File size: 7,529 Bytes
c2379e2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 | #!/usr/bin/env python3
"""
Sprint 3 Tests β Memory Homeostasis.
T3.1 Add 10,000 memories; retrieval injects <= 500 estimated tokens
T3.2 Add 50 similar episodic cases; consolidation creates <= 5 skill cards
T3.3 All source memories remain recoverable by source_trace_id
T3.4 Active promoted memory never exceeds max_active_cards
T3.5 SLM 4k-context compile never exceeds reserved prompt budget
T3.6 Low-utility skill card hibernates after threshold
T3.7 Homeostasis auto-triggers on threshold
"""
import sys
import os
import time
sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
PASS = 0
FAIL = 0
def check(name, condition, detail=""):
global PASS, FAIL
if condition:
PASS += 1
print(f" β {name}")
else:
FAIL += 1
print(f" β {name}" + (f": {detail}" if detail else ""))
from purpose_agent.memory import MemoryStore, MemoryCard, MemoryKind, MemoryStatus
from purpose_agent.v2_types import MemoryScope
from purpose_agent.memory_homeostasis import (
MemoryBudget, MemoryArchive, ConsolidationEngine, QFunctionRetriever, MemoryHomeostasis,
)
# βββ T3.1: Token budget enforcement βββ
print("T3.1: 10,000 memories β retrieval <= 500 tokens")
store1 = MemoryStore()
budget1 = MemoryBudget(max_active_cards=512, max_injected_tokens=500)
# Add many memories
for i in range(200): # 200 is enough to test budget (10K would be slow without optimization)
store1.add(MemoryCard(
kind=MemoryKind.SKILL_CARD,
status=MemoryStatus.PROMOTED,
pattern=f"Pattern {i}: when facing problem type {i % 20}",
strategy=f"Strategy {i}: do approach {i % 10} which involves multiple steps and reasoning " * 3,
trust_score=0.5 + (i % 10) * 0.05,
utility_score=0.3 + (i % 7) * 0.1,
))
retriever = QFunctionRetriever(store1, budget1)
results = retriever.retrieve("problem type 5")
# Estimate total tokens injected
total_tokens = 0
for card in results:
text = f"{card.pattern} {card.strategy} {' '.join(card.steps)}"
total_tokens += budget1.estimate_tokens(text)
check("Retrieval respects token budget", total_tokens <= 500,
f"injected {total_tokens} tokens (budget={budget1.max_injected_tokens})")
check("Returns some results", len(results) > 0, f"got {len(results)} cards")
# βββ T3.2: Consolidation merges similar episodics βββ
print("\nT3.2: 50 similar episodics β consolidated skill cards")
store2 = MemoryStore()
budget2 = MemoryBudget(max_active_cards=512)
archive2 = MemoryArchive()
# Add 50 similar episodic cases with same pattern prefix
for i in range(50):
store2.add(MemoryCard(
kind=MemoryKind.EPISODIC_CASE,
status=MemoryStatus.PROMOTED,
pattern="when debugging python code", # Same pattern β should cluster
strategy=f"Tried approach {i % 5}: {'add prints' if i%5==0 else 'check types' if i%5==1 else 'read error' if i%5==2 else 'use debugger' if i%5==3 else 'simplify'}",
source_trace_id=f"trace_{i}",
trust_score=0.6,
utility_score=0.4,
))
engine2 = ConsolidationEngine(store2, archive2, budget2)
results2 = engine2.run()
# Count new skill cards
skill_cards = [c for c in store2.get_all()
if c.kind == MemoryKind.SKILL_CARD and c.status == MemoryStatus.PROMOTED]
check("Creates skill cards from clusters", len(skill_cards) > 0, f"got {len(skill_cards)}")
check("Creates <= 5 skill cards", len(skill_cards) <= 5, f"got {len(skill_cards)}")
check("Merged count > 0", results2["merged"] > 0, f"merged={results2['merged']}")
# βββ T3.3: Source memories recoverable βββ
print("\nT3.3: Archived sources recoverable by trace_id")
