Skip to content

Evidence

Evidence-based reasoning for architecture decisions.

Overview

The evidence system supports confidence-weighted reasoning about architectural properties.

Class Documentation

upir.core.evidence.Evidence dataclass

A piece of evidence supporting or refuting an architectural decision.

Evidence can come from various sources (benchmarks, tests, production data, formal proofs) and carries a Bayesian confidence level that can be updated as new observations arrive.

Based on the TD Commons disclosure, UPIR tracks evidence with confidence levels that propagate through reasoning graphs. The Bayesian update implements a simple beta-binomial conjugate prior approach.

Attributes:

Name Type Description
source str

Where the evidence came from (e.g., "load_test_2024-01", "formal_verification", "production_metrics")
type str

Type of evidence - one of: benchmark, test, production, formal_proof
data Dict[str, Any]

The actual evidence data (metrics, test results, proof artifacts)
confidence float

Bayesian confidence level in [0, 1]
timestamp datetime

When the evidence was collected (UTC)
Example

evidence = Evidence( ... source="load_test_2024-01", ... type="benchmark", ... data={"latency_p99": 95, "throughput": 10000}, ... confidence=0.8, ... timestamp=datetime.utcnow() ... ) evidence.update_confidence(new_observation=True) evidence.confidence > 0.8 # Confidence increased True

References: - TD Commons: Evidence tracking structure - Bayesian inference: Beta-binomial conjugate prior

Source code in upir/core/evidence.py 
 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
@dataclass
class Evidence:
    """
    A piece of evidence supporting or refuting an architectural decision.

    Evidence can come from various sources (benchmarks, tests, production data,
    formal proofs) and carries a Bayesian confidence level that can be updated
    as new observations arrive.

    Based on the TD Commons disclosure, UPIR tracks evidence with confidence
    levels that propagate through reasoning graphs. The Bayesian update
    implements a simple beta-binomial conjugate prior approach.

    Attributes:
        source: Where the evidence came from (e.g., "load_test_2024-01",
                "formal_verification", "production_metrics")
        type: Type of evidence - one of: benchmark, test, production, formal_proof
        data: The actual evidence data (metrics, test results, proof artifacts)
        confidence: Bayesian confidence level in [0, 1]
        timestamp: When the evidence was collected (UTC)

    Example:
        >>> evidence = Evidence(
        ...     source="load_test_2024-01",
        ...     type="benchmark",
        ...     data={"latency_p99": 95, "throughput": 10000},
        ...     confidence=0.8,
        ...     timestamp=datetime.utcnow()
        ... )
        >>> evidence.update_confidence(new_observation=True)
        >>> evidence.confidence > 0.8  # Confidence increased
        True

    References:
    - TD Commons: Evidence tracking structure
    - Bayesian inference: Beta-binomial conjugate prior
    """
    source: str
    type: str
    data: Dict[str, Any]
    confidence: float
    timestamp: datetime = field(default_factory=datetime.utcnow)

    def __post_init__(self):
        """Validate evidence fields."""
        valid_types = {"benchmark", "test", "production", "formal_proof"}
        if self.type not in valid_types:
            raise ValueError(
                f"Invalid evidence type '{self.type}'. "
                f"Must be one of: {valid_types}"
            )

        if not 0 <= self.confidence <= 1:
            raise ValueError(
                f"Confidence must be in [0, 1], got {self.confidence}"
            )

        if not self.source:
            raise ValueError("Source cannot be empty")

    def update_confidence(
        self,
        new_observation: bool,
        prior_weight: float = 0.1
    ) -> None:
        """
        Update confidence using Bayesian update based on new observation.

        This implements a simple Bayesian update using a beta-binomial conjugate
        prior. The prior_weight controls how much the new observation affects
        the current confidence.

        Update rules:
        - Positive observation: confidence += prior_weight * (1 - confidence)
        - Negative observation: confidence *= (1 - prior_weight)

        These rules ensure:
        1. Confidence stays in [0, 1]
        2. Positive observations increase confidence (asymptotically to 1)
        3. Negative observations decrease confidence (multiplicatively)
        4. Higher current confidence is harder to change (conservative)

        Args:
            new_observation: True if observation supports the evidence,
                           False if it contradicts it
            prior_weight: Weight of the prior in [0, 1]. Higher values mean
                         new observations have more impact. Default 0.1.

