trading-pipeline

name	trading-pipeline
description	Use when the user asks to run the full trading workflow end-to-end, from finding markets to paper trading — covers discover, backtest, and paper-run in sequence.

Trading Pipeline

discover → backtest → paper-run. Each phase gates the next. Be terse — do the work, report ONE summary table at the end.

Execution Style

• Batch everything. Run discovers in parallel, add relations in parallel, check statuses in parallel.
• Don't narrate. Skip analysis monologues. The user can read the table output.
• One summary at the end. Don't report after each sub-step. Deliver a single final summary with: relations found, backtest results, deploy status.
• Skip redundant checks. Don't relations list right after adding — you already know what you added.

Phase 1: Discover & Add Relations

Step 1: News → keywords (optional, skip if user provides topic)

coinjure market news --limit 10

Extract 2-4 keywords. Move on immediately.

Step 2: Parallel discovers

Run 2-4 discover commands in a single parallel batch:

# Run ALL of these in parallel — never sequentially
coinjure market discover -q "keyword1" --exchange polymarket --limit 40 --auto-discover
coinjure market discover -q "keyword2" --exchange polymarket --limit 40 --auto-discover
coinjure market discover -q "keyword3" --exchange polymarket --limit 40

Use --auto-discover to auto-create intra-event relations (implication, exclusivity, complementary). Do NOT use --json — read the table to spot cross-event relations.

Step 3: Add relations in parallel

After scanning all discover results, add all identified relations in a single parallel batch:

# Run ALL adds in parallel
coinjure market relations add -m <id1> -m <id2> --spread-type <type> --hypothesis "<constraint>" --reasoning "why"
coinjure market relations add -m <id3> -m <id4> --spread-type <type> --hypothesis "<constraint>" --reasoning "why"
# ... etc

Relation types (each auto-selects a builtin strategy):

Type	Strategy	Hypothesis
implication	ImplicationArbStrategy	A <= B (e.g. "by March" implies "by June")
exclusivity	GroupArbStrategy	sum <= 1 (mutually exclusive outcomes)
complementary	GroupArbStrategy	sum == 1 (all outcomes in one event)
same_event	DirectArbStrategy	A ~= B (same question, two platforms)
correlated	CointSpreadStrategy	spread mean-reverts
structural	StructuralArbStrategy	f(A) = g(B)
conditional	ConditionalArbStrategy	A|B bounded
temporal	LeadLagStrategy	lead-lag

Quick pattern recognition — look for these in discover output:

• Same question, nested dates → implication (earlier ≤ later)
• Same underlying, nested thresholds → implication (harder ≤ easier)
• Same topic, different angles, same timeframe → correlated
• Same question across exchanges → same_event

Skip markets with bid/ask at 0.99/1.00 or 0.00/0.01 — already resolved.

Phase 2: Backtest (one command)

coinjure engine backtest --all-relations --json

Results auto-update relation status. Don't stop to discuss failures — they're filtered out by --all-relations in the next phase.

Phase 3: Paper-Run

# Stop any old hub, start fresh, disable killswitch, deploy — chain it
coinjure hub stop 2>&1; coinjure hub start --detach --json       # or --demo for synthetic data
coinjure engine killswitch --off
coinjure engine paper-run --all-relations --detach --json

Use --demo on hub start when user wants synthetic/demo data.

After deploy, wait ~8s then check all statuses in parallel:

coinjure engine status --id <id1> --json
coinjure engine status --id <id2> --json
# ... etc

Final Summary Format

Deliver ONE table at the end:

| Relation | Type | Backtest PnL | Trades | Paper Status | Portfolio |
|----------|------|-------------|--------|--------------|-----------|
| ...      | ...  | ...         | ...    | ...          | ...       |

Plus monitoring commands for the user.

Gate Rules

Gate	Requirement
Discover → Backtest	At least 1 relation exists
Backtest → Paper	At least 1 backtest_passed
Paper → Live	Explicit user approval

Hard Rules

• Never use --json for market discover — table output is needed for cross-event reasoning.
• Never skip backtest before paper-run.
• Always start hub before paper-run (unless --no-hub).
• --all-relations on paper-run only deploys backtest_passed relations.

Reset (if needed)

# Remove all relations
coinjure market relations list 2>&1 \
  | grep '│' \
  | awk -F'│' '{gsub(/^[ \t]+|[ \t]+$/, "", $2); if ($2 != "" && $2 != "Relation ID") print $2}' \
  | while read id; do coinjure market relations remove "$id"; done