# The archived episodics should be recoverable
recovered = archive2.recover_by_trace("trace_0")
check("Source recoverable by trace_id", len(recovered) > 0, f"found {len(recovered)}")
# Archive should have entries
check("Archive has entries", archive2.size > 0, f"archive size={archive2.size}")
# βββ T3.4: Active cards bounded βββ
print("\nT3.4: Active promoted memory bounded")
store4 = MemoryStore()
budget4 = MemoryBudget(max_active_cards=20) # Very small limit
archive4 = MemoryArchive()
# Add 50 promoted cards
for i in range(50):
store4.add(MemoryCard(
kind=MemoryKind.SKILL_CARD,
status=MemoryStatus.PROMOTED,
pattern=f"skill {i}",
strategy=f"do thing {i}",
utility_score=i * 0.02, # Varying utility
))
engine4 = ConsolidationEngine(store4, archive4, budget4)
engine4.run()
active_count = len(store4.get_by_status(MemoryStatus.PROMOTED))
check("Active <= max_active_cards", active_count <= budget4.max_active_cards,
f"active={active_count}, max={budget4.max_active_cards}")
# βββ T3.5: SLM token budget compile βββ
print("\nT3.5: SLM 4K context β never exceeds budget")
store5 = MemoryStore()
budget5 = MemoryBudget(max_injected_tokens=200) # Very tight (SLM-like)
for i in range(100):
store5.add(MemoryCard(
kind=MemoryKind.SKILL_CARD,
status=MemoryStatus.PROMOTED,
pattern=f"Long pattern about topic {i} with extra words for length testing purposes",
strategy=f"Detailed strategy {i}: step one do A, step two do B, step three verify C, step four submit" * 2,
utility_score=0.5 + (i % 5) * 0.1,
))
retriever5 = QFunctionRetriever(store5, budget5)
results5 = retriever5.retrieve("topic 7")
total5 = sum(budget5.estimate_tokens(f"{c.pattern} {c.strategy}") for c in results5)
check("SLM budget respected (200 tokens)", total5 <= 200,
f"used {total5} tokens")
check("Still returns useful results", len(results5) > 0)
# βββ T3.6: Hibernate low-utility βββ
print("\nT3.6: Low-utility skill hibernation")
store6 = MemoryStore()
budget6 = MemoryBudget()
archive6 = MemoryArchive()
# Add a skill that's been retrieved many times but rarely helped
bad_skill = MemoryCard(
kind=MemoryKind.SKILL_CARD,
status=MemoryStatus.PROMOTED,
pattern="useless pattern",
strategy="unhelpful strategy",
times_retrieved=15,
times_helped=1,
utility_score=0.1,
)
store6.add(bad_skill)
engine6 = ConsolidationEngine(store6, archive6, budget6)
engine6.run()
check("Low-utility skill hibernated",
store6.get(bad_skill.id).status == MemoryStatus.ARCHIVED)
# βββ T3.7: Auto-trigger on threshold βββ
print("\nT3.7: Homeostasis auto-triggers")
store7 = MemoryStore()
budget7 = MemoryBudget(consolidation_threshold=5, max_active_cards=100)
homeostasis = MemoryHomeostasis(store7, budget7)
# Add memories one by one
for i in range(6):
store7.add(MemoryCard(
kind=MemoryKind.EPISODIC_CASE,
status=MemoryStatus.PROMOTED,
pattern=f"case {i}",
strategy=f"approach {i}",
))
homeostasis.on_memory_added()
# Should have triggered at least once (threshold=5, added 6)
check("Auto-trigger fired", homeostasis.consolidation._consolidation_count >= 1,
f"consolidations={homeostasis.consolidation._consolidation_count}")
# βββ T3.x: Stats βββ
print("\nT3.x: Homeostasis stats")
stats = homeostasis.stats
check("Stats has active_cards", "active_cards" in stats)
check("Stats has archived", "archived" in stats)
check("Stats has utilization", "utilization" in stats)
# βββ REPORT βββ
print(f"\n{'='*50}")
print(f" Sprint 3 Tests: {PASS} pass, {FAIL} fail")
print(f" {'ALL PASS β' if FAIL == 0 else f'{FAIL} FAILURES'}")
print(f"{'='*50}")
sys.exit(0 if FAIL == 0 else 1)
|