From knowledge to practice: sources and learning evidence
Private reference retrieval, source versions and explicit read, practice and verification records.
🎯 For Developers, integrators and teams evaluating an AI pilot
- Choose a bounded task
- Distinguish reference material from measured product evidence
Status, 29 October 2026: internal reference retrieval works locally. This is not a cloud knowledge base connected to every client or a competency certificate.
Three kinds of knowledge
| Layer | Store | Does not establish |
|---|---|---|
| External references | URL, text, retrieval date, version and provenance | That advice is correct or implemented in Tvijo |
| Project evidence | Specifications, code, tests, release and runtime receipts | That a feature benefits every customer |
| Customer data | A separate boundary with access and retention rules | Permission to share with other customers or publish |
No customer data was loaded into this reference index. Retrieved text provides context, not authority to execute commands it contains.
What happened to the database
The supplied AI Wiki UA database contains 1,255 pages; stored word_count fields sum to 1,709,857. These describe the export, not verified facts or proven site coverage.
A separate local snapshot was created. Retrieval uses the project's existing SQLite FTS5 EvidenceIndex: 1,249 pages and 5,806 chunks were indexed. Six pages were excluded by the credential-pattern check. They may contain educational examples, but the check was not disabled. No indexed page truncation was recorded.
Results carry the URL, title, export date, source-text hash and line range in the normalized chunk. Context budgets use UTF-8 bytes, not model tokens. No embeddings, answer generation or paid calls were run. Lexical retrieval does not guarantee synonym matching or query translation.
Full external texts are not published through this wiki. Bulk redistribution rights have not been established. Public Tvijo pages provide our original method and source links. Indexing does not replace checking usage rights before commercial product integration.
Read, practiced, verified
Read is an explicit personal acknowledgment for a particular source version. Importing a page does not create it.
Practiced means an exercise was performed and an artifact supplied: a test, diff, query or report.
Verified records another named reviewer and review evidence. The local ledger treats these as operator attestations; it does not authenticate reviewer identity or award an independent certificate.
When source text changes, earlier progress is marked stale and retained in history. This currently works through a local CLI ledger. Personal progress buttons and account synchronization on the website are not implemented.
Practice by role
| Role | First exercise | Reviewable artifact |
|---|---|---|
| Developer | Complete a bounded task and inspect the diff | Changes, tests and a human explanation |
| QA | Test invalid input, refusal and recovery | Reproducible scenarios and outcomes |
| Integrator | Check permissions, company and read limits | Allowed and denied request cases |
| Manager | Compare the existing process with a pilot | Time, quality, workload and a decision |
| Analyst | Build a reproducible baseline | Data versions, SQL, time split and errors |
Learning stages may be labelled L0–L3: understanding, independent practice, integration and adoption leadership. This is our working rubric, not an official certification or proof of legal compliance.
Priorities for the Tvijo family
- A common task-and-evidence contract: source, version, outcome, limitations and access.
- A repeatable useful task through an existing extension or terminal, observing the first run.
- An agreed Odoo pilot tested against an authorized ERP environment.
- Shared client knowledge retrieval after access, relevance and source-use checks.
- Built-in TvijoCode profiles and new ERP adapters after demonstrated demand.
TvijoCode, the VS Code extension and terminal should share the contract while retaining separate interface acceptance. For the terminal route, record the harness, DeepSeek model, limits and verified provider route separately. Working retrieval does not establish that route works.
Continue: adoption paths · evidence scorecards · Odoo pilot.
Sources
Each source was opened by a human on the date shown. Product behaviour is described from this deployment, not from marketing material.
🔗 Article connections
Open the full knowledge graph ↗Next steps
Related reading
Evidence scorecards: compare outcomes fairly
A proposed evaluation method with explicit measurements, missing data and independence. No invented product scores.
Adoption and evidenceAdoption paths: roles, tasks and shared integrations
Choose a first pilot, distinguish adapters from products, and build partnerships around a measurable result.
💬 Comments