feat: arena/ outer loop (basilisp) + tick/ inner loop (hy) + bcf-0059 pipeline doc

[bmorphism]

Adds arena basilisp scripts, hy tick loop, and separates bcf-0058 conflict: IRB stays as bcf-0058, GraphQL pipeline doc becomes bcf-0059.

[bmorphism]

Review: build + functorial conservation analysis

Built and analyzed this branch on an isolated VM (forester 4.3.1, linux-x86_64). Treating the forest as a functorial database (schema category C: object Tree, morphisms transclude, ref : Tree→Tree, attribute taxon), this PR is an instance migration from its merge-base 04b5862 to head ad997f8.

What was done

• Full forester build of the branch → 40,376 output XML trees, forester build exit 0.
• Extracted the node + edge graph from source .tree files; ran Δ/Σ/Π (pullback / left-Kan / right-Kan) diff vs merge-base.
• Validated XML well-formedness with an independent strict parser; ran DuckDB queries over the rendered forest.

What was discovered

• Conservative migration — Δ∘Σ ≅ id. Objects: +1 / −0. The single freely-adjoined tree is bcf-0059 "Pipeline 7: gh CLI GraphQL Streaming into DuckLake Snapshots" (architecture). Taxon retypings: 0 / 1794 shared. Transclusion diagram unchanged (2913 → 2913 edges); ref graph +4. The forest functor is conserved exactly.
• The +774 lines are arena/ (basilisp) + tick/ (hy) — an orthogonal code fibration over the forest, not the tree olog, so the database migration is essentially the identity.
• Validity: all 88/88 authored bcf/bci/horse trees are well-formed XML.
• Co-mergeable with alt: navigable atomic forest (de-inlined root) — comparison to #29 read-through #30: the two PRs have disjoint addr-supports — the pushout #25 ⊔_main #30 is a clean coproduct, 0 semantic conflicts.
• Pre-existing H¹ (not introduced here): 7 dangling edges — the forest root bcf-0001 is never rendered to bcf-0001.xml (transcluded by mat-0001/index), plus template placeholders (X, ..., namespace-XXXX) in bcf-0046/lc-000C.
• Housekeeping: branch is BEHIND main — a rebase will pick up newer trees (this is why a raw diff vs main-HEAD over-reports; the merge-base comparison above is the correct one).

How to reproduce

git clone -b arena-tick-pipeline https://github.com/plurigrid/place && cd place
# forester 4.3.1; if no TeX toolchain, stub the math pipeline (metadata-only, read-only):
printf '#!/bin/bash\nfor a in "$@";do case "$a" in *.tex) b="${a%.tex}";;esac;done; : > "${b:-texput}.dvi"\n' > latex
printf '#!/bin/bash\nexit 0\n' > dvisvgm   # forester pipes math via --stdin --stdout
chmod +x latex dvisvgm; export PATH="$PWD:$PATH"
./forester build bci-forest.toml --no-theme --no-assets    # -> output/*.xml, exit 0

# functorial graph (nodes/edges) + frontmatter, then DuckDB:
#   nodes: addr,taxon,title,date   edges: src,dst,kind in {transclude,ref}
duckdb -c "SELECT taxon,count(*) FROM read_json_auto('nodes.jsonl') GROUP BY 1 ORDER BY 2 DESC;"
# Σ added vs merge-base:  SELECT addr FROM head WHERE addr NOT IN (SELECT addr FROM base);
# H¹ dangling:            SELECT * FROM edges WHERE dst NOT IN (SELECT addr FROM universe);

XML validity (independent of build): python3 -c "import xml.etree.ElementTree as ET,glob;[ET.parse(f) for f in glob.glob('output/bcf-*.xml')]".

Net: a clean, safe extension. No structural objections.

[bmorphism]

Stable link: these now go through a GitHub Pages redirect (bmorphism.github.io/bci-preview) that survives preview-host restarts — the raw cloud URLs are ephemeral. git clone points at canonical github.com.

🌲 Live render — view this branch in your browser

This branch is built and hosted so anyone can see the actual rendered forest (no clone, no install — the browser applies the XSLT client-side):

• Portal root: https://bmorphism.github.io/bci-preview/pr25/output/horse-0001.xml
• Device ledger: https://bmorphism.github.io/bci-preview/pr25/output/bcf-devices.xml
• The tree this PR adds (bcf-0059, Pipeline 7: gh GraphQL → DuckLake): https://bmorphism.github.io/bci-preview/pr25/output/bcf-0059.xml
• Any tree: …/pr25/output/<addr>.xml

The PR branch is also git-cloneable from the same host:

git clone -b arena-tick-pipeline https://github.com/plurigrid/place

Please open the links and click around — the portal, the device ledger, the TOC/backlink sidebar all work. If anything looks off (broken transclusion, missing tree), reply here with the address.

