Tvijo wiki
Catalog

Learn Tvijo: six exercises from task to evidence

Original exercises for scoped tasks, budgets, source retrieval, review, publication and product measurement.

These are original Tvijo exercises. They use public or synthetic material and do not require importing another publisher's course. They are reading-and-practice lessons; automated enrolment, saved progress and certificates are not implemented here.

Technology map Β· Workspace Β· Knowledge search Β· Evaluation method

1. Define a task that can finish

Choose a small documentation change or a harmless local fixture. Write the desired outcome, allowed paths, forbidden actions, numerical budget and acceptance command. Save the initial file hashes. Success means another reviewer can decide whether the result passes without asking the author what was intended.

Deliverable: a task record and a passing/failing check. Do not use live trading, production keys or customer records as exercises.

2. Choose an entry point and a route

Separate editor file execution from an MCP consultation. Record requested model, provider route, admitted status, input/output bounds and failure behaviour. A model in OpenRouter's catalogue is not automatically a tested editor option.

Deliverable: a route card. With no known budget or unsupported output ceiling, the exercise ends with a preserved blocked receipt rather than an unbounded call.

3. Retrieve with sources and access scope

Use the public knowledge map to find a method such as ORBITA or agent evaluation. Open the source and record observation date, evidence level and reuse rights. URL labels in the external map are navigation metadata, not proof that a page's full contents were read.

For an internal project, put a synthetic document in tenant A and a different one in tenant B. Query A for B's unique marker. Passing means no B result, and a valid A result when asked for A. The server must establish tenant identity; a client-supplied string is insufficient authorization.

4. Reconcile cost and judge output

Compare the worker's narrative with the meter. Sum all attempts, retries and review calls. Divide by the number of accepted tasks; if none pass, cost per accepted task is undefined, not zero. Record subscription usage separately when no monetary allocation is available.

Deliverable: one result table with provider usage, held/pending spend, acceptance and uncertainty. A model's statement that it spent zero is not a receipt.

5. Turn evidence into a useful article

Write the problem, scope, method, measured result, source and limitation. Publish only approved original or licensed content. Link private evidence through an authorized service rather than exposing it on a public page. Add a useful title, one primary heading, canonical URL, language links and links to related work.

Deliverable: an anonymous reader can reach the article, source summary and relevant product in both languages. A successful HTTP response proves availability; Search Console and repeated citation checks answer different questions.

6. Test value with a real workflow

Choose one narrow studio or company task. Measure time to first accepted result, total cost, rework and the user's decision to continue. A synthetic inquiry tests plumbing; a paid settlement and repeat usage are stronger commercial evidence.

Deliverable: a case report with consent, aggregate results and explicit missing data. Use the pilot inquiry to request a defined scope, not a vague promise of autonomous work.

External reading, not a copied LMS

The ABAI Wiki provides a public taxonomy useful for identifying missing topics. Our map records links to docs, learn, journey and CAIO sections. Full-text reuse rights have not been established. External brands retain their names and ownership; listing them implies no affiliation.

Priority gaps for future Tvijo lessons: a measured three-minute onboarding attempt; retrieval evaluation with citations and tenant tests; model route admission; accounting reconciliation; rollback in isolated fixtures; original portal-go/HTMX and Django case walkthroughs; accessible bilingual publication; and a repeatable AI-citation panel. Each new lesson should require a tangible output and a clear check, not only a completion click.

Next product phase: authenticated progress and assessment records, lesson versioning and instructor review. Those features are proposed, not available in this article.

πŸ”— Article connections

Open the full knowledge graph β†—

πŸ’¬ Comments

Comments are public and moderated. Do not post personal data.
  • No comments yet. Be the first!

Next steps