Teaching and Source Guide

Hands-on: Three Operator Cases — notebook-aligned tutorial. Evidence reviewed 24 October 2026. The full reading guide is separate from the timed main talk. Corporate issuers are listed when no individual author is recorded. Original figures, page locations and image-source credits are included. Attribution alone is not a blanket reuse license.

Using the case and source guide

Format
50 self-contained concept slides and three what-if discussions. No software or account is required.
Learning route
Understand the problem, change one condition, follow the agent request and numerical result, then identify the operator decision.
Evidence
Numerical figures are prepared results. Recorded language-model replies retain their own case inputs and scope. A different recorded study is explicitly labeled. These examples do not demonstrate field deployment or general performance gains.
Practice afterward
The participant lab guide and notebooks are separate optional practice material. Their setup and cell labels are not needed to understand this presentation.
Credits
External Stanford material retains its credits. Full source and calculation provenance are collected here; the authors’ own case slides have no source footer.

1. Agentic AI: Three Operator Cases

Timing
00:00–01:00 (1 minutes)
Slide content
Understand the problem, follow the agent and review the evidence. 90 minutes / three what-if discussions
Teaching notes
The examples are self-contained. No computer or software setup is needed. We will follow a goal, the agent’s requests, numerical evidence and a human decision. The results come from reduced or synthetic studies; they are examples for learning.
[H00] Classroom Labs notebook. Setup: COLAB_GUIDE.md; Classroom_Labs.ipynb. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/COLAB_GUIDE.md
[B01] IEEE Communications Society. IEEE Communications Society: official resources and network presentation template. Reviewed 17 October 2026.
https://www.comsoc.org/about/brand-experience/resources

2. An agent chooses what to try next

Timing
01:00–02:00 (1 minutes)
Slide content
Give the agent a goal. It requests a tool, reads the result and chooses the next step.
Teaching notes
The agent does not calculate the electrical result by writing a confident answer. It chooses a permitted tool, reads the result and decides whether another request or human review is needed. Trace the feedback arrow once with the audience.
Figure scope
Teaching diagram: the model requests a tool and uses the returned evidence.
Editable figure
diagrams/plain_agent_loop.tex
Notebook / source locator
Monica Lam. CS224V Agentic AI, Lecture 1: Introduction. Stanford University, Fall 2026, p.2.
[ST01] Monica Lam. CS224V Agentic AI, Lecture 1: Introduction. Stanford University. Fall 2026.
https://web.stanford.edu/class/cs224v/lectures_2026/l-introduction.pdf
[DEF03] Erik S., Barry Zhang. Building effective agents. Anthropic Engineering. 2024-12-19.
https://www.anthropic.com/research/building-effective-agents
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

3. The agent asks; the tool calculates

Timing
02:00–02:30 (0.5 minutes)
Slide content
A tool does one job. An agent can connect several jobs using their results.
Teaching notes
The Stanford figure distinguishes one function call from a longer agentic workflow. A function can perform one calculation. An agentic workflow connects requests and uses returned results to decide what comes next.
Figure scope
Original Stanford figure: one function call and a multi-step workflow.
Figure asset
images/stanford_agentic_capabilities.png
Notebook / source locator
Monica Lam. CS224V Agentic AI, Lecture 1: Introduction. Stanford University, Fall 2026, p.2.
[ST01] Monica Lam. CS224V Agentic AI, Lecture 1: Introduction. Stanford University. Fall 2026.
https://web.stanford.edu/class/cs224v/lectures_2026/l-introduction.pdf
[DEF03] Erik S., Barry Zhang. Building effective agents. Anthropic Engineering. 2024-12-19.
https://www.anthropic.com/research/building-effective-agents
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

4. Solar settings: which option should we try?

Timing
02:30–03:00 (0.5 minutes)
Slide content
Compare possible settings, then check voltage and line current. The agent can organize these trials and report the checked options.
Teaching notes
Suppose an operator needs new solar settings. Different settings affect voltage, current and useful solar output. The agent organizes a bounded study; the numerical tool supplies the engineering evidence.
Figure scope
The tool calculates; the agent coordinates the next request.
Editable figure
diagrams/stanford_colab_voltage.tex
[ST01] Monica Lam. CS224V Agentic AI, Lecture 1: Introduction. Stanford University. Fall 2026.
https://web.stanford.edu/class/cs224v/lectures_2026/l-introduction.pdf
[ST02] Sina J. Semnani, Violet Z. Yao, Heidi C. Zhang, Monica S. Lam. WikiChat: Stopping the Hallucination of Large Language Model Chatbots by Few-Shot Grounding on Wikipedia. Findings of the Association for Computational Linguistics: EMNLP 2023, pp.2387–2413. December 2023.
https://aclanthology.org/2023.findings-emnlp.157/
[H02] Classroom Labs notebook. Classroom_Labs.ipynb, L2; cell classroom-007; src/continuous_control.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

5. Give the agent a goal and clear limits

Timing
03:00–04:00 (1 minutes)
Slide content
State the goal, allowed tools and stopping rules. Keep the operator’s review step explicit.
Teaching notes
A useful request states the operating goal, available tools and stopping conditions. The agent needs to know when it must ask for review rather than continue trying options. The operator defines these boundaries.
Figure scope
Grid teaching adaptation of Stanford’s task-specification concept.
Editable figure
diagrams/plain_rules.tex
Notebook / source locator
Monica Lam. CS224V Agentic AI, Lecture 1: Introduction. Stanford University, Fall 2026, p.40.
[ST01] Monica Lam. CS224V Agentic AI, Lecture 1: Introduction. Stanford University. Fall 2026.
https://web.stanford.edu/class/cs224v/lectures_2026/l-introduction.pdf
[ST03] Harshit Joshi, Shicheng Liu, James Chen, Larsen Weigle, Monica Lam. Controllable and Reliable Knowledge-Intensive Task-Oriented Conversational Agents with Declarative Genie Worksheets. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, Volume 1: Long Papers, pp.27264–27308. July 2025.
https://aclanthology.org/2025.acl-long.1323/
[H01] Classroom Labs notebook. Classroom_Labs.ipynb, L0–L1; cells classroom-002 and classroom-004. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

6. Restoration: who should get power first?

Timing
04:00–04:30 (0.5 minutes)
Slide content
The operator sets the service priority. That priority changes the study the agent requests.
Teaching notes
When supply is limited, restoring the greatest total power can differ from protecting essential services. The operator chooses that priority. The agent must preserve it when requesting and comparing studies.
Figure scope
Same network and limits, different service goals.
Editable figure
diagrams/stanford_colab_restoration.tex
[ST01] Monica Lam. CS224V Agentic AI, Lecture 1: Introduction. Stanford University. Fall 2026.
https://web.stanford.edu/class/cs224v/lectures_2026/l-introduction.pdf
[ST03] Harshit Joshi, Shicheng Liu, James Chen, Larsen Weigle, Monica Lam. Controllable and Reliable Knowledge-Intensive Task-Oriented Conversational Agents with Declarative Genie Worksheets. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, Volume 1: Long Papers, pp.27264–27308. July 2025.
https://aclanthology.org/2025.acl-long.1323/
[H06] Classroom Labs notebook. Classroom_Labs.ipynb, L6; cell classroom-018; src/standard_formulations.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

7. Keep three jobs clear

Timing
04:30–05:00 (0.5 minutes)
Slide content
The agent requests a study. The tool calculates. The operator reviews the result.
Teaching notes
Keep the jobs separate. The agent coordinates requests. The numerical tool calculates. The operator reviews the evidence and decides the next authorized step. A tool passing its checks is not operating permission.
Figure scope
Adapted principle: language, calculation and human review have different jobs.
Editable figure
diagrams/plain_three_jobs.tex
Notebook / source locator
Monica Lam. CS224V Agentic AI, Lecture 1: Introduction. Stanford University, Fall 2026, p.53.
[ST01] Monica Lam. CS224V Agentic AI, Lecture 1: Introduction. Stanford University. Fall 2026.
https://web.stanford.edu/class/cs224v/lectures_2026/l-introduction.pdf
[H09] Classroom Labs notebook. Classroom_Labs.ipynb, L9; cell classroom-026; fixed trusted-recovery fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

