# Slice 0 — First Complete TASI Trading-Day Loop

Slice 0 is complete only when one real TASI trading day can run end-to-end without manual engineering intervention.

## Inputs
- TASI market profile and exchange calendar.
- Configurable virtual SAR budget.
- Investment and Swing strategies.
- Point-in-time market/evidence data.
- Versioned engine and RuleRegistry.

## Required sequence
1. Data Agent ingests and quality-checks available TASI evidence.
2. Analyzer produces independent Investment and Swing Top-X rankings.
3. Every surfaced stock has action, score, confidence, next target, exit/invalidation, horizon, reasons, contradictions and evidence provenance.
4. Recommendation batch is frozen in the immutable ledger.
5. Risk/Portfolio Agent treats the configured budget as real money and decides BUY/HOLD/ADD/TRIM/SELL/SKIP/DO NOTHING under cited rules.
6. Simulator records deterministic virtual orders/fills/positions/cash/costs without any brokerage call.
7. Luxury executive pre-open HTML email artifact is finalized for delivery exactly T-2h.
8. After verified close data, outcomes are appended without changing morning records.
9. Observer classifies recommendations RIGHT/WRONG/STILL OPEN where meaningful and records evidence-backed causes/findings.
10. Diagnoser prepares structured next-session context without mutating production weights.
11. Luxury executive post-close HTML accountability email is generated from the immutable morning batch plus realized outcomes.
12. Entire run can be replayed and traced from source data → features → rules → recommendation → simulation decision → email → outcome → finding.

## Failure conditions
Slice 0 fails if any of the following occurs:
- future-data leakage;
- fabricated/missing provenance for numeric facts;
- market mixing;
- recommendation overwritten after outcome;
- simulation action without rule citations/reason;
- optimistic hidden fill;
- duplicate email for the same market/session/report type;
- material stale/conflicting data silently accepted;
- Observer edits the production engine;
- manual database correction is required to finish the normal loop.

## Success evidence
A single run package must contain IDs/hashes for the analysis run, engine version, data cutoff, recommendation batch, evidence snapshots, feature snapshots, RuleRegistry version, simulation actions, portfolio snapshots, pre-open report, close observations, Observer findings, post-close report and replay verification.