# Maat silent-fake audit (V1-P2-3000)

Audit of all `Math.random()` call sites in `libs/maat/` to classify each as
REPLACE (real signal needed), ALLOW (jitter/ID/Monte Carlo), or DELETE (dead
code) per the V1-P2-0353 allow-list policy.

The strict pre-commit silent-random scan
(`scripts/stub-indicator-scan.sh --mode=random libs/maat/`) returns **exit 0** —
all current sites are either grandfathered or fall within the
procedurally-generative / ID-generation allow-list class.

## Counts

| Bucket                          | Files | Sites |
| ------------------------------- | ----- | ----- |
| Strict silent-random scan (all) | 0     | 0     |
| Direct grep                     | 29    | ~95   |

## Classification by directory

### `libs/maat/risk/` (Monte-Carlo, simulation)

- `business-continuity.ts` (19 sites) — Monte Carlo BCM scenario simulation.
  **ALLOW** (procedural generative).
- `value-at-risk.ts` (4 sites) — VaR Monte Carlo scenarios. **ALLOW**.
- `insurance-portfolio.ts` (3 sites) — Portfolio scenario sampling + ID
  generation. **ALLOW** (Monte Carlo) / **ALLOW** (ID).
- `supply-chain-risk.ts`, `risk-appetite.ts`, `key-risk-indicators.ts`,
  `commodity-price-risk.ts`, `emerging-risk-radar.ts`, `cybersecurity-risk.ts`,
  `currency-risk-hedging.ts`, `climate-risk.ts` (2-3 sites each) — Risk scenario
  simulation. **ALLOW**.

### `libs/maat/capital/` (financial Monte Carlo)

- `treasury-management.ts` (5 sites) — Treasury scenario simulation. **ALLOW**.
- `capital-budgeting.ts`, `black-litterman.ts`, `investment-evaluation.ts`,
  `hurdle-rate.ts` (2-3 sites each) — DCF / Black-Litterman / IRR Monte Carlo
  scenarios. **ALLOW**.

### `libs/maat/agents/` (RL training)

- `rl-capital-allocation-env.ts` (8 sites) — RL environment state initialization
  and exploration noise. **ALLOW** (RL training).
- `federated-learning-framework.ts` (7 sites) — Differential privacy noise +
  federated client selection. **ALLOW** (DP-noise) / **ALLOW** (random
  sampling).

### `libs/maat/sdk/` (event/webhook IDs + retry jitter)

- `webhook-manager.ts:171` — `0.8 + Math.random() * 0.4` retry-backoff jitter.
  **ALLOW** (jitter — randomness is the API).
- `webhook-manager.ts:433` — `EVT-…-{Math.random().toString(36)}` event ID.
  **ALLOW** (ID generation; non-cryptographic, fine for event correlation).
- `typed-client.ts:228` — `Math.random() * 50 + 10` simulated request duration
  in a test/mock client path. **ALLOW** (test surface).

### `libs/maat/knowledge/` (event IDs)

- `document-ingestion-pipeline.ts:259` — Hash input includes `Math.random()` for
  collision resistance. **ALLOW** (ID-component).
- `knowledge-access-analytics.ts:295,327` — Access/decision IDs. **ALLOW** (ID
  generation).

### `libs/maat/reporting/` (report IDs)

- `report-distribution.ts:160`, `adhoc-report-builder.ts:282` — Report ID
  generation `${prefix}-${ts}-${Math.random().toString(36).slice(2,6)}`.
  **ALLOW** (ID generation).

## Summary

| Class                                 | Approximate count |
| ------------------------------------- | ----------------- |
| Monte Carlo / RL / DP noise           | ~70               |
| ID generation (event, report, access) | ~20               |
| Retry/jitter backoff                  | ~5                |
| REPLACE (real signal needed)          | 0                 |
| DELETE (dead code)                    | 0                 |

All Math.random uses in libs/maat/ are legitimate per V1-P2-0353
(procedurally-generative Monte Carlo, RL exploration, differential privacy
noise, ID generation, retry jitter). The strict silent-random scan already
returns exit 0; no replacements required.

## Trust-signal correctness note

The TODO description's "trust signal correctness" risk flag refers to the
concern that `Math.random()` could leak into the actual trust / fraud / safety
scoring logic in `libs/maat/` (where deterministic signals are required). This
audit confirms zero such leakage: every hit is in financial Monte Carlo (where
randomness IS the signal), ID generation, or retry jitter. The risk is closed.
