Learning system · Experiment 03
Browser onlySmall ideas become useful when they return at the right moment. Each visit draws a fresh mix and works through the full deck before starting over.
Preparing a new mix…
Working model
A problem can be expensive and remain tolerated. A compelling event connects the impact to a date or condition that changes the consequence of waiting.
Try it
For one live deal, write the consequence, owner, and date in a single sentence. Mark every unverified word.
Working model
A champion has influence, a personal reason to act, and a history of taking risk on your behalf. Access and action are stronger evidence than enthusiasm.
Try it
Ask for one reasonable action that requires influence, then watch what happens.
Working model
Good preparation produces informed hypotheses, not a longer questionnaire. A hypothesis gives the buyer something concrete to correct and lets the conversation find signal faster.
Try it
Bring three “we often see…” hypotheses to your next first call and retire any that do not survive contact.
Working model
A contact list is not a decision network. Useful multi-threading maps distinct roles, incentives, influence, and risks across the buying process.
Try it
Label each stakeholder by decision role and write the unique question only that person can answer.
Working model
A concession should purchase something that improves the deal: scope clarity, timing, term, referenceability, access, or reduced risk. Unilateral concessions teach the other side to wait.
Try it
Before the next negotiation, list three tradable items and the reciprocal value required for each.
Working model
It becomes mutual when the buyer uses it to manage their own decision and implementation—not when they politely accept the seller’s timeline.
Try it
Rewrite one milestone in customer language: their owner, decision, evidence, and consequence.
Working model
The browser checks local caches, a recursive resolver follows the DNS hierarchy to an authoritative answer, and the returned address lets the client begin a network connection.
Try it
Use dig on one domain and identify the recursive resolver, record type, TTL, and authoritative nameservers.
Working model
It authenticates the server, agrees on cryptographic parameters, and derives short-lived session keys before application data is exchanged.
Try it
Explain TLS to a buyer once in technical language and once as a trust-and-privacy outcome.
Working model
A reverse proxy can terminate connections, enforce security, cache content, balance traffic, hide origin details, and apply policy close to users.
Try it
Draw the request path for direct-to-origin traffic and proxied traffic. Label where policy can be applied.
Working model
An API is a behavioral contract: schemas, authentication, errors, rate limits, idempotency, versioning, and expectations about change all shape whether an integration is dependable.
Try it
Open one API reference and find how it documents errors, limits, and backward compatibility.
Working model
Observability is the ability to infer internal system state from outputs. Tools help, but useful telemetry, shared context, and questions you can answer are the real system.
Try it
Take one user complaint and name the logs, metrics, and traces that would narrow its cause.
Working model
Zero trust is an architecture for continuously evaluating identity, device, resource, and context while limiting access. It is not a product or a demand to distrust every request equally.
Try it
Pick one internal application and define its user, device, policy decision, enforcement point, and minimum access.
Working model
Working backward from failure reveals hidden dependencies, fragile assumptions, and behaviors to avoid. It complements—but does not replace—planning toward success.
Try it
Pre-mortem one important meeting: list three failure modes and one prevention for each.
Working model
Base rates anchor a forecast in how similar cases usually turn out. The inside view can then adjust that anchor with specific, defensible evidence.
Try it
Forecast one deal from stage conversion and cycle data before adding account-specific judgment.
Working model
First-order thinking asks whether an action works now. Second-order thinking asks what the action teaches, incentivizes, constrains, or compounds next.
Try it
For one tempting shortcut, complete: “And then what?” three times.
Working model
If no observation can count against a belief, it cannot guide learning. Naming disconfirming evidence turns an opinion into a testable model.
Try it
Write one strong belief and the smallest credible signal that would force you to update it.
Working model
A model compresses reality so you can act, but its categories are not the world itself. Trouble starts when evidence is forced to fit the model instead of updating it.
Try it
Find one field in your sales framework that looks complete but rests on weak evidence.
Working model
Outcomes are delayed and noisy. A tighter loop connects an action to a proximal signal, a review, and a deliberate adjustment while the context is still available.
Try it
Choose one weekly skill and define its attempt, observable signal, review time, and next adjustment.
Working model
Specific nouns and observable actions let a reader test the claim. Abstract language can hide missing thought behind a polished surface.
Try it
Replace three abstract nouns in your last note with a person, action, number, or example.
Working model
A strong paragraph advances one move: claim, evidence, complication, example, or consequence. Mixing several moves makes the reader organize the draft for you.
Try it
Label the job of every paragraph in a draft. Split or delete the ones with two labels or none.
Working model
Distance weakens your memory of intended meaning, so you can see what the words actually communicate. Even a small context switch can restore some reader perspective.
