Week 11 · lesson

Lab: Audit an Objective Function

Mission

Run the local Objective Function Audit Lab. Rank the same supplied behaviors under naive and safety-aware objectives, preserve every operand, and export an Objective Function Audit Record.

Guided start

Select Naive objective and calculate rankings.
Confirm the shortcut ranks first with score 11.
Inspect why repeated checkpoint hits help the loop.
Switch to Safety-aware objective without changing behavior evidence.
Confirm the safe route ranks first with score 10.

Procedure

Preserve all candidate measurements and both formulas.
Predict each ranking before running it.
Calculate every naive-objective score.
Calculate every safety-aware score.
Identify the ranking reversal and the loophole each added term addresses.
Preserve your first highest-score claim and write a bounded correction.
Complete Observation, Measurement, Decision, and Limitation.
Export the record and confirm it names ai-for-beginners-objective-function-audit-lab.

Success criteria and boundary

candidate evidence remains fixed across objectives;
each score is reproducible from displayed operands;
ranking changes are attributed to formula changes;
loopholes and omitted consequences remain visible; and
the limitation rejects safety, optimality, transfer, deployment, and real-world authorization claims.

Use only supplied fictional behaviors. The Lab controls no agent or device.

Robotnix Lab · contract v1

Objective Function Audit Lab

Compare fixed behaviors under naive and safety-aware objectives and expose ranking reversals and reward loopholes.

Help and boundaries

A deterministic browser simulation scores invented behavior measurements and controls no agent, environment, or device.

No personal data, private prompts, or API keys. This Lab makes no hidden remote writes and does not persist your work after the page session.

No result trains, controls, or deploys an agent or authorizes action.

Lab draft is active for this page session only.

Objectives

Calculate two objective functions from preserved operands.
Explain ranking reversal and repeated-reward loopholes.
Separate score optimization from safety, true optimality, transfer, and authorization.

Procedure

Inspect candidatesRecord goal, steps, hits, unique checkpoints, and hazards for all behaviors.
Calculate objectivesPreserve every operand and score under both formulas.
Compare rankingsIdentify the reversal and explain which formula terms caused it.
Bound optimization claimSeparate ranking evidence from safety, true optimality, transfer, deployment, and authorization.

Workbench

Inspect the objective before trusting the ranking.

Objective

Select an objective and calculate the deterministic rankings.

Boundary: These fictional scores do not establish safety, true optimality, transfer to another environment, or permission to train or deploy a real agent.

Text and static fallback

Every measurement, formula, operand, score, ranking, loophole, and boundary appears in accessible text and tables.

Naive ranks shortcut 11 above safe route 10 and loop 8; safety-aware ranks safe route 10 above shortcut -1 and loop -6.

Evidence record

Record only the evidence requested. Do not enter names, personal information, private prompts, or credentials. This draft stays in memory only unless you explicitly export it.

Candidate evidence *Record every candidate measurement and both objective formulas.Scores and rankings *Record all six calculations, both rankings, and the reversal.Optimization claim repair *Preserve and repair one highest-score-means-best claim.Limitation *State what the fictional audit cannot establish about a trained or deployed optimizer.

Success criteria

Candidate evidence and formulas remain explicit.
Every score and ranking is reproducible from displayed operands.
The correction follows objective-specific evidence only.
The limitation rejects unsupported safety, optimality, transfer, and deployment claims.

Complete every procedure step and required evidence field before export.

Import evidence JSON

Vocabulary lab

Flip the idea, not just the card

Explain the term before you reveal the back. Then compare your explanation with the definition, example, and warning.

Read all terms without animation

objective function: A numeric rule an optimization process attempts to improve. Example: The Lab provides naive and safety-aware formulas. Do not confuse it with: It is a model of priorities, not the priorities themselves.
constraint: A requirement or boundary that limits acceptable candidates or actions. Example: Zero hazard contacts could be enforced separately from reward. Do not confuse it with: A penalty may not be equivalent to a hard constraint.
ranking reversal: A change in candidate order caused by a changed evaluation rule. Example: Shortcut leads naively; safe route leads with hazard penalties. Do not confuse it with: The candidate evidence did not change.
exploration: Trying less-known actions or candidates to gather information or find alternatives. Example: A policy tests a different route. Do not confuse it with: Exploration is not permission for unbounded real-world risk.
Objective Function Audit Record: Evidence preserving candidates, formulas, scores, rankings, claim repair, and limitation. Example: The Lab export. Do not confuse it with: It is not evidence from a trained agent or deployed optimizer.