The library

Start with the controls. Go in order.

203 free lessons. The whole bench, raw instrument file through published figure, plus building your own tools.

Tap one, or describe what you're stuck on below.

New here? Start with Run your controls and go in order. Jump anywhere once it clicks. The Builder track is its own path. Start it whenever you want AI building FOR you.

⬡ Ask Protocol for a lesson on anything

Describe a real thing you need to get done. Protocol builds the drill from your exact situation.

The lab work

Instrument files to figures to manuscripts. 19 areas · 112 lessons on the lab work, all free.

1Run your controls6 lessons›

Before capability, verification: how a model fails, what fluency hides, and the sixty-second check before any output gets used.

Your first control: a question you already know the answer to›Why a fluent answer is not a checked answer›The three failure shapes: fabrication, drift, and confident interpolation›Calibration - asking for uncertainty and getting something usable›When to distrust - novelty, specificity, numbers, citations›The 60-second check before you use any output›

2What you cannot paste6 lessons›

Unpublished data, sponsor material, human-subject data. What disclosure costs, and who to ask - your TTO, your sponsor, your IRB.

Unpublished data, and why disclosure is not undoable›Patent disclosure - what counts as publication before you file›CDAs, MTAs and sponsor agreements - whose data is in your file›Human-subject data, identifiers, and what de-identified actually requires›Local, cloud and institutional deployments - which one are you actually using›Your institution's data classification policy, read once and applied always›

3What the model can and cannot see6 lessons›

The instrument file no model can open, the converter it writes instead, and the reasoning slice with a control inside it.

The wall - why the model cannot open your instrument file›Move 1: the model writes the converter, it does not read the file›Move 2: the reasoning slice - shape, schema, sample, control›Move 3: anything done twice becomes a script›Metadata is the experiment - what conversion destroys›Using the format inventory - finding your format and your path›

4Spreadsheets and instrument exports6 lessons›

Real header rows, units hiding in row two, joins that lose rows, and the decimal comma that moves your decimal point.

The human-formatted spreadsheet problem›Finding the real header row, and the units row underneath it›Joins that lose rows, and how to prove they did not›Encoding, decimal commas, and silent order-of-magnitude errors›Excel's gene-symbol problem and other auto-coercions›From a formatted sheet to a tidy table you can trust›

5Cleaning and QC6 lessons›

Blanks and controls inside the slice, batch effects, replicate structure, and a filter log that reproduces cleaned data from raw.

What QC means when the model cannot see the sample›Blanks, controls and reference samples inside the slice›Outliers - detecting versus deciding, and who decides›Batch effects, plate position and run order›Replicate structure - technical, biological, and what the model assumes›The audit trail - recording every filter you applied›

6Statistics you can defend6 lessons›

The confidently wrong test and why it looks right. Design first, assumptions checked, and the decisions that are not the model's.

The confidently wrong test, and why it looks right›Stating your design before asking for a test›Assumptions - normality, independence, variance, and what breaks›Multiple comparisons, and the correction you did not know you needed›Power and effect size, before the experiment rather than after›What the model may not decide for you›

7Figures6 lessons›

Journal specs applied, palettes that survive colorblind readers, error bars you can name, and a figure script that regenerates from raw.

Journal figure specifications, read and applied›From a tidy table to a first plot›Colorblind-safe palettes, and why the default is not one›Error bars - which kind, and saying which›Panels, alignment, and the multi-panel figure›A figure script that regenerates from raw data›

8Your instrument, end to end6 lessons›

Pick your track - spectra, sequencing, expression, imaging, cytometry - and run one full pipeline with every decision logged.

Choosing your track - spectra, separations, omics, imaging, cytometry›Spectra and separations - NMR, MS and chromatography end to end›Sequencing and variants - FASTQ to a filtered call set›Expression and single cell - counts to clusters, decisions named›Imaging - Bio-Formats, metadata, and measurement you can defend›Cytometry - compensation, gating, and the readable workspace›

9From chat to script6 lessons›

A chat transcript is not a method section. Notebooks, versioning, environments, and analysis a colleague can reproduce.

Why the chat is not the record›Your first notebook, and when a notebook is the wrong answer›Version control without becoming a software engineer›Environments, and why the analysis broke six months later›The methods section written from the script›Handing the analysis to someone else, including future you›

10Literature at scale6 lessons›

Triage what to actually read, extract into a table you can query, and catch a summary of an abstract the model never read past.

Triage - which papers to actually read›Structured extraction into a table you can query›Systematic review support, and its limits›Catching a summary of an abstract the model never read past›A working bibliography that stays accurate›Staying current without drowning›

11Citations that survive6 lessons›

The invented reference and why it is so plausible. DOI verification, retraction checks, and a pass across a whole reference list.

The invented reference, and why it is so plausible›Verifying a DOI, a title and an author list›Retraction and correction checks before you cite›Quoting and paraphrasing without misattributing›Preprints, versions, and what you are actually citing›A verification pass you can run across a whole reference list›

12Methods at the bench6 lessons›

Adapt a published method to your reagents, find the substitution that changes the result, and write a protocol someone else can run.

