# TIKKI.WIKI

> KI-lesbare Wissensdatenbank. Datenschutzgeprüfte vorhandene Fragen und Antworten werden zuerst in Originalsprache auffindbar gemacht. Quellen werden danach ergänzt, Übersetzungen erst nach abgeschlossener Quellenphase. Rohdaten und Uploads sind nicht öffentlich.

Canonical homepage: https://tikki.wiki/
Knowledge-area catalog: https://tikki.wiki/catalog.json
AI discovery manifest: https://tikki.wiki/.well-known/ai-discovery.json
OpenAPI specification: https://tikki.wiki/openapi.json
Human-readable knowledge hub: https://tikki.wiki/wissen/
XML sitemap index: https://tikki.wiki/sitemap.xml
Merchant product schema: https://tikki.wiki/schemas/merchant-product-feed.schema.json

## Public search API

Search endpoint: https://tikki.wiki/api/v1/search?q={query}
Index metadata: https://tikki.wiki/api/v1/index
Method: GET
Authentication: none
Index boundary: public records only; raw uploads, quarantine data and private server paths are excluded.
Required parameter: q (maximum 256 characters)
Optional parameters: area=gesundheit|oele, kind=qa, lang={language-code}, limit=1..50
Response schema: tikki.search-response.v1
Each result contains a published question and answer, its canonical human-readable `url`, its explicit privacy and verification fields, and every public source currently available. For privacy-screened existing health answers, `sources` may be empty while the slower source-enrichment phase is pending. A linked source is not automatically an independently verified claim. The complete raw snapshot is indexed separately for the private server-side pipeline and is never returned by this HTTP API.

## Human-readable public pages

Knowledge hub: https://tikki.wiki/wissen/
Health hub: https://tikki.wiki/wissen/gesundheit/
Stable record pattern: https://tikki.wiki/wissen/{area}/frage/{public-record-id}/{slug}/
Every record page contains the same complete public question, answer and verification boundary as the search API. Pages are server-rendered and do not require JavaScript.

## Optional human voice mode

- The homepage offers browser/OS speech input and speech output after explicit user activation.
- A final speech segment is submitted after a short pause; the best public answer can then be read aloud at a selectable speed.
- Microphone permission, audio processing and voice availability remain controlled by the visitor's browser and device. Text search stays available and no raw audio is published.

## Published knowledge areas

### Ätherische Öle

Canonical specialist publisher: https://essentialoils.academy/
Manifest: https://tikki.wiki/oele/
Source-linked Q&A: https://tikki.wiki/oele/qa/source-linked.jsonl
Knowledge index: https://tikki.wiki/oele/ai/index.json
Question discovery routes: https://tikki.wiki/oele/ai/topic-question-frontier.json
Knowledge chunks: https://tikki.wiki/oele/ai/chunks.jsonl
Source-linked chunks: https://tikki.wiki/oele/ai/source-linked-chunks.jsonl
Source registry: https://tikki.wiki/oele/ai/sources.jsonl
Evidence map: https://tikki.wiki/oele/ai/evidence-map.json
Knowledge graph: https://tikki.wiki/oele/ai/knowledge-graph.jsonld

### Gesundheit

Manifest: https://tikki.wiki/gesundheit/
Inventory: 50,000 existing questions; 49,999 have an existing answer; one question has no answer.
Immediate privacy-screened native search tier: 49,643 pairs (28,729 German; 20,914 English).
Quarantined from immediate search: 356 pairs after privacy, completeness, structure and native-language checks.
Sequential source-verification shard manifest for existing base Q&A: https://tikki.wiki/gesundheit/qa/prebuilt-source-and-quote-checked/manifest.json
Read machine source-and-quote-checked records from the JSONL shards listed in that manifest, in manifest order.
Existing source-declared Q&A corpus manifest: https://tikki.wiki/gesundheit/qa/prebuilt/manifest.json
Separate legacy document-pipeline Q&A (not progress in the serial base-pair chain): https://tikki.wiki/gesundheit/qa/source-linked.jsonl

The later source-verification chain takes exactly one existing question-answer
pair at a time. It does not generate or rewrite a question or answer. It
searches for high-authority sources, verifies retrieved supporting quotations
and then moves to the next existing pair. Translation remains gated until the
source phase is complete.

The prebuilt corpus is a separate publication tier. Every published row has a
resolvable source URL, but source identity and supporting quotation are marked
as not independently verified until the verification chain completes.

The immediate Health search tier contains each existing base question in its
native seed language (German or English) with the matching existing answer,
only after a deterministic privacy gate. This tier does not create or translate
text and is explicitly labelled
`existing_native_answer_privacy_screened_sources_pending` until sources are
added.

## Publication boundary

- Raw documents and private uploads are never public interfaces.
- Credentials and deterministic personal-data signals are quarantined locally.
- A deterministic privacy and format gate plus an independent full-corpus audit must pass before an existing native pair enters public search.
- An existing answer requires independently retrieved support before it enters the checked shard tier.
- The cited evidence quote must occur in the fetched source text.
- A final challenger must approve the complete answer before publication.
- Discovery question forms are retrieval routes, not proof of a claim.
- Health content is educational information, not medical advice.
- Merchant feeds are private uploads until schema, domain, rights and claim review pass. Uploaded JSON-LD is never published directly.