Bounded Delegation: Unit A Exercises

Student edition · 90 minutes of dedicated work · 2026-09-09

Runs on Google Colab as it ships today, or on any local Python 3.12+ kernel.

This is one of two practical units for Chapter 18. Unit A constructs and connects the mechanism; Unit B investigates a controlled failure, repairs it and transfers the invariant. Each is a complete ninety-minute session, with its own setup and required conceptual introductions. Basic Python variables, conditions, loops, functions, lists and dictionaries are the starting knowledge. Libraries and specialized concepts used here are introduced below before the main task.

By the end you should be able to:

1. Explain the chapter's mechanism using a prediction and an observed intermediate result.
2. Build a deterministic catering quote from the catalog and a validated Inquiry. Round portions upward, validate the selling price and return an explicit draft with no stock reservation.
3. Solve change the supplier's portions per tub using changed inputs and an independent expectation.
4. Retain your implementation, failed/corrected observations, causal explanation and limits.

Installation is preparation time. These are planning estimates, not measured completion times. Use the reference primers when a term is unfamiliar; in Unit B retrieve an explanation before re-reading it. Run All checks that the artifact executes. Unfinished student functions deliberately produce NEEDS_WORK. Keep your first attempt before opening answers.

This notebook belongs to the construction edition. Its supplied teaching runtime is embedded, so it can run without the textbook or another notebook. Where code uses REFERENCE_LESSON, that is the frozen runtime exercise identifier; the reader-facing chapter and saved unit identifiers use the current edition. Building against a supplied runtime is not proof that you have constructed all of its dependencies.

Run the self-contained setup

Open this notebook in Google Colab with the badge above, or use a local Python 3.12+ Jupyter kernel, with Pydantic 2. If needed, run %pip install "pydantic==2.13.4" once in a separate cell and restart the kernel. Package installation needs internet; the lesson itself needs no repository download, API key or prior notebook. The closing extension can use an OpenAI key from Colab's Secrets pane to put a live model behind the same tools; without a key it replays a recorded transcript and says so.

The collapsed cell contains 90 frozen teaching files. Base85 represents compressed bytes as text; zlib decompresses them; SHA-256 checks that the decoded files match this edition. These are supplied packaging operations, not learner algorithms. tempfile creates an isolated working copy; Path handles file locations; sys.path tells Python where the supplied modules live; a .pth file, the mechanism an editable install uses, tells reviewed local subprocesses the same. The code is available for inspection below and performs no package installation itself. The subsequent lesson teaches the libraries used by the mechanisms you will implement.

Run setup on every fresh kernel. It writes scratch runtime files separately from your retained practical-work/ch18-a folder. Rerunning setup restores the frozen support files and keeps your saved work. Restarting a kernel clears variables, not saved submission files. Source basis: Sovereign Agent 444c5f6. Some tasks use reviewed local subprocesses; they are not an OS sandbox.

Supplied offline setup and teaching files

The complete supplied setup cell is in the downloadable notebook. Run it there before attempting the cells below. This web edition omits only that compressed setup payload.

Commit to a prediction before the examples

prediction_notes = {
    "prediction": "Write the expected behavior before running the worked example.",
    "reason": "Name the input and rule behind that prediction.",
    "falsifier": "Name an observation that would prove the explanation wrong.",
    "revision": "After execution, explain what changed in your understanding.",
}

Reading the Python vocabulary used in this notebook

You need basic assignments, if, loops, functions, lists and dictionaries. The less familiar features used by the supplied code are introduced here. A library is reusable code that Python can import. The standard library ships with Python; Pydantic is an additional package. An import makes a name available, but does not mean that you have completed the exercise.

JSON is text for exchanging structured values. A Python dictionary is an in-memory object; the JSON representation is a string. Use json.dumps to encode and json.loads to decode. Decoding proves that text has valid JSON syntax, not that its fields match our business contract. Predict which of the following two decoded objects could describe a stock count.

import json

intro_data = {"sku": "MANGO", "count": 3}
intro_text = json.dumps(intro_data, sort_keys=True)
print(type(intro_data).__name__, type(intro_text).__name__, intro_text)
print(json.loads(intro_text))
print("Also valid JSON:", json.loads('["not", "a", "stock", "record"]'))
assert json.loads(intro_text) == intro_data

The first result is a dictionary; the second is a list. Before indexing a decoded object, check the shape that your function promises to accept. An exception interrupts the normal path. raise ValueError(...) refuses an invalid value; try/except lets a caller inspect that expected refusal. Catch the expected class, rather than turning every programming error into apparent success. finally runs cleanup even when an earlier operation raises.

An annotation, such as count: int, documents the expected type. It does not by itself enforce the type at runtime. A class defines a kind of object; an instance holds one object's data. @dataclass asks Python to generate routine construction and comparison methods from annotated fields. frozen=True prevents ordinary reassignment of the instance's fields; it does not make every object nested inside those fields immutable. A method is a function attached to a class; self refers to the instance receiving the call.

from dataclasses import dataclass


@dataclass(frozen=True)
class IntroObservation:
    operation: str
    count: int


intro_observation = IntroObservation("count-mango", 3)
print(intro_observation.operation, intro_observation.count)
assert intro_observation == IntroObservation("count-mango", 3)

A callback is a function passed to another function. This is how the classroom harness invokes your implementation. The argument candidate below is a function object; parentheses perform the call. Predict the two answers before execution, then trace the result to the callback.

def intro_apply(candidate, value):
    return {"input": value, "observed": candidate(value)}


def intro_double(value):
    return value * 2


print(intro_apply(intro_double, 3))
print(intro_apply(lambda value: value + 2, 3))
assert intro_apply(intro_double, 3)["observed"] == 6

lambda value: value + 2 is a small anonymous function. A closure is a function that retains access to values from its surrounding scope. It can bind a tool to a shop snapshot. A shallow copy duplicates only the outer container; copy.deepcopy also copies nested containers used in these fixtures. A set stores distinct values; required <= allowed asks whether every required item is allowed. frozenset is the corresponding immutable set. A tuple groups ordered values; (value,) is a one-item tuple, including the comma.

