Correction loci — where the fixes land in the printed book

Every accepted correction to a Cologne dictionary's digital text is recorded as a change-file line in csl-corrections; its correction_loci.tsv feed resolves each record to the printed <pc> page it touches. This page reads that feed as a spatial and editorial signal: where in each printed edition the reported errors sit, and how actively maintained each dictionary is — the editorial axes of the per-dictionary radar (METALEXICOGRAPHY_ROADMAP §2.6/§3).

Read the process split first. 76% of all records are machine batches (the BOR digitization completion and the LRV markup-homogenization sweep), not human proofreading. Every view below separates human from bulk; cross-dictionary maintenance comparisons should default to the human series.

The feed carries correction records across dictionaries — human, bulk; are resolved to a printed page.

Editorial KPIs — corrections per 1,000 entries

Corrections per 1k current entries (<L> records in csl-orig v02), stacked by process. This is the "how actively maintained" axis: BOR's enormous per-1k value is its one-shot digitization completion batch, not ongoing editorial attention — which is exactly why the human slice is drawn separately.

Editorial radar — the correction-fed axes

Five editorial axes per dictionary, each normalized to the maximum across the dictionaries currently shown (per-1k axes on a log scale, so BOR's batch spike does not flatten everyone else). These are the axes this feed contributes to the roadmap's per-dictionary radar; markup-density and citation axes stay on HOLD in the roadmap and are not drawn here.

Locus heatmap — normalized page position × dictionary

Each row is one dictionary's printed edition, left edge = page 1, right edge = its last corrected page (see the trust block: the feed's own maximum, not the true page count). Colour is corrections per position bin ( bins). Toggle the process series; "row share" rescales each row by its own total so sparsely-corrected dictionaries stay readable.

One dictionary in detail — position × column

For dictionaries whose change files carry a regular column notation (), the same positions split by printed column.

The most-corrected printed pages per dictionary. Each row links into the source viewer at a sample corrected line, so a cell of the heatmap is one click from the actual dictionary text (headwords shown in IAST).

Batch cadence

Records per batch month, by process. Change files land in dated batches, so this is delivery cadence (when fixes were filed into csl-orig), not discovery dates.

Maker QA — correction-pressure column

Prioritization signal for the H5 maker-QA worksheet: how much human correction traffic has already touched the candidate lemma or its nearest-real neighbour (form-keyed via slp1_norm). No new queue — one column on the existing H5 packet (docs/H5_MAKER_QA_CANDIDATES.md). Bulk machine batches are excluded.

Chart Trust Block