8. Wrong readings can still fit a model

Timing
05:00–05:30 (0.5 minutes)
Slide content
A good fit alone does not prove that the readings are true. The agent can organize checks and flag missing evidence.
Teaching notes
Readings may fit a model and still be wrong. An alarm test measures one kind of mismatch. The agent can request additional checks and identify missing information, but its explanation cannot make a data source trustworthy.
Figure scope
Model fit and trusted information answer different questions.
Editable figure
diagrams/stanford_colab_cyber.tex
[ST01] Monica Lam. CS224V Agentic AI, Lecture 1: Introduction. Stanford University. Fall 2026.
https://web.stanford.edu/class/cs224v/lectures_2026/l-introduction.pdf
[H08] Classroom Labs notebook. Classroom_Labs.ipynb, L8; cell classroom-023; four-bus DC fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb
[H09] Classroom Labs notebook. Classroom_Labs.ipynb, L9; cell classroom-026; fixed trusted-recovery fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

9. Case 1: which solar settings should we use?

Timing
05:30–07:30 (2.0 minutes)
Slide content
Use solar generation while keeping voltage and line current within the allowed range. An agent can organize the trials and return checked options.
Teaching notes
Start with the physical problem: choose solar controls that respect voltage, current and equipment capacity while serving the stated generation goal. The agent can organize trials and return a checked option. It does not replace the solver or the operator.
Figure scope
Concept sketch: choose settings, then calculate their effects.
Editable figure
diagrams/plain_voltage_problem.tex
[H02] Classroom Labs notebook. Classroom_Labs.ipynb, L2; cell classroom-007; src/continuous_control.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb
[VCASE] Existing GridGuard workshop / SMART-DS synthetic-feeder extraction. Synthetic Greensboro feeder uhs20_1247–udt8894: model map and declared study scope. Existing teaching snapshot reviewed 18 October 2026.
https://data.openei.org/submissions/2981

10. Two tools suggest different settings

Timing
07:30–08:30 (1 minutes)
Slide content
MPC plans with a model. SAC uses a previously trained controller. Treat both outputs as proposals that need the same electrical checks.
Teaching notes
MPC means model predictive control: it plans with a model. SAC means soft actor-critic: it uses a previously trained policy. Both produce proposals. Apply the same electrical checks to each before comparing cost or usefulness.
Figure scope
MPC: model predictive control. SAC: soft actor-critic. Both proposals need the same checks.
Editable figure
diagrams/plain_two_proposals.tex
[H02] Classroom Labs notebook. Classroom_Labs.ipynb, L2; cell classroom-007; src/continuous_control.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb
[VMPC] James B. Rawlings; David Q. Mayne; Moritz M. Diehl. Model Predictive Control: Theory, Computation, and Design, second edition. Nob Hill Publishing. 2017.
https://sites.engineering.ucsb.edu/~jbraw/mpc/MPC-book-2nd-edition-1st-printing.pdf
[VSAC] Tuomas Haarnoja; Aurick Zhou; Pieter Abbeel; Sergey Levine. Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor. Proceedings of the 35th International Conference on Machine Learning, PMLR 80, pp. 1861–1870. 2018.
https://proceedings.mlr.press/v80/haarnoja18b.html
[LRV] Workshop recorded model replies and numerical study records. Voltage P3_current_reserve: Llama 3.1:8b planner, recovery and review calls 1–3. Public teaching archive. 27 September 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/continuous_llama_opendss_20260927T161223523552Z.json

11. How the agent helps with solar settings

Timing
08:30–11:30 (3.0 minutes)
Slide content
The agent requests a study and reads the returned checks. It can try another allowed option or ask the operator for help.
Teaching notes
Read the loop from left to right. The operator supplies a goal. The agent requests an allowed trial. The tool calculates voltage and current. The checks decide whether it can be compared. A failed trial can lead to another permitted request or a request for help.
Figure scope
The agent uses returned checks to choose the next permitted step.
Editable figure
diagrams/plain_voltage_agent.tex
[LRV] Workshop recorded model replies and numerical study records. Voltage P3_current_reserve: Llama 3.1:8b planner, recovery and review calls 1–3. Public teaching archive. 27 September 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/continuous_llama_opendss_20260927T161223523552Z.json
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

12. Two solar settings change the result

Timing
11:30–12:30 (1 minutes)
Slide content
One setting controls solar power output. The other provides voltage support. The tool checks voltage, line current and the solar unit’s capacity.
Teaching notes
The two controls are active-power output and voltage support. The source model represents them as a and b. They share inverter capacity, so reducing one effect can change another. Use the tool to calculate those effects rather than infer safety from the control names.
Figure scope
Active power and voltage support share equipment capacity.
Editable figure
diagrams/plain_solar_controls.tex
[H02] Classroom Labs notebook. Classroom_Labs.ipynb, L2; cell classroom-007; src/continuous_control.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

13. Both proposals face the same checks

Timing
12:30–16:30 (4 minutes)
Slide content
The tool calculates the effect of each proposal. Compare their scores only after voltage, current and capacity checks pass.
Teaching notes
The same checks apply to both candidates. Check voltage, line current and inverter capacity first. Only a passing proposal is eligible for cost comparison. The agent’s next request should be based on the returned checks.
Figure scope
Both proposals need the same voltage, current and capacity checks.
Editable figure
diagrams/concept_solar_checks.tex
[H02] Classroom Labs notebook. Classroom_Labs.ipynb, L2; cell classroom-007; src/continuous_control.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb
[VMPC] James B. Rawlings; David Q. Mayne; Moritz M. Diehl. Model Predictive Control: Theory, Computation, and Design, second edition. Nob Hill Publishing. 2017.
https://sites.engineering.ucsb.edu/~jbraw/mpc/MPC-book-2nd-edition-1st-printing.pdf
[VSAC] Tuomas Haarnoja; Aurick Zhou; Pieter Abbeel; Sergey Levine. Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor. Proceedings of the 35th International Conference on Machine Learning, PMLR 80, pp. 1861–1870. 2018.
https://proceedings.mlr.press/v80/haarnoja18b.html

14. Read the voltage and current checks

Timing
16:30–17:30 (1 minutes)
Slide content
Left: voltages should stay in the allowed band. Right: current should stay below 100% of the allowance.
Teaching notes
The left plot shows the voltage distribution. The right compares current use with the allowed current. Both pass this model’s checks. The curves on the left are sorted separately, so their horizontal rank is not a matched physical location.
Figure scope
Calculated example at the original current allowance.
Figure asset
images/hands_on_L2_plot.png
[H02] Classroom Labs notebook. Classroom_Labs.ipynb, L2; cell classroom-007; src/continuous_control.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

15. Both pass, but the trade-off is different

Timing
17:30–18:30 (1 minutes)
Slide content
MPC uses 73.62% of the current allowance; SAC uses 51.68%. MPC has the lower overall score because current is only one part of the goal.
Teaching notes
The current-use values are 73.62% for MPC and 51.68% for saved SAC. Their scores are approximately 0.03110 and 0.43643. Lower score is preferred after all checks pass. SAC uses less current here partly because it also uses less active power.
Figure scope
Calculated example: both electrical checks pass.
Editable figure
diagrams/concept_solar_result.tex
[H02] Classroom Labs notebook. Classroom_Labs.ipynb, L2; cell classroom-007; src/continuous_control.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