Paths and cleanup. Path represents a filesystem location. path / "file.json" constructs a child path; read_text and write_text read and write text. A context manager, used with with, manages entry and exit. A temporary-directory context removes its contents on exit. Save your submission outside temporary runtime directories. Reopening a file is different from reusing a Python variable: the former tests retained bytes, while the latter only tests this kernel.

from pathlib import Path
from tempfile import TemporaryDirectory

with TemporaryDirectory() as intro_folder:
    intro_path = Path(intro_folder) / "observation.json"
    intro_path.write_text(json.dumps(intro_data), encoding="utf-8")
    intro_reopened = json.loads(intro_path.read_text(encoding="utf-8"))
    assert intro_reopened == intro_data
    print("Read from a file:", intro_reopened)

Retrieval check: explain JSON versus a dictionary, annotation versus validation, class versus instance, and defining a callback versus invoking it. Change the callback above so an incorrect implementation visibly changes the observed output. This distinction will matter when grading your connected work. Reference: Python's JSON, dataclasses, and pathlib documentation.

Pydantic: turn an input dictionary into a checked object

Pydantic is an additional Python library for validating data. Its model is a class describing fields, not a neural network. Inherit from BaseModel, declare annotated fields, then call model_validate on incoming data. A field without a default is required. A field with a default can be omitted. The result is an instance whose values you read with dot notation.

from pydantic import BaseModel, ConfigDict, Field, ValidationError


class IntroCourseRequest(BaseModel):
    model_config = ConfigDict(strict=True, extra="forbid")
    name: str = Field(min_length=1)
    quantity: int = Field(gt=0, le=1000)
    note: str = ""


intro_request = IntroCourseRequest.model_validate({"name": "mango", "quantity": 4})
print(intro_request.name, intro_request.quantity, repr(intro_request.note))
assert intro_request.note == ""

The annotation says the field's type. Field supplies constraints: gt=0 means greater than zero, le=1000 means at most 1000, and min_length=1 excludes an empty name. ConfigDict sets model-wide behavior. strict=True rejects conversions for this integer field, including "4", 4.0 and True; extra="forbid" rejects undeclared keys. Pydantic can otherwise convert some compatible inputs, so choose this boundary deliberately rather than assuming every accepted input arrived in the expected type.

Predict which rule refuses each payload. ValidationError reports a failed contract. Its errors() entries contain loc, the field location, and type, the failure category. Catching that expected exception lets the notebook inspect the failure and continue.

intro_bad_requests = [
    {"name": "mango", "quantity": "4"},
    {"name": "mango", "quantity": True},
    {"name": "", "quantity": 4},
    {"name": "mango", "quantity": 0},
    {"name": "mango", "quantity": 4, "approved": True},
    {"quantity": 4},
]
for intro_bad_request in intro_bad_requests:
    try:
        IntroCourseRequest.model_validate(intro_bad_request)
    except ValidationError as intro_error:
        print([(item["loc"], item["type"]) for item in intro_error.errors(include_input=False)])
    else:
        raise AssertionError("An invalid input crossed the declared contract")

Use model_dump() for a Python dictionary, model_dump_json() for JSON text, and model_validate_json() to parse and validate JSON. model_json_schema() describes the contract; it is neither an instance's current values nor an invocation of the business handler.

intro_serialized = intro_request.model_dump_json()
intro_schema = IntroCourseRequest.model_json_schema()
assert IntroCourseRequest.model_validate_json(intro_serialized) == intro_request
print("Actual values:", intro_request.model_dump())
print("Quantity contract:", intro_schema["properties"]["quantity"])
assert intro_schema["properties"]["quantity"]["exclusiveMinimum"] == 0

Four is valid input to this schema even if the shop needs six. Pydantic checks the declared shape and constraints; the handler still needs authoritative stock, price and permission. Ordinary assignments to an existing instance are not automatically revalidated unless configured for assignment validation. This lesson validates new input at the boundary and uses the resulting values. Explain these limits before relying on a model object in a transaction or tool call.

Chapter 2's full introduction expands this pattern with a separate data-repair checkpoint. This notebook contains the required pattern here so prior Pydantic experience is not needed. References: models, fields, and strict mode.

SQLite from the first row to an atomic change

A dictionary disappears when its process ends. A database can retain records so a later process can resume from evidence. SQLite is an embedded database: Python's sqlite3 library opens a local database file without starting a separate database server. SQL is the language used to define, select and change its records. A table has named columns and rows. A primary key identifies a row; a query asks for rows satisfying a condition.

Start with a deliberately small preference table. Read the SQL as instructions: create the table, insert one named value, then select the value for one session. ? is a parameter placeholder; the values are passed separately so they are data, not SQL instructions. fetchone() returns one row or None; it does not guarantee that a matching row exists.

import sqlite3
from pathlib import Path
from tempfile import TemporaryDirectory

with TemporaryDirectory() as intro_sql_folder:
    intro_db_path = Path(intro_sql_folder) / "example.sqlite"
    intro_db = sqlite3.connect(intro_db_path, autocommit=True)
    intro_db.execute("CREATE TABLE preference (id INTEGER PRIMARY KEY, session TEXT, value TEXT)")
    intro_db.execute("INSERT INTO preference (session,value) VALUES (?,?)", ("lucy", "09:00"))
    intro_row = intro_db.execute(
        "SELECT value FROM preference WHERE session=?", ("lucy",)
    ).fetchone()
    print("Matching row:", intro_row)
    assert intro_row == ("09:00",)
    assert (
        intro_db.execute(
            "SELECT value FROM preference WHERE session=?", ("another-session",)
        ).fetchone()
        is None
    )
    intro_db.close()
    intro_reopen = sqlite3.connect(intro_db_path, autocommit=True)
    assert intro_reopen.execute("SELECT count(*) FROM preference").fetchone()[0] == 1
    intro_reopen.close()
print("The row survived closing and reopening its connection.")

