Ottili LD3

LD3 topic expansion

How the topic expansion engine in Ottili LD3 turns a verified main topic, through a fact map, into citable content angles and a topical content cluster plan — Beta.

Overview

The topic expansion engine is the part of Ottili LD3 that grows a verified main topic into several related subtopics (content angles) and plans them as a topical content cluster. It builds on what source intake, verification and fact extraction have already produced — it does not invent material out of nowhere.

LD3 is currently in public Beta (BETA)*; topic expansion is available as part of it on the public LD3 surface (/ld3). The engine is deterministic and traceable: every proposed angle carries the facts it was derived from.

Where topic expansion stands

Capability	Status
Topic expansion (basic, TEE Basic)	Beta (BETA)
Fact map	Beta (BETA)
Angle detection (10 angle types)	Beta (BETA)
Duplicate check	Beta (BETA)
Content cluster plan	Beta (BETA)
TEE Pro add-on (advanced expansion)	Add-on within LD3 Beta

The flow

The engine follows a fixed chain — from the main topic to the cluster plan:

1. Main topic* — a detected and (after [LD3 topic verification](/docs/ld3-topic-verification)) checked topic.

2. Fact map* — LD3 aggregates the topic, confirmed facts, evidence, sources, claims and a coverage breakdown.

3. Angle detection* — viable angle types are determined from the fact map.

4. Subtopic proposals* — for each viable angle a concrete subtopic proposal is produced.

5. Duplicate check* — near-duplicate titles are filtered out.

6. Content cluster plan* — the remaining subtopics are assembled into a cluster with internal links.

7. Generation* — the main article plus the subtopic articles are generated.

Fact map

Before the engine detects angles, LD3 builds a fact map from the topic. It contains:

Topic* — the verified main-topic record.
Confirmed facts* — extracted, citable facts (types such as date, location, price, statistic, entity, event, quote, product, person, organisation).
Evidence* — the verification evidence, including source item IDs.
Sources* — the original sources the topic rests on.
Claims* — from the topic's fact sheet.
Coverage areas* — per area a status of well_covered or sparse plus a 0–1 score.

Areas without enough facts stay visible as sparse — the engine does not hide gaps, it makes them transparent.

Angle detection

From the fact map the engine derives ten angle types. Each type has its own eligibility threshold so nothing is produced without a foundation:

Angle type	Minimum requirement
Hauptnews	≥ 3 facts and* a canonical headline
Hintergrund (background)	≥ 2 context facts (entity/organisation/person/event)
Vergleich (comparison)	≥ 2 comparable entities
Liste (list)	≥ 3 enumerable facts
FAQ	≥ 3 derivable question-answer pairs
Ratgeber (guide)	≥ 1 fact or headline (commentary exemption)
Pro/Contra	≥ 2 opposing viewpoints in the evidence
Lokale Einordnung (local)	≥ 1 location fact (also locale-aware)
Produktbezug (product)	≥ 1 product fact
Social-only	≥ 1 fact or headline (reaction/commentary)

Detection is purely heuristic and deterministic — no AI model call — and returns a viability score (0–1) per angle. The most viable angles are preferred.

Subtopic proposals and limits

For each viable angle the engine produces a concrete subtopic proposal with a title, summary, the supporting fact IDs, a derivation type (fact_based, commentary, evergreen, derivable) and an evidence score.

The engine deliberately does not produce junk. A subtopic is allowed only if:

enough confirmed facts are present, or*
it is clearly marked as commentary / opinion / guide, or*
it is logically derivable from the main topic.

Subtopics with no fact foundation and no recognisable derivation type are never produced.

Duplicate check

Before proposals enter the cluster plan, LD3 filters near-duplicate titles (Jaccard similarity ≥ 0.82) — against already-stored expansion candidates, against existing topics in the same project, and against other proposals in the same run. This keeps the planning queue free of redundancy.

Content cluster plan

The validated, de-duplicated subtopics are assembled into a cluster plan:

Main topic* and subtopics* as a structured list.
Internal links* — the main article links to every subtopic, every subtopic links back to the main article, and subtopics link to each other (sibling links).
Estimated content count* — 1 (main article) + number of subtopics.
Cluster status* — planned.

TEE Pro add-on

Basic expansion (TEE Basic) is included in LD3 Beta. An optional TEE Pro add-on* unlocks advanced topic expansion (entitlement advanced_topic_expansion) — including a higher number of expansion candidates (up to 10 instead of the default limit). The rest of the chain is unchanged.

[LD3 topics](/docs/ld3-topics) — detecting and expanding topics into content angles.
[LD3 topic verification](/docs/ld3-topic-verification) — how LD3 checks topics before expansion.
[LD3 fact extraction](/docs/ld3-facts) — where the fact map gets its confirmed facts.
[LD3 reviews](/docs/ld3-reviews) — the human review before publishing.
[LD3 content](/docs/ld3-content) — getting started with the LD3 module.

Was this article helpful?