Try it
Before sending an important note, change medium or font size and read it aloud once.
Working model
Turning actions into nouns often removes the actor and bloats the sentence. A direct verb restores who did what.
Try it
Search a draft for words ending in “-tion” and convert three back into verbs.
Working model
Readers keep moving when the draft anticipates what they now need to know. Structure is the sequence of those questions, not a collection of headings.
Try it
Write the reader’s question beside each section and reorder any section that answers too early or late.
Working model
Writers use introductory sentences to discover what they mean. Once the point exists, the private warm-up can usually go.
Try it
Delete the first two sentences of a draft temporarily. See whether the third is the real opening.
Working model
Complementary skills multiply: technical fluency makes commercial judgment more credible, while writing makes both more visible and reusable.
Try it
Name three skills whose combination differentiates you and one project that demonstrates all three.
Working model
A useful artifact turns a claim about curiosity or ability into something inspectable. It also creates a tighter learning loop than private intention.
Try it
Publish one small tool or note that answers a real question you encountered this week.
Working model
Skills, relationships, reputation, savings, and shipped work can open paths without requiring an immediate commitment to one. Optionality is useful when paired with direction, not permanent indecision.
Try it
Compare two next steps by what each teaches, signals, and makes possible six months later.
Working model
Preparation lets you choose the clearest evidence, recover precise details, and distinguish your contribution from the team’s. It creates honesty and brevity, not canned delivery.
Try it
Build one story card with context, stakes, your decisions, result, and what changed afterward.
Working model
Look for urgency, budget ownership, switching friction, competitive structure, product truth, and whether customers can actually adopt what they buy.
Try it
Choose one target company and map its buyer, trigger, status quo, budget, and deployment risk.
Working model
The tasks that reliably create focus, curiosity, or momentum are behavioral evidence about fit. Energy is not the only criterion, but it is useful data.
Try it
For one week, note which activities leave you more alert afterward. Look for the repeated conditions.
Working model
Strong discovery connects the customer’s current state, desired future state, required capabilities, and measurable success criteria. Skipping a bucket leaves the business case or solution fit resting on assumptions.
Try it
Divide a page into the four buckets for one opportunity. Mark every statement that the customer has not directly verified.
Working model
Teaching earns attention with a differentiated insight, tailoring makes it matter to each stakeholder, and constructive control turns agreement into a navigable buying process.
Try it
For your next meeting, write one thing you will teach, one stakeholder-specific implication, and one next step you will confidently request.
Working model
A commercial insight changes how the customer understands an important problem or opportunity and leads naturally toward a capability the seller can credibly defend.
Try it
Write an insight as four lines: accepted belief, overlooked evidence, business implication, and the capability now required.
Working model
A reframe changes the customer’s understanding of the problem. The solution becomes meaningful only after the buyer sees why the old frame produces an incomplete or costly response.
Try it
Take one product pitch and remove the product. Rewrite it as a new way for the customer to understand the underlying problem.
Working model
Constructive control makes hard issues discussable, clarifies tradeoffs, and guides the next decision around customer value. Pressure demands seller timing without improving the customer’s decision.
Try it
Find one stalled next step. Draft a direct request that names the customer benefit, owner, date, and consequence of delay.
Working model
The framework tests Metrics, Economic Buyer, Decision Criteria, Decision Process, Paper Process, Identify Pain, Champion, and Competition. Its value comes from evidence, not completed labels.
Try it
Score each element in one opportunity as verified, inferred, or unknown. Choose the unknown that creates the greatest forecast risk.
Working model
A useful metric establishes a credible baseline, quantifies the desired change, and connects that change to a business outcome the customer values and owns.
Try it
Write one metric as baseline, target, time horizon, financial or operational value, and customer owner.
Working model
The economic buyer has final authority over the funds or business tradeoff required for the purchase. A senior stakeholder may influence the decision without owning that authority.
Try it
For one deal, write who can approve the spend, who can redirect it, and what evidence confirms those powers.
Working model
Decision criteria define what the customer will evaluate. The decision process defines how, when, and by whom the evaluation and approval will happen.
Try it
Make two columns for a live deal: evaluation requirements and decision steps. Move any item that is sitting in the wrong column.
Working model
A real paper process maps procurement, legal, privacy, security, finance, and signature requirements, including owners, dependencies, and realistic review time.
Try it
Work backward from a target signature date and add every approval, owner, handoff, and buffer the customer confirms.
Working model
Why change anything tests pain, why choose this approach tests differentiation, and why act now tests urgency. Weakness in any one can leave the status quo as the safest choice.
Try it
Answer all three for one opportunity in customer language. Underline every answer supported by direct evidence.
