research: Jupyter Notebooks deep dive; UC-84 (SHARD-WP-0004 T3)

.ipynb JSON cells + embedded computed outputs with fragile execution
provenance; derived output stored inside the source. Non-Markdown/lossy;
kernel = capability, default = present snapshot + static render.
Enriches UC-54/55/59/35; links UC-32/83/79. Marks T3 done.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

spec/UseCaseCatalog.md

@@ -1082,6 +1082,34 @@ build system (driving derivation is capability-gated). Feeds SHARD-WP-0002 T12
 named-chunk transclusion).  
 **Priority:** Later
 
+---
+
+### UC-84 — Attach or project a computational notebook with computed-output provenance
+
+**Actor:** Orchestrator / adapter  
+**Goal:** Attach/project a **computational notebook** (`.ipynb`) preserving its **cell
+structure** (markdown / code / output) and its **embedded computed outputs**, surfacing each
+output **as a snapshot carrying its (weak) execution provenance** (run count, environment
+not guaranteed); re-execution is **capability-gated** — the default is to present the stored
+snapshot plus an offered static rendered projection.  
+**Source:** wikiengines, intent  
+**Notes:** A Jupyter notebook is a **JSON document** of ordered cells where code cells own
+**MIME-bundle outputs captured from the last run**
+(`research/260614-jupyter-deep-dive/findings.md`). Its defining trait is that the **derived
+output is stored *inside* the canonical source** — the source/projection line runs *through*
+the document — and its provenance is **fragile** (`execution_count` may be out-of-order;
+environment/data/versions are not captured by `nbformat`). Treatment: keep the JSON as
+canonical payload (non-Markdown; any Markdown projection is lossy and directional, UC-55/
+UC-59); present captured outputs as **derivation-projection snapshots** (UC-83) marked "run
+N, environment unguaranteed" (never as live/authoritative, union-without-erasure); offer
+nbconvert/nbviewer static render; **re-execute or parameterize (papermill) only as a
+capability**, never assumed (shard-wiki is not a kernel host). Cell `id` (nbformat 4.5+) is
+the sub-page address for anchoring/transclusion (UC-32/UC-35); JSON diffs are noisy →
+text-pairing (Jupytext) / cell-aware merge (nbdime) for history (T13). Feeds SHARD-WP-0002
+T12 (notebook page shape: outputs embedded in source), T15 (lossy non-Markdown translation;
+MIME opacity), T13 (paired-text history), T16 (output = derivation-projection snapshot).  
+**Priority:** Later
+
 ## D. Addressing, identity & query
 
 *Span / identity addressing, transclusion-as-reference, dimensional navigation, and query over the union.*
@@ -1583,7 +1611,7 @@ marker's true lineage is its per-source mapping subsection below):
 | ⚓ | wiki.js | ⊞ | federated-wiki | ⬡ | wikibase |
 | ⎇ | forge-wikis (Gitea/GitLab/GitHub) | ⊡ | tiddlywiki | ⊟ | ikiwiki |
 | ◧ | quip | ⊙ | mojomojo | ⊚ | oddmuse (UseModWiki reinforces, no marker) |
-| ⊛ | literate-programming (WEB/weave/tangle) | | | | |
+| ⊛ | literate-programming (WEB/weave/tangle) | ⊜ | jupyter (notebooks) | | |
 
 | UC | c2 research | yawex research | federation research | wikiengines research | INTENT |
 |----|-------------|----------------|---------------------|----------------------|--------|
@@ -1652,6 +1680,7 @@ marker's true lineage is its per-source mapping subsection below):
 | UC-81 | | | | ⊙ | ✓ |
 | UC-82 | ✓ | | | ⊚ | ✓ |
 | UC-83 | | | ✓⊛ | | ✓ |
+| UC-84 | | | | ⊜ | ✓ |
 | UC-08 | ✓ | | |
 | UC-09 | ✓ | | |
 | UC-10 | ✓ | | |
@@ -2341,6 +2370,37 @@ snapshot** — never present a derivation without its source link (union-without
 Architecture logged for `SHARD-WP-0002` (T12/T16): one-source-many-projections page shape,
 replication- vs derivation-projection, named-chunk transclusion.
 
+### jupyter mapping
+
+(⊜ UC-84 lineage = the **Jupyter deep dive**, `research/260614-jupyter-deep-dive/findings.md`;
+a concrete computational content type, not a wiki engine.)
+
+| Jupyter mechanism (findings §) | Catalog UC |
+|--------------------------------|------------|
+| `.ipynb` JSON: ordered cells, code cells own embedded MIME-bundle outputs (§1) | UC-84 (new) |
+| Captured outputs stored *inside* the source; weak `execution_count` provenance (§2) | UC-84; links UC-83 (derivation-projection snapshot) |
+| JSON + MIME bundles = non-Markdown; nbconvert→MD lossy & directional (§3) | UC-55 / UC-59 (enriched) |
+| Code cell = computation-defined content (§1) | UC-54 (enriched) |
+| Cell `id` (nbformat 4.5+) = sub-page address (§4) | UC-35 (enriched); links UC-32 |
+| Noisy JSON diffs → Jupytext pairing / nbdime cell-merge (§3) | links UC-43, T13 |
+| nbstripout/pairing = source-canonical, outputs-derived (§3) | links UC-79, UC-83 |
+
+Note: Jupyter is the **dominant computational document** and the concrete test of T1's
+projection split: nbconvert/nbviewer are **derivation-projections**, but the twist is that
+**captured outputs are stored back inside the canonical source** (the source/projection line
+runs *through* the document) with **fragile provenance** (out-of-order `execution_count`;
+environment uncaptured). The page model (T12) gains a **notebook shape** — ordered cells with
+code cells owning embedded computed outputs — distinct from prose, typed records,
+query-defined, inline-embedded objects (Quip/Notion), typed-graph (Wikibase), and the
+literate one-source-many-projection shape (UC-83); its new attribute is *derived output
+embedded in source*. **Boundary recorded:** shard-wiki is **not a kernel host** — default to
+attach + present captured outputs as snapshots marked "run N, environment unguaranteed"
+(never live, UC-84) + offer static render; re-execution/papermill are **capabilities**. Keep
+JSON canonical (Markdown projection is lossy, UC-55/59); prefer paired-text storage / nbdime
+for history when the shard is git. Architecture logged for `SHARD-WP-0002` (T12/T15/T13/T16):
+notebook page shape, lossy non-Markdown + MIME opacity, paired-text history, output =
+derivation-projection snapshot.
+
 ---
 
 ## Open questions