All 28 skills, each showing the exact artifact you hand
in and the checklist it is graded against. A skill is done as one week's
research-log entry, never as an extra submission — the slot named on each card is the
assignment it replaces. Every lesson exists in full in the handbook; this is the index.
M1 The Workbench
Year 1 · 5 skills
The five surfaces you actually work on, plus the habit that makes all of them safe.
01
Spreadsheet discipline
arithmetic3 hoursprereq: none
You hand in
An .xlsx attached to the entry, plus a text-box write-up: raw sheet nothing writes to, transform sheet, one pivot, one chart with units, and your line-by-line annotation of the AI-written formula
Submitted as that week's WA entry — not as an extra assignment.
Spreadsheets are where most lab data first lands and where most silent errors are born: a formula dragged one row short, a sort that moved one column and not its neighbours, a constant typed into a cell six months ago that nobody can now explain.
It passes when — all 6 of these are true
Raw sheet exists and no formula writes to it
At least one named range used in a formula
No bare numeric constant inside a formula except 0, 1, and documented unit conversions
One pivot answering a stated question; one chart with axes labelled and units given
An AI-generated formula, with every argument annotated in the student's own words
A deliberate break, with the prediction shown before the result
02
Notebooks and AI assistance
arithmetic4 hoursprereq: none
You hand in
A notebook share link (view access on) plus a verification table pasted in the text box: one row per factual claim — claim, verified / unverified / wrong, and how you checked it
Submitted as that week's WA entry — not as an extra assignment.
The failure mode is not a bad answer.
It passes when — all 7 of these are true
Link opens for someone who is not the author
First cell shows interpreter version and pinned package versions; seed set
The hand-computed answer appears before the code that recomputes it
Any divergence reconciled in writing, not removed
Three structurally different prompts, not rewordings
Every factual claim is a row, each naming a specific verification method
At least one claim identified as wrong or unverifiable
03
The command line and a remote machine
arithmetic5 hoursprereq: none
You hand in
A pasted terminal transcript (not a screenshot — it must be searchable) plus the output file you pulled back off the remote machine
Submitted as that week's WA entry — not as an extra assignment.
This lab runs Linux servers alongside Windows workstations, and the jobs that matter — a ReEDS run, a large MILP, a parameter sweep — do not belong on your laptop.
It passes when — all 5 of these are true
Local: navigation, a pipe, a redirect, and a recursive search, all shown
ssh to a lab server or HPC node, shown
A file moved to the remote machine and a result moved back
A job that survived a disconnection, with evidence it did — the submission, then the output
The log read, with one line quoted that you had to look up
An Overleaf share link plus the compiled PDF, in the target journal's own template, zero errors
Submitted as that week's WW entry — not as an extra assignment.
Submissions stall on formatting, and they stall late — at the point where the science is done and the only thing between you and a submitted paper is a template you have never opened.
It passes when — all 5 of these are true
The template is the journal's official one, named
Compiles with zero errors
Warnings read, and any overfull box in a figure or table resolved
Bibliography generated from a .bib file, not hand-written
Page count stated against the journal's limit
05
Sanity checks and orders of magnitude
arithmetic3 hoursprereq: none
You hand in
A text-box entry: the magnitude you predicted before running, the actual, a unit audit of every input and output, one published reference value with its source, and one extreme case
Submitted as that week's WA entry — not as an extra assignment.
The errors that survive code review are the ones where the code is correct and the number is wrong. A diff will not catch a capacity in MW that should have been GW, a price in $/tonne read as $/kg, or a discount rate entered as 7 instead of 0.07.
It passes when — all 6 of these are true
Predicted magnitude written before the run, with a timestamp or commit that proves it
Units stated for every input and output involved
One published reference value, cited with a source, and the comparison
Any discrepancy over a factor of two explained, not waved through
One extreme case run and its behaviour described
If a solver was used: status word, achieved gap and residual all reported — or an explicit note that no solver was involved
M2 Reproducible Code
Year 1 · 4 skills
The milestone that prevents "I cannot reproduce my own Chapter 3." All four are performed on your actual dissertation repository — not a sample, not a copy.
06
Git and GitHub
arithmetic4 hoursprereq: none
You hand in
A repo URL plus pasted git check-ignore -v output — one line per data file and credential path, each showing the rule that matched
Submitted as that week's WA entry — not as an extra assignment.
Do this one before anything else in M2. Git history is permanent.