Index [0] selects the first column of a returned tuple. sqlite3.Row is an alternative row factory that also permits named-column access. dict(row) then produces an ordinary dictionary. The book's Database wrapper supplies that configuration and its schema; the wrapper is course code, while sqlite3 is the standard library. You will use the public connection and transaction methods explained at the exercise boundary, rather than needing to reconstruct the wrapper.

Now consider a budget. Moving five cents from reserved to spent requires two values to change together. A transaction makes a group of local changes commit together or roll back together. The example explicitly controls SQL transactions with autocommit=True and SQL statements. BEGIN IMMEDIATE starts a write transaction; COMMIT keeps its changes; ROLLBACK discards them. Predict the row after the deliberately raised exception. Catching an error alone would not undo the first update; the rollback is the operation that restores the prior state.

intro_ledger = sqlite3.connect(":memory:", autocommit=True)
intro_ledger.execute(
    "CREATE TABLE budget (id INTEGER PRIMARY KEY, reserved INTEGER, spent INTEGER)"
)
intro_ledger.execute("INSERT INTO budget VALUES (1,5,0)")
intro_ledger.execute("BEGIN IMMEDIATE")
try:
    intro_ledger.execute("UPDATE budget SET reserved=0 WHERE id=1")
    raise ValueError("injected failure before the matching spend update")
except ValueError:
    intro_ledger.execute("ROLLBACK")
assert intro_ledger.execute("SELECT reserved,spent FROM budget").fetchone() == (5, 0)
intro_ledger.execute("BEGIN IMMEDIATE")
intro_ledger.execute("UPDATE budget SET reserved=0,spent=5 WHERE id=1")
intro_ledger.execute("COMMIT")
print(
    "After a complete change:", intro_ledger.execute("SELECT reserved,spent FROM budget").fetchone()
)
intro_ledger.close()

The literal :memory: creates a temporary database inside this connection; it is useful for the small experiment, but the earlier file example establishes persistence. Neither example proves that a remote supplier rolls back when the local transaction rolls back. An external operation has its own state and evidence.

SQL you will meet later. UPDATE ... SET ... WHERE ... changes selected rows. AND combines conditions. ORDER BY makes an ordering explicit; absent that clause, do not rely on row order. count(*) counts rows; sum(amount) totals a column and can be NULL on an empty input; coalesce(sum(amount),0) uses zero for that empty aggregate. A UNIQUE constraint rejects duplicate identities. GROUP BY status computes one aggregate per status.

An invariant is a condition that must remain true across operations, such as nonnegative reserved money. A snapshot is a consistent view at one point; two separate reads can describe different moments unless their transaction contract binds them. In the book, with db.immediate() groups related writes. It is a course-defined context manager with the commit/rollback purpose you just observed. Do not assume that an arbitrary with connection has identical behavior under every SQLite autocommit setting.

Your prediction: two workers both read ten remaining cents outside a transaction and each approve seven. Why can both believe the next order fits? Explain what must be checked together with the write. Then change the example's initial reserved amount and repeat the failure. Reference: Python's SQLite tutorial and transaction control.

Delegation is a bounded assignment, not a reason to use another model

Lucy asks for a catering quote. One part is fixed arithmetic: divide guests by portions per tub, round up, multiply by a validated price. Another part might require research or judgment. A delegated assignment gives a worker a specific objective, inputs, authority, budget and a result contract. A handoff returns evidence the parent can inspect. More workers do not automatically improve either answer quality or operating cost.

Start with the deterministic part so you can defend whether delegation adds value. A function that calculates a quote can be easier to verify than a second model describing the calculation. The chapter's architecture comparison retains the function when that is the better-supported choice. The exercise is not scored by how many agents it creates.

Derive ceiling division without floating point

Ten guests fit one ten-portion tub. Eleven guests require two. Integer floor division // would return one for eleven divided by ten and underquote the order. For positive integers, (guests + portions - 1) // portions rounds upward. Work the boundaries 1, 10, 11 and 20 by hand.

intro_portions = 10
for intro_guests in (1, 10, 11, 17, 20):
    intro_tubs = (intro_guests + intro_portions - 1) // intro_portions
    intro_total = intro_tubs * 325
    print(intro_guests, "guests:", intro_tubs, "tubs:", intro_total, "cents")
assert (11 + 10 - 1) // 10 == 2

The formula's precondition matters. Zero portions would divide by zero; negative or boolean guest counts are not valid inquiries under the strict contract. A Pydantic Inquiry represents validated input. The quote function looks up the requested catalog product and validates its selling price. It returns an explicit draft and does not reserve stock or authorize a purchase.

A replacement must inherit used allowance

Suppose an assignment has a three-attempt budget. Its first worker uses two attempts and fails. A replacement inherits one remaining attempt, not a fresh budget of three. Otherwise repeatedly replacing a worker makes a bounded assignment unbounded. Keep assignment identity separate from worker identity and keep usage attached to the assignment.

intro_assignment = {"id": "quote-research", "limit": 3, "used": 2}
intro_workers = ["first-worker", "replacement-worker"]
for intro_worker in intro_workers:
    intro_remaining = max(0, intro_assignment["limit"] - intro_assignment["used"])
    print(intro_worker, "remaining assignment allowance", intro_remaining)
assert intro_remaining == 1

This table illustrates retained accounting, not a full worker implementation. The surrounding runtime supplies claims, leases, parent cancellation and result identity. A child may produce a result after its parent cancels; accepting it must still depend on current authority and the assignment's state. Duplicate results should not duplicate downstream effects.