        Raises:
            ValueError: If prior_weight not in [0, 1]

        Example:
            >>> evidence = Evidence(
            ...     source="test",
            ...     type="benchmark",
            ...     data={},
            ...     confidence=0.5
            ... )
            >>> evidence.update_confidence(new_observation=True)
            >>> evidence.confidence
            0.55
            >>> evidence.update_confidence(new_observation=False)
            >>> evidence.confidence
            0.495

        References:
        - Beta-binomial conjugate prior: Standard Bayesian approach for
          binary observations
        - Murphy (2006): Bayesian inference for Bernoulli distribution
        """
        if not 0 <= prior_weight <= 1:
            raise ValueError(
                f"prior_weight must be in [0, 1], got {prior_weight}"
            )

        if new_observation:
            # Positive observation: move confidence towards 1
            # confidence += learning_rate * (target - current)
            self.confidence = self.confidence + prior_weight * (1 - self.confidence)
        else:
            # Negative observation: decay confidence
            # confidence *= (1 - decay_rate)
            self.confidence = self.confidence * (1 - prior_weight)

        # Ensure confidence stays in valid range (should be guaranteed by math)
        self.confidence = max(0.0, min(1.0, self.confidence))

    def to_dict(self) -> Dict[str, Any]:
        """
        Serialize evidence to JSON-compatible dictionary.

        Returns:
            Dictionary with all evidence fields

        Example:
            >>> evidence = Evidence(
            ...     source="test",
            ...     type="benchmark",
            ...     data={"metric": 100},
            ...     confidence=0.8,
            ...     timestamp=datetime(2024, 1, 1, 12, 0, 0)
            ... )
            >>> d = evidence.to_dict()
            >>> d["source"]
            'test'
        """
        return {
            "source": self.source,
            "type": self.type,
            "data": self.data.copy(),
            "confidence": self.confidence,
            "timestamp": self.timestamp.isoformat()
        }

    @classmethod
    def from_dict(cls, data: Dict[str, Any]) -> "Evidence":
        """
        Deserialize evidence from dictionary.

        Args:
            data: Dictionary containing evidence fields

        Returns:
            Evidence instance

        Example:
            >>> data = {
            ...     "source": "test",
            ...     "type": "benchmark",
            ...     "data": {"metric": 100},
            ...     "confidence": 0.8,
            ...     "timestamp": "2024-01-01T12:00:00"
            ... }
            >>> evidence = Evidence.from_dict(data)
            >>> evidence.source
            'test'
        """
        return cls(
            source=data["source"],
            type=data["type"],
            data=data["data"],
            confidence=data["confidence"],
            timestamp=datetime.fromisoformat(data["timestamp"])
        )

    def __str__(self) -> str:
        """Human-readable string representation."""
        return (
            f"Evidence({self.type} from {self.source}, "
            f"confidence={self.confidence:.2f})"
        )

Functions

__post_init__()

Validate evidence fields.

Source code in upir/core/evidence.py 
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
def __post_init__(self):
    """Validate evidence fields."""
    valid_types = {"benchmark", "test", "production", "formal_proof"}
    if self.type not in valid_types:
        raise ValueError(
            f"Invalid evidence type '{self.type}'. "
            f"Must be one of: {valid_types}"
        )

    if not 0 <= self.confidence <= 1:
        raise ValueError(
            f"Confidence must be in [0, 1], got {self.confidence}"
        )

    if not self.source:
        raise ValueError("Source cannot be empty")

update_confidence(new_observation, prior_weight=0.1)

Update confidence using Bayesian update based on new observation.

This implements a simple Bayesian update using a beta-binomial conjugate prior. The prior_weight controls how much the new observation affects the current confidence.

Update rules: - Positive observation: confidence += prior_weight * (1 - confidence) - Negative observation: confidence *= (1 - prior_weight)

These rules ensure: 1. Confidence stays in [0, 1] 2. Positive observations increase confidence (asymptotically to 1) 3. Negative observations decrease confidence (multiplicatively) 4. Higher current confidence is harder to change (conservative)

Parameters:

Name Type Description Default
new_observation bool

True if observation supports the evidence, False if it contradicts it

 required
prior_weight float

Weight of the prior in [0, 1]. Higher values mean new observations have more impact. Default 0.1.