It passes when — all 5 of these are true
Repo URL resolves
git check-ignore -v output pasted, showing a matched rule for every data and credential path
Searching the full history for your own credential variable names returns nothing
At least five commits whose messages state a reason
git status on a clean checkout is empty
07
Code folder structure
arithmetic4 hoursprereq: 06-git-github
You hand in
A repo URL plus one paragraph on what moved where, plus the two places you found the standard stricter than Wilson et al. and why
Submitted as that week's WA entry — not as an extra assignment.
"I cannot reproduce my own Chapter 3" is the concrete failure this prevents, and it has a specific mechanism: processed data overwrote raw data, or a script's output became its own input, and now there is no path back to the numbers in the draft.
It passes when — all 4 of these are true
Repo layout matches the archetype
Nothing under the raw tier is written by any script (verify, do not assert)
README is usable by someone who has never seen the project
Two places the standard is stricter than Wilson et al., identified and explained
08
Reproducible environments
arithmetic4 hours (plus your partner's 1 hour)prereq: 07-code-folder-structure
You hand in
A repo URL plus your partner's run log, signed by them — what they ran, what broke, what they had to guess, and whether the figure reproduced
Submitted as that week's WA entry — not as an extra assignment.
There is exactly one test of whether an environment is reproducible, and it is not a file listing.
It passes when — all 5 of these are true
Partner reproduced the named figure on their own machine, or the divergence is explained
Stated tolerance appears before the comparison, not after
Solver version and interpreter version both recorded in the repo
Every question the partner had to ask is listed as a README defect
A transcript excerpt plus the diff you accepted, with your one-sentence task statement timestamped before the session began
Submitted as that week's WA entry — not as an extra assignment.
An agent with filesystem access is fast, confident, and wrong often enough to matter — and a research repo is the most expensive place for that combination, because a plausible edit to an analysis script produces plausible numbers.
It passes when — all 5 of these are true
Task stated in one sentence, timestamped before the session
Work done on a branch, not the default branch
The full diff is shown and was reviewed line by line
At least one thing the agent got wrong is recorded
Whatever check the repo already has (tests, an agreement assertion) still passes after the merge
M3 Reading and Reasoning
Year 1-2 · 5 skills
Evaluating somebody else's argument, finding where the field stops, spotting an inference that fails, building a claim of your own, and saying it no more strongly than your evidence allows.
10
Reading and critiquing papers
reading3 hoursprereq: none
You hand in
300–500 words in the text box: the paper's claim quoted verbatim with a locator, the load-bearing assumption named, the gap between claimed and shown, and one alternative with its cost
Submitted as that week's WR entry — not as an extra assignment.
Summarising is not critiquing, and the weekly log makes the difference easy to hide.
It passes when — all 4 of these are true
The central claim is quoted verbatim, with a location
One load-bearing assumption is named
The gap between what is claimed and what is shown is stated explicitly
A specific alternative is proposed, with its cost
11
Literature review and gap-finding
reading8 hoursprereq: 10-critiquing-papers
You hand in
A PDF: your three candidate axes with a reason for rejecting two, fifteen cited papers positioned on the one you chose, the empty region, and the three-part gap sentence
Submitted as that week's WR entry — not as an extra assignment.
"No one has done X" is not a gap.
It passes when — all 7 of these are true
Fifteen papers, cited
Three candidate axes written down before one was chosen, each rejected one with a stated reason
One named axis, with units or categories stated
Every paper positioned on it
The empty region identified
One sentence on why it is empty, distinguishing "not yet possible" from "not worth doing"
The three-part sentence, written out
12
Logic and its failures
reading3 hoursprereq: 13-claim-reason-evidence
You hand in
A text-box entry: three quoted passages — one from a published paper, one from your own draft, one from AI output — each with the failure named and a concrete repair
Submitted as that week's WR entry — not as an extra assignment.
An argument can be perfectly valid and completely false, or true in every part and still invalid.
It passes when — all 5 of these are true
Three examples, one from each of the three sources
Each passage quoted, with a locator
The specific failure named, not just "this is weak"
A concrete repair for each — what evidence or qualifier would fix it
The one from your own draft is actually fixed, and the draft link shows it
13
Claim, reason, evidence
writing3 hoursprereq: none
You hand in
One page: the claim sentence with all three slots filled, the warrant written out as a general principle, and three objections a competent person would actually raise
Submitted as that week's WW entry — not as an extra assignment.
Most students arrive with a topic and believe it is a claim.
It passes when — all 5 of these are true
A single claim sentence with all three slots filled
The claim is contestable — someone competent could disagree with it
The warrant is stated as a general principle, not restated as the evidence
Three objections, none of them trivial
For each objection, one line on whether the paper currently answers it
14
Saying exactly what you claim
writing2 hoursprereq: 13-claim-reason-evidence
You hand in
A three-column table — verb used / what the evidence licenses / the revision — covering both your draft and your CV, plus links to both revised documents
Submitted as that week's WW entry — not as an extra assignment.