16. What if we allow less current?

Timing
18:30–24:30 (6 minutes)
Slide content
Reduce the allowance from 80% to 75% of nominal current. Which proposal might fail? Which would you choose if both pass? What must the agent ask the tool to check again?
Teaching notes
Ask people to predict without doing any calculation. Make the rule change explicit: allowed current decreases from 80% to 75% of nominal. Give pairs time to explain what must be checked again and why the lowest current is not automatically the best proposal. Advance only after the discussion.
Figure scope
Discuss the scenario, then advance to the calculated result.
Editable figure
diagrams/concept_current_change.tex
[H02] Classroom Labs notebook. Classroom_Labs.ipynb, L2; cell classroom-007; src/continuous_control.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

17. Both proposals pass the new current check

Timing
24:30–26:30 (2 minutes)
Slide content
MPC has the lower score. SAC uses less current and less solar output. These results cover the medium-voltage model. A wider grid needs a wider study.
Teaching notes
The numerical study has already been calculated for this changed allowance. MPC uses about 78.53% of the new allowance and SAC about 54.85%; both pass. Their scores are about 0.03110 and 0.43246. MPC has the lower score. These checks cover the medium-voltage model, not the entire distribution system.
Figure scope
Same load; the allowed current has been reduced to 75% of nominal.
Editable figure
diagrams/exercise_voltage_solution.tex
[H02] Classroom Labs notebook. Classroom_Labs.ipynb, L2; cell classroom-007; src/continuous_control.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

18. Less current does not mean the best option

Timing
26:30–27:30 (1 minutes)
Slide content
Here SAC also uses less solar output. The operator’s goal values solar generation as well as operating limits.
Teaching notes
In the changed-allowance example, the active-power setting is about 96% of inverter rating for MPC and 34% for SAC. These percentages use inverter rating as the base. Less current comes with a generation trade-off, so the operator’s objective matters.
Figure scope
Active-power settings use inverter rating as their base.
Editable figure
diagrams/plain_voltage_tradeoff.tex
[H02] Classroom Labs notebook. Classroom_Labs.ipynb, L2; cell classroom-007; src/continuous_control.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

19. A PASS covers only the checked model

Timing
27:30–28:00 (0.5 minutes)
Slide content
A study of the medium-voltage network cannot confirm a whole-grid claim. The agent should request the missing study or return HOLD.
Teaching notes
The model’s scope travels with its result. A passing medium-voltage study does not establish low-voltage or whole-feeder safety. When the request is wider than the evidence, the useful response is a missing-study request or HOLD.
Figure scope
A passing result applies to the model and conditions actually checked.
Editable figure
diagrams/concept_scope.tex
[H05] Classroom Labs notebook. Classroom_Labs.ipynb, L5; cell classroom-016; MV-only default model. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

20. Recorded example: the first option fails

Timing
28:00–30:00 (2 minutes)
Slide content
Llama requests a check of the saved SAC proposal. The numerical tool reports too much current and rejects it.
Teaching notes
This is a separate recorded full-feeder study, with different inputs from the earlier numerical comparison. Read the actual Llama request and returned tool result. The SAC trial reaches about 111.97% of allowed current, so it fails. The tool result determines eligibility.
Figure scope
Recorded full-feeder study; different inputs from the earlier comparison.
Figure asset
images/llama_voltage_1.png
[LRV] Workshop recorded model replies and numerical study records. Voltage P3_current_reserve: Llama 3.1:8b planner, recovery and review calls 1–3. Public teaching archive. 27 September 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/continuous_llama_opendss_20260927T161223523552Z.json
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

21. Recorded example: another option is checked

Timing
30:00–32:00 (2 minutes)
Slide content
Llama requests a check of the existing MPC fallback. The returned electrical checks pass for this recorded case.
Teaching notes
The agent requests a check of an existing MPC fallback. It does not invent or solve MPC. The historical electrical check returns about 99.60% of allowed current and no listed violations. That makes the option eligible for review within this study’s stated scope.
Figure scope
Recorded full-feeder study: the requested fallback passes the checks.
Figure asset
images/llama_voltage_2.png
[LRV] Workshop recorded model replies and numerical study records. Voltage P3_current_reserve: Llama 3.1:8b planner, recovery and review calls 1–3. Public teaching archive. 27 September 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/continuous_llama_opendss_20260927T161223523552Z.json
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

22. Recorded example: the operator receives a proposal

Timing
32:00–34:00 (2 minutes)
Slide content
The agent returns the checked proposal and its conditions. The operator still reviews it before any authorized action.
Teaching notes
The recorded workflow returns the checked proposal, while human acknowledgement remains pending and no equipment command is sent. Ask what evidence the operator receives and what remains outside the study, including permissions and protection checks.
Figure scope
Recorded workflow result; human review remains pending.
Figure asset
images/llama_voltage_3.png
[LRV] Workshop recorded model replies and numerical study records. Voltage P3_current_reserve: Llama 3.1:8b planner, recovery and review calls 1–3. Public teaching archive. 27 September 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/continuous_llama_opendss_20260927T161223523552Z.json
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

23. Case 2: who gets power after a fault?

Timing
34:00–36:00 (2.0 minutes)
Slide content
A fault removes a supply path. Available capacity limits what can be restored. The operator sets priorities; the agent can request and check restoration options.
Teaching notes
A fault removes one supply path in the six-node network. The available backup capacity limits which loads can be restored. Cyan marks the critical loads. The operator defines priorities, while the agent can request alternative restoration studies.
Figure scope
Six-node example. Cyan marks critical loads; dashed links are backup paths.
Editable figure
diagrams/plain_restoration_problem.tex
[H06] Classroom Labs notebook. Classroom_Labs.ipynb, L6; cell classroom-018; src/standard_formulations.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

24. How the agent helps restore service

Timing
36:00–39:00 (3 minutes)
Slide content
Choose the goal: critical loads first, or the most total power. The agent can request a study, read the trade-off and ask for a new study when limits change.
Teaching notes
The goal changes the requested optimization. The agent should keep the operator’s priority, ask the tool for a plan, read service and capacity checks, and request a new study when inputs change. The solver decides which whole loads fit the network constraints.
Figure scope
The operator sets the priority. The agent requests the matching study.
Editable figure
diagrams/plain_restoration_agent.tex
[H06] Classroom Labs notebook. Classroom_Labs.ipynb, L6; cell classroom-018; src/standard_formulations.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

25. The service goal changes the study request

Timing
39:00–42:00 (3 minutes)
Slide content
Critical-first gives more weight to essential service. Most-total-power aims to restore the largest amount of power.
Teaching notes
Critical-first assigns more value to essential service. Most-total-power maximizes total restored kW. Both still need the same network constraints. The agent translates the selected goal into the appropriate permitted study request; the solver performs the optimization.
Figure scope
Two possible operator goals for the same fault and available network.
Editable figure
diagrams/concept_service_request.tex
[H06] Classroom Labs notebook. Classroom_Labs.ipynb, L6; cell classroom-018; src/standard_formulations.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

26. The goal changes who receives power

Timing
42:00–43:00 (1 minutes)
Slide content
Critical-first: 130 kW total, including 90 kW for critical loads. Most-total-power: 140 kW total, including 30 kW for critical loads.
Teaching notes
At a 40 kW backup tie, the critical-first result restores 130 kW total, including 90 kW critical. The total-power result restores 140 kW, including only 30 kW critical. Neither objective removes the need to explain the service trade-off. In the plots, overlapping markers indicate equal outcomes.
Figure scope
Calculated restoration results for the same network and limits.
Figure asset
images/hands_on_L6_plot.png
[H06] Classroom Labs notebook. Classroom_Labs.ipynb, L6; cell classroom-018; src/standard_formulations.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

