Carry evidence safely through delegation

Published and revised: October 13, 2026 · Version 1.0.0

Connect source grounding, injection boundaries, and specialist return checks into one manager-owned decision.

Goal and proposed time

Decide when a source can support a fact without allowing it to choose tools, recipients, or permissions.

Allow 120–160 minutes, including selected reading, making the outputs, and review. This is a proposed study budget, not measured completion time. Split the path across sessions if needed; stop at a checkpoint rather than rushing an evidence check.

Prerequisites and inputs

• Complete the first bounded-experiment path, or arrive with an equivalent permission matrix
• Read simple structured records and keep source IDs with claims

Use only the linked guides and their public or synthetic fixtures. A text editor, calculator, and timer are enough. No provider connection, account change, publication, or outside communication is required.

Work through the path

1. Ground an answer in a small, inspectable source set

Read the guide. Complete the guest-led policy question and its conflict variant. Keep authority and applicability separate from whether a passage looks relevant.

Carry forward: A claim ledger and an explicit unresolved-conflict outcome.

2. Rehearse the boundary between source text and instructions

Read the guide. Inspect the three clean/contaminated pairs. Record factual contamination separately from prohibited action proposals and false blocks.

Carry forward: A boundary checklist for text entering the next role.

3. Bound a multi-agent task with a handoff packet

Read the guide. Prepare manager-owned assignments. Define what a complete specialist result means, and validate returns before joining them by room ID.

Carry forward: Two minimum-context packets and a manager acceptance checklist.

4. Evaluate an agent and learn from its failures

Read the guide. Adapt the four observable checks to the capstone: outcome, evidence, boundaries, and truthful action reporting. Review first attempts without coaching them into a pass.

Carry forward: A scoring sheet that retains rejected returns and precise stop reasons.

Worked practice

Complete Keep an untrusted instruction out of a specialist handoff. Minimize specialist inputs, validate returns, and stop source text from acquiring authority during a manager-led room comparison.

In the capstone, R1 costs 4800 cents and R2 costs 6300 cents. The manager must combine cost and suitability, reject the claim that a room is booked, and preserve unknown availability. A plausible specialist summary is not a substitute for that final check.

Output

A source-to-claim ledger, two restricted assignment packets, checked specialist returns, and a manager decision with a boundary-failure log.

Exit checks

• Current applicable evidence supports each material claim
• Untrusted notes do not become instructions in summaries or handoffs
• Specialists receive only fields needed for their questions
• Manager validates task IDs, versions, calculations, and combined constraints
• No booking, new recipient, or recursive delegation is inferred

Decide what happens next

Keep the result when the checks pass. Revise a specific failed check in a new draft, preserving the first attempt. Hold any part that needs missing evidence or authority; an honest hold can complete this learning path even though the wider real-world task remains unfinished. Do not call an authored answer key an observed result.

IIEN means infer intent and extrapolate needs here: use it to notice a useful next question and label assumptions. It does not authorize new actions, expand access, or tune model weights.

Sources and scope

This is an editorial sequence of the linked existing guides. Their source sections support the individual methods; the sequence, checkpoints, and study budget are original teaching choices. This path does not claim certification, deployed capability, or measured performance improvement.