Reviewers reject papers for overclaiming more often than for bad work.
It passes when — all 7 of these are true
Every claim-making sentence in the draft appears as a row
Each row states what the evidence licenses, not just whether the verb "sounds strong"
Every instance of proves is either removed or justified as genuine mathematics
At least one claim gains an explicit scope condition
The revised draft is linked and the changes are in it
The same audit run over every bullet of your own CV, with the revised CV attached
At least one CV verb downgraded to what the evidence actually supports
M4 The Manuscript
Year 2 · 5 skills
Everything that goes into assembling a document somebody else can read, trust and cite. The argument itself was M3; this is the production of the artifact that carries it.
15
IMRAD structure
writing3 hoursprereq: none
You hand in
An Overleaf link: one contribution-claim sentence, and one topic sentence for every IMRAD section, each traceable to the claim in one step
Submitted as that week's WW entry — not as an extra assignment.
Most rejected drafts in this lab are not rejected for the research.
It passes when — all 5 of these are true
The claim sentence exists, with both blanks filled
"Matters because" gives a reason other than novelty
One topic sentence per IMRAD section
Each topic sentence traceable to the claim in one step
Any section that failed the test is named, with the resolution stated
16
Reference manager and citation verification
reading3 hoursprereq: none
You hand in
A ten-row table — reference / does the DOI resolve / where the claim sits by page or figure / verdict — plus a link to the draft with the failures struck
Submitted as that week's WR entry — not as an extra assignment.
Fabricated references are a live failure mode, not a hypothetical one — a language model will produce a plausible author list, a plausible title, a plausible year, and a DOI that resolves to something else entirely.
It passes when — all 5 of these are true
Ten rows
Every row has a resolving DOI or a stable URL
Every row locates the claim by page, section, or figure
Any reference that failed is struck from the draft, and the draft link shows it
The count of failures is stated plainly
17
Research ethics and authorship
writing2 hoursprereq: none
You hand in
A text-box entry: everyone who touched the paper, CRediT roles per person, author vs. acknowledgement with reasons, the order, the corresponding author — plus proof you circulated it
Submitted as that week's WW entry — not as an extra assignment.
Authorship disputes happen at submission, which is the one moment when they cannot be resolved calmly.
It passes when — all 6 of these are true
Every contributor named, including students at other tiers
CRediT roles assigned per person
Author versus acknowledgement decided with a stated reason
Order justified in one sentence
Corresponding author named
Evidence of circulation
18
Figure and table design
writing4 hoursprereq: none
You hand in
A PDF showing the figure before, after, and converted to greyscale, with a caption that makes sense to someone who has read no body text
Submitted as that week's WW entry — not as an extra assignment.
Figures are read first and often alone.
It passes when — all 6 of these are true
Before and after both shown
Caption comprehensible with no body text — test it on someone outside your subfield
Units on both axes
Greyscale conversion shown, series still distinguishable
Colour is never the only channel carrying meaning
Every plotted series is named in the caption
19
Accessibility of notebooks and figures
writing3 hoursprereq: 18-figures-tables
You hand in
A repo URL — alt text naming the trend on every figure, no skipped heading levels, nothing encoded by colour alone, and a measured contrast ratio
Submitted as that week's WW entry — not as an extra assignment.
This one has a date.
It passes when — all 5 of these are true
Every figure has alt text naming the trend or takeaway, not the chart type
No heading level is skipped anywhere in the notebook
Greyscale check passed
Tables have header rows and are not images
Body text contrast measured at 4.5:1 or better, with the measurement stated
M5 The Review Cycle
Year 3+ · 5 skills
Everything between "the draft is finished" and "it is out in the world." Review someone else's work, answer a review of your own, choose where to send it, present it, and defend it under questioning.
20
Reviewing papers
reading4 hoursprereq: 10-critiquing-papers
You hand in
A PDF referee report: the paper summarised in your words, two or more specific strengths, two or more weaknesses with line numbers split into must-fix and would-improve, and an explicit recommendation
Submitted as that week's WR entry — not as an extra assignment.
You will be asked to review before anyone teaches you how, and the default is to produce something either useless ("interesting paper, minor comments") or destructive.