Working model
Net revenue retention shows how recurring revenue from an existing customer cohort changes after expansion, contraction, and churn, excluding revenue from new customers.
Try it
Start with a fictional $100 cohort, add $15 expansion, subtract $5 contraction and $10 churn, then calculate and interpret the NRR.
Working model
It estimates how many months of gross-margin-adjusted customer contribution are needed to recover acquisition cost. A longer period ties up more capital and raises execution risk.
Try it
Choose hypothetical CAC, monthly recurring revenue, and gross margin inputs. Calculate the payback period and name the assumption most likely to move it.
Working model
Bookings record contracted business, billings record amounts invoiced, and revenue is recognized as the company fulfills its obligations. The three can occur on different schedules.
Try it
Take a one-year prepaid contract and mark the booking date, invoice date, cash timing, and monthly revenue-recognition schedule.
Working model
An LLM learns statistical patterns in token sequences so it can predict and generate language from an input. Fluent output reflects learned patterns, not guaranteed factual understanding.
Try it
Explain next-token prediction with a three-word sentence, then name one reason that plausible continuation can still be wrong.
Working model
A tokenizer divides text into model-specific units that may be words, word fragments, punctuation, or symbols. Model limits and usage are measured in these units rather than ordinary word count.
Try it
Take a sentence with a long word, punctuation, and a number. Predict where a tokenizer might split it, then check your assumptions with any tokenizer.
Working model
Instructions, conversation history, retrieved material, tool results, and generated output all consume tokens in the finite context the model can consider at one time.
Try it
Budget a fictional context window across instructions, source material, examples, and answer space. Decide what you would compress first.
Working model
An operational prompt gives the model a clear objective, relevant context, constraints, output shape, and—when useful—examples or evaluation criteria.
Try it
Rewrite one vague request using five lines: objective, context, constraints, output format, and definition of a good result.
Working model
RAG finds relevant external information at request time and places it in the model’s context, helping answers use knowledge that may be private, detailed, or absent from training.
Try it
Sketch query, retrieval, context assembly, answer, and citation. Add one failure mode and one check at every step.
Working model
An agent combines a model with instructions, tools, state, and a feedback loop so it can choose and carry out multiple actions toward a goal.
Try it
Map one agent workflow as observe, decide, act, verify, and stop. Define the permission and failure boundary at each tool call.
Working model
Training adjusts model parameters using data and an optimization objective. Inference keeps those learned parameters fixed while producing an output for new input.
Try it
Classify data preparation, parameter updates, prompt processing, and response generation as training or inference, then compare their main costs.
Working model
A data lake stores large volumes of raw data for flexible future use. A data warehouse stores curated, structured data optimized for consistent analytics and reporting.
Try it
Place raw application logs and a finance KPI dashboard in the better starting system. Explain the schema, governance, and query tradeoff.
Working model
ETL transforms data before loading it into the destination. ELT loads first and uses the destination platform’s compute to transform the data there.
Try it
Draw both pipelines for the same source. Mark where raw data exists, where quality checks run, and which system pays the compute cost.
Working model
OLTP supports frequent, small operational transactions with fast writes and consistency. OLAP supports complex analytical queries over larger historical datasets.
Try it
Classify five actions from an online store as OLTP or OLAP and explain the latency and data-shape requirement behind each.
Working model
IaaS provides infrastructure you configure, PaaS provides a managed application platform, and SaaS provides a complete application operated by the vendor.
Try it
Draw the stack from facilities to application and mark who manages each layer under IaaS, PaaS, and SaaS.
Working model
A VPC is a logically isolated network in a public cloud where you control address ranges, subnets, routing, and network access rules.
Try it
Draw a minimal VPC with public and private subnets, an internet path, an application tier, and a database tier.
Working model
Subnets segment a network so routing, access, fault isolation, and address use can be managed separately for different workloads or trust zones.
Try it
Take a fictional application with web, application, and database tiers. Assign each tier a subnet and justify the routes it should have.
Working model
It distributes requests across healthy targets, allowing horizontal capacity while avoiding instances that fail health checks.
Try it
Draw three application instances behind a load balancer. Walk through what happens when one instance becomes slow but does not fully crash.
Working model
A CDN caches content at geographically distributed edge locations, reducing network distance and the work required from the origin server.
Try it
Trace one request through a cache hit and a cache miss. Label latency, origin work, TTL, and invalidation behavior.
Working model
The provider manages server provisioning, capacity, and much of the scaling. The team still owns code, data, permissions, observability, cost behavior, and application reliability.
Try it
Choose one bursty workload and one steady workload. Decide whether serverless fits each and name the operational tradeoff.
Working model
A container packages an application and its runtime dependencies for consistent execution while sharing the host operating-system kernel.
