Topic 12 · Prompt strategy + source safety

Prompt Strategy and Source BoundariesChoose a prompting strategy, test it on fixed cases, and keep untrusted context inside its evidence boundary.

A longer prompt is not automatically a better prompt. Students select a strategy for a bounded task, compare behavior on fixed synthetic cases, revise from evidence, and constrain source use so retrieved text cannot silently change the task.

Before class

Bring or recreate one bounded project task, one weak prompt, and one human-check rule from Topic 11, or use the supplied Study Sprint task. No prior packet, live model, AI account, paid tool, public URL, private upload, live web, or prompt history is required.

By the end

Four outcomes you can demonstrate.

The focus is judgment students can show, not vocabulary they can merely repeat.

01 · ROUTE

Choose direct instruction, examples, rubric critique, or source-bound prompting for a task.

02 · TEST

Compare weak and improved prompts using observable criteria and fixed synthetic cases.

03 · BOUND

Treat supplied or retrieved text as inspectable evidence rather than hidden authority or instructions.

04 · HANDOFF

Record prompt, evidence, failure, revision, human checkpoint, and one Topic 13 security risk.

Class route

Two meetings with separate learning arcs.

Meeting 1 plans 68 taught minutes plus a 12-minute flex and access buffer; Meeting 2 plans 80 with no separate buffer.

MEETING 1

Choose and Test a Prompt Strategy

Recover a bounded task, route it to a strategy, work through three strategy rounds, build a selective prompt contract, and run a three-case mini-evaluation before revising.

80 MIN

00:00–00:17
Task before wordingRecover or select a bounded task; identify the job and the behavior that would count as success.
00:17–00:45
Three strategy roundsCompare direct instruction, examples, and neutral rubric critique through commit-before-reveal cases.
00:45–01:04
Contract + fixed-case mini-evalChoose only load-bearing prompt fields, test three fixed cases, and revise one component.
01:04–01:08
Preserve the experimentRecord prompt, test evidence, failure, revision, and human-check rule.
01:08–01:20
Flex/access bufferUse projected outputs, the account-free lab, paper, private writing, speech-to-text, or LMS text.

Start Meeting 1

MEETING 2

Build a Source Boundary and Security Handoff

Inspect a synthetic source pack, separate evidence from instructions, complete three source-bound rounds, audit an answer claim-by-claim, add permissions and stop rules, and package the Topic 13 handoff.

80 MIN

00:00–00:04
LaunchOne vague rule, six cases, four roles, one laptop per table.
00:04–00:20
Cases 1–2 · source role, embedded instructionWhat may enter the prompt; what a line inside a source may not make it do.
00:20–00:36
Cases 3–4 · not stated, empty ruleWhen the pack cannot answer; when an instruction decides nothing.
00:36–00:52
Cases 5–6 · permission, the handoff claimWhat a workflow may do unasked; what you may honestly say you found.
00:52–01:12
Audit + Experiment CardMap claims to sources against the rules the table wrote, then assemble three pages of evidence.
01:12–01:20
Security handoffName the untrusted-context risk Topic 13 should test; this screen also holds the access and transfer time.

In-class exerciseSource Boundary Simulation

Six cases, four roles, one laptop per table. Your table relays through prompter, reviewer, source owner and security, then writes the rule the case needed. Open it when your instructor names the case; it also works alone.

Open the simulation

Start Meeting 2

Evidence

Prompt Experiment Card

Define one bounded task; compare a weak and improved prompt; justify the selected strategy; run three fixed synthetic cases; revise from evidence; add a source rule, assistance disclosure, limitation, human-review checkpoint, and Topic 13 untrusted-context handoff.

What to submit

Submit the three-page card through the course LMS only if the instructor assigns it. Preferred filename: Lastname-Topic12-PromptExperiment.pdf. LMS text, DOCX/export, scanned paper, speech-to-text, projected/transcript evidence, or an instructor-approved accessible alternative is accepted. The LMS supplies the exact deadline, points, late rule, and resubmission rule.
Include one bounded task, observable success criterion, weak prompt, selected strategy and reason, improved prompt contract, and one prompt element intentionally omitted with a reason.
Compare the same three synthetic or approved class-safe cases before and after the repair. Record expected behavior, observed behavior, pass/revise/not tested status, one failure, one targeted revision, and one tradeoff or remaining limit.
Include a source-bound rule that treats supplied text as evidence rather than authority, rejects embedded instructions, uses not-stated behavior for missing evidence, and names a human-review checkpoint before any consequential external action.
No live model, AI account, paid tool, public URL, prior packet, private prompt, private upload, prompt length, generated volume, or polished output determines core credit.

Open assignment

Boundary

Use fictional, public, instructor-provided, or approved class-safe material. Do not upload or submit private prompts, full chats, credentials, account/device screens, private source packs, restricted course files, grades, health/identity data, consent/assent records, unpublished peer work, or unnecessary identity evidence. Your own name is permitted only in the card header and is never saved by the browser draft feature.

Access: The account-free Prompt Strategy Lab and course-owned synthetic outputs/source cards are the required route. Own-task, supplied-task, projected/transcript, paper, LMS text, speech-to-text, and approved accessible routes earn equal credit for equivalent decisions and evidence. Prior Topic 11 work, live browsing, a live model, AI access, public deployment, and external certificates are not gates.

AI disclosure: Name any tool or person that materially shaped the submitted prompt, output analysis, or revision; state what help was used and what you accepted, changed, rejected, or personally verified. Do not submit full chats or private prompt histories.

References

Core sources and optional extensions.

The class deck contains the required explanation. These links support verification, deeper reading, or alternate access.

Editable Prompt Experiment CardPrimary three-page route for task, strategy, mini-eval, source boundary, disclosure, and Topic 13 handoff.Printable Prompt Experiment Card PDFPaper/print route. Use HTML, LMS text, DOCX/export, speech-to-text, or an approved accessible route when needed.Prompt Strategy LabAccount-free strategy and source-bound rounds with deterministic synthetic outputs; no storage or network requests.Andrew Ng Prompting CourseOptional enrichment. The course-owned cases are sufficient; video access and completion proof are not required.AI Workflow Skill MapOptional course map for planning, implementation, review, and handoff skills.Writing for AgentsOptional source on load-bearing instructions. Course rules remain vendor-neutral, and a skill is not required.Self-Improving Agent SystemsOptional advanced enrichment. It is ungraded and not required for Topic 13.20-Minute Advanced Visual LessonOptional advanced PDF; not part of the required Topic 12 route.

Next week

Topic 13: Security Review and Evidence

Carry one untrusted-context entry point, possible task distortion, control, and synthetic test into the security review.

Open next week