Compare alternatives using the same task

Hold inputs and required output fixed. Record the ordinary function's result, observed model attempts and cost, then compare a delegated path under the same contract. A zero-model-call function is a legitimate baseline. If the delegated result is different, inspect which difference is useful and what evidence supports it. Avoid comparing a simple arithmetic task with a larger research task and attributing the entire difference to delegation.

Unit A builds the real quote function used by the delegated research path. Unit B injects floor division, then checks partial-tub cases against the actual quote tool. The changed-constraint exercise varies portions per tub and price so a memorized ten-portion expression is insufficient. You must validate the new contract, calculate the draft, and retain physical stock unchanged.

Before coding, distinguish inquiry, quote, reservation, approval and purchase. Draw their order and mark which transitions this notebook implements. At the end, write a short decision record: keep the function or delegate, observed evidence, cost/latency limits, and one future condition that would justify reconsidering. A defensible decision is a learning outcome, not an optional essay after “the real coding.”

Understand the supplied execution interface

The course runtime is provided so your implementation can be connected to real callers and storage. SourceTask(ROOT, chapter) makes a private copy. install(source) replaces only the declared function; visible() invokes the real chapter probe; save(path, result) retains a successful implementation and its evidence. load(path) checks the saved identities and hashes. inject_failure() changes the declared boundary; repair(fragment) replaces that broken fragment. close() removes the scratch copy after you retain evidence. These methods are supplied harness operations, not additional packages you must discover or install.

RuntimeLab provides the same copied-source failure experiment without the complete-function construction layer. Its run method records exit status, observations and the compared expectation. A subprocess log from an unfinished learner implementation is feedback about that implementation; it is not a successful connection. A syntax error in the notebook cell itself is a separate issue to fix. The task below names which interface it uses.

For direct-function units, the visible driver calls your callback without installing a source string. In either case, trace where your code is invoked. Supplied fixtures, database wrappers and replay models are labeled infrastructure; your own implementation and changed-case explanation are the evidence of learning.

Main practical: construct, connect and challenge

Design question: Eleven guests need two tubs?

Build a deterministic catering quote from the catalog and a validated Inquiry. Round portions upward, validate the selling price and return an explicit draft with no stock reservation.

You edit a complete function in a temporary copy of src/reference_organizations/store/delegation.py. The real callers, database and tool boundaries remain connected. The self-contained setup above supplies the frozen development runtime. The notebook runs reviewed local subprocesses; it is not a security sandbox. No live account is required.

Predict before running

Invoke the actual quote tool used by the delegated research path and inspect product identity, total and unchanged stock.

Write the expected result and a falsifying observation before running. Include one legal action, one refusal, and the exact-empty case where the interface permits it. Explain the consequence for Lucy if your prediction is wrong.

import json
import os
import runpy
from pathlib import Path

ROOT = COURSE_ROOT

SourceTask = runpy.run_path(str(ROOT / "book/always_on/exercises/source_tasks_v1.py"))["SourceTask"]
REFERENCE_LESSON = 14
HANDOFF = Path("ch18-unit-a-handoff-v1.json")

Construct the complete mechanism

Implement quote in the string below. Keep the named signature and existing helper interfaces. The starter is intentionally incomplete; its failed connection is reported separately from whether the notebook itself executed. A constant answer cannot stand in for the real mechanism.

Inspect the supplied caller around the function in src/reference_organizations/store/delegation.py. Draw the data path from the observed output through this function to its actual input or query. Then write your implementation from the contract.

implementation_source = """
def quote(db: Database, inquiry: Inquiry) -> dict[str, Any]:
    raise NotImplementedError("Construct this chapter mechanism")
"""

Hint 1: the design

Floor division underquotes any partial tub.

Hint 2: the boundary

Inspect the parameters and the caller in src/reference_organizations/store/delegation.py. Identify validation, durable state and the first externally observable effect. Preserve the existing surrounding helper contracts.

Hint 3: the structure

Read the requested product record, validate positive integer price and use integer ceiling arithmetic for ten portions per tub.

Connect to the cumulative runtime

The following installs your complete function into the copied runtime and executes the chapter probe against it. It saves your implementation and the observed connection for Unit B only after that connection succeeds.

def connect_build(source):
    task = SourceTask(ROOT, REFERENCE_LESSON)
    try:
        task.install(source)
        result = task.visible()
        if result["status"] == "PASS":
            task.save(HANDOFF, result)
        return result
    finally:
        task.close()


build_result = connect_build(implementation_source)
print("CONNECTION", build_result["status"])
print("OBSERVATION", build_result["observation"])

Challenge and transfer

Try 1, 10, 17 and 200 guests, a changed product price and invalid Inquiry shapes. Explain why fixed arithmetic earns zero model calls.

Use a fresh SourceTask, install your implementation and edit only its copied probe to run the changed input. Keep the expected result in your prediction notes, independent of your implementation. Call task.run("MY_TRANSFER", expected=your_expected) and close the task in finally. Retain both a valid and a refused case so rejecting everything cannot pass.

The instructor runs additional cases with different identities and boundaries against the real source. Passing the visible connection alone is not the transfer verdict. Do not put instructor solutions or holdouts into a student submission.

Save and explain

After success, submit ch15-unit-a-handoff-v1.json, your source, prediction notes and changed-input observations. Unit B checks the chapter, runtime hash and exact implementation hash and re-executes your code.

