Limits & Defaults - TealTiger Docs

Quick reference for every default value, size limit, and configurable quota in TealTiger.

Engine Limits

Limit	Default	Configurable	Notes
Max modules per engine	No hard limit	—	Practical limit ~20; each adds evaluation latency
Max policy file size	No hard limit	—	Recommend < 1 MB for fast parsing
Max FREEZE rules	No hard limit	—	Evaluated sequentially in pre-evaluation
Max reason codes per decision	No hard limit	—	Typically 1–5 per evaluation
Risk score range	0–100	—	Computed from action severity
Correlation ID format	UUID v4	—	OpenTelemetry-compatible
Evaluation timeout	None	—	Bounded by slowest module
TEEC version	2.0.0	—	Evidence envelope schema
Sidecar max body size	1 MB	`TEALTIGER_MAX_BODY_BYTES`	HTTP 413 on exceed

TEEC Registry (v2.0.0)

Catalog	v1.2 Count	v1.3 Count	Description
Reason codes	32	60+	Across 12 categories
Event types	18	30+	Across 14 modules
Decision actions	12	16	Including PENDING
Governance domains	6	10	SEC, COST, REL, MEM, REG, EVID, NHI, FLOW, TEMP, DRIFT

Module Defaults

TealEngine v1.3

Setting	Default	Description
`mode`	`MONITOR`	Policy enforcement mode
`fail_behavior`	`fail_closed`	Behavior on internal error
`plan_only_mode`	`false`	Block all side-effects
Pre-evaluation order	FREEZE → NHI → Scope → JIT → Temporal	Sequential, short-circuit on deny

TealSecrets

Setting	Default	Description
`confidence_threshold`	0.8	Minimum confidence (0–1) to report
`max_content_size`	1 MB	Larger content skips scan
Pattern count	500+	Built-in detection patterns

TealClassifier (v1.3)

Setting	Default	Description
`ensemble_mode`	`regex_only`	v1.2 compatible default
`confidence_threshold`	0.7	ML classification threshold
`max_tokens`	512	Max input tokens for ONNX inference
Inference latency target	< 5ms	CPU, 512 tokens
Fallback on model failure	`regex_only`	Emits `CLASSIFIER_FALLBACK`

TealDrift (v1.3)

Setting	Default	Description
`min_samples`	100	Samples before alerting activates
`baseline_window`	7 days	Rolling window for baseline
`sensitivity`	`medium`	low / medium / high
`threshold_sigma`	3.0	Standard deviations for alert

TealTemporal (v1.3)

Setting	Default	Description
`session_ttl`	8 hours	Default session lifetime
`max_session_ttl`	24 hours	Maximum configurable TTL
`cooldown_default`	None	No cooldown unless configured
`timezone`	`UTC`	Default for time-of-day restrictions

TealState (v1.3)

Setting	Default	Description
`max_context_tokens`	8192	Default context size limit
`warning_threshold`	0.8	Warn at 80% capacity
`on_exceeded`	`deny`	deny / truncate_oldest / truncate_lowest_priority
`default_ttl`	1 hour	Default entry TTL

TealMonitor v2 (v1.3)

Setting	Default	Description
`per_request_max`	None	No per-request limit unless set
`daily_per_agent_max`	None	No per-agent daily limit unless set
`reasoning_token_budget`	None	No reasoning budget unless set
`anomaly_sensitivity`	`medium`	low / medium / high
`anomaly_baseline_window`	7 days	Rolling baseline for anomaly detection

TealProof (v1.3)

Setting	Default	Description
`merkle_algorithm`	SHA-256	Hash algorithm for Merkle tree
`max_leaves`	10,000	Leaves before tree rotation
`tsa_schedule`	Every 1000 decisions or 5 min	RFC 3161 anchor frequency
`passport_window`	1 hour	Governance Passport window size

TealFlow (v1.3)

Setting	Default	Description
Max jobs per workflow	No hard limit	Practical limit ~50
Max dependency depth	No hard limit	Circular deps rejected at validation
Floor enforcement	Enabled	Org controls cannot be weakened

NHI Governance (v1.3)

Setting	Default	Description
`enforce_scopes`	`true`	Scope enforcement enabled
`enforce_environment`	`true`	Environment binding enabled
`require_attestation`	`false`	Attestation optional by default
`max_grant_duration`	1 hour	Maximum JIT grant lifetime
`clock_skew_tolerance`	30 seconds	Attestation token clock skew

TealMemory

Setting	Default	Description
`default_scope`	`session`	Default memory scope
`classification`	`internal`	Default classification level
Available scopes	5	request, session, agent, tenant, global
Available classifications	4	public, internal, confidential, restricted

TealReliability

Setting	Default	Description
`retry_budget`	3	Max retries before deny
`circuit_threshold`	5	Failures before circuit opens
`circuit_reset_timeout`	30,000 ms	Time before half-open

Performance Targets

Metric	Target	Notes
Basic evaluation (4 modules)	< 15ms p99	Parallel via `Promise.allSettled`
Full v1.3 pipeline (FREEZE + NHI + modules + hooks)	< 25ms p99	End-to-end
PLAN_ONLY mode	< 10ms p99	Skips evidence emission
TealClassifier (ONNX, CPU)	< 5ms p99	512 tokens
TealDrift evaluation	< 3ms p99	Statistical comparison
TealProof receipt generation	< 2ms p99	SHA-256 + Merkle append
Pre-evaluation (FREEZE + NHI)	< 2ms	In-memory checks
Post-hooks (async)	< 5ms	Non-blocking
Sidecar health check	< 5ms	`/health` endpoint
Module lazy-init	< 50ms	Per module, first evaluation only

Provider Support (12)

Provider	TypeScript	Python	New in v1.3
OpenAI	✅	✅
Anthropic	✅	✅
Google Gemini	✅	✅
AWS Bedrock	✅	✅
Azure OpenAI	✅	✅
Cohere	✅	✅
Mistral	✅	✅
DeepSeek	✅	✅	✅
Groq	✅	✅	✅
Together AI	✅	✅	✅
xAI (Grok)	✅	✅	✅
HuggingFace TGI	✅	✅	✅

Action Severity Scores

Used by the “most restrictive wins” merge strategy:

Action	Severity	Dimensions
`DENY`	100	all
`DENY_WRITE`	100	memory
`DENY_READ`	100	memory
`REQUIRE_APPROVAL` / `PENDING`	80	all
`REDACT`	70	security, content
`REDACT_AND_WRITE`	70	memory
`DEGRADE`	60	cost, reliability
`STORE_SUMMARY_ONLY`	60	memory
`MODIFY`	50	security, content
`ALLOW_WRITE`	0	memory
`ALLOW`	0	all