LEARNING PLAN/ Progress
Context Engineering
Use one original set of club documents to implement real token budgets, evidence selection, faithful compression, versions and validity periods, persistent memory, citation support, and failure evaluation.
Start with a question, derive the mechanism, then test your implementation with new inputs.
Times are author estimates that include derivation and practice. They have not been measured with learners; work through the sections at your own pace.
WHAT YOU WILL BUILD
What will you build to demonstrate understanding?
- Actual encoded and certified volume variable for complete wire containing instructions, sources, formats and text
- Separating source truth, budget selection, visible support and reality
- A state-of-the-state set-up for building version, entry into force, revocation, range memory and historical playback
- Run complete CLI from the new directory, problem-by-issue failure assessment and source list of evidence
What should you know before starting?
- Runs Python functions, dictionaries, strings and local files
- Understand limited aggregations, Boolean conditions and basic scores; words, versions, retrievals and indicators are progressively defined in their instructions
- Model API, visible cards or external weights are not required and the project is based on Python standard library only
Use the prerequisites to choose a starting point. You can read freely and return to earlier lessons whenever you need them.
THE LEARNING ROUTE
A clear path, one step at a time.
Saved in this browser only. Marking a lesson complete does not mean the project has passed verification.
Budget and truth
Select the evidence and examine the compression and regression under the full sequence count.
Real Token Budgets and Complete Packing Contracts
If the documents fit, does the complete context still fit after adding instructions, source metadata and an output reserve?
Selection and Truncation with True Incremental Costs
If the first evidence item is too long, should we stop, skip it or reconsider the remaining candidates?
Faithful Compression: Source Text, Qualifications and Fallbacks
If compression removes a qualification from a fact, how can we detect the loss and return to the original?
Version & Answer
To maintain historical status, integrate search and check visible support and conflict.
Versioned Memory: Activation, Revocation and Replay
A notice is published, takes effect and is later revoked. Which version should a query about a historical date see?
Retrieval and Fusion: Rank Evidence from Source Documents
When retrievers use different score scales, how can we merge rankings without treating rank values as answer confidence?
Grounded Answers: Visibility, Citations and Conflict Gates
If a fact exists in the database but is absent from the actual context, can the system use it to support an answer?
Errors and deliveries
Joint evaluation budgets, fields, references and re-opening.
Evaluation and Faults: Preserve Every Failure
Why can a cited answer within budget still omit required fields or become incorrect after a failure?
Complete Delivery: Run and Modify from a Clean Directory
After importing a new notice, how do we verify that budgets, versions, citations and replay after reopening remain consistent?
BUILD / TEST / TRANSFER
A replayable community information compiler.
Real file import, SQLite event, date view, freeze BPE, word combination, budget wrapping, reference extracts, conflict door, durability and 18 tasks joint evaluation; student package included.
Runtime Requirements
Python 3.9+ and the standard library only. No model download or API key is required.
Read the README after extracting the archive, then follow its environment and run instructions. Reference implementations and learning tasks are provided separately.
Project Checklist
This is a personal checklist, not an automatic grade. Keep your inputs, results, failures and reproduction commands.