27. Check the restoration plan

Timing
43:00–43:30 (0.5 minutes)
Slide content
Keep the faulted branch out of service. Check supply paths and capacity. Voltage, protection and field switching need separate studies.
Teaching notes
Check that the fault stays isolated, supply paths connect and capacities are respected. This teaching exercise uses an active-power service model. AC voltages, protection coordination and the order of physical switching require separate studies and procedures.
Figure scope
Active-power model checks. Voltage, protection and field switching require separate work.
Editable figure
diagrams/plain_restoration_checks.tex
[H06] Classroom Labs notebook. Classroom_Labs.ipynb, L6; cell classroom-018; src/standard_formulations.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb
[RSTL01] Ying Wang, Yin Xu, Jiaxu Li, Jinghan He, Xiaojun Wang. On the Radiality Constraints for Distribution System Restoration and Reconfiguration Problems. Author preprint, arXiv:1912.05185v3. 15 March2020(v3); first submitted11 December2019.
https://arxiv.org/abs/1912.05185v3
[RSTL02] Bo Chen, Chen Chen, Jianhui Wang, Karen L. Butler-Purry. Sequential Service Restoration for Unbalanced Distribution Systems and Microgrids. IEEE Transactions on Power Systems33(2),1507–1520. March2018.
https://doi.org/10.1109/TPWRS.2017.2720122

28. What if the backup path can carry more power?

Timing
43:30–49:30 (6 minutes)
Slide content
Increase its capacity from 40 kW to 180 kW. Keep the fault and loads the same. Would the two service goals still give different results? What should the agent recalculate before reporting the new plan?
Teaching notes
Change only backup capacity in the scenario, from 40 to 180 kW. Ask whether the two service goals must always disagree. Invite participants to explain what the agent should recalculate before presenting a new plan. No software is needed to make the prediction.
Figure scope
Discuss the scenario, then advance to the calculated result.
Editable figure
diagrams/concept_capacity_change.tex
[H06] Classroom Labs notebook. Classroom_Labs.ipynb, L6; cell classroom-018; src/standard_formulations.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

29. More capacity removes this service trade-off

Timing
49:30–51:30 (2 minutes)
Slide content
At 40 kW, critical-first serves 90 kW of critical load instead of 30 kW. At 180 kW, both goals serve 250 kW total, including all 90 kW of critical load.
Teaching notes
At 180 kW the two goals both restore 250 kW total, including 90 kW critical. Among the sampled capacities, the first agreement is at 80 kW. That discrete sweep does not establish an exact continuous crossover. More capacity removes the trade-off in this example.
Figure scope
Calculated results with the same fault, loads and service goals.
Editable figure
diagrams/exercise_restoration_solution.tex
[H06] Classroom Labs notebook. Classroom_Labs.ipynb, L6; cell classroom-018; src/standard_formulations.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

30. A new limit needs a new study

Timing
51:30–52:00 (0.5 minutes)
Slide content
Now reduce the allowed backup capacity from 40 kW to 20 kW. The old result no longer supports the request. The agent must request a new study.
Teaching notes
Now imagine a stricter request: the backup capacity falls from 40 to 20 kW. Evidence calculated for the old capacity no longer matches the request. The agent should hold the old proposal and obtain a new constrained study. A failed evidence match does not itself prove that every new plan will fail.
Figure scope
A changed limit requires evidence calculated for the new request.
Editable figure
diagrams/plain_stale_result.tex
[H07] Classroom Labs notebook. Classroom_Labs.ipynb, L7; cell classroom-021; request/evidence binding. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

31. The agent keeps the operator’s goal

Timing
52:00–52:30 (0.5 minutes)
Slide content
The operator sets priorities. The solver selects loads. The agent can request the solve and explain the returned service trade-off.
Teaching notes
In the illustrated priority plan, loads 1, 2 and 5 receive service: 130 kW total and 90 kW critical. The operator sets the goal. The numerical solver selects loads. The agent can request the solve and explain the returned trade-off without changing the goal.
Figure scope
The operator’s priority stays fixed while the tool selects a feasible plan.
Editable figure
diagrams/restoration_plan.tex
[H06] Classroom Labs notebook. Classroom_Labs.ipynb, L6; cell classroom-018; src/standard_formulations.py. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

32. Recorded example: another plan restores more service

Timing
52:30–54:30 (2 minutes)
Slide content
In this separate larger-feeder study, Llama requests another plan. The tool reports more service and passing electrical checks.
Teaching notes
This recorded larger-feeder example is separate from the six-node illustration. Llama requests another catalog option. The tool returns about 1,843.61 kW total and 474.04 kW critical service, with current at about 91.87% of its allowance. A proposed improvement becomes evidence only after the check.
Figure scope
Recorded larger-feeder example, separate from the six-node network.
Figure asset
images/llama_restoration_14.png
[LRR] Workshop recorded model replies and numerical study records. Feeder-catalog restoration: Llama 3.1:8b refinement and ranking calls 14–16. Public teaching archive. 27 September 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/proposal_run_20260927T184402Z_53e0cf/llama_calls.json
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

33. Recorded example: the tool rejects a trial

Timing
54:30–56:00 (1.5 minutes)
Slide content
The next trial carries too much current. It cannot replace the previous passing option.
Teaching notes
The next recorded trial restores more load but reaches about 111.01% of allowed current. It is rejected. The agent must keep the earlier passing option eligible rather than replace it with an overloaded plan.
Figure scope
Recorded larger-feeder example: the overloaded trial is rejected.
Figure asset
images/llama_restoration_15.png
[LRR] Workshop recorded model replies and numerical study records. Feeder-catalog restoration: Llama 3.1:8b refinement and ranking calls 14–16. Public teaching archive. 27 September 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/proposal_run_20260927T184402Z_53e0cf/llama_calls.json
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

34. Recorded example: a wrong ranking is corrected

Timing
56:00–58:00 (2 minutes)
Slide content
Llama puts two passing plans in the wrong cost order. Python keeps the lower-cost plan and checks it again.
Teaching notes
The model ranks two passing plans in the wrong cost order. Since smaller cost is preferred, negative 6109 is lower than negative 5509. Python keeps the lower-cost passing plan and rechecks it. This example shows why the workflow needs a numerical guard.
Figure scope
Recorded larger-feeder example. Lower cost is preferred, including negative costs.
Figure asset
images/llama_restoration_16.png
[LRR] Workshop recorded model replies and numerical study records. Feeder-catalog restoration: Llama 3.1:8b refinement and ranking calls 14–16. Public teaching archive. 27 September 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/proposal_run_20260927T184402Z_53e0cf/llama_calls.json
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

35. Case 3: can we trust these readings?

Timing
58:00–60:00 (2.0 minutes)
Slide content
Wrong readings can still look consistent with the model. The agent can request checks, but action still needs a trusted basis.
Teaching notes
Wrong measurements can look consistent with a grid model. The operating question is whether the evidence supports a correction, not whether the agent can write a convincing explanation. Data trust and model coverage remain explicit conditions.
Figure scope
Concept sketch using a small four-bus model.
Editable figure
diagrams/plain_cyber_problem.tex
[H08] Classroom Labs notebook. Classroom_Labs.ipynb, L8; cell classroom-023; four-bus DC fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

36. How the agent helps with suspicious readings

Timing
60:00–63:00 (3.0 minutes)
Slide content
The agent can request the checks and keep missing evidence visible. The operator reviews trust, model coverage and the proposed correction.
Teaching notes
The agent can request the permitted checks, read the results and report missing evidence. A required missing trust condition should lead to HOLD. The operator reviews the data basis, model scope and proposed adjustment.
Figure scope
The agent requests checks and preserves missing evidence.
Editable figure
diagrams/plain_cyber_agent.tex
[H09] Classroom Labs notebook. Classroom_Labs.ipynb, L9; cell classroom-026; fixed trusted-recovery fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

