DARPA CLARA TA1 – tri build & demo pipeline before Apr 17

[gHashTag]

DARPA CLARA (Compositional Learning‑And‑Reasoning for AI Complex Systems) full proposal is due April 17, 2026, 4:00 PM ET. darpa
CLARA TA1 targets new approaches for high‑assurance ML+AR with provable polynomial‑time behavior, verifiability, and compositional reasoning, with open‑source software deliverables. jm-aq

Trinity already has the core pieces (HSLM ternary NN, VSA symbolic layer, Queen Lotus reasoning, FPGA path), but we need a clean, reproducible tri build + CLARA demo pipeline that can be referenced in the proposal and used in hackathons.

Goal

By the CLARA full‑proposal deadline:

• tri builds cleanly (via Docker and/or Railway).
• A single command (e.g. tri clara demo) runs an end‑to‑end CLARA‑style pipeline:
  • HSLM forward pass on example inputs.
  • VSA binding / similarity reasoning.
  • (Optionally) Datalog/Zodd layer for symbolic rules.
  • Human‑readable proof‑style explanation (max ~10 steps).
• There is a short, copy‑pasteable "How to run the CLARA demo" section for the proposal / reviewers.

Tasks

• Fix remaining Zig 0.15 build issues so that zig build tri succeeds (no missing modules, no broken imports).
• Ensure tri builds inside Docker:
  • docker-compose -f deploy/docker/docker-compose.yml build tri passes.
  • Document minimal Docker build command in CLAUDE.md / docs/.
• Wire tri railway to the build pipeline:
  • Confirm tri railway build can trigger a remote tri build (Railway CLI wrapper is working).
  • Confirm tri railway status and tri railway logs give usable feedback for CLARA builds.
• Implement a CLARA demo command (initially local):
  • Add tri clara demo (or similar) entry in tri_register.zig.
  • Demo path: HSLM → VSA → (optional Zodd/Datalog) → explanation.
  • Use a small, deterministic example (e.g. "kill‑web" style toy scenario, medical toy, or supply‑chain toy) that matches CLARA use cases. darpa
• Implement an explanation output:
  • Print a bounded proof trace (~3–10 steps) with clear natural‑language labels.
  • Ensure the explanation format aligns with CLARA's emphasis on hierarchical, human‑readable explanations. darpa
• Write a minimal demo README:
  • New file: docs/clara_demo.md (or similar).
  • Include: prerequisites, build instructions (Docker / Railway), how to run tri clara demo, what the output means.
  • Keep it proposal‑ready (can be quoted / paraphrased in the CLARA technical volume).

Out of scope (for this issue)

• Full formalization of all Trinity theorems.
• Complete CLARA proposal text (handled in separate docs/proposals issues).
• Non‑essential refactors unrelated to getting a working demo and build pipeline.

[gHashTag]

AutoLoop Cycle 2026-04-02T16:35 ✅

Состояние:

• Build: ✅ GREEN
• Tests: ✅ GREEN
• Format: ✅ OK

Что сделано:

• ✅ Обновлён docs/clara_demo.md — добавлены Docker build инструкции
• ✅ Коммит: f1e2c8c342

Прогресс по issue #486:

• Fix Zig 0.15 build issues — DONE
• Register tri clara command — DONE
• Register tri railway command — DONE
• Create docs/clara_demo.md — DONE ✅
• Implement CLARA demo pipeline — DONE ✅
• Implement explanation output — DONE ✅
• Add Docker build instructions — DONE ✅
• Verify Docker build (Docker daemon not running — needs manual test)

Осталось:

• Docker build verification (требуется запущенный Docker daemon)
• Sacred AI fix/bypass — инфраструктурный блокер для live тестов

Примечание: Все основные задачи выполнены.只剩下 Docker verification (требует запущенного Docker) и Sacred AI fix/bypass.

[gHashTag]

AutoLoop Status Update

Build: ✅ GREEN
Tests: ✅ GREEN
Format: ✅ OK

Completed Tasks:

• Fix Zig 0.15 build issues — DONE
• Wire tri railway — DONE (works directly)
• Implement tri clara demo — DONE (proof trace working)
• Write docs/clara_demo.md — DONE
• Docker build instructions — ADDED to docs

Remaining:

• Docker build verification (requires running Docker daemon)
• Sacred AI REPL fix (infrastructure, separate issue)

Discovery: tri railway and tri clara demo work when called directly from binary. Sacred AI only blocks REPL, not direct calls.

Related Issues Status:

• Batch 2: Transformer + Scheduler specs (CLARA-critical) #490 (Batch 2): All 8 specs exist, ready to close
• Batch 1: RNN/Sequence + Optimizer specs #489 (Batch 1): All 6 specs exist, ready to close