# Recorded uses in source and mixed-interface readers Candidate integration; deployment is a separate check. Open the occurrence reader with `diagram=construction&group=interfaces`, then select **Recorded variable uses** or **Recorded uses of …** on a context extension. It uses the same source inspector, type geometry, loading switch and navigation history. The optional panel lists bounded batches of recorded source uses within the current frame. Each row preserves the use's own context and observed role. Binder names are compiled presentation payloads; source routes determine identity. **Inspect use** and **Inspect binder** select those exact occurrences. The diagram pans to the selected region or binder reference within its viewport. Each **Type** button opens one original native observation, preserving separate direct-query and construction-premise origins even when their types agree. **Focus source use** returns to its original occurrence geometry. The dashed reference link joins an existing source region to a retained binder slot or an explicit context-reference label. A binder inside the hidden region is labelled internal and does not become an outer-context requirement. The reference contributes zero source occurrences. Selecting an ordinary source port clears that reference selection; changing frames retires its controls. Every retained context unit has two separate coverage measures after the panel loads: recorded typed uses and original expanded syntax occurrences. A complete syntax count is not additional typing coverage. An unavailable syntax count is displayed as unavailable, including the seven current large-proof cases. Empty recorded batches do not establish absence of source uses. The list remains explicitly sparse; counts do not claim logical necessity. `public/js/proof-atlas-variable-uses.js` is the portable consumer and bounded gzip transport. Its reference results are compared with `index.cjs` across all 51 calibration/extraction profiles. The published delivery covers 26 extraction profiles: 152 uses, 304 existing observations and 269 binder counts. Core calibrations remain reference tests. Both complete and paged readers use this sparse delivery; paged mode does not fall back to complete source/type archives. `public/js/proof-atlas-variable-use-view.js` places the controls and references inside the existing mixed-interface view. The source/type bridge validates the actual source occurrence, lexical target, original context, evidence hash, type-table/component identity and native recording before displaying an observation. Lookup entries and supplemental construction owners stay distinct. Bounded input bytes are copied before asynchronous hashing, so later mutation of a caller's buffer cannot change the parsed catalogue after its hash was checked. Late observations cannot replace a newer selection. Optional use-data failure keeps the original interface and source navigation available for retry. ## Checks From the repository root: ```sh node scripts/build-proof-atlas-variable-uses.cjs node scripts/test-proof-atlas-variable-uses.cjs node scripts/test-proof-atlas-variable-uses-paged.cjs ``` The producer's default checks the exact canonical decoded metadata and its content address. Gzip encodings may differ across runtimes; existing compressed assets are retained byte for byte, and their compressed hashes are recorded separately for deployment checks. Portable compression controls accept three equivalent encodings and reject changed decoded content, damaged/truncated gzip and output exceeding the expected byte bound. `--write` generates only missing compiled metadata under `variable-use-pages` and its inventory. It neither reads private mathematical source nor adds a native Lean job. After the production site build, run `scripts/test-proof-atlas-variable-uses-browser.cjs` with Playwright/axe and the existing `ATLAS_ORIGIN` setting. Its default covers all 269 current mixed-frame starts and each reachable distinct recorded use/observation, source and binder selection, exact partitions, reference geometry, both loading modes, keyboard access, four theme/viewport configurations and optional-data/stale-response controls. Recorded uses outside these admitted context frames remain outside this browser scope. The current admitted frames reach 124 of the 152 catalogued uses; the remaining 28 are excluded from this browser coverage. The optional `ATLAS_VARIABLE_USES_FOCUSED=1` run writes a separate report with `fullCorpus:false`; it does not establish corpus coverage. These checks establish finite correspondence and application behavior. They add no native judgments, verified codec/renderer, general preservation theorem, semantic abstraction or measured human inspection benefit. The general program remains active; no blog result is claimed. ## Ordinary source cuts In **Source occurrences**, **Recorded variable uses** opens the same optional panel for the current source cut. Each use belongs either to one retained `bvar` occurrence or one disjoint hidden frontier. A retained use highlights its existing lexical edge; it adds no duplicate arrow. A hidden use links its original region to a retained binder, an explicit original-context reference, or a labelled binder inside that hidden region. Reference labels contribute zero occurrences and do not change the cut's retained/hidden partition. Retained `lam`, `forallE` and `letE` binders can filter the list. These are source constructors; `forallE` is not reclassified as a proof introduction. Syntax counts are available only for the binders included in the existing count inventory. The source scene and its other nodes keep `typing: not-queried`. Each type button attaches only the use's selected direct or kernel-premise observation, preserving its recording owner. Alternate Meta observations remain outside this catalogue. Source, binder and recorded-type actions use the existing inspector. Changing source cuts, representations or declarations retires the panel and pending loads. Selecting an ordinary source card clears the reference selection. Both source loading modes use the same sparse catalogue; paged mode reads bounded source and type pages without loading complete archives. Unsupported profiles keep their source graph available and report the missing use inventory. The independent reference rule is `source-cuts.cjs`. Reproduce its comparison with `node scripts/test-proof-atlas-source-variable-uses.cjs`. The finite model checks 1,209 cuts across 51 profiles, all 170 indexed uses and 13,923 reference placements, including the 28 public uses outside mixed frames. The browser test `scripts/test-proof-atlas-source-variable-uses-browser.cjs` exercises all 152 public uses, retained lexical-edge reuse, actual source/binder/type selections, failure recovery, stale controls and accessibility. `ATLAS_SOURCE_USES_FOCUSED=1` selects one Riccati profile and produces a separate `fullCorpus: false` report. A test description is not a passing result; inspect its emitted report.