Explain which input or state caused the output, which observation would refute the explanation and what remains outside the guarantee: A delegated draft neither reserves stock nor authorizes a purchase.

exercise_report = {
    "unit": "ch18-a",
    "attempted": 1,
    "completed": int(build_result["status"] == "PASS"),
    "failed": int(build_result["status"] != "PASS"),
    "skipped": 0,
    "connection": build_result["status"],
    "handoff": "WRITTEN" if build_result["status"] == "PASS" else "NOT_READY",
}
print("EXERCISE_REPORT=" + json.dumps(exercise_report, sort_keys=True))

Changed-constraint construction: Change the supplier's portions per tub

Allow twenty minutes. Spend three minutes predicting, ten implementing and tracing, five on a new case of your own, and two explaining the surviving limitation. This is dedicated work, not an invitation to run a supplied answer. Both units revisit the same invariant after different core experiences; in Unit B, attempt this task from memory before consulting Unit A.

Implement transfer_check(guests, price_cents, portions). Require exact positive integers for all three inputs, otherwise raise ValueError. Return [tubs, total_cents], using ceiling division. The new portions argument replaces the earlier fixed ten-portion assumption. No stock or approval state changes.

Write your expected values before running the table. Keep one accepted case and one refusal. Your function is passed directly into the driver below. The driver copies inputs and checks they remain unchanged; it does not replace your implementation with the reference answer.

Hint 1: identify the authoritative inputs Name the source field for each output value. Which input changes while the rule remains the same?

Hint 2: choose the boundary cases Start with exact empty, exact equality and one value on each side of the boundary where valid. Do not add a special case for a visible product name or operation identity.

def transfer_check(guests, price_cents, portions):
    raise NotImplementedError("Validate and round the changed pack size upward")

import copy
import json

TRANSFER_CASES = [
    ("eleven guests", [11, 325, 10], [2, 650]),
    ("exact eight portion tub", [16, 275, 8], [2, 550]),
    ("partial new tub", [17, 275, 8], [3, 825]),
    ("zero portions", [17, 275, 0], {"raises": "ValueError"}),
]


def same_transfer_value(actual, expected):
    if type(actual) is not type(expected):
        return False
    if isinstance(expected, dict):
        return actual.keys() == expected.keys() and all(
            same_transfer_value(actual[key], value) for key, value in expected.items()
        )
    if isinstance(expected, list):
        return len(actual) == len(expected) and all(
            same_transfer_value(a, e) for a, e in zip(actual, expected, strict=True)
        )
    return actual == expected


def run_transfer(candidate, cases):
    observations = []
    for label, arguments, expected in cases:
        supplied = copy.deepcopy(arguments)
        before = copy.deepcopy(supplied)
        raised = None
        try:
            actual = candidate(*supplied)
        except NotImplementedError:
            raised = "NotImplementedError"
            actual = {"unfinished": True}
        except Exception as error:
            raised = type(error).__name__
            actual = {"raises": raised}
        expects_error = isinstance(expected, dict) and set(expected) == {"raises"}
        correct = (
            raised == expected["raises"]
            if expects_error
            else (raised is None and same_transfer_value(actual, expected))
        )
        passed = correct and same_transfer_value(supplied, before)
        observations.append(
            {"case": label, "expected": expected, "observed": actual, "passed": passed}
        )
        print("PASS" if passed else "NEEDS_WORK", label, "expected", expected, "observed", actual)
    return observations


transfer_observations = run_transfer(transfer_check, TRANSFER_CASES)
TRANSFER_PASSED = all(row["passed"] for row in transfer_observations)
print("TRANSFER_STATUS", "PASS" if TRANSFER_PASSED else "NEEDS_WORK")

Design a counterexample and retrieve the mechanism

Add one new case with an independently calculated expected outcome to TRANSFER_CASES and rerun the driver. Change one condition at a time. Then deliberately replace your candidate with a constant answer in a temporary copy and show a case that rejects it. Restore your implementation. Explain why that counterexample is stronger than repeating the original example with a new name.

Without viewing the worked example, write the invariant in words and trace one observed value back to its input. Identify which part is a local fixture result and which claim would need a live provider, host or external-system observation. Keep a first attempt even if you used a hint.

Save your evidence and explain the result

Fill the prediction notes and your explanation before saving. Include the exact observed value, the input or retained row that caused it, your code's invocation point, one failed hypothesis, and the strongest claim the evidence still cannot support. A completed code cell alone does not earn explanation credit. Do not label reference-start behavior as your own Unit A construction.

Keep this edited notebook, the Markdown if used for notes, saved handoff files, and the JSON record below. Your work folder survives scratch cleanup and can be reopened in a new kernel. An instructor can ask for an unseen case after the visible checks; keep your implementation general.

explanation_notes = {
    "causal_trace": "Explain the input, learner invocation and observed result.",
    "failed_hypothesis": "Describe a prediction the evidence changed.",
    "remaining_limit": "Name the guarantee not established by this experiment.",
}
course_submission = {
    "unit": "ch18-a",
    "planned_minutes": 90,
    "starting_evidence": globals().get("HANDOFF_ORIGIN", "INDEPENDENT_UNIT_A"),
    "prediction": prediction_notes,
    "explanation": explanation_notes,
    "core_report": exercise_report,
    "transfer": transfer_observations,
    "explanation_review": "HUMAN_REVIEW_REQUIRED",
}
submission_path = COURSE_WORK / "ch18-a-submission-v1.json"
submission_path.write_text(
    json.dumps(course_submission, indent=2, sort_keys=True), encoding="utf-8"
)
print("Saved evidence:", submission_path)
print(
    "COURSE_REPORT="
    + json.dumps(
        {
            "unit": "ch18-a",
            "transfer_passed": TRANSFER_PASSED,
            "starting_evidence": course_submission["starting_evidence"],
            "edition": "student",
        },
        sort_keys=True,
    )
)

Extension: let a model hand out the bounded assignment

Every quote and every delegation above was yours. In production the request to delegate comes from the shop assistant itself, when a model decides a catering inquiry deserves a second look. The boundary this chapter built is a contract: delegate refuses any assignment without a finite deadline, one to eight model calls and a bounded budget, refuses to recurse from anything but an eligible shop parent, and records exactly one immutable assignment per parent. run_once then runs the worker under the book's own loop with that contract as its Limits, and grades the worker against the function it was asked to check.

