The tiers of the parameters

Every parameter of the build has a tier that says how prominent it is on the screens; it changes nothing else. A page’s « Tune » panel shows its essential parameters first: the few a curator turns. Intermediate ones are folded under « More », advanced ones under « Advanced ». The assignment is one table, PARAM_TIERS in cartolex/build/stages.py, which must list every parameter of every stage once (the registry refuses to load otherwise); the layout parameters of a map version carry theirs in LAYOUT_DEFAULTS (cartolex/app/method.py).

This table was approved by the owner on 1 October 2026.

The build’s parameters

stage	parameter	tier	reason
corpus.assemble	parts	essential	which parts of a text are read decides what the keywords come from
corpus.assemble	recency_years	essential	the year window of the keywords, the first question of a field’s map
corpus.assemble	doc_types	intermediate	the defaults read texts and skip datasets and software; changed for an unusual corpus
corpus.assemble	provider_priority	advanced	only matters when several providers give the same part
corpus.assemble	duplicate_min_title	advanced	an internal threshold of the same-work detection
corpus.assemble	duplicate_year_gap	advanced	an internal threshold of the same-work detection
keywords.extract	min_people	essential	the main filter: how widely a keyword must be used
keywords.extract	min_texts	essential	the other main filter: in how many texts
keywords.extract	counting_unit	intermediate	a real choice (people, texts, organisations), but the default fits most fields
keywords.extract	max_share	intermediate	the generality cut; rarely turned
keywords.extract	rejects	intermediate	switching the rejection lists off is occasional
keywords.extract	max_words	intermediate	the length of the phrases; understandable, rarely changed
keywords.extract	length_bonus	intermediate	favours longer phrases; a visible effect, rarely changed
keywords.extract	max_candidates	advanced	a cap for very large corpora
keywords.extract	vote	advanced	a scoring internal
keywords.extract	of_complement	advanced	a grammar rule of English phrases
keywords.extract	fragment_share	advanced	a rule of the keyword filters (bands)
keywords.extract	drop_share	advanced	a rule of the keyword filters (bands)
keywords.extract	keep_share	advanced	a rule of the keyword filters (bands)
keywords.extract	name_share	advanced	a rule of the keyword filters (bands)
keywords.extract	stop_words	advanced	a rule of the keyword filters (closed words)
keywords.extract	closed_word_edges	advanced	a rule of the keyword filters (closed words)
keywords.extract	foreign_reading	advanced	a rule of the keyword filters (closed words)
keywords.extract	even_spread	advanced	a rule of the keyword filters (generic words)
keywords.extract	even_people	advanced	a rule of the keyword filters (generic words)
keywords.extract	common_modifier	advanced	a rule of the keyword filters (generic words)
keywords.triage	enabled	intermediate	the AI clean-up is switched on from the keywords screen; shown here for completeness
keywords.build	max_keywords	essential	the size of the vocabulary
keywords.build	keywords_per_person	advanced	empty by default: each person’s row of the space holds every keyword of the lexicon they use; a number keeps only their best ones
keywords.build	keywords_per_organisation	intermediate	the profiles of organisations
keywords.build	keywords_of_field	intermediate	the profile of the whole field
keywords.build	weights_basis	intermediate	a real choice of how themes weigh a person’s keywords
keywords.build	nested_threshold	advanced	an internal of the vocabulary’s nesting
keywords.build	ngram_range	advanced	the counting of forms; follows the longest phrase anyway
themes.space	space_unit	essential	people or texts: what “near” means for keywords
themes.space	dimensions	intermediate	the resolution of the space; a rule sets it from the project size
themes.space	svd_seed	advanced	an SVD internal
themes.space	svd_iterations	advanced	an SVD internal
themes.space	svd_algorithm	advanced	an SVD internal
themes.group	depth	essential	how many levels of themes
themes.group	level_sizes	essential	the size of each level, when set by hand
themes.group	comb	essential	the comb on or off: keywords on the level their texts support
themes.group	top_groups	intermediate	the size of the top level when the sizes are computed
themes.group	keywords_per_group	intermediate	the size of the finest groups when the sizes are computed
themes.group	comb_theta	intermediate	the comb’s one threshold, calibrated by default
themes.group	cluster_dimensions	advanced	a clustering internal
themes.group	exact_ward_limit	advanced	a memory limit of the clustering
themes.group	micro_clusters	advanced	a clustering internal
themes.group	micro_seed	advanced	a clustering internal
themes.group	comb_grid	advanced	the comb’s calibration grid
themes.group	comb_theta_one_level	advanced	the comb on a tree of one level
themes.group	comb_min_texts	advanced	the comb’s evidence floor
themes.group	comb_max_cells	advanced	a memory limit of the comb
themes.group	own_name_floor	advanced	a naming internal
map.layout	neighbours	intermediate	how keywords and projected people are placed; the default fits most maps
map.layout	link_radius	advanced	an internal of the placement
map.trajectories	window_years	essential	the time window of the trajectories
map.trajectories	min_texts_per_window	intermediate	which windows are placed
map.trajectories	spans	advanced	every run of consecutive windows (all): the map draws the windows themselves

The seed and the pinned year are set once for the whole build, in the texts’ « Tune » panel (the People screen’s Texts tab). Every parameter also has a short label in the interface’s catalogues (param.label.<stage>.<name>, and param.label.layout.<name> for a map version’s), shown with its code name.

The map’s layout (a map version, maps.json)

The map’s « Tune » panel shows them beside map.layout’s parameters; they are saved as a new map version, pinned, not in params.json.

method	parameter	tier	reason
all	the method	essential	UMAP, t-SNE or the theme tree: the look of the map
all	the seed	essential	draws another map with the same settings
umap	n_neighbors	essential	local detail against the broad picture
umap	min_dist	essential	tight clusters against an even spread
umap	metric, n_epochs, spread, set_op_mix_ratio, local_connectivity, repulsion_strength, negative_sample_rate	advanced	UMAP internals
tsne	perplexity	essential	the one knob of t-SNE
tsne	metric	advanced	the distance; cosine fits the space
tree	fill	essential	how much of each theme’s area its people fill
tree	gap, lean	intermediate	the spacing of themes and how people lean to their other themes
tree	sharp	advanced	an internal of the lean