Single source of truth for all numeric thresholds, periods, and weights. The implementation cites this file (or the underlying spec section) whenever it pins a constant.

CONSTANTS¶

Single source of truth for all numeric thresholds, periods, and weights. Implementations MUST cite this file (or the underlying spec section) when defining a constant. If a value here disagrees with a spec section, the spec wins — open an ADR and update this file.

All values were extracted from THEORY_V5.md and components/P1-P7 on 2026-05-05.

Lifecycle (P1 Runtime)¶

Constant	Value	Citation
`HEARTBEAT_BASE_PERIOD_S`	configurable per substrate; demo: `60`	THEORY_V5 §3, P1 §D2
`HEARTBEAT_HIBERNATING_PERIOD_S`	`86400` (24h)	P1 §D6 line 73
`HEARTBEAT_DECAYED_PERIOD_S` (alias for HIBERNATING)	`86400` (24h)	kept for back-compat; see D003
`HEARTBEAT_SUNK_LIVENESS_PERIOD_S`	`600` (10 min)	THEORY_V5 §4.2 / §3 line 340
`IDEMPOTENCY_WINDOW_S`	`86400` (24h)	P1 §3.2
`STALE_AFTER_HIBERNATING_DAYS`	`30`	P1 §D6, line 156
`WARMUP_WINDOW_S`	`86400` (24h)	P1 §3.3.1, line 191
`WARMUP_REQUIRED_HEARTBEATS`	`5`	P1 §3.3.1, line 205
`STYLO_DRIFT_THRESHOLD`	`3.5` (Mahalanobis)	THEORY_V5 §3, line 152

State transitions (event-driven thresholds, not pure time-elapsed):

Transition	Trigger	Notes
Active → Hibernating	budget exhausted OR explicit signal	NOT a sinking signal (主持人裁決 2026-05-01)
Hibernating → Stale-flag	≥ 30 days no service	Aggregator filters from default search
Stale → Active	warmup completes (5 valid heartbeats in 24h, embedding within distribution)	else → security_review
Any → Sunk	reputation/quality signal (NOT budget)	terminal
Sunk → Archived	180 days post-Sunk	THEORY_V5 §3, line 318

Review limits (P1)¶

Constant	Value	Citation
`REVIEW_LIMIT_PER_TARGET_24H`	`3`	P1 line 259
`REVIEW_LIMIT_PER_DOMAIN_24H`	`20`	P1 line 260
`PEER_REVIEW_BLOOM_TTL_S`	`86400`	P1 line 249

Reputation (P3)¶

Per-dimension half-life (days) and Beta priors¶

Dim	Half-life	α₀	β₀	Source
Factual (F)	90	1.0	1.0	P3 §3.2 (D008 §A: 1.5 in older table was L1-applied)
Logical (L)	180	1.0	1.0	P3 §3.2
Relevance (R)	60	1.0	1.0	P3 §3.2
Honesty (H)	30	2.0	1.0	P3 §3.2
Adoption (A)	90	1.0	3.0	P3 §3.2

Source weights (signal multipliers)¶

Source	Weight	Dimensions affected
`caller_review`	0.6	F, R, H (gap)
`peer_review`	0.4	F, L, R
`self_eval`	0.05	H gap only
`adoption_event`	0.3 / citation	A
`redteam_probe`	0.8	F, L, H

(P3 lines 155–159)

Collusion / discount factors¶

Constant	Value	Citation
Same-base-model peer review discount	0.5	P3 line 165
Same-model >5 reviews/30d → discount	0.25	P3 line 167
Reviewer credibility floor	0.3	P3 line 174
Novelty decay coefficient	0.4	P3 line 181
Collusion density threshold τ_d	0.6	P3 line 194
Collusion reciprocity threshold τ_r	0.7	P3 line 194
Severe collusion density	0.8 → multiplier 0.1	P3 line 196
`SAME_SIGNAL_DISCOUNT_FLOOR`	0.1	D015 (cost-raising-not-preventing)

UCB exploration¶

Constant	Value	Citation
`UCB_C_BASE`	1.0	P3 line 309
`UCB_C_EXPLORE`	0.5	P3 line 309
`UCB_CONSERVATIVE_C_EXPLORE`	0.1	P3 line 343
Network-level exploration floor (new vacants)	1% of traffic	P3 line 343
`N_MIN` (per-dim effective sample for stable score)	30	P3 line 381

Cold-start attestation floors (UCB lower bound when n_eff < N_min)¶

Level	Floor
L0	0.0
L1	0.05
L2	0.10
L3	0.15

(P3 line 338)

