Batch Operations

Glacis supports batch operations where multiple attestations are linked together. This is useful for scenarios like batch processing, decomposing results into individual items, and creating revision chains.

Overview

Three fields on Attestation enable batch operations:

Field	Type	Description
`operation_id`	`str`	UUID linking all attestations in the same operation
`operation_sequence`	`int`	Ordinal position within the operation (0-based)
`supersedes`	`str \| None`	Attestation ID this replaces (for revision chains)

When you call attest() without specifying an operation_id, a new UUID is auto-generated in offline mode. In online mode, the field is omitted from the request and the server assigns it. The operation_sequence defaults to 0.

OperationContext

Use glacis.operation() to create an OperationContext that manages the operation_id and auto-increments the sequence number:

import os
from glacis import Glacis

glacis = Glacis(mode="offline", signing_seed=os.urandom(32))

op = glacis.operation()

# First attestation: sequence 0
r1 = glacis.attest(
    service_id="my-service",
    operation_type="inference",
    input={"prompt": "Step 1"},
    output={"response": "Result 1"},
    operation_id=op.operation_id,
    operation_sequence=op.next_sequence(),  # 0
)

# Second attestation: sequence 1
r2 = glacis.attest(
    service_id="my-service",
    operation_type="inference",
    input={"prompt": "Step 2"},
    output={"response": "Result 2"},
    operation_id=op.operation_id,
    operation_sequence=op.next_sequence(),  # 1
)

print(r1.operation_id == r2.operation_id)  # True
print(r1.operation_sequence)               # 0
print(r2.operation_sequence)               # 1

You can also provide an explicit operation_id:

op = glacis.operation(operation_id="my-custom-operation-id")

Decompose

The decompose() method breaks a batch attestation into individual item attestations. All decomposed items share the parent’s operation_id, with incrementing operation_sequence values starting after the parent’s sequence.

import os
from glacis import Glacis

glacis = Glacis(mode="offline", signing_seed=os.urandom(32))

# Create a batch attestation
batch_receipt = glacis.attest(
    service_id="my-service",
    operation_type="batch",
    input={"source": "knowledge-base.pdf"},
    output={"qa_pairs": [
        {"question": "What is X?", "answer": "X is..."},
        {"question": "How does Y work?", "answer": "Y works by..."},
        {"question": "Why use Z?", "answer": "Z is useful for..."},
    ]},
)

# Decompose into individual item attestations
items = [
    {"question": "What is X?", "answer": "X is..."},
    {"question": "How does Y work?", "answer": "Y works by..."},
    {"question": "Why use Z?", "answer": "Z is useful for..."},
]

item_receipts = glacis.decompose(
    attestation=batch_receipt,
    items=items,
    operation_type="item",
    source_data={"source": "knowledge-base.pdf"},
)

# All items share the same operation_id
for r in item_receipts:
    print(f"{r.id}: op={r.operation_id}, seq={r.operation_sequence}")

decompose() Parameters

Parameter	Type	Default	Description
`attestation`	`Attestation`	(required)	Parent batch attestation
`items`	`list[dict]`	(required)	Individual items to attest
`operation_type`	`str`	`"item"`	Operation type for each decomposed item
`source_data`	`Any`	`None`	Shared input data for all items. When falsy (`None`, `{}`, etc.), defaults to `{"parent_attestation_id": attestation.id}`

Sequence Numbering

Decomposed items start their sequence at parent.operation_sequence + 1:

Attestation	`operation_sequence`
Parent (batch)	`0`
Item 1	`1`
Item 2	`2`
Item 3	`3`

Revision Chains with supersedes

Use the supersedes parameter to create revision chains, linking a new attestation to the one it replaces:

import os
from glacis import Glacis

glacis = Glacis(mode="offline", signing_seed=os.urandom(32))

# Original attestation
original = glacis.attest(
    service_id="my-service",
    operation_type="inference",
    input={"prompt": "Explain quantum computing"},
    output={"response": "Quantum computing uses qubits..."},
)

# Revised attestation (supersedes the original)
revised = glacis.attest(
    service_id="my-service",
    operation_type="inference",
    input={"prompt": "Explain quantum computing"},
    output={"response": "Quantum computing leverages quantum mechanics..."},
    supersedes=original.id,
)

print(f"Original: {original.id}")
print(f"Revised: {revised.id}")
print(f"Supersedes: {revised.supersedes}")  # original.id

This creates an auditable revision history where you can trace from any attestation back through its predecessors.