 0.1

Raises:

Type Description
ValueError

If prior_weight not in [0, 1]
Example

evidence = Evidence( ... source="test", ... type="benchmark", ... data={}, ... confidence=0.5 ... ) evidence.update_confidence(new_observation=True) evidence.confidence 0.55 evidence.update_confidence(new_observation=False) evidence.confidence 0.495

References: - Beta-binomial conjugate prior: Standard Bayesian approach for binary observations - Murphy (2006): Bayesian inference for Bernoulli distribution

Source code in upir/core/evidence.py 
 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
def update_confidence(
    self,
    new_observation: bool,
    prior_weight: float = 0.1
) -> None:
    """
    Update confidence using Bayesian update based on new observation.

    This implements a simple Bayesian update using a beta-binomial conjugate
    prior. The prior_weight controls how much the new observation affects
    the current confidence.

    Update rules:
    - Positive observation: confidence += prior_weight * (1 - confidence)
    - Negative observation: confidence *= (1 - prior_weight)

    These rules ensure:
    1. Confidence stays in [0, 1]
    2. Positive observations increase confidence (asymptotically to 1)
    3. Negative observations decrease confidence (multiplicatively)
    4. Higher current confidence is harder to change (conservative)

    Args:
        new_observation: True if observation supports the evidence,
                       False if it contradicts it
        prior_weight: Weight of the prior in [0, 1]. Higher values mean
                     new observations have more impact. Default 0.1.

    Raises:
        ValueError: If prior_weight not in [0, 1]

    Example:
        >>> evidence = Evidence(
        ...     source="test",
        ...     type="benchmark",
        ...     data={},
        ...     confidence=0.5
        ... )
        >>> evidence.update_confidence(new_observation=True)
        >>> evidence.confidence
        0.55
        >>> evidence.update_confidence(new_observation=False)
        >>> evidence.confidence
        0.495

    References:
    - Beta-binomial conjugate prior: Standard Bayesian approach for
      binary observations
    - Murphy (2006): Bayesian inference for Bernoulli distribution
    """
    if not 0 <= prior_weight <= 1:
        raise ValueError(
            f"prior_weight must be in [0, 1], got {prior_weight}"
        )

    if new_observation:
        # Positive observation: move confidence towards 1
        # confidence += learning_rate * (target - current)
        self.confidence = self.confidence + prior_weight * (1 - self.confidence)
    else:
        # Negative observation: decay confidence
        # confidence *= (1 - decay_rate)
        self.confidence = self.confidence * (1 - prior_weight)

    # Ensure confidence stays in valid range (should be guaranteed by math)
    self.confidence = max(0.0, min(1.0, self.confidence))

to_dict()

Serialize evidence to JSON-compatible dictionary.

Returns:

Type Description
Dict[str, Any]

Dictionary with all evidence fields
Example

evidence = Evidence( ... source="test", ... type="benchmark", ... data={"metric": 100}, ... confidence=0.8, ... timestamp=datetime(2024, 1, 1, 12, 0, 0) ... ) d = evidence.to_dict() d["source"] 'test'

Source code in upir/core/evidence.py 
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
def to_dict(self) -> Dict[str, Any]:
    """
    Serialize evidence to JSON-compatible dictionary.

    Returns:
        Dictionary with all evidence fields

    Example:
        >>> evidence = Evidence(
        ...     source="test",
        ...     type="benchmark",
        ...     data={"metric": 100},
        ...     confidence=0.8,
        ...     timestamp=datetime(2024, 1, 1, 12, 0, 0)
        ... )
        >>> d = evidence.to_dict()
        >>> d["source"]
        'test'
    """
    return {
        "source": self.source,
        "type": self.type,
        "data": self.data.copy(),
        "confidence": self.confidence,
        "timestamp": self.timestamp.isoformat()
    }

from_dict(data) classmethod

Deserialize evidence from dictionary.

Parameters:

Name Type Description Default
data Dict[str, Any]

Dictionary containing evidence fields

 required

Returns:

Type Description
Evidence

Evidence instance
Example

data = { ... "source": "test", ... "type": "benchmark", ... "data": {"metric": 100}, ... "confidence": 0.8, ... "timestamp": "2024-01-01T12:00:00" ... } evidence = Evidence.from_dict(data) evidence.source 'test'

Source code in upir/core/evidence.py 
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
@classmethod
def from_dict(cls, data: Dict[str, Any]) -> "Evidence":
    """
    Deserialize evidence from dictionary.

