Algorithms & libraries — ecosystem plan (full vertical spectrum)¶

Status: Active (rev. 1 — 2026-05-21)
Audience: Architects, package owners, agents
Canonical product vision: world-studio-vision.md
Compiler truth: master plan · provability-gaps.md
Perf / validity: competitive-engines-plan.md · benchmarks/competitive/registry.toml

This page is the algorithms-and-libraries layer under World Studio — not the studio shell, not lic internals alone.

1. Executive assessment (2026-05-21)¶

Verdict: The studio + profile plan is detailed enough to steer agents. The algorithms/libraries plan was implicit across RFCs — this document makes it explicit and adds missing competitor + scheduling discipline.

2. What we already have (do not duplicate)¶

3. Vertical coverage matrix (applications → libraries)¶

Legend: UI = Studio shell · Algo = proved kernels · Bench = timed/validity row · Intel = competitor map maintained

4. Competitor intelligence — two layers (required)¶

Layer A — HPC / language runtimes (mature)¶

Source: benchmarks/competitive/registry.toml, competitive-engines-plan.md.

Layer B — Domain tools & algorithm libraries (incomplete)¶

Each vertical needs a maintained row in benchmarks/competitive/verticals.toml (new file — see §7) with:

UX intel (Layer C) stays in competitive-intel/ — it does not replace Layer B algorithm tracking.

Verticals missing Layer B tables (add in rev. 2)¶

• Engineering FEA (mesh solve, linear elasticity)
• CFD (pressure-velocity coupling, turbulence models)
• Computational chemistry (basis sets, DFT functionals — beyond stub API)
• CAD/B-rep (predicates, booleans)
• Slicer / toolpath (infill, support, thermal compensation)
• Cinematic (encode pipeline, color, audio sync)
• DCC mesh ops (subdiv, UV, rig skinning)

Template: copy structure from competitive-bioengineering-plan.md §5 (benchmark registry slice).

5. lic maturity — gate before large-scale libraries¶

Do not interpret composable-import green or World Studio stub milestones as proof-certificate readiness.

Wave A — compiler proof gate (blocking all “production” domain libs)¶

Until Wave A: domain packages stay stub/composable, workload_class = stub, no “on-par with GROMACS/Gaussian” marketing.

Wave B — runtime perf gate (parallel Wave A tail)¶

Wave C — package ecosystem gate (before scaling repo count)¶

Current snapshot (li-language / lic workspace)¶

Honesty rule: PH-world-studio-program.md “Done” on stubs ≠ algorithm parity. Rename mentally to “interface landed” until Wave A passes.

6. Implementation waves (when to start large-scale work)¶

flowchart TB
  A[Wave A: lic 2e 2f 7b 2i 7e]
  B[Wave B: tier-2 perf + external oracles]
  C[Wave C: lip lit 8x registry]
  D[Wave D: vertical library scale-up]
  E[Wave E: Studio native + GPU LKIR]

  A --> B
  A --> C
  B --> D
  C --> D
  D --> E

Recommended vertical order for Wave D (algorithm depth)¶

1. Shared numerics — math, math.numerics, linalg (all verticals depend on this)
2. Tier-2 physics core — MD, PDE, N-body (already started; harden + external oracle)
3. Gaming rigid/soft body — v1 gaming kernels per competitive-engines-plan
4. QM (chem) — Psi4/ORCA trusted driver + one native small basis set
5. Drug / bioeng — LITL + DBTL on real assay/oracle subset
6. Additive — thermal + warp on heat_equation + export path
7. CAD/geometry — predicates + mesh booleans (Manifold-class), not OCCT monolith
8. CFD/FEA — single canonical cavity flow + linear elasticity (engineering)
9. Cinematic — deterministic seq + ffmpeg-class trusted encode
10. DCC — glTF + procedural mesh ops only (no Maya feature parity goal)

Parallelism: Studio UX/agent work continues in stubs during Wave A; do not staff >2 FTE-equivalent agents on kernel implementation until Wave A gate is green.

7. Package gap register (what to add vs implement)¶

Monorepo index: packages/li.toml (40+ members today). This table is the execution backlog for packages — not the Studio product vision alone.