It passes when — all 5 of these are true
Summary written in the reviewer's own words
At least two strengths, each pointing at a specific section
At least two weaknesses, each tied to a line or section number
Must-fix and would-improve separated into distinct lists
An explicit recommendation
21
Response to reviewers
writing4 hoursprereq: 20-reviewing-papers
You hand in
A PDF letter built on a table — reviewer point / your response / the change made / exactly where it is in the revision — with every point given a row
Submitted as that week's WW entry — not as an extra assignment.
Rebuttals fail in two directions and both lose the paper.
It passes when — all 5 of these are true
Every reviewer point has a row
Every claimed change is located by section or line
At least one point respectfully declined, with a reason rather than an apology
The revised draft actually contains the changes claimed — spot-checked against three rows
No unattached apologies
22
Venue and audience selection
writing3 hoursprereq: 15-imrad
You hand in
A one-page memo (three venues or three named groups × four attributes, and your choice justified against the two you rejected) plus the draft cover letter or outreach message
Submitted as that week's WW entry — not as an extra assignment.
A paper sent to the wrong venue does not get rejected quickly — it gets desk-rejected after six weeks, or sent to reviewers who evaluate it against criteria it was never trying to meet.
It passes when — all 7 of these are true
Three named venues
All four attributes present for each, with sources
Choice justified against the two rejected, on grounds specific to this paper
Cover letter states the contribution in its first paragraph
AI-disclosure requirement identified and a draft disclosure written
If run on an employer: three named groups, not companies, each with the specific thing it builds
The recipient demonstrably works on the thing the message is about — or the ask is an introduction instead
23
Conference talk and poster
writing6 hoursprereq: 15-imrad, 18-figures-tables
You hand in
A slides PDF plus the three questions you were actually asked and the answers you wish you had given
Submitted as that week's WW entry — not as an extra assignment.
A conference talk is twelve minutes and you will be asked three questions you did not anticipate.
It passes when — all 5 of these are true
Delivered within twelve minutes, plus or minus one
A slide carries the contribution claim, matching the manuscript word for word
Figures meet the Figure and table design checks
Three questions recorded verbatim
A revised answer for each, and a note on whether it implies a change to the paper
24
Defending your work
writing4 hoursprereq: 13-claim-reason-evidence
You hand in
A PDF: three objections written in the critic's voice at full strength, your response to each, and each response classified as wrong / right-and-costly / right-and-unfixable
Submitted as that week's WW entry — not as an extra assignment.
A defence fails in two directions and both lose the room.
It passes when — all 6 of these are true
Three objections, each written in the critic's voice at full strength
Each steelman would be recognised as fair by someone who holds that view
Each response classified as one of the three types above
At least one objection conceded — if none are, the steelmen are too weak
At least one objection is the kind an employer raises, not a reviewer
Delivered live, with a note on which question the lab asked that you had not anticipated
M6 Stewardship
Year 3+ · 4 skills
The loosest grouping of the six, and worth saying so plainly. What joins these four is that each produces something that outlasts you in this lab: a citable deposit, a dataset somebody can still find terms for in four years, an idea written well enough to be funded, and a person who learned to do research because you scoped their first task properly.
A Zenodo record URL — a tagged release archived, both DOIs placed where the standard says, and a CITATION.cff that parses
Submitted as that week's WA entry — not as an extra assignment.
A paper that says "code available on request" is a paper whose code will not be available in four years, because the student has graduated and the laptop is gone.
It passes when — all 4 of these are true
Record URL resolves and the archive contains the tagged release
Both DOIs present and placed per the standard
CITATION.cff parses — the GitHub "Cite this repository" button appears
The paper's data-availability statement cites the correct one of the two
26
Research data management
arithmetic3 hoursprereq: 25-zenodo
You hand in
A one-page PDF data statement — per dataset: source, the instrument it arrived under, licence, any restriction quoted from the actual agreement, and what is withheld and where it lives
Submitted as that week's WA entry — not as an extra assignment.
"Can this be deposited?" has to be answered before the deposit, because a deposit is public and permanent and the undo button does not fully exist.
It passes when — all 5 of these are true
Every dataset used in your work appears
Source named with the instrument it arrived under
Licence named explicitly per dataset
Any restriction quoted from the agreement, with the agreement located
Withheld items named, with a reason and a location
27
Proposal writing fragments
writing6 hoursprereq: 11-lit-review-gap, 15-imrad
You hand in
A PDF within the call's stated page limit: project summary under the headings the call requires, and a broader-impacts section naming an activity, an audience, and how you would measure it
Submitted as that week's WW entry — not as an extra assignment.
Writing a full proposal is not a course assignment — it is a months-long effort with a real deadline attached to somebody's salary.
