Competition Task Intelligence

Part of the Exo suite — literature (hfpclawer) → experiments (expflow-pde) → proofs (omega-architect). Three independent CLIs that meet through files and CLI calls, never imports. Entry skill: exo-suite-linkage (wiring, cost tiers low → medium → high, degradation ladder). Install: uv tool install hfpclawer · uv tool install expflow-pde · uv tool install "omega-architect @ git+https://github.com/diamond2nv/omega-architect@v0.2.3"

Overview

System for structured PDE equation management and competition task analysis. Provides:

1. PDE Equation Registry — structured metadata (LaTeX, dimensions, params, datasets) for 11+ PDEs
2. Task Analysis — per-task difficulty assessment, bottleneck identification, proven strategy catalog
3. Score Projection — optimistic/expected/conservative score estimates with confidence levels
4. Strategic Advising — which task to focus on, suggested schedule, rationale
5. CLI + MCP — expflow analyze command group and MCP tools

Installation

pip install expflow-pde

Version floor: install expflow-pde>=0.7.0 (public/PyPI line). The development line is ahead (0.7.3): anything marked "development line" below is not in the PyPI 0.7.0 artifact.

uv tool install expflow-pde                  # 1) isolated CLI, `expflow` on PATH (recommended)
# pin the version — `uvx`/`uv tool run` reuse an installed tool env (may run an older
# release), and an unpinned launch is a supply-chain (rug-pull) risk
uvx --from "expflow-pde==0.7.0" expflow --help        # 2) run it without installing anything
uv pip install expflow-pde                   # 3) inside an existing project / venv
pip install expflow-pde                      # 4) no uv — pip and pipx both work
pipx install expflow-pde                     #    (pipx needs a recent version; 1.0.0 cannot parse specs)
# extras ride along with any form, e.g.: uv tool install "expflow-pde[clearml]"

All five forms above were run and verified (uv tool install / uvx --from / uv pip / pip / pipx).

Architecture

expflow_pde/equations.py     ──── PDE equation static registry (11+ equations)
expflow_pde/analyze.py       ──── Analysis engine (task intelligence, strategy)
expflow_pde/cli_analyze.py   ──── CLI: analyze task/equations/status/advise
expflow_pde/mcp_server.py    ──── MCP: exp_compare_scores, exp_list_workers

1. PDE Equation Registry

Each equation entry in EQUATIONS dict includes: full name, LaTeX, dimensions, parameters, competition task mapping, metrics, solver, data samples, and competition info.

API

from expflow_pde.equations import (
    get_equations(),                    # All 11+ equations
    get_equation(name),                 # Single equation
    list_equations_for_task(task_id),   # task1/task2/task3
    get_equation_metrics(name, task),   # Relevant STANDARD_METRICS
    list_equation_names(),              # Sorted names
    list_competition_equations(),       # Only competition equations
)

2. Task-Level Intelligence

CLI

# Strategic advising (primary entry point)
expflow analyze advise

# Per-task analysis
expflow analyze task task1
expflow analyze task task3

# Equation reference
expflow analyze equations --task competition

# Competition overview
expflow analyze status

Example Output

expflow analyze status

Task     Score              Difficulty     Status         Priority
  ────────────────────────────────────────────────────────────────────
  task1    142/150            🟡 medium       🔴 In Progress  high
  task2    -/150              🔴 hard         ⚪ Not Started  low
  task3    -/350              🔥 very_hard    ⚪ Not Started  medium

  总分: 142/650  (508 pts remaining)

Score Estimation

from expflow_pde.analyze import estimate_score_potential, get_strategic_recommendation

estimates = estimate_score_potential("task1")
# Returns: {"optimistic": 148, "expected": 145, "conservative": 140, "confidence": "high"}

rec = get_strategic_recommendation()
# Returns: {"primary_focus": "task1", "remaining_headroom": {...}, "suggested_schedule": {...}}

Difficulty Classification

Integration with Other Systems

With experiment-lifecycle-governance

compare-scores gating builds on equation metrics from this system. When adding a new equation, its metrics must exist in STANDARD_METRICS for gating to work.

With analyze-experiment-autoregressive-degradation

Chain: analyze advise → decide task → run experiment → analyze degradation → feed back to _TASK_META.

Hermes Agent Environment

• MCP server — expflow mcp starts a FastMCP server (18+ tools) for agent integration:

  # ~/.hermes/config.yaml
  mcp:
    servers:
      expflow:
        command: "expflow"
        args: ["mcp"]
  

• Config — expflow init writes the toolkit config; real credentials (ClearML API, Langfuse keys, dataset paths) belong in .env / a local config.yaml, never in tracked files.

• Environment overrides — EXPFLOW_HOME, EXPFLOW_PIN_HASH, EXPFLOW_COMPETITION_DEADLINE, EXPFLOW_SEMANTIC_URL.

• Cost profile (not zero-cost) — the orchestration layer itself makes no LLM call, but runs cost real money and hardware: ClearML workers/queue time, GPU hours for the submitted experiments, and storage for artifacts. Quote costs from recorded runs, never an estimate.

Pitfalls

1. _TASK_META becomes stale — hardcoded scores must be updated after each submission
2. Competition deadline hardcoded — get_strategic_recommendation() has remaining_days from 2026-05-27
3. Scoring formula duplication — Task 3 formulae are in both equations.py and analyze.py; keep synced
4. No clearml import in analyze — analyze.py uses only pure Python/stdlib for fast CLI startup