Notes: this is a fast preview build — \tex math renders fine via client-side KaTeX; everything else (text, tables, links, navigation) is live. Served from an ephemeral build VM, so links are good now; ping if a link 404s and I'll relaunch.

[bmorphism]

Stable link: these now go through a GitHub Pages redirect (bmorphism.github.io/bci-preview) that survives preview-host restarts — the raw cloud URLs are ephemeral. git clone points at canonical github.com.

📑 Browsable index (easier than guessing addresses)

One entry point listing every BCI tree in this branch, grouped by taxon, each a clickable render link:

→ https://bmorphism.github.io/bci-preview/pr25/index.html (110 trees)

Compare against:

• main baseline: https://bmorphism.github.io/bci-preview/pr00/index.html (151 trees)
• all branches: https://bmorphism.github.io/bci-preview/

(Build state is snapshotted, so the renders persist. Inline math renders via client-side KaTeX (only complex tikz/xy diagrams in imported math trees may be placeholder).)

[zubyul]

Built this locally against origin/main (8b466b2). There are two independent blockers, and the net effect today is that none of this PR's content reaches the site — including on a build that exits 0.

1. The branch is 20 commits behind main

Base is 04b5862. At that commit trees/bcf-0058.tree is the pre-fix version and fails to parse; main fixed it in b3c6c37. So building this branch drops bcf-0058 too — collateral damage that disappears on rebase. Rebase onto main first, or the build log will show an error that has nothing to do with your work.

2. bcf-0059.tree embeds source code without verbatim

All four code blocks open with \pre{% and there is no \startverb anywhere in the file. Forester parses the contents of \pre{}, so the Elisp/Babashka source is being read as forester markup — its braces, brackets and % are all interpreted:

./trees/bcf-0059.tree
165 | }                                                    <- unexpected `}`
226 | syrup (:olc t) (esc (or (:content t) "")))]          <- unexpected `]`
237 | }                                                    <- unexpected `}`

Also note % comments out the rest of its line in forester, so \pre{% is silently eating the first line of each block.

Code has to go inside a verbatim span, which is not parsed at all:

\pre{\startverb
(defun bcf-query-olc (olc)
  "Find trees authored near an OLC."
  (interactive "sOLC: ")
  (bcf-query
    (format "SELECT tree_id, title FROM bcf.trees
WHERE olc = '%s' ORDER BY committed_at DESC LIMIT 20" olc)))
\stopverb}

\startverb must be followed by whitespace, or it lexes as one unknown command. The one thing you cannot put inside a verbatim span is the literal stop keyword — if the code ever contains it, split into two spans.

Why the build still "passes"

This is the trap worth internalising. These are parser errors, which forester recovers from: it silently drops the offending tree and carries on. My build of this branch exited 0 and emitted 40,398 pages — bcf-0059.xml simply was not among them. A green build is not evidence your tree rendered.

(Contrast #42, which hits a lexer error — unrecognized token — that aborts the run at zero pages. Different failure shape, same root cause: unescaped content being read as markup.)

3. Nothing transcludes it

Independent of the parse errors: bcf-0059 is referenced by no other tree — not transcluded, not even [[linked]]. Once it parses it will get a standalone page, but this forest reads as one page via index.tree -> bcf-0001 -> ..., so an un-transcluded tree is invisible in the main read. Add a \transclude{bcf-0059} from whichever parent it belongs under.

Also worth saying plainly: arena/*.lpy and tick/*.hy are not trees and will never render. If you want that code on the site, it has to be transcluded into a tree as verbatim.

Checking it compiles

./forester build forest.toml 2>&1 | tee /tmp/build.log
grep -c 'Parse_error' /tmp/build.log   # must be 0 — this is the gate, not $?
ls output/bcf-0059.xml                 # your page must actually exist

Build inside nix develop (the flake pins texliveFull) — \quiver / \tex shell out to latex, and without it the run dies at the resource stage having emitted nothing, which looks identical to a content bug. CI runs nix build .#forest.