Try it
For one application, list what belongs in the image, what arrives at runtime, and what remains the host’s responsibility.
Working model
Kubernetes continuously works to match declared desired state by scheduling workloads and managing deployment, scaling, service discovery, configuration, and recovery.
Try it
Describe a desired state of three replicas, then walk through how the system should react when one container exits.
Working model
A CI/CD pipeline automates building, testing, and preparing or releasing software so changes move from version control to users quickly and repeatably.
Try it
Map commit, build, unit test, integration test, security check, deploy, and rollback for one small application.
Working model
Infrastructure as code makes system configuration versioned, reviewable, repeatable, testable, and automatable while reducing configuration drift.
Try it
Choose one manual infrastructure change and describe how a code review, automated plan, and rollback would alter its risk.
Working model
Confidentiality prevents unauthorized disclosure, integrity prevents unauthorized alteration, and availability keeps systems and data accessible when needed.
Try it
Take one incident and write its confidentiality, integrity, and availability impact separately. Do not assume all three are affected.
Working model
Authentication verifies identity, authorization determines permitted actions, and accounting records activity for audit and accountability.
Try it
Trace one user from sign-in to a sensitive action. Identify the evidence produced at each of the three stages.
Working model
Authentication establishes who or what an actor is. Authorization separately decides which resources and actions that authenticated actor may use.
Try it
Write one scenario where authentication succeeds but authorization correctly fails, then name the policy being enforced.
Working model
Access may be excessive in action, resource, scope, or duration. Least privilege limits all four to what the current task actually requires.
Try it
Take one broad administrator role and reduce its actions, resources, environment scope, and time window.
Working model
The attack surface is the full set of exposed entry points. An attack vector is a specific path or technique used to reach and exploit one of them.
Try it
Inventory five entry points for a web application, then pair each with one plausible attack vector and one control.
Working model
Encryption is reversible with the correct key to preserve confidentiality. Hashing produces a fixed-size digest designed to be one-way for integrity checks and comparisons.
Try it
Choose the right tool for storing a password verifier, protecting a message, and checking a downloaded file. Explain each choice.
Working model
PKI binds public keys to verified identities through digital certificates and trusted issuers, enabling authentication and encrypted communication at scale.
Try it
Trace a website certificate from subject to issuing authority to trusted root. Mark what the browser verifies at each step.
Working model
Untrusted content can instruct an AI system to ignore intended rules, reveal protected context, or misuse connected tools and data.
Try it
Threat-model a document-reading assistant. Separate data from instructions and add permission, output, and human-review boundaries.
Working model
OAuth delegates authorization to access resources. OpenID Connect adds an identity layer on top of OAuth so a client can authenticate the user.
Try it
Trace an access token and an identity token through a sign-in flow. Label their audience and the question each token answers.
Working model
SQL injection manipulates database queries through unsafe input. Cross-site scripting injects executable content that runs in another user’s browser.
Try it
Match parameterized queries and output encoding to the attack they primarily mitigate, then explain why input validation still helps both.
Working model
Expected value weighs every meaningful outcome by both its probability and impact. It exposes choices where a less likely result matters enough to change the decision.
Try it
Estimate three possible outcomes for one decision, assign rough probabilities and values, and compare the weighted result with your intuition.
Working model
People optimize the metric, often discovering ways to improve the number without improving the underlying outcome. The measure loses value when incentives detach it from reality.
Try it
Choose one team metric and list two ways it could improve while the real outcome worsens. Add a balancing measure.
Working model
In decision-oriented writing, the reader usually needs the recommendation, request, or result first. Evidence can then explain and qualify the conclusion without making the reader hunt for it.
Try it
Rewrite one status update so its first sentence contains the outcome, significance, and requested action.
Working model
A claim states the point, evidence makes it credible, and implication explains why the reader should care. Missing any one weakens the reasoning.
Try it
Take one paragraph and label its claim, evidence, and implication. Write the missing move if one is absent.
Working model
Deliberate practice isolates a specific subskill, works near the edge of current ability, produces timely feedback, and repeats with adjustment.
Try it
Choose one subskill you use at work and design a 20-minute drill with a scoring rule and immediate feedback.
Working model
An experience builds career capital when it creates rare skills, credible evidence, trusted relationships, or judgment that increases future leverage and choice.
Try it
Score one opportunity on skills, evidence, relationships, judgment, and optionality. Name what would make the experience compound.
Under the hood
The page uses a shuffle bag, not a fresh coin flip for every card. Once a card appears, it leaves the bag until the rest of that topic has had a turn.
The remaining order is stored in this browser. Refreshing the page advances the deck instead of accidentally serving the same favorites again.