37. Mismatch is only one check

Timing
63:00–64:00 (1 minutes)
Slide content
Mismatch J measures how well the model fits the readings. The simulation lets us measure how wrong the estimate is (state error).
Teaching notes
Mismatch J measures how well the model fits the readings. It is different from the control-cost score used in the solar case. The simulation also supplies a known true state, allowing an estimation error to be measured. That true state is not normally available to a real operator.
Figure scope
Mismatch measures model fit; state error measures estimation accuracy.
Editable figure
diagrams/plain_mismatch.tex
[H08] Classroom Labs notebook. Classroom_Labs.ipynb, L8; cell classroom-023; four-bus DC fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb
[CY01] Fred C. Schweppe, J. Wildes. Power System Static-State Estimation, Part I: Exact Model. IEEE Transactions on Power Apparatus and Systems, PAS-89(1), pp. 120–125. January 1970.
https://doi.org/10.1109/TPAS.1970.292678

38. Two patterns of wrong readings

Timing
64:00–67:00 (3 minutes)
Slide content
One wrong reading may conflict with the other readings. Coordinated changes may still look consistent with the model.
Teaching notes
One biased reading can disagree with the remaining data. Coordinated changes can agree with one another and still move the estimated state away from truth. This is a conceptual illustration of the two tested patterns; it is not an instruction for creating an attack.
Figure scope
Concept illustration of two patterns used in the numerical comparison.
Editable figure
diagrams/concept_bad_readings.tex
[H08] Classroom Labs notebook. Classroom_Labs.ipynb, L8; cell classroom-023; four-bus DC fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

39. Read the alarm and the estimation error

Timing
67:00–68:00 (1 minutes)
Slide content
Left: mismatch and its alarm threshold. Right: error against the simulation’s known truth. A quiet alarm can appear alongside a growing error.
Teaching notes
The left plot compares mismatch with the alarm threshold. The right compares the state estimate with known simulation truth. A quiet alarm can coexist with a growing error. The agent should preserve this distinction when summarizing results.
Figure scope
Calculated model-fit and state-error results.
Figure asset
images/hands_on_L8_plot.png
[H08] Classroom Labs notebook. Classroom_Labs.ipynb, L8; cell classroom-023; four-bus DC fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

40. A coordinated change can escape the alarm

Timing
68:00–69:00 (1 minutes)
Slide content
At scale 1, one biased reading gives mismatch J=277.50 and an alarm. A coordinated change gives J=3.92, no alarm, and 0.02678 rad of state error.
Teaching notes
At the reference change size, an isolated bias gives mismatch about 277.50 and triggers the alarm. A coordinated change gives about 3.92, no alarm, and about 0.02678 radians of state error. A low mismatch is incomplete evidence about data trust.
Figure scope
The known state is available inside the simulation for this comparison.
Editable figure
diagrams/plain_hidden_error.tex
[H08] Classroom Labs notebook. Classroom_Labs.ipynb, L8; cell classroom-023; four-bus DC fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb
[CY02] Yao Liu, Peng Ning, Michael K. Reiter. False Data Injection Attacks against State Estimation in Electric Power Grids. ACM Transactions on Information and System Security, 14(1), Article 13, 33 pages. May 2011.
https://doi.org/10.1145/1952982.1952995

41. What if the data change is small?

Timing
69:00–75:00 (6 minutes)
Slide content
Reduce the change to one tenth of the reference size. Will the alarm always detect the error? What else would the operator need before trusting the readings?
Teaching notes
Ask pairs to predict what a smaller error does to the alarm. Use one tenth of the reference change size. The question is whether a quiet alarm proves correct readings and what independent information an operator would need. Reveal the calculated example on the next slide.
Figure scope
Discuss the scenario, then advance to the calculated result.
Editable figure
diagrams/concept_bias_change.tex
[H08] Classroom Labs notebook. Classroom_Labs.ipynb, L8; cell classroom-023; four-bus DC fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

42. A quiet alarm can still hide an error

Timing
75:00–77:00 (2 minutes)
Slide content
Neither example triggers this alarm at one tenth of the reference size. Both estimates still contain error. No alarm alone does not establish data trust.
Teaching notes
For the smaller change, mismatch is about 10.14 for isolated bias and 3.92 for the coordinated pattern. Both are below the threshold of about 15.09. Both estimates still contain error. No alarm means this test has not crossed its threshold; it does not authenticate the readings.
Figure scope
Calculated results at one tenth of the reference change size.
Editable figure
diagrams/exercise_cyber_solution.tex
[H08] Classroom Labs notebook. Classroom_Labs.ipynb, L8; cell classroom-023; four-bus DC fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

43. Trusted information is needed before recovery

Timing
77:00–77:30 (0.5 minutes)
Slide content
Without the required trusted readings, the response is HOLD. With readings assumed trustworthy, the tool can calculate a proposal for review.
Teaching notes
The recovery example assumes additional trustworthy readings are available. Without that required condition the result is HOLD. With the assumption, the tool can calculate a bounded proposal. Neither the fitting algorithm nor Llama authenticates those channels.
Figure scope
Additional trustworthy readings are assumed in this example.
Editable figure
diagrams/plain_trust_choice.tex
[H09] Classroom Labs notebook. Classroom_Labs.ipynb, L9; cell classroom-026; fixed trusted-recovery fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

44. The tool calculates a small correction

Timing
77:30–78:00 (0.5 minutes)
Slide content
The solver finds the smallest allowed adjustment that meets the checked branch limits. Predicted flow: 0.17000 p.u. Offline check: 0.17046 p.u. Limit: 0.18000 p.u.
Teaching notes
The fixed recovery example proposes an adjustment of about 0.121694 per unit. The model predicts a monitored flow of 0.17000, while the offline synthetic-truth check gives 0.17046; both are below the 0.18000 limit. Per unit expresses values relative to a common model base. The offline truth check is an experimental aid, not guaranteed field information.
Figure scope
Per-unit values share the same model base.
Editable figure
diagrams/plain_correction.tex
[H09] Classroom Labs notebook. Classroom_Labs.ipynb, L9; cell classroom-026; fixed trusted-recovery fixture. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

45. Recorded example: request the recovery tool

Timing
78:00–79:30 (1.5 minutes)
Slide content
Llama requests the supplied recovery tool. Python calculates a correction using the assumed trusted readings.
Teaching notes
Read the actual recorded Llama request. The task already supplies a permitted recovery tool and the trusted-reading assumption. Python estimates the state and calculates the correction. The language model routes the request; it does not provide the numerical estimate.
Figure scope
Recorded recovery request and numerical tool result.
Figure asset
images/llama_cyber_7.png
[LRC] Workshop recorded model replies and numerical study records. Standard four-bus cyber study: Llama 3.1:8b tool selection and REVIEW calls 7–8. Public teaching archive. 21 September 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/standard_stage_20260921T143826Z_42d19a61/llama_calls.json
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

46. Recorded example: keep human review

Timing
79:30–81:30 (2 minutes)
Slide content
The model checks pass; human review is still pending. Data trust and operating permission still need review.
Teaching notes
The recorded model checks pass, but the overall result remains REVIEW and human acknowledgement is pending. The operator still needs a trusted data basis and operating authority. The fixed workflow obtains the same numerical correction, so this is not evidence that Llama improves estimation.
Figure scope
Recorded workflow result; human review remains pending.
Figure asset
images/llama_cyber_8.png
[LRC] Workshop recorded model replies and numerical study records. Standard four-bus cyber study: Llama 3.1:8b tool selection and REVIEW calls 7–8. Public teaching archive. 21 September 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/standard_stage_20260921T143826Z_42d19a61/llama_calls.json
[LR0] Classroom Labs notebook and Python helper. L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py. Public reproducible teaching material. 24 October 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py

47. Review must refer to the same proposal

Timing
81:30–82:30 (1 minutes)
Slide content
The proposal ID and model scope must match the checked study. A different proposal or wider scope needs a new review.
Teaching notes
A reviewer must refer to the same proposal and scope that were checked. A changed proposal ID or a wider request cannot inherit an earlier passing result. In the teaching example, readback is simulated and sends no equipment command.
Figure scope
Illustrative proposal IDs. Review is separate from equipment operation.
Editable figure
diagrams/concept_review_match.tex
[H12] Classroom Labs notebook. Classroom_Labs.ipynb, L12; cell classroom-033; simulated readback only. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