Adapting a published method to your reagents and equipment›The substitution that changes the result›Troubleshooting a failed prep, systematically›Writing a protocol someone else can actually run›Scaling up and scaling down›Safety and waste, checked before the bench›

13Manuscripts and reviewer responses6 lessons›

The methods section as a reproducibility artifact, results without overclaiming, and a response letter that knows what to concede.

The methods section as a reproducibility artifact›Drafting results without overclaiming›Abstract, title, and fit to the journal›Reading a review honestly before responding›The response letter - what to concede and what to defend›Disclosing AI use, by journal policy›

14Grants, talks and posters6 lessons›

Specific aims, significance without inflation, the summary statement read honestly, and a poster people stop at.

Specific aims - the page everything hangs on›Significance and innovation, without inflation›Budget justification and the biosketch›Resubmission - reading the summary statement›The conference talk, and the three-minute version›A poster people stop at›

15Running a group6 lessons›

Onboarding a student in a week, SOPs the lab follows, ELN hygiene, and a lab AI policy you can actually enforce.

Onboarding a new student in a week rather than a term›SOPs the lab will actually follow›ELN hygiene and the searchable lab›Ordering, inventory, and the reagent that ran out›Recommendation letters and reviews, written honestly›Setting the lab's AI policy, and enforcing it›

16The working week7 lessons›

Audit your week, wrap the day into the notebook, cost a scripted run, and write a core-facility request line that hands off what it must not answer.

Audit your bench week before you hand anything over›Wrap up the day into the notebook›Know what a scripted run costs before you start it›The core-facility request line›The PI meeting, rehearsed before the real one›Audit an answer before it enters the record›What the funder and the data agreement already decided›

17Compliance & safe AI use (free)7 lessons›

Grounding, unpublished data, and what never gets pasted. The skeleton, practiced to reflex.

Show it one methods paragraph that passed review›What lab data is safe to paste›Build your lab's grounding master prompt›Delegate a full pre-submission review›Make AI defer to your data, not its guess›Force the answer into a decision matrix›Convene an AI board of advisors›

18Workflows & automation5 lessons›

Turn a procedure you run every week into a workflow. With your check exactly where it belongs. Nothing leaves the bench unread.

Map a repeatable task into a workflow›Design the human-in-the-loop approval gate›Spec the follow-up cadence›Automate the pre-submission checklist›Set the control dial: where a human stays in the loop›

19Claude & Cowork mastery3 lessons›

One platform, mapped to the bench: a Project holding your standing method context, a Cowork skill for the analysis you repeat, MCP connectors for the live inbox and calendar.

Give Claude your bench once›Turn your best prompt into a skill›Let Claude read your inbox and calendar›

Builder track

Stop using AI. Start building with it. Claude Code, skills, agents, and systems that run without you. 14 areas · 91 lessons, capability-first, all free.

B1Setup & first contactInstall Claude Code, meet the terminal, and have your first real conversation about the analysis on your bench.›

Four words before you startBuild →Pick a plan & installBuild →Terminal tour: read your dashboardBuild →IDE setup: give it a windowBuild →Your first real task, iteratedBuild →Your files as contextBuild →Talk to it: voice inputBuild →

B2The project brainCLAUDE.md sets the trajectory: your instruments, your file formats, your lab's traceability and disclosure rules, baked in.›

CLAUDE.md: set the trajectoryBuild →/init on a real folderBuild →The 200–500 line ruleBuild →Split into rules/Build →Auto-memoryBuild →Secrets & .env hygieneBuild →

B3Steering & safetyA minute of planning saves ten of building, and it is the whole difference where unpublished data is involved.›

Permission modesBuild →Grill, then planBuild →/permissions rulesBuild →Bypass mode, safelyBuild →Cost capsBuild →The human-in-the-loop matrixBuild →

B4Context engineeringKnow what fills the window before you paste a 40,000-row export into it.›

/context anatomyBuild →Handoff, clear, compactBuild →Token densityBuild →Specific beats vagueBuild →Pick model & effortBuild →The overhead auditBuild →

B5The verify loopTask -> do -> verify. On data, the verify step IS the analysis; skip it and you have published a guess.›

Task → do → verifyBuild →Screenshot loopsBuild →Tests as verificationBuild →Fresh-data testingBuild →Cold-start runsBuild →Self-annealingBuild →

B6Skills: package your processThe analyses you repeat (the tidy, the QC pass, the figure spec) become software the AI runs the same way every time.›

SOP → SKILL.mdBuild →Frontmatter anatomyBuild →Scripts as musiciansBuild →Iterate to 99%Build →Self-healing skillsBuild →Your skill libraryBuild →

B7Connect your tools (instrument exports, your ELN, your data)Connectors that let it read the export folder and the notebook, plus skills other people built for you: install, vet, audit.›

What MCP isBuild →Install & vet a serverBuild →Audit the token costBuild →Paste the API docs insteadBuild →MCP → skill conversionBuild →Build a tiny serverBuild →