    Args:
        data: Dictionary containing evidence fields

    Returns:
        Evidence instance

    Example:
        >>> data = {
        ...     "source": "test",
        ...     "type": "benchmark",
        ...     "data": {"metric": 100},
        ...     "confidence": 0.8,
        ...     "timestamp": "2024-01-01T12:00:00"
        ... }
        >>> evidence = Evidence.from_dict(data)
        >>> evidence.source
        'test'
    """
    return cls(
        source=data["source"],
        type=data["type"],
        data=data["data"],
        confidence=data["confidence"],
        timestamp=datetime.fromisoformat(data["timestamp"])
    )

__str__()

Human-readable string representation.

Source code in upir/core/evidence.py 
210
211
212
213
214
215
def __str__(self) -> str:
    """Human-readable string representation."""
    return (
        f"Evidence({self.type} from {self.source}, "
        f"confidence={self.confidence:.2f})"
    )

upir.core.evidence.ReasoningNode dataclass

A node in the reasoning graph representing an architectural decision.

Reasoning nodes form a directed acyclic graph (DAG) where each node represents a decision or conclusion, supported by evidence and potentially dependent on other decisions (parent nodes).

Based on the TD Commons disclosure, UPIR maintains a reasoning graph to track decision provenance and propagate confidence through the architecture.

Attributes:

Name Type Description
id str

Unique identifier (UUID)
decision str

The decision or conclusion made
rationale str

Explanation of why this decision was made
evidence_ids List[str]

IDs of Evidence objects supporting this decision
parent_ids List[str]

IDs of other ReasoningNodes this depends on (for DAG)
alternatives List[Dict[str, Any]]

Other options that were considered but not chosen
confidence float

Computed confidence in this decision [0, 1]
Example

node = ReasoningNode( ... id=str(uuid.uuid4()), ... decision="Use PostgreSQL for primary database", ... rationale="Strong consistency needed for financial transactions", ... evidence_ids=["evidence-1", "evidence-2"], ... parent_ids=["node-consistency-requirement"], ... alternatives=[ ... {"option": "MongoDB", "rejected_because": "eventual consistency"} ... ], ... confidence=0.0 # Will be computed from evidence ... )

References: - TD Commons: Reasoning graph structure - DAG: Directed acyclic graph for decision dependencies

Source code in upir/core/evidence.py 
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
@dataclass
class ReasoningNode:
    """
    A node in the reasoning graph representing an architectural decision.

    Reasoning nodes form a directed acyclic graph (DAG) where each node
    represents a decision or conclusion, supported by evidence and potentially
    dependent on other decisions (parent nodes).

    Based on the TD Commons disclosure, UPIR maintains a reasoning graph to
    track decision provenance and propagate confidence through the architecture.

    Attributes:
        id: Unique identifier (UUID)
        decision: The decision or conclusion made
        rationale: Explanation of why this decision was made
        evidence_ids: IDs of Evidence objects supporting this decision
        parent_ids: IDs of other ReasoningNodes this depends on (for DAG)
        alternatives: Other options that were considered but not chosen
        confidence: Computed confidence in this decision [0, 1]

    Example:
        >>> node = ReasoningNode(
        ...     id=str(uuid.uuid4()),
        ...     decision="Use PostgreSQL for primary database",
        ...     rationale="Strong consistency needed for financial transactions",
        ...     evidence_ids=["evidence-1", "evidence-2"],
        ...     parent_ids=["node-consistency-requirement"],
        ...     alternatives=[
        ...         {"option": "MongoDB", "rejected_because": "eventual consistency"}
        ...     ],
        ...     confidence=0.0  # Will be computed from evidence
        ... )

    References:
    - TD Commons: Reasoning graph structure
    - DAG: Directed acyclic graph for decision dependencies
    """
    id: str
    decision: str
    rationale: str
    evidence_ids: List[str] = field(default_factory=list)
    parent_ids: List[str] = field(default_factory=list)
    alternatives: List[Dict[str, Any]] = field(default_factory=list)
    confidence: float = 0.0

    def __post_init__(self):
        """Validate reasoning node fields."""
        if not self.id:
            raise ValueError("ID cannot be empty")

        if not self.decision:
            raise ValueError("Decision cannot be empty")

        if not 0 <= self.confidence <= 1:
            raise ValueError(
                f"Confidence must be in [0, 1], got {self.confidence}"
            )

    @staticmethod
    def generate_id() -> str:
        """
        Generate a unique ID for a reasoning node.