This section puts a model behind that contract through four tools. catering_quote wraps quote, the fixed arithmetic. delegate_research wraps delegate for the parent claim Lucy's message created. run_worker wraps run_once with the runtime's OfflineCateringModel, so the child's work is a real research work row with a real transcript, not a stub. research_report reads the child's row back. The model chooses when to delegate and what allowance to ask for; it cannot widen the contract, reserve stock, or mark the child as done.

Prediction: the next cell prints the four schemas and the seeded stock. Write down which two tools are consequential, which argument class is the chapter's own Inquiry, and what on_hand is for vanilla, before you run it.

import json
import time
from pathlib import Path

from pydantic import Field

from reference_organizations.store.agent import NoArguments, seed_lucy
from reference_organizations.store.delegation import (
    Inquiry,
    OfflineCateringModel,
    delegate,
    quote,
    run_once,
)
from sovereign_agent.agent_loop import Limits, run_loop
from sovereign_agent.assistant_work import claim, enqueue
from sovereign_agent.database import Database
from sovereign_agent.llm_lab import ScriptedModel, describe, scripted_turn, tool_trace
from sovereign_agent.tool_dispatch import Dispatcher, ExecutableTool

LLM_CATERING_DB = COURSE_WORK / "ch18-llm-catering.sqlite"
LUCY_REQUEST = (
    "A customer wants vanilla for eleven guests on Saturday. Quote it, then have the research "
    "worker double-check the draft; give it a generous allowance of twelve model calls."
)


class Assignment(Inquiry):
    model_calls: int = Field(ge=1)


def open_catering_desk(path):
    """A fresh ledger, Lucy's stock, and one claimed shop work item to delegate from."""
    for stale in (path, Path(f"{path}-wal"), Path(f"{path}-shm"), path.with_suffix(".authority")):
        stale.unlink(missing_ok=True)
    db = Database(path)
    seed_lucy(db)
    enqueue(db, "lab:catering", "lucy", LUCY_REQUEST)
    parent = claim(db, "lab-desk")
    assert parent is not None and parent.role == "shop"
    return db, parent


def catering_desk(db, parent):
    """Four tools over the real runtime; the ledger keeps the reason of every refusal."""
    refusals = []

    def draft_quote(args):
        return quote(db, args)

    def assign_research(args):
        inquiry = Inquiry(sku=args.sku, guests=args.guests)
        try:
            child = delegate(
                db,
                parent.id,
                inquiry,
                deadline=time.time() + 600,
                model_calls=args.model_calls,
                estimated_call_cents=1,
                budget_cents=20,
            )
        except (ValueError, PermissionError) as error:
            refusals.append(("delegate_research", str(error)))
            raise
        return {"work": child, "model_calls_limit": args.model_calls}

    def run_worker(_):
        outcome = run_once(db, OfflineCateringModel())
        report = outcome.get("report") or {}
        return {
            "status": outcome["status"],
            "passed": report.get("passed"),
            "quote": report.get("quote"),
            "model_calls": (report.get("assignment_usage") or {}).get("model_calls"),
        }

    def research_report(_):
        row = db.connection.execute(
            "SELECT w.status,w.result,d.model_calls,d.model_calls_limit "
            "FROM assistant_delegations d JOIN assistant_work w ON w.id=d.work_id "
            "WHERE d.parent_id=?",
            (parent.id,),
        ).fetchone()
        return {"assigned": row is not None, **(dict(row) if row else {})}

    def before_write(call):
        paused = db.connection.execute("SELECT paused FROM assistant_control WHERE id=1").fetchone()
        if paused[0]:
            refusals.append((call.name, "runtime paused"))
            raise PermissionError("runtime paused")

    dispatcher = Dispatcher(
        [
            ExecutableTool(
                "catering_quote",
                "Draft a catering quote: ten portions per tub, rounded up. Reserves nothing.",
                Inquiry,
                draft_quote,
            ),
            ExecutableTool(
                "delegate_research",
                "Assign one bounded research task to check a quote. model_calls is the worker's "
                "whole allowance.",
                Assignment,
                assign_research,
                consequential=True,
            ),
            ExecutableTool(
                "run_worker",
                "Run one research worker pass on the assigned task.",
                NoArguments,
                run_worker,
                consequential=True,
            ),
            ExecutableTool(
                "research_report",
                "Read the assigned research task's status, result and model calls used.",
                NoArguments,
                research_report,
            ),
        ],
        allowed=frozenset({"catering_quote", "delegate_research", "run_worker", "research_report"}),
        before_write=before_write,
    )
    return dispatcher, refusals


catering_db, catering_parent = open_catering_desk(LLM_CATERING_DB)
catering_dispatcher, catering_refusals = catering_desk(catering_db, catering_parent)
for tool_schema in catering_dispatcher.schemas():
    function = tool_schema["function"]
    print(function["name"], "requires", function["parameters"].get("required", []))
    print("   ", function["description"])
SEEDED_STOCK = {
    row["sku"]: (row["on_hand"], row["reserved"])
    for row in catering_db.connection.execute("SELECT sku,on_hand,reserved FROM inventory")
}
print("Seeded stock:", SEEDED_STOCK, "| parent work:", catering_parent.id[:8])
CATERING_MESSAGES = [
    {
        "role": "system",
        "content": (
            "You run the counter of Lucy's ice cream shop. Answer catering inquiries with "
            "catering_quote, which is fixed arithmetic over the catalog: ten portions per tub, "
            "rounded up, draft only. When Lucy asks for a quote to be checked, assign exactly "
            "one research task with delegate_research, run the worker, and read its report. "
            "The shop's contract allows a research task at most eight model calls. Finish with "
            "a short plain-words summary: tubs, USD total, whether the worker verified the "
            "draft, and that nothing was reserved or purchased."
        ),
    },
    {"role": "user", "content": LUCY_REQUEST},
]

catering_quote takes the chapter's own Inquiry, so guests must be an integer from 1 to 200 and any extra field is refused as invalid_arguments. delegate_research extends it with model_calls, bounded below by the schema and above by the runtime: the contract's one-to-eight rule lives in delegate, not in the prompt and not in the Pydantic class, so a request for twelve calls is a business refusal, tool_failed, with the reason in the ledger. The gate refuses every write while the runtime is paused.