B8The DO frameworkCode executes, the LLM judges. Compound error math explains why a long pipeline drifts, and why the count lives in the script, not the chat.›

Docs → chatbots → agentsBuild →The PTMRO loopBuild →Directive anatomyBuild →Code executes, the LLM judgesBuild →Compound-error mathBuild →MetadirectivesBuild →

B9Subagents & parallel workFan out when it pays: three normalisations run in parallel is a comparison; feeling productive is not.›

When to fan outBuild →Reviewer & QA agentsBuild →Summaries, not raw textBuild →Git worktreesBuild →3 approaches in parallelBuild →The 3–4 agent capBuild →

B10Design: figures & postersReference-driven figures and posters with a screenshot-compare loop, against a journal's own figure as the reference.›

The screenshot-compare loopBuild →Reference-driven designBuild →Component librariesBuild →Decks & one-pagersBuild →Brand consistencyBuild →The design QA criticBuild →

B11Lab & data agentsSpreadsheets, QC reports, and the weekly summary to your PI that produce themselves.›

Excel that explains itselfBuild →docx & pptx skillsBuild →Sheet-connected pipelinesBuild →The weekly market report factoryBuild →Ask-your-workbook agentBuild →Deliverable factoriesBuild →Build your market second brainBuild →

B12Ship a real toolWebhooks, schedules, observability: one durable tool for the lab, the export watcher that tidies as it lands, demoed.›

Deploy a static siteBuild →Wrap a skill in a URLBuild →WebhooksBuild →Cron in plain EnglishBuild →Secrets & observabilityBuild →Design pass: ten checks on a phoneBuild →Capstone: ship your bench toolBuild →One real build, start to finishBuild →

B15The workspaceThe folder around the file: what is protocol and what is this run's, what loads and what never does, and how to catch the file that is confidently out of date.›

Is a workspace worth building?Build →Rules and materialBuild →What never goes inBuild →The routing tableBuild →File it at creation, then write the short versionBuild →The file that is confidently wrongBuild →

B16The work folderFrom chat to a folder of your own files that one agent reads and writes. Keep a list of your corrections, turn it into a protocol skill, shape the folder around your runs, and hand it on.›

Which layer are you onBuild →Test your setup with one tiny jobBuild →Keep a list of your correctionsBuild →Turn corrections into a skillBuild →Start with the outcomeBuild →Shape the folder around the workBuild →Give each stage a fileBuild →One model, different jobsBuild →Keep it current and hand it onBuild →

Vendor notes, dated

Prices and plan names move. Each note says when it was looked up, whether a person has confirmed it, and the day it comes due. Past that day, treat it as a question.

Claude. Claude Pro is $20 a month billed monthly, or $17 a month billed annually. Researched Oct 6, 2026, not yet confirmed by a person, recheck by Dec 15, 2026.
ChatGPT. ChatGPT Plus is $20 a month. Researched Oct 6, 2026, not yet confirmed by a person, recheck by Dec 15, 2026.
Google Gemini. Google Gemini's paid consumer plan is Google AI Pro at $19.99 a month; Google AI Ultra starts at $99.99 a month. Researched Oct 6, 2026, not yet confirmed by a person, recheck by Dec 15, 2026.
Anthropic. Anthropic's consumer plans (Free, Pro, Max) may use chats to train models under a setting the user controls, with five-year retention when training is on and 30 days when it is off; Claude for Work, Education, Government and the API are excluded. [P] Researched Oct 6, 2026, not yet confirmed by a person, recheck by Dec 15, 2026.
Anthropic. Anthropic's usage policy, effective 2025-09-15, requires a qualified human reviewer and disclosure of AI use for housing, legal and financial decisions. [P] Researched Oct 6, 2026, not yet confirmed by a person, recheck by Dec 15, 2026.
OpenAI. OpenAI's consumer plans train on chats unless the Improve the model setting is turned off, and a Temporary Chat is not used for training; Enterprise, Business, Edu and API data are not used by default. [S] Researched Oct 6, 2026, not yet confirmed by a person, recheck by Dec 15, 2026.
Google Gemini. Google Gemini Apps keeps activity 18 months by default, keeps human-reviewed chats up to three years, and tells users not to enter confidential information. [P] Researched Oct 6, 2026, not yet confirmed by a person, recheck by Dec 15, 2026.
Microsoft Copilot. Microsoft Copilot on a personal account may use chats to train models unless the user opts out; Entra and Microsoft 365 work accounts are excluded. [S] Researched Oct 6, 2026, not yet confirmed by a person, recheck by Dec 15, 2026.
Benchling. Benchling's academic plan is free; its pricing page shows no paid prices and routes labs to a demo. Researched Oct 6, 2026, not yet confirmed by a person, recheck by Dec 15, 2026.
protocols.io. protocols.io's Open Research plan is free, with unlimited public protocols and two private ones. Researched Oct 6, 2026, not yet confirmed by a person, recheck by Dec 15, 2026.

Every area gives away its first moves free. Each track spins endless personalized lessons.