        Returns:
            UUID string

        Example:
            >>> node_id = ReasoningNode.generate_id()
            >>> len(node_id) == 36  # UUID format
            True
        """
        return str(uuid.uuid4())

    def compute_confidence(self, evidence_map: Dict[str, Evidence]) -> float:
        """
        Compute aggregate confidence from supporting evidence using geometric mean.

        The geometric mean is more conservative than arithmetic mean - a single
        piece of low-confidence evidence significantly reduces overall confidence.
        This matches how engineers actually reason: one weak piece of evidence
        can't be compensated by many strong pieces.

        Formula: exp(mean(log(c_i))) for confidences c_i

        If no evidence is available, returns 0.0 (no confidence).
        If any evidence has confidence 0, returns 0.0 (geometric mean property).

        Args:
            evidence_map: Mapping from evidence IDs to Evidence objects

        Returns:
            Computed confidence in [0, 1]

        Example:
            >>> evidence_map = {
            ...     "e1": Evidence("src1", "test", {}, 0.8, datetime.utcnow()),
            ...     "e2": Evidence("src2", "test", {}, 0.9, datetime.utcnow())
            ... }
            >>> node = ReasoningNode(
            ...     id="node-1",
            ...     decision="test",
            ...     rationale="test",
            ...     evidence_ids=["e1", "e2"]
            ... )
            >>> conf = node.compute_confidence(evidence_map)
            >>> 0.84 < conf < 0.85  # sqrt(0.8 * 0.9) ≈ 0.8485
            True

        References:
        - Geometric mean: https://en.wikipedia.org/wiki/Geometric_mean
        - More conservative than arithmetic mean for combining confidences
        """
        # Filter evidence to only those referenced by this node
        relevant_evidence = [
            evidence_map[eid]
            for eid in self.evidence_ids
            if eid in evidence_map
        ]

        if not relevant_evidence:
            return 0.0

        # Check for any zero confidence (geometric mean would be 0)
        confidences = [e.confidence for e in relevant_evidence]
        if any(c == 0.0 for c in confidences):
            return 0.0

        # Compute geometric mean: exp(mean(log(c_i)))
        log_sum = sum(math.log(c) for c in confidences)
        log_mean = log_sum / len(confidences)
        geometric_mean = math.exp(log_mean)

        return geometric_mean

    def to_dict(self) -> Dict[str, Any]:
        """
        Serialize reasoning node to JSON-compatible dictionary.

        Returns:
            Dictionary with all node fields

        Example:
            >>> node = ReasoningNode(
            ...     id="node-1",
            ...     decision="Use caching",
            ...     rationale="Reduce latency",
            ...     evidence_ids=["e1"],
            ...     parent_ids=[],
            ...     alternatives=[{"option": "No cache"}],
            ...     confidence=0.8
            ... )
            >>> d = node.to_dict()
            >>> d["decision"]
            'Use caching'
        """
        return {
            "id": self.id,
            "decision": self.decision,
            "rationale": self.rationale,
            "evidence_ids": self.evidence_ids.copy(),
            "parent_ids": self.parent_ids.copy(),
            "alternatives": [alt.copy() for alt in self.alternatives],
            "confidence": self.confidence
        }

    @classmethod
    def from_dict(cls, data: Dict[str, Any]) -> "ReasoningNode":
        """
        Deserialize reasoning node from dictionary.

        Args:
            data: Dictionary containing node fields

        Returns:
            ReasoningNode instance

        Example:
            >>> data = {
            ...     "id": "node-1",
            ...     "decision": "Use caching",
            ...     "rationale": "Reduce latency",
            ...     "evidence_ids": ["e1"],
            ...     "parent_ids": [],
            ...     "alternatives": [],
            ...     "confidence": 0.8
            ... }
            >>> node = ReasoningNode.from_dict(data)
            >>> node.decision
            'Use caching'
        """
        return cls(
            id=data["id"],
            decision=data["decision"],
            rationale=data["rationale"],
            evidence_ids=data.get("evidence_ids", []),
            parent_ids=data.get("parent_ids", []),
            alternatives=data.get("alternatives", []),
            confidence=data.get("confidence", 0.0)
        )

    def __str__(self) -> str:
        """Human-readable string representation."""
        return (
            f"ReasoningNode({self.decision}, "
            f"confidence={self.confidence:.2f}, "
            f"evidence={len(self.evidence_ids)})"
        )

    def __repr__(self) -> str:
        """Developer-friendly representation."""
        return (
            f"ReasoningNode(id='{self.id}', "
            f"decision='{self.decision}', "
            f"confidence={self.confidence}, "
            f"evidence_count={len(self.evidence_ids)}, "
            f"parent_count={len(self.parent_ids)})"
        )

Functions

__post_init__()

Validate reasoning node fields.