Run the loop against a recorded transcript

The turns below are what a model replied, written down. Turn one quotes. Turn two asks for the allowance Lucy suggested. Turn three asks for four. Turn four runs the worker and reads the report in one turn. Turn five is the answer, which ends the loop. The transcript is built by a function so the challenge can change one number without touching the story.

Prediction: how many tubs, what total in cents, and how many model calls does the worker spend? OfflineCateringModel needs one call to ask for the quote and one to answer. Write down what research_report returns for status and model_calls before you run it.

def recorded_turns(model_calls):
    return [
        scripted_turn(
            calls=[{"name": "catering_quote", "arguments": {"sku": "SKU-VANILLA", "guests": 11}}]
        ),
        scripted_turn(
            calls=[
                {
                    "name": "delegate_research",
                    "arguments": {"sku": "SKU-VANILLA", "guests": 11, "model_calls": 12},
                }
            ]
        ),
        scripted_turn(
            calls=[
                {
                    "name": "delegate_research",
                    "arguments": {"sku": "SKU-VANILLA", "guests": 11, "model_calls": model_calls},
                }
            ]
        ),
        scripted_turn(
            calls=[
                {"name": "run_worker", "arguments": {}},
                {"name": "research_report", "arguments": {}},
            ]
        ),
        scripted_turn(
            "Eleven guests need 2 tubs of vanilla, USD 10.00 as a draft. Twelve model calls was "
            "over the contract, so the worker ran with four and verified the same draft. "
            "Nothing was reserved or purchased."
        ),
    ]


RECORDED_TURNS = recorded_turns(4)
LOOP_LIMITS = Limits(model_calls=8, tool_calls=16, seconds=120, output_tokens=4096)

scripted_result = run_loop(
    ScriptedModel(RECORDED_TURNS), catering_dispatcher, CATERING_MESSAGES, limits=LOOP_LIMITS
)
print(describe(scripted_result, "scripted"))
print("Refusal ledger:", catering_refusals)
scripted_trace = tool_trace(scripted_result)
assert scripted_result.status == "COMPLETED"
assert [step["ok"] for step in scripted_trace] == [True, False, True, True, True]
assert scripted_trace[0]["result"]["tubs"] == 2
assert scripted_trace[0]["result"]["total_cents"] == 1000
assert scripted_trace[0]["result"]["stock_reserved"] is False
assert scripted_trace[1]["result"] == "tool_failed"
assert catering_refusals == [("delegate_research", "bounded delegation contract required")]
assert scripted_trace[3]["result"]["status"] == "DONE"
assert scripted_trace[3]["result"]["quote"] == scripted_trace[0]["result"]
scripted_report = scripted_trace[4]["result"]
assert scripted_report["status"] == "DONE" and scripted_report["model_calls"] == 2
assert scripted_report["result"].startswith("Catering draft: 2 tubs for 11 guests, USD 10.00.")
# Durable state: one immutable assignment for the parent, and stock untouched.
assignments = catering_db.connection.execute(
    "SELECT count(*) FROM assistant_delegations WHERE parent_id=?", (catering_parent.id,)
).fetchone()[0]
assert assignments == 1
stock_now = {
    row["sku"]: (row["on_hand"], row["reserved"])
    for row in catering_db.connection.execute("SELECT sku,on_hand,reserved FROM inventory")
}
assert stock_now == SEEDED_STOCK
# Every observation in the transcript carries the identifier of the request it answers.
assert all(m.get("tool_call_id") for m in scripted_result.messages if m["role"] == "tool")

The refused assignment is an ordinary failed observation; the loop carries on and the model adapts. The child's row is real: assistant_work holds it with role research, observe wrote its transcript, and assistant_delegations.model_calls counted the two calls the worker spent against its limit of four. Stock did not move, because nothing in this path can move it.

Run the same loop against a live model

Nothing changes except the model and a fresh desk. build_model looks for an OPENAI_API_KEY in Colab's Secrets pane (the key icon in the left sidebar: add a secret with that name and switch on notebook access for it), then in the process environment for a local kernel. Without a key it returns the scripted model again and the record says so.

The live model is gpt-5.1 through the openai library, which Colab ships preinstalled; on a local kernel run %pip install openai once. The model may cap the allowance at eight on its own, ask for twelve and adapt, or read the report before running the worker. What it cannot do is create a second assignment for the same parent, widen the contract, or change stock.