48. The operator chooses the next step

Timing
82:30–83:30 (1 minutes)
Slide content
Review the case, tool results, checked limits and missing work. Request another study, reject the proposal or take the next authorized step.
Teaching notes
The operator can request more evidence, reject the proposal or continue with a defined authorized step. Review the case, returned limits, model coverage and missing studies. Acceptance of a study is separate from issuing a field command.
Figure scope
The operator reviews the study and chooses the next authorized step.
Editable figure
diagrams/ops_review.tex
[H12] Classroom Labs notebook. Classroom_Labs.ipynb, L12; cell classroom-033; simulated readback only. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb
[OT01] Keith Stouffer, Michael Pease, CheeYee Tang, Timothy Zimmerman, Victoria Pillitteri, Suzanne Lightman, Adam Hahn, Stephanie Saravia, Aslam Sherule, Michael Thompson. Guide to Operational Technology (OT) Security. National Institute of Standards and Technology, NIST SP 800-82 Revision 3. September 2023 (final).
https://doi.org/10.6028/NIST.SP.800-82r3
[EVAL01] National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework (AI RMF 1.0). NIST AI 100-1. 26 January 2023.
https://doi.org/10.6028/NIST.AI.100-1

49. Multi-agent framework used in these cases

Timing
83:30–87:30 (4.0 minutes)
Slide content
The five roles share the same case facts and checked results. One Llama model handles planning and refinement. Python calculates and checks.
Teaching notes
Awareness reads the shared case. Planning prepares permitted options and uses Llama to choose a study. Simulation calculates. Verification checks limits and case identity. The Llama optimizer requests a bounded change or ranks passing options. Every changed trial returns for simulation and verification. Python guards the final choice before operator review. One Llama model fills two roles; this is not five independent language models. The implementation uses case-specific numerical tools behind the same stages.
Figure scope
Implemented five-role workflow with a case-specific numerical tool.
Editable figure
diagrams/workshop_multi_agent.tex
[WF01] Alaa Selim and Mohamed Chehada. Five-role operator study workflow: implementation and recorded cases. GridGuard Agentic AI Tutorials, reproducible Python source and saved study archive. 2026.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/proposal_workflow.py
[H10] Classroom Labs notebook. Classroom_Labs.ipynb, L10–L11; cells classroom-028 and classroom-030. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

50. Explain one case in your own words

Timing
87:30–90:00 (2.5 minutes)
Slide content
What is the operator’s problem, and what should the agent request next? Which tool result supports the proposal, and what must the operator still decide?
Teaching notes
Ask participants to choose one case and explain the complete loop in their own words. They should name the problem, an agent request, a returned numerical fact and a remaining human decision. Optional practice materials can be used after the concept tutorial, but no account or code is needed for this discussion.
Figure scope
Explain the idea first. Practice material is available separately.
Editable figure
diagrams/concept_case_summary.tex
[H00] Classroom Labs notebook. Setup: COLAB_GUIDE.md; Classroom_Labs.ipynb. Public notebook and Python source. 24 October 2026; stable notebook cell IDs checked.
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/COLAB_GUIDE.md

[H00] Setup: COLAB_GUIDE.md; Classroom_Labs.ipynb

Authors / issuer
Classroom Labs notebook
Publisher / venue
Public notebook and Python source
Date
24 October 2026; stable notebook cell IDs checked
Type
Executable teaching material
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/COLAB_GUIDE.md

[B01] IEEE Communications Society: official resources and network presentation template

Authors / issuer
IEEE Communications Society
Publisher / venue
IEEE Communications Society
Date
Reviewed 17 October 2026
Theme assets
comsoc-lockup.png and image7.png; independent educational adaptation
Full source URL
https://www.comsoc.org/about/brand-experience/resources

[ST01] CS224V Agentic AI, Lecture 1: Introduction

Authors / issuer
Monica Lam
Publisher / venue
Stanford University
Date
Fall 2026
Type
University lecture slides, supplied by the presenter
Full source URL
https://web.stanford.edu/class/cs224v/lectures_2026/l-introduction.pdf
Publication
https://web.stanford.edu/class/cs224v/

[DEF03] Building effective agents

Authors / issuer
Erik S., Barry Zhang
Publisher / venue
Anthropic Engineering
Date
2024-12-19
Type
Official practitioner architecture account by the system maintainers
Location
What are agents?; Agents; Workflow: Routing; Workflow: Orchestrator-workers; Workflow: Evaluator-optimizer
Version note
The live page now warns that tooling has changed since the original publication. Use its conceptual distinction, not its present SDK list as a claim about the 2024 implementation.
Full source URL
https://www.anthropic.com/research/building-effective-agents

[LR0] L10B: Watch how Llama chooses a tool; src/llama_walkthrough.py

Authors / issuer
Classroom Labs notebook and Python helper
Publisher / venue
Public reproducible teaching material
Date
24 October 2026
Type
Verified historical reply replay
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/llama_walkthrough.py
Source code
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

[ST02] WikiChat: Stopping the Hallucination of Large Language Model Chatbots by Few-Shot Grounding on Wikipedia

Authors / issuer
Sina J. Semnani, Violet Z. Yao, Heidi C. Zhang, Monica S. Lam
Publisher / venue
Findings of the Association for Computational Linguistics: EMNLP 2023, pp.2387–2413
Date
December 2023
Type
Peer-reviewed dialogue research
Full source URL
https://aclanthology.org/2023.findings-emnlp.157/
Full paper PDF
https://aclanthology.org/2023.findings-emnlp.157.pdf
Source code
https://github.com/stanford-oval/WikiChat

[H02] Classroom_Labs.ipynb, L2; cell classroom-007; src/continuous_control.py

Authors / issuer
Classroom Labs notebook
Publisher / venue
Public notebook and Python source
Date
24 October 2026; stable notebook cell IDs checked
Type
Executable teaching material
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

[ST03] Controllable and Reliable Knowledge-Intensive Task-Oriented Conversational Agents with Declarative Genie Worksheets

Authors / issuer
Harshit Joshi, Shicheng Liu, James Chen, Larsen Weigle, Monica Lam
Publisher / venue
Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, Volume 1: Long Papers, pp.27264–27308
Date
July 2025
Type
Peer-reviewed task-oriented dialogue framework
Full source URL
https://aclanthology.org/2025.acl-long.1323/
Full paper PDF
https://aclanthology.org/2025.acl-long.1323.pdf
Source code
https://github.com/stanford-oval/genie-worksheets

[H01] Classroom_Labs.ipynb, L0–L1; cells classroom-002 and classroom-004

Authors / issuer
Classroom Labs notebook
Publisher / venue
Public notebook and Python source
Date
24 October 2026; stable notebook cell IDs checked
Type
Executable teaching material
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

[H06] Classroom_Labs.ipynb, L6; cell classroom-018; src/standard_formulations.py

Authors / issuer
Classroom Labs notebook
Publisher / venue
Public notebook and Python source
Date
24 October 2026; stable notebook cell IDs checked
Type
Executable teaching material
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

[H09] Classroom_Labs.ipynb, L9; cell classroom-026; fixed trusted-recovery fixture

Authors / issuer
Classroom Labs notebook
Publisher / venue
Public notebook and Python source
Date
24 October 2026; stable notebook cell IDs checked
Type
Executable teaching material
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

[H08] Classroom_Labs.ipynb, L8; cell classroom-023; four-bus DC fixture

Authors / issuer
Classroom Labs notebook
Publisher / venue
Public notebook and Python source
Date
24 October 2026; stable notebook cell IDs checked
Type
Executable teaching material
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

[VCASE] Synthetic Greensboro feeder uhs20_1247–udt8894: model map and declared study scope

Authors / issuer
Existing GridGuard workshop / SMART-DS synthetic-feeder extraction
Publisher / venue
data.openei.org
Date
Existing teaching snapshot reviewed 18 October 2026
Type
Existing local scientific model figure; external URL is the dataset reference retained by the original workshop
Full source URL
https://data.openei.org/submissions/2981