Source code in upir/core/evidence.py 
264
265
266
267
268
269
270
271
272
273
274
275
def __post_init__(self):
    """Validate reasoning node fields."""
    if not self.id:
        raise ValueError("ID cannot be empty")

    if not self.decision:
        raise ValueError("Decision cannot be empty")

    if not 0 <= self.confidence <= 1:
        raise ValueError(
            f"Confidence must be in [0, 1], got {self.confidence}"
        )

generate_id() staticmethod

Generate a unique ID for a reasoning node.

Returns:

Type Description
str

UUID string
Example

node_id = ReasoningNode.generate_id() len(node_id) == 36 # UUID format True

Source code in upir/core/evidence.py 
277
278
279
280
281
282
283
284
285
286
287
288
289
290
@staticmethod
def generate_id() -> str:
    """
    Generate a unique ID for a reasoning node.

    Returns:
        UUID string

    Example:
        >>> node_id = ReasoningNode.generate_id()
        >>> len(node_id) == 36  # UUID format
        True
    """
    return str(uuid.uuid4())

compute_confidence(evidence_map)

Compute aggregate confidence from supporting evidence using geometric mean.

The geometric mean is more conservative than arithmetic mean - a single piece of low-confidence evidence significantly reduces overall confidence. This matches how engineers actually reason: one weak piece of evidence can't be compensated by many strong pieces.

Formula: exp(mean(log(c_i))) for confidences c_i

If no evidence is available, returns 0.0 (no confidence). If any evidence has confidence 0, returns 0.0 (geometric mean property).

Parameters:

Name Type Description Default
evidence_map Dict[str, Evidence]

Mapping from evidence IDs to Evidence objects

 required

Returns:

Type Description
float

Computed confidence in [0, 1]
Example

evidence_map = { ... "e1": Evidence("src1", "test", {}, 0.8, datetime.utcnow()), ... "e2": Evidence("src2", "test", {}, 0.9, datetime.utcnow()) ... } node = ReasoningNode( ... id="node-1", ... decision="test", ... rationale="test", ... evidence_ids=["e1", "e2"] ... ) conf = node.compute_confidence(evidence_map) 0.84 < conf < 0.85 # sqrt(0.8 * 0.9) ≈ 0.8485 True

References: - Geometric mean: https://en.wikipedia.org/wiki/Geometric_mean - More conservative than arithmetic mean for combining confidences

Source code in upir/core/evidence.py 
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
def compute_confidence(self, evidence_map: Dict[str, Evidence]) -> float:
    """
    Compute aggregate confidence from supporting evidence using geometric mean.

    The geometric mean is more conservative than arithmetic mean - a single
    piece of low-confidence evidence significantly reduces overall confidence.
    This matches how engineers actually reason: one weak piece of evidence
    can't be compensated by many strong pieces.

    Formula: exp(mean(log(c_i))) for confidences c_i

    If no evidence is available, returns 0.0 (no confidence).
    If any evidence has confidence 0, returns 0.0 (geometric mean property).

    Args:
        evidence_map: Mapping from evidence IDs to Evidence objects

    Returns:
        Computed confidence in [0, 1]

    Example:
        >>> evidence_map = {
        ...     "e1": Evidence("src1", "test", {}, 0.8, datetime.utcnow()),
        ...     "e2": Evidence("src2", "test", {}, 0.9, datetime.utcnow())
        ... }
        >>> node = ReasoningNode(
        ...     id="node-1",
        ...     decision="test",
        ...     rationale="test",
        ...     evidence_ids=["e1", "e2"]
        ... )
        >>> conf = node.compute_confidence(evidence_map)
        >>> 0.84 < conf < 0.85  # sqrt(0.8 * 0.9) ≈ 0.8485
        True

    References:
    - Geometric mean: https://en.wikipedia.org/wiki/Geometric_mean
    - More conservative than arithmetic mean for combining confidences
    """
    # Filter evidence to only those referenced by this node
    relevant_evidence = [
        evidence_map[eid]
        for eid in self.evidence_ids
        if eid in evidence_map
    ]

    if not relevant_evidence:
        return 0.0

    # Check for any zero confidence (geometric mean would be 0)
    confidences = [e.confidence for e in relevant_evidence]
    if any(c == 0.0 for c in confidences):
        return 0.0

    # Compute geometric mean: exp(mean(log(c_i)))
    log_sum = sum(math.log(c) for c in confidences)
    log_mean = log_sum / len(confidences)
    geometric_mean = math.exp(log_mean)

    return geometric_mean

to_dict()

Serialize reasoning node to JSON-compatible dictionary.