Legend: exists-stub = folder + composable smoke · exists-partial = real code, incomplete · missing = not in workspace · extend = no new repo, expand in place

7.1 Numerics & spatial math (quaternions live here)¶

Quaternions: do not add quaternion or li-quat package. math already has Quat, quat_identity, quat_mul. Add in math:

• quat_dot, quat_conj, quat_normalize, quat_from_axis_angle, quat_slerp, quat_rotate_vec3, quat_to_mat4
• mat4_mul, mat4_mul_mat4 (full 4×4)
• Wire scene.Transform3 to use math.Quat (replace raw qx..qw helpers)

7.2 Graphics & UI stack (no proper GPU UI yet)¶

Rule: ui = editor chrome / MCP IDs; gui = widgets + draw lists consumed by render. Do not merge into one package.

7.3 Creative authoring (animation & cinematic)¶

Start anim under scene if package count is a concern; split when import anim stabilizes.

7.4 CAD & geometry¶

CAD gap doc (li-language): merge into lic/docs/ecosystem/cad-fundamentals.md — AL-4.

7.5 Already exist — deepen in place (no new package name)¶

7.6 Do not add (common mistakes)¶

7.7 Package implementation order (agent checklist)¶

P0  math quat/Mat4 + scene transforms
P0  lic Wave A (2e/2f) — parallel with thin stubs only
P1  linalg package + li-tests/math_linalg green
P1  lig → render present path (wgpu smoke)
P2  gui package + ui/studio wire-up
P2  assets glTF smoke
P3  anim → seq → studio.publish encode
P3  geometry mesh ops
P4  deepen physics.* / sim.* / chem per verticals.toml

8. Deliverables to add (plan maintenance)¶

9. Package taxonomy (avoid studio-only trap)¶

Rule: T3+ features must cite a benchmarks/competitive/* row or be labeled stub in docs.

10. Agent routing (read order)¶

1. vision-and-roadmap.md — north star
2. This file — algorithms/libraries scheduling
3. world-studio-vision.md — product architecture
4. provability-gaps.md — what is not proved today
5. competitive-engines-plan.md — how to run benches
6. Vertical RFC under docs/game-dev/specs/ when implementing one domain

11. Related links¶

• Engineering standards
• Official packages
• CAD fundamentals (li-language)
• Master plan phase map

Maintainers: Update §7 when adding/removing packages/* members. Bump updated in verticals.toml / registry.toml on quarterly SOTA review.

12. Fresh-machine agent handoff¶

Copy this block into the agent prompt on a new checkout.

## Li ecosystem agent — fresh machine bootstrap

**Repos (clone `li-langverse`):**
- `lic` — compiler + `packages/*` + `benchmarks/` (primary workspace)
- `roadmap` — governance + agent-kit
- `lip`, `lit` — package manager + tests (optional until phase 8 work)
- `benchmarks` — dashboard ingest (optional)

**Read first (30 min):**
1. `lic/docs/ecosystem/algorithms-and-libraries-plan.md` — §7 package gap register + Wave A–E
2. `lic/docs/verification/provability-gaps.md` — do not overclaim proof
3. `lic/docs/superpowers/plans/2026-05-14-li-master-plan.md` — phase tracker (2e/2f partial)
4. `lic/docs/game-dev/world-studio-vision.md` — Studio profiles (if product-facing)

**Build `lic` (macOS example):**
export LLVM_DIR="$(brew --prefix llvm@22)/lib/cmake/llvm"
export CC=clang CXX=clang++
cd lic && ./scripts/build.sh
./build/compiler/lic/lic --version
./li-tests/run_all.sh

**Default task queue (unless user specifies otherwise):**
- Wave A only: contracts/Lean/compiler — do NOT scale domain kernels
- P0 package: extend `packages/math` quaternions + `Mat4`; wire `packages/scene`
- Next new packages: `gui`, `linalg`, then harden `lig`→`render`
- New package: `./scripts/li-new-package <name> --kind library` then add to `packages/li.toml`

**Proof rule:** `lic build` certificate not complete (G-VERIFY-01). Composable green = interface only.

**Bench:** `python3 benchmarks/harness/bench.py --tier 0` then tier 12 when changing perf.

**Branch:** `feat/agent-first-gui` on `lic` has latest ecosystem plan; merge via PR to `main`.

**Do not:** create `li-quaternion` repo; port OCCT; claim GROMACS parity without `verticals.toml` oracle row.

Maintainers: Keep §12 in sync when default branch or bootstrap commands change.