[VMPC] Model Predictive Control: Theory, Computation, and Design, second edition

Authors / issuer
James B. Rawlings; David Q. Mayne; Moritz M. Diehl
Publisher / venue
Nob Hill Publishing
Date
2017
Type
Primary author-hosted textbook
Full source URL
https://sites.engineering.ucsb.edu/~jbraw/mpc/MPC-book-2nd-edition-1st-printing.pdf

[VSAC] Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor

Authors / issuer
Tuomas Haarnoja; Aurick Zhou; Pieter Abbeel; Sergey Levine
Publisher / venue
Proceedings of the 35th International Conference on Machine Learning, PMLR 80, pp. 1861–1870
Date
2018
Type
Primary research paper
Full source URL
https://proceedings.mlr.press/v80/haarnoja18b.html

[LRV] Voltage P3_current_reserve: Llama 3.1:8b planner, recovery and review calls 1–3

Authors / issuer
Workshop recorded model replies and numerical study records
Publisher / venue
Public teaching archive
Date
27 September 2026
Type
Recorded Llama reply and engineering result
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/continuous_llama_opendss_20260927T161223523552Z.json
Source code
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/workshop_run_opendss.json

[H05] Classroom_Labs.ipynb, L5; cell classroom-016; MV-only default model

Authors / issuer
Classroom Labs notebook
Publisher / venue
Public notebook and Python source
Date
24 October 2026; stable notebook cell IDs checked
Type
Executable teaching material
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

[RSTL01] On the Radiality Constraints for Distribution System Restoration and Reconfiguration Problems

Authors / issuer
Ying Wang, Yin Xu, Jiaxu Li, Jinghan He, Xiaojun Wang
Publisher / venue
Author preprint, arXiv:1912.05185v3
Date
15 March2020(v3); first submitted11 December2019
Full source URL
https://arxiv.org/abs/1912.05185v3
Full paper PDF
https://arxiv.org/pdf/1912.05185v3

[RSTL02] Sequential Service Restoration for Unbalanced Distribution Systems and Microgrids

Authors / issuer
Bo Chen, Chen Chen, Jianhui Wang, Karen L. Butler-Purry
Publisher / venue
IEEE Transactions on Power Systems33(2),1507–1520
Date
March2018
Full source URL
https://doi.org/10.1109/TPWRS.2017.2720122

[H07] Classroom_Labs.ipynb, L7; cell classroom-021; request/evidence binding

Authors / issuer
Classroom Labs notebook
Publisher / venue
Public notebook and Python source
Date
24 October 2026; stable notebook cell IDs checked
Type
Executable teaching material
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

[LRR] Feeder-catalog restoration: Llama 3.1:8b refinement and ranking calls 14–16

Authors / issuer
Workshop recorded model replies and numerical study records
Publisher / venue
Public teaching archive
Date
27 September 2026
Type
Recorded Llama reply and engineering result
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/proposal_run_20260927T184402Z_53e0cf/llama_calls.json
Source code
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/proposal_complete_workflow.json

[CY01] Power System Static-State Estimation, Part I: Exact Model

Authors / issuer
Fred C. Schweppe, J. Wildes
Publisher / venue
IEEE Transactions on Power Apparatus and Systems, PAS-89(1), pp. 120–125
Date
January 1970
Full source URL
https://doi.org/10.1109/TPAS.1970.292678

[CY02] False Data Injection Attacks against State Estimation in Electric Power Grids

Authors / issuer
Yao Liu, Peng Ning, Michael K. Reiter
Publisher / venue
ACM Transactions on Information and System Security, 14(1), Article 13, 33 pages
Date
May 2011
Full source URL
https://doi.org/10.1145/1952982.1952995
Full paper PDF
https://reitermk.github.io/papers/2011/TISSEC2.pdf

[LRC] Standard four-bus cyber study: Llama 3.1:8b tool selection and REVIEW calls 7–8

Authors / issuer
Workshop recorded model replies and numerical study records
Publisher / venue
Public teaching archive
Date
21 September 2026
Type
Recorded Llama reply and engineering result
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/standard_stage_20260921T143826Z_42d19a61/llama_calls.json
Source code
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/standard_agent_workflow.json

[H12] Classroom_Labs.ipynb, L12; cell classroom-033; simulated readback only

Authors / issuer
Classroom Labs notebook
Publisher / venue
Public notebook and Python source
Date
24 October 2026; stable notebook cell IDs checked
Type
Executable teaching material
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

[OT01] Guide to Operational Technology (OT) Security

Authors / issuer
Keith Stouffer, Michael Pease, CheeYee Tang, Timothy Zimmerman, Victoria Pillitteri, Suzanne Lightman, Adam Hahn, Stephanie Saravia, Aslam Sherule, Michael Thompson
Publisher / venue
National Institute of Standards and Technology, NIST SP 800-82 Revision 3
Date
September 2023 (final)
Type
Published OT security guidance
Full source URL
https://doi.org/10.6028/NIST.SP.800-82r3
Full paper PDF
https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-82r3.pdf
Publication
https://csrc.nist.gov/pubs/sp/800/82/r3/final

[EVAL01] Artificial Intelligence Risk Management Framework (AI RMF 1.0)

Authors / issuer
National Institute of Standards and Technology
Publisher / venue
NIST AI 100-1
Date
26 January 2023
Type
Voluntary risk-management guidance
Full source URL
https://doi.org/10.6028/NIST.AI.100-1
Full paper PDF
https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf

[WF01] Five-role operator study workflow: implementation and recorded cases

Authors / issuer
Alaa Selim and Mohamed Chehada
Publisher / venue
GridGuard Agentic AI Tutorials, reproducible Python source and saved study archive
Date
Implementation inspected 25 October 2026; records dated 27 September 2026
Type
Executable teaching implementation and recorded evidence
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/proposal_workflow.py
Source code
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/src/proposal_case_tools.py
Publication
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/results/proposal_complete_workflow.json

[H10] Classroom_Labs.ipynb, L10–L11; cells classroom-028 and classroom-030

Authors / issuer
Classroom Labs notebook
Publisher / venue
Public notebook and Python source
Date
24 October 2026; stable notebook cell IDs checked
Type
Executable teaching material
Full source URL
https://github.com/AlaaSelim-sudo/GridGuard-Agentic-AI-Tutorials/blob/main/agentic_grid_workflow_tutorial/Classroom_Labs.ipynb

Visual credit: stanford_agentic_capabilities.png

Caption
Original Stanford figure: one function call and a multi-step workflow.
Credit
Monica Lam. CS224V Agentic AI, Lecture 1: Introduction. Stanford University, Fall 2026, p.2.
Used on
3. The agent asks; the tool calculates
Asset type
Original Stanford lecture figure excerpt
Transformation
Unchanged vector selection from the presenter-supplied Stanford lecture, p.2; Poppler PNG rendering. Exact crop and source/image hashes: research/stanford_sources/excerpt_manifest.json.
Provenance note
Cropped from the supplied Stanford lecture PDF. Original author and page retained in the caption and full source guide.
Evidence limit
Original Stanford figure: one function call and a multi-step workflow.
Original source / page
https://web.stanford.edu/class/cs224v/lectures_2026/l-introduction.pdf#page=2
Original figure / image URL
research/evidence/replots/stanford_agentic_capabilities.png

Visual credit: hands_on_L2_plot.png

Caption
Calculated example at the original current allowance.
Credit
Classroom Labs: actual executed notebook output, Python 3.12
Used on
14. Read the voltage and current checks
Asset type
Executed notebook plot or browser capture
Transformation
Exact PNG output or documented HTML output selection; see research/hands_on/hands_on_manifest.json.
Evidence limit
Calculated example at the original current allowance.
Original source / page
research/hands_on/index.html
Original figure / image URL
research/evidence/replots/hands_on_L2_plot.png