Prediction: write down the tubs and total a careful assistant should report, and whether it should accept Lucy's twelve, then run the cell and compare with the model's judgment, or with the recorded transcript when the run is scripted.

from sovereign_agent.llm_lab import DEFAULT_MODEL, build_model, resolve_api_key

live_db, live_parent = open_catering_desk(LLM_CATERING_DB)
live_dispatcher, live_refusals = catering_desk(live_db, live_parent)
chosen_model = build_model(RECORDED_TURNS, model=DEFAULT_MODEL)
print("Model source:", chosen_model.source, "| key found:", resolve_api_key() is not None)
live_result = run_loop(chosen_model, live_dispatcher, CATERING_MESSAGES, limits=LOOP_LIMITS)
print(describe(live_result, chosen_model.source))
print("Refusal ledger:", live_refusals)
live_trace = tool_trace(live_result)
assert live_result.status in {"COMPLETED", "MODEL_CALL_LIMIT", "TOOL_LIMIT", "MODEL_FAILED"}
# Whatever the model did: every quote is the fixed arithmetic, every assignment is bounded.
for step in live_trace:
    if step["tool"] == "catering_quote" and step["ok"]:
        guests = step["arguments"]["guests"]
        assert step["result"]["tubs"] == (guests + 9) // 10, step
        assert step["result"]["total_cents"] == step["result"]["tubs"] * 500, step
        assert step["result"]["stock_reserved"] is False, step
    if step["tool"] == "delegate_research" and step["ok"]:
        assert 1 <= step["arguments"]["model_calls"] <= 8, step
live_children = live_db.connection.execute(
    "SELECT w.status,d.model_calls,d.model_calls_limit FROM assistant_delegations d "
    "JOIN assistant_work w ON w.id=d.work_id WHERE d.parent_id=?",
    (live_parent.id,),
).fetchall()
assert len(live_children) <= 1
for child in live_children:
    assert child["model_calls"] <= child["model_calls_limit"], dict(child)
live_stock = {
    row["sku"]: (row["on_hand"], row["reserved"])
    for row in live_db.connection.execute("SELECT sku,on_hand,reserved FROM inventory")
}
assert live_stock == SEEDED_STOCK
assert live_db.connection.execute("SELECT count(*) FROM assistant_orders").fetchone()[0] == 0
assert all(m.get("tool_call_id") for m in live_result.messages if m["role"] == "tool")

Challenge: an allowance of one, then a paused runtime

The contract is data. recorded_turns(1) keeps every turn the same and changes only the allowance the model settles on. The worker's loop admits one model call, OfflineCateringModel asks for the quote with it, and the loop ends with MODEL_CALL_LIMIT before an answer exists. run_once grades that as not passed and finishes the child as BLOCKED.

Predict before running: what does research_report return for status and model_calls, and does quote called directly still say two tubs? The scripted answer still claims the worker verified the draft; the ledger and the child's row say otherwise. Fixed arithmetic earned the right answer with zero model calls; the delegation spent one and proved nothing.

Then pause the runtime with one UPDATE assistant_control SET paused=1 and rerun the transcript on the same desk: the gate refuses both assignments before delegate runs, and the worker finds no claimable work. With a live key, rerun the live cell after either change and explain whether the model noticed. In your notes, separate what the scripted run proves (the contract, the grading, the gate, the rows) from what only a live run can show (the model's choice of allowance). The saved record keeps the two apart.

tight_db, tight_parent = open_catering_desk(LLM_CATERING_DB)
tight_dispatcher, tight_refusals = catering_desk(tight_db, tight_parent)
tight_result = run_loop(
    ScriptedModel(recorded_turns(1)), tight_dispatcher, CATERING_MESSAGES, limits=LOOP_LIMITS
)
print(describe(tight_result, "scripted"))
tight_trace = tool_trace(tight_result)
assert tight_result.status == "COMPLETED"
assert tight_trace[3]["result"]["status"] == "BLOCKED"
assert tight_trace[3]["result"]["passed"] is False
tight_report = tight_trace[4]["result"]
assert tight_report["status"] == "BLOCKED" and tight_report["model_calls"] == 1
assert quote(tight_db, Inquiry(sku="SKU-VANILLA", guests=11))["tubs"] == 2

with tight_db.immediate() as connection:
    connection.execute("UPDATE assistant_control SET paused=1 WHERE id=1")
paused_dispatcher, paused_refusals = catering_desk(tight_db, tight_parent)
paused_result = run_loop(
    ScriptedModel(RECORDED_TURNS), paused_dispatcher, CATERING_MESSAGES, limits=LOOP_LIMITS
)
paused_trace = tool_trace(paused_result)
print("Paused:", [step["ok"] for step in paused_trace], paused_refusals)
assert [step["ok"] for step in paused_trace] == [True, False, False, False, True]
assert paused_refusals == [
    ("delegate_research", "runtime paused"),
    ("delegate_research", "runtime paused"),
    ("run_worker", "runtime paused"),
]

llm_lab_report = {
    "unit": "ch18-a",
    "scripted": {
        "status": scripted_result.status,
        "tool_calls": scripted_result.tool_calls,
        "refused": [list(entry) for entry in catering_refusals],
        "quote": scripted_trace[0]["result"],
        "research": scripted_report,
    },
    "session": {
        "source": chosen_model.source,
        "model": DEFAULT_MODEL if chosen_model.source == "live" else None,
        "status": live_result.status,
        "model_calls": live_result.model_calls,
        "tool_calls": live_result.tool_calls,
        "refused": [list(entry) for entry in live_refusals],
        "research": [dict(child) for child in live_children],
        "answer": live_result.answer[:400],
    },
    "allowance_of_one": {"research": tight_report, "quote_tubs": 2},
    "paused": {"refused": [list(entry) for entry in paused_refusals]},
}
llm_report_path = COURSE_WORK / "ch18-a-llm-lab-report-v1.json"
llm_report_path.write_text(json.dumps(llm_lab_report, indent=2, sort_keys=True), encoding="utf-8")
print("Saved evidence:", llm_report_path)
print("LLM_LAB_REPORT=" + json.dumps(llm_lab_report["session"], sort_keys=True))
for open_db in (catering_db, live_db, tight_db):
    open_db.close()

Keep building with Prof Rod

Found this material through a colleague, classroom or shared download? Get the complete book at profrod.ai/book and join the Prof Rod learner community. Bring one result, one question or one failure you learned from. Share this resource with another learner and keep its source links with it so they can find the full course and future updates.