit USD 2,500 seed / 6,000–10,000 established; pilot USD 12,000–30,000; portfolio package USD 12,000–18,000; fractional lead USD 4,000–8,000 per month. Validate scope, demand and capacity before presenting a quote. Measure time savings with a stated baseline; do not translate days saved into cash/runway without explicit assumptions.

Each week passes only against its specific artifact checks and gate, with links in evidence.md. Use states NOT_STARTED, PREPARED, IN_PROGRESS, BLOCKED, COMPLETE; external events additionally distinguish DRAFT, SENT/SUBMITTED, PUBLIC, ACCEPTED and PAID where applicable.

If a prerequisite is missing, complete independent preparation and record the blocked dependent work. Use the week's minimum action and v5 P14 after a lost week. Preserve original gate targets and log any reforecast/approved scope decision; never silently move the gate or call a fallback equivalent to an unmet milestone.

At handoff, update review.md, PROGRESS.md, LOG.md and TODAY.md. Explain measured outcome, verification, limitations and next action in plain language. Do not report an evidence level from intent.