Returns:

Type Description
Dict[str, Any]

Dictionary with all node fields
Example

node = ReasoningNode( ... id="node-1", ... decision="Use caching", ... rationale="Reduce latency", ... evidence_ids=["e1"], ... parent_ids=[], ... alternatives=[{"option": "No cache"}], ... confidence=0.8 ... ) d = node.to_dict() d["decision"] 'Use caching'

Source code in upir/core/evidence.py 
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
def to_dict(self) -> Dict[str, Any]:
    """
    Serialize reasoning node to JSON-compatible dictionary.

    Returns:
        Dictionary with all node fields

    Example:
        >>> node = ReasoningNode(
        ...     id="node-1",
        ...     decision="Use caching",
        ...     rationale="Reduce latency",
        ...     evidence_ids=["e1"],
        ...     parent_ids=[],
        ...     alternatives=[{"option": "No cache"}],
        ...     confidence=0.8
        ... )
        >>> d = node.to_dict()
        >>> d["decision"]
        'Use caching'
    """
    return {
        "id": self.id,
        "decision": self.decision,
        "rationale": self.rationale,
        "evidence_ids": self.evidence_ids.copy(),
        "parent_ids": self.parent_ids.copy(),
        "alternatives": [alt.copy() for alt in self.alternatives],
        "confidence": self.confidence
    }

from_dict(data) classmethod

Deserialize reasoning node from dictionary.

Parameters:

Name Type Description Default
data Dict[str, Any]

Dictionary containing node fields

 required

Returns:

Type Description
ReasoningNode

ReasoningNode instance
Example

data = { ... "id": "node-1", ... "decision": "Use caching", ... "rationale": "Reduce latency", ... "evidence_ids": ["e1"], ... "parent_ids": [], ... "alternatives": [], ... "confidence": 0.8 ... } node = ReasoningNode.from_dict(data) node.decision 'Use caching'

Source code in upir/core/evidence.py 
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
@classmethod
def from_dict(cls, data: Dict[str, Any]) -> "ReasoningNode":
    """
    Deserialize reasoning node from dictionary.

    Args:
        data: Dictionary containing node fields

    Returns:
        ReasoningNode instance

    Example:
        >>> data = {
        ...     "id": "node-1",
        ...     "decision": "Use caching",
        ...     "rationale": "Reduce latency",
        ...     "evidence_ids": ["e1"],
        ...     "parent_ids": [],
        ...     "alternatives": [],
        ...     "confidence": 0.8
        ... }
        >>> node = ReasoningNode.from_dict(data)
        >>> node.decision
        'Use caching'
    """
    return cls(
        id=data["id"],
        decision=data["decision"],
        rationale=data["rationale"],
        evidence_ids=data.get("evidence_ids", []),
        parent_ids=data.get("parent_ids", []),
        alternatives=data.get("alternatives", []),
        confidence=data.get("confidence", 0.0)
    )

__str__()

Human-readable string representation.

Source code in upir/core/evidence.py 
419
420
421
422
423
424
425
def __str__(self) -> str:
    """Human-readable string representation."""
    return (
        f"ReasoningNode({self.decision}, "
        f"confidence={self.confidence:.2f}, "
        f"evidence={len(self.evidence_ids)})"
    )

__repr__()

Developer-friendly representation.

Source code in upir/core/evidence.py 
427
428
429
430
431
432
433
434
435
def __repr__(self) -> str:
    """Developer-friendly representation."""
    return (
        f"ReasoningNode(id='{self.id}', "
        f"decision='{self.decision}', "
        f"confidence={self.confidence}, "
        f"evidence_count={len(self.evidence_ids)}, "
        f"parent_count={len(self.parent_ids)})"
    )

See Also

  • UPIR - Main UPIR class