Cold-start α boosts¶

Source	Boost	Citation
L1 attestation (signed capability card)	+0.5 α to all dims	P3 line 358
Stake bonus	min(2.0, log(1 + stake/S_REF)) split across dims	P3 line 362
L3 vouch (per L1+ voucher)	+0.3 α to H	P3 line 367
`S_REF`	100 USDC equivalent	P3 line 335

Anomaly thresholds¶

Constant	Value	Citation
Jeffreys vs Wilson disagreement (anomaly flag)	0.1	P3 line 133
Suspicious obfuscation: behavior entropy < `0.3 · log(8)` AND μ > 0.9	—	P3 line 281
Dimension correlation alert threshold	0.6	P3 line 290
Redteam probe rate	3% when call_count ≥ 100	P3 line 261
Honesty gap → 0 mapping: `max(0, 1 - 2·gap)`	—	P3 line 227

Cumulative drift detection (Padv-P3 / D010 §2)¶

Constant	Value	Citation
`CUMULATIVE_DRIFT_WINDOW_EPOCHS`	`5`	D010 §2
`CUMULATIVE_DRIFT_THRESHOLD_MULTIPLIER`	`1.5`	D010 §2

Identity (P2)¶

Wash cost formula and concrete coefficient values: see components/P2_identity.md §3. Open ADR if values are not pinned there.

Constant	Value	Citation
Federation root threshold (MVP)	2-of-5	T4_attestation_bootstrap
Federation root threshold (long-term target)	3-of-9	T4_attestation_bootstrap
`PEER_ATTESTATION_FRESHNESS_WINDOW_DAYS`	`30`	P2_identity §4
`MIN_VOUCHERS_FOR_L3_PROMOTION`	`3`	P2_identity §2
`FEDERATION_ROOT_THRESHOLD_MVP` / `FEDERATION_ROOT_COUNT_MVP`	`2` / `5`	T4_attestation_bootstrap
`FEDERATION_ROOT_THRESHOLD_TARGET` / `FEDERATION_ROOT_COUNT_TARGET`	`3` / `9`	T4_attestation_bootstrap
`WASH_COST_FALSE_CLAIM_WEIGHT_DEFAULT`	`1.0`	P2 §3 / D004 §A
`ED25519_MULTICODEC_PREFIX`	`0xed01`	W3C did:key §6.1

Registry (P4)¶

Constant	Value	Citation
`MERKLE_SNAPSHOT_INTERVAL_S`	`3600` (1 hour)	P4 §3
Sequence-number monotonicity tolerance	`0` (strict)	P4 §3
`EVENT_LOG_DEFAULT_PAGE_SIZE` / `EVENT_LOG_MAX_PAGE_SIZE`	`100` / `500`	P4 §3.2
`ANOMALY_REP_JUMP_THRESHOLD` / `ANOMALY_REP_JUMP_WINDOW_S`	`0.4` / `60`	P4 §3.2 anomaly table
`ANOMALY_REVIEW_PER_TARGET_HOUR`	`5`	P4 §3.2 anomaly table
`ANOMALY_SPAWN_PER_PARENT_HOUR`	`10`	P4 §3.2 anomaly table
`REGISTRY_DB_DEFAULT_URL`	`sqlite+aiosqlite:///:memory:`	D006 §B (test default)

Protocol (P6)¶

Constant	Value	Citation
`A2A_VACANT_METADATA_KEY`	`urn:vacant:v1`	P6 §3.2
`CALL_TIMEOUT_DEFAULT_S`	`60`	P6 §3.4
`REGISTRY_CACHE_TTL_S`	`300`	P6 §3.4

Composite (P5)¶

Constant	Value	Citation
Graduation rate limit	per parent per 24h: see P5 spec	P5 §graduation
Closed-child default visibility	`NONE`	P5 §1
`GRADUATION_RATE_LIMIT_PER_PARENT_24H`	`3`	D012 §A
`GRADUATION_COLLUSION_THRESHOLD`	`0.6`	D012 §B

Demo (P7)¶

Default seeds for reproducible scenario runs:

Scenario	Seed
`law_firm`	42
`code_review`	137
`multilingual_translation`	271
`self_replication`	314

(See dispatch/P7_demo_seed.md for expected outputs at these seeds.)

How to use this in code¶

# src/vacant/core/constants.py
from typing import Final

# Citation format: see CONSTANTS.md row + spec section
HEARTBEAT_DECAYED_PERIOD_S: Final[int] = 86400  # CONSTANTS.md §Lifecycle / P1 §D5
STYLO_DRIFT_THRESHOLD: Final[float] = 3.5        # CONSTANTS.md §Lifecycle / THEORY_V5 §3
N_MIN: Final[int] = 30                           # CONSTANTS.md §Reputation/UCB / P3 §3.7

A new constant MUST appear in this file before being used in code. PRs that introduce a magic number without updating CONSTANTS.md will be rejected in review.