Baseline: what the topic template must beat
This file names the baseline a typed topic template is compared against, and how the comparison would be measured.
The named baseline
| Role | Baseline | Why this one |
|---|---|---|
| Typed units with derivation (the one to beat) | the DITA 1.3 base topic (OASIS 2015) with the concept, task and reference specializations of its technical content edition (OASIS 2015b) | the only system found that declares a base, derives types from it, and lets a program read any derived unit as its base; an OASIS Standard since 2005, listed by 802 firms (DITAWriter 2026) |
| Historical anchor | Information Mapping's typed maps, as in Horn 1969 | the first typed system found with derivation (a map type taking every block of another), and the only one with a controlled test against untyped text |
| Control | the same content as untyped prose, Horn's "P version" | removes the visible typing and keeps the words, so a gain belongs to the type |
- The DITA base is the right rival because it already holds what the topic claims. It has one base shape (title, short description, prolog, body, related links, nested topics; OASIS 2015, 2.2.1.6) and derived types that keep it.
- The topic must therefore beat DITA on readers, not on structure: DITA has no reader study in this search, so the topic's claim is testable where DITA's is not yet tested.
- Kubernetes 2026b (concept, task, tutorial, reference) is a second, lighter baseline in Markdown, without declared derivation.
What "beat" means
The topic beats the baseline if, on the same content, it does at least one of these without losing on the others:
- readers find a fact faster;
- readers recall the gist better, at once and after a delay;
- readers who learned the base read an unseen derived type better than readers who learned only sibling types;
- a program checks more of the reader-ergonomics rules on it, at any depth.
How a comparison would be measured
| Reader task | Outcome | Instrument | Prior study to borrow from |
|---|---|---|---|
| Find | time to locate a stated fact; share found | fixed search questions on one document set in three forms | Kools 2008 (search speed with and without headings) |
| Understand | gist questions against local questions, separately | main-idea and summary items; cloze or verbatim items | Kintsch 1982 (global against local comprehension) |
| Recall, delayed | recall at once and after one week or more | free recall scored by idea units | Bogaerds-Hazenberg 2020 (effects gone at delay) |
| Transfer to a derived type | performance on a type never studied, derived from the base | train on the base and two derived types, test on a third derived type | Hebert 2016 (near transfer to untaught structures) |
| Cost to the writer | time to write; errors against the type | the same content written in each template | none found for typed templates |
| Program check | share of units that validate as their base type at every depth | a checker that reads each topic as its base, as DITA's @class fallback does |
OASIS 2015 (2.5.3.6, 2.5.4) |
- Forms compared: untyped prose (the control), DITA concept, task and reference, and the topic template; the words held constant, as in Horn 1969.
- Design: within-subject crossover, counterbalanced order, a delayed session after one week.
- Effect sizes: Hedges' g with confidence intervals, as in the meta-analyses; report each outcome apart, since effects differ by outcome (Bogaerds-Hazenberg 2020).
- Readers: adults reading documentation, not school students; most evidence above comes from students, a threat to generalisation.
- Readers are not taught the types beforehand in one arm, and are taught them in another: the positive results so far all rest on taught or known types.
- For a model as reader: the same three forms, with and without the type label, scored on retrieval and answer accuracy, as in Jimeno Yepes 2024.
A checklist in the baseline's own terms
Per unit, from the DITA base (OASIS 2015):
- The unit has an identifier and a title.
- A short description says what the unit is about; the specification treats it as optional and strongly advised.
- The body holds only what the unit's type permits.
- Related units are linked from a related-links part.
- Child units are nested whole, each with its own title.
- A derived type declares its ancestry, so it can be read as its base.
Per map, from Information Mapping (Horn 1969):
- Each block carries a label naming its kind.
- Each kind of block keeps one format across the document.
- A map begins on a new page and fits one page where possible.
- Feedback questions sit next to the block they test.