docs: honest 'not yet ported' audit (concept engine, viz panels, session, cache)

Author: glendonjklassen

PROGRESS.md

@@ -112,6 +112,38 @@ the architecture and the locked decisions; see
 
 Dropped by decision: signed **patches** and **rules** (not ported).
 
+## Not yet ported (real overlay features, neither done nor dropped)
+
+An honest audit of the reference modules — these belong here and are all
+pure-code doable (no new data, no blocked toolchain):
+
+- **Symbolic concept engine** (`Concept.hs` 1042 + `ConceptGraph.hs` 242 +
+  `ConceptMap.hs` 124): derived co-occurrence statistics, the **radial concept
+  neighbourhood graph**, and the **dispersion strip** ("where across the 66
+  books does this concept occur"). We ported the *embedding* concept layer; this
+  *symbolic/statistical* one is separate and unported.
+- **Leitwort / burst discovery** (`Burst.hs` 429): corpus-wide "bursty,
+  deliberately-repeated" concepts (theme-words). Pure statistics.
+- **Weave visualizations**: the **constellation** (`Constellation.hs` 286, the
+  weave-library overview) and the **chord/arc map** (`ChordMap.hs` 145,
+  book-to-book weave density). Cairo drawing like the canon strip.
+- **Text-as-witness** (`Witness.hs` 144): the lexicon-free alignment that can
+  disagree with the bridge sources. We ported the trust *priors* it feeds, not
+  the witness model itself.
+- **Session restore** (`Session.hs` 168): reopen the last panes / position. Today
+  the app always opens to John 3 (only mode + body size persist).
+- **Startup index cache** (`IndexCache.hs` 136): a gzip snapshot of the built
+  indices for faster launch. We rebuild each launch (fast except the ~1s SIF
+  build). A pure perf optimization.
+- **Panel breadth** (`Panels.hs`): the GTK study panel covers word study /
+  search / concordance / threads / tags / suggested / weave-compare, but not the
+  concept-graph, dispersion-strip, constellation, chord-map, or burst panels
+  above (they depend on those unported engines).
+
+Also offline-only (not runtime): the analysis harvesters (`Tasks.hs`, quotation
+detection, `audit_sources.py`) — their outputs are consumed, the harvesting is
+not a reader feature.
+
 ## Toolchain status (updated 2026-07-09)
 
 `rustup` is in and the cross **targets are added** (android arm64/armv7/x86_64,