It passes when — all 5 of these are true
A real call is named, with its deadline
Within the stated page limit — verified, not estimated
Required headings present exactly as the call words them
Broader impacts names an activity, an audience, and a measure
Intellectual merit traceable to the Literature review and gap-finding gap statement
28
Mentoring an undergraduate
writingongoing, plus 2 hours to write upprereq: 07-code-folder-structure
You hand in
The task brief (definition of done, input and output paths, checkpoint date) plus your retrospective naming at least two things you left underspecified, with the mentee's output linked
Submitted as that week's WW entry — not as an extra assignment.
This lab runs three tiers, so PhD students supervise before anyone has asked whether they can.
It passes when — all 6 of these are true
Task brief contains a definition of done that names an artifact
Input and output paths specified, consistent with the repo archetype
A checkpoint date, before the deadline
Retrospective names at least two underspecified items
The mentee's actual output is linked
If they used git, their .gitignore was verified before their first commit
ungraded SEAR Lab Domain Knowledge
read in week one
What this lab actually studies. Infrastructure has three layers — physical, digital, institutional — and a study that addresses only one of them gets the answer wrong. Surface level on purpose: each area names the course that goes deep.
Physical systems, hardware, experiments
Digital models, analytics, decision tools
Institutional policy, programs, deployment
Energy systems
shipboard power and HESM · NAVSEA land-based testing · microgrids from repurposed EV batteries and solar
ReEDS and GCAM · data-centre and AI load growth · transmission expansion under renewable buildout · surrogate optimisation and DACE
Green Warriors weatherization in South Dallas · energy burden and DERs · Powerhouse Texas EPAC · cited in IPCC AR6 WGIII
Critical minerals
500+ kWh of recovered EV batteries and motors · shred, grind, sieve · electrical, chemical and powder analysis · 3R-BET
global lithium supply chain optimisation · centralized vs. decentralized black mass · reverse logistics and recycling network design
trade and tariff policy · domestic supply chain security · NAATBatt Track and Trace · geothermal on tribal lands and HUBZones
RFID and asset tracking
antennas, readers, printers · real-time location systems · circular conveyor and robotic arm · operational hospital beds
plant-level inventory systems · asset performance analytics · set-cover optimisation for reader placement · blockchain provenance
battery track-and-trace standards · community naloxone distribution · pharmaceutical anti-diversion · health asset management
The thesis: infrastructure has three layers
Societies thrive or struggle on the infrastructure they build, and that infrastructure is never only physical. It spans three inseparable layers:
1. Energy systems
The system that keeps the lights on, and the system this lab models most.
2. Critical mineral supply chains
End to end and back. This phrase is the lab's central intellectual claim and it is worth understanding precisely.
3. RFID and asset tracking
The short version. RFID tags carry an identifier that a reader can interrogate without line of sight. That single property — no line of sight — is what separates it from a barcode and is the reason it works for pallets, hospital beds, and battery packs moving through a facility.
Numbers worth carrying
Order-of-magnitude anchors for energy work. Not for precision — for catching a result that is wrong by a factor of a thousand, which is the failure that survives code review.
Working across borders
The supply chains here cross a dozen jurisdictions with different rules, different data conventions, and genuinely different ideas about what a deadline means. The lab collaborates internationally and most of its students trained somewhere other than the United States.
The questions behind the methods rather than inside them. What counts as evidence when you cannot run an experiment, and why a discount rate is a claim about how much future people matter.
Philosophy of science
You build models of systems you cannot run experiments on. You will never have a control group for the Texas grid.
The normative content of your modelling choices
A discount rate is a claim about how much future people matter. Three percent and seven percent are different moral positions, and the choice between them was argued out by economists and philosophers, not engineers.
Policy and institutional literacy
You model ERCOT. Somebody decided ERCOT would be an energy-only market with no capacity payment, and that decision was political, recent, and revisable.
History of your own field
The grid looks the way it does for reasons, and most of them are historical accidents that hardened. Rural electrification, PUHCA in 1935, the 1978 PURPA opening, restructuring in the 1990s, and Texas's decision to stay outside the interconnections.
The economics of what you model
Externalities, market failure, and why a carbon price is a policy instrument rather than a moral statement. If your optimisation minimises cost, you have adopted a welfare framework — worth knowing which one, and what it does not capture.
Pencil, spreadsheet, simulation. Solve the smallest version of your problem by hand, reproduce it numerically, then switch the full model down until it must agree. It is how the lab checks a model, including one an AI wrote.
Why this exists
A large model that returns a number has not explained anything. It has produced a number, and the number is only as