Visual credit: llama_voltage_1.png

Caption
Recorded full-feeder study; different inputs from the earlier comparison.
Credit
Recorded Llama 3.1:8b replies and numerical study records; exact public reasons preserved
Used on
20. Recorded example: the first option fails
Asset type
Browser-rendered selection of verified recorded fields
Transformation
Presentation selection of verified recorded fields. Exact returned reason is unchanged. Full data, hashes and source paths: research/llama_replay/episodes.json.
Provenance note
Raw JSON, parsed response and historical trace cross-checked by src/llama_walkthrough.py. Selected fields are not a screenshot of a deployed product. capture_manifest.json records source and image SHA-256 values.
Evidence limit
Recorded full-feeder study; different inputs from the earlier comparison.
Original source / page
research/llama_replay/index.html
Original figure / image URL
research/evidence/replots/llama_voltage_1.png

Visual credit: llama_voltage_2.png

Caption
Recorded full-feeder study: the requested fallback passes the checks.
Credit
Recorded Llama 3.1:8b replies and numerical study records; exact public reasons preserved
Used on
21. Recorded example: another option is checked
Asset type
Browser-rendered selection of verified recorded fields
Transformation
Presentation selection of verified recorded fields. Exact returned reason is unchanged. Full data, hashes and source paths: research/llama_replay/episodes.json.
Provenance note
Raw JSON, parsed response and historical trace cross-checked by src/llama_walkthrough.py. Selected fields are not a screenshot of a deployed product. capture_manifest.json records source and image SHA-256 values.
Evidence limit
Recorded full-feeder study: the requested fallback passes the checks.
Original source / page
research/llama_replay/index.html
Original figure / image URL
research/evidence/replots/llama_voltage_2.png

Visual credit: llama_voltage_3.png

Caption
Recorded workflow result; human review remains pending.
Credit
Recorded Llama 3.1:8b replies and numerical study records; exact public reasons preserved
Used on
22. Recorded example: the operator receives a proposal
Asset type
Browser-rendered selection of verified recorded fields
Transformation
Presentation selection of verified recorded fields. Exact returned reason is unchanged. Full data, hashes and source paths: research/llama_replay/episodes.json.
Provenance note
Raw JSON, parsed response and historical trace cross-checked by src/llama_walkthrough.py. Selected fields are not a screenshot of a deployed product. capture_manifest.json records source and image SHA-256 values.
Evidence limit
Recorded workflow result; human review remains pending.
Original source / page
research/llama_replay/index.html
Original figure / image URL
research/evidence/replots/llama_voltage_3.png

Visual credit: hands_on_L6_plot.png

Caption
Calculated restoration results for the same network and limits.
Credit
Classroom Labs: actual executed notebook output, Python 3.12
Used on
26. The goal changes who receives power
Asset type
Executed notebook plot or browser capture
Transformation
Exact PNG output or documented HTML output selection; see research/hands_on/hands_on_manifest.json.
Evidence limit
Calculated restoration results for the same network and limits.
Original source / page
research/hands_on/index.html
Original figure / image URL
research/evidence/replots/hands_on_L6_plot.png

Visual credit: llama_restoration_14.png

Caption
Recorded larger-feeder example, separate from the six-node network.
Credit
Recorded Llama 3.1:8b replies and numerical study records; exact public reasons preserved
Used on
32. Recorded example: another plan restores more service
Asset type
Browser-rendered selection of verified recorded fields
Transformation
Presentation selection of verified recorded fields. Exact returned reason is unchanged. Full data, hashes and source paths: research/llama_replay/episodes.json.
Provenance note
Raw JSON, parsed response and historical trace cross-checked by src/llama_walkthrough.py. Selected fields are not a screenshot of a deployed product. capture_manifest.json records source and image SHA-256 values.
Evidence limit
Recorded larger-feeder example, separate from the six-node network.
Original source / page
research/llama_replay/index.html
Original figure / image URL
research/evidence/replots/llama_restoration_14.png

Visual credit: llama_restoration_15.png

Caption
Recorded larger-feeder example: the overloaded trial is rejected.
Credit
Recorded Llama 3.1:8b replies and numerical study records; exact public reasons preserved
Used on
33. Recorded example: the tool rejects a trial
Asset type
Browser-rendered selection of verified recorded fields
Transformation
Presentation selection of verified recorded fields. Exact returned reason is unchanged. Full data, hashes and source paths: research/llama_replay/episodes.json.
Provenance note
Raw JSON, parsed response and historical trace cross-checked by src/llama_walkthrough.py. Selected fields are not a screenshot of a deployed product. capture_manifest.json records source and image SHA-256 values.
Evidence limit
Recorded larger-feeder example: the overloaded trial is rejected.
Original source / page
research/llama_replay/index.html
Original figure / image URL
research/evidence/replots/llama_restoration_15.png

Visual credit: llama_restoration_16.png

Caption
Recorded larger-feeder example. Lower cost is preferred, including negative costs.
Credit
Recorded Llama 3.1:8b replies and numerical study records; exact public reasons preserved
Used on
34. Recorded example: a wrong ranking is corrected
Asset type
Browser-rendered selection of verified recorded fields
Transformation
Presentation selection of verified recorded fields. Exact returned reason is unchanged. Full data, hashes and source paths: research/llama_replay/episodes.json.
Provenance note
Raw JSON, parsed response and historical trace cross-checked by src/llama_walkthrough.py. Selected fields are not a screenshot of a deployed product. capture_manifest.json records source and image SHA-256 values.
Evidence limit
Recorded larger-feeder example. Lower cost is preferred, including negative costs.
Original source / page
research/llama_replay/index.html
Original figure / image URL
research/evidence/replots/llama_restoration_16.png

Visual credit: hands_on_L8_plot.png

Caption
Calculated model-fit and state-error results.
Credit
Classroom Labs: actual executed notebook output, Python 3.12
Used on
39. Read the alarm and the estimation error
Asset type
Executed notebook plot or browser capture
Transformation
Exact PNG output or documented HTML output selection; see research/hands_on/hands_on_manifest.json.
Evidence limit
Calculated model-fit and state-error results.
Original source / page
research/hands_on/index.html
Original figure / image URL
research/evidence/replots/hands_on_L8_plot.png

Visual credit: llama_cyber_7.png

Caption
Recorded recovery request and numerical tool result.
Credit
Recorded Llama 3.1:8b replies and numerical study records; exact public reasons preserved
Used on
45. Recorded example: request the recovery tool
Asset type
Browser-rendered selection of verified recorded fields
Transformation
Presentation selection of verified recorded fields. Exact returned reason is unchanged. Full data, hashes and source paths: research/llama_replay/episodes.json.
Provenance note
Raw JSON, parsed response and historical trace cross-checked by src/llama_walkthrough.py. Selected fields are not a screenshot of a deployed product. capture_manifest.json records source and image SHA-256 values.
Evidence limit
Recorded recovery request and numerical tool result.
Original source / page
research/llama_replay/index.html
Original figure / image URL
research/evidence/replots/llama_cyber_7.png

Visual credit: llama_cyber_8.png

Caption
Recorded workflow result; human review remains pending.
Credit
Recorded Llama 3.1:8b replies and numerical study records; exact public reasons preserved
Used on
46. Recorded example: keep human review
Asset type
Browser-rendered selection of verified recorded fields
Transformation
Presentation selection of verified recorded fields. Exact returned reason is unchanged. Full data, hashes and source paths: research/llama_replay/episodes.json.
Provenance note
Raw JSON, parsed response and historical trace cross-checked by src/llama_walkthrough.py. Selected fields are not a screenshot of a deployed product. capture_manifest.json records source and image SHA-256 values.
Evidence limit
Recorded workflow result; human review remains pending.
Original source / page
research/llama_replay/index.html
Original figure / image URL
research/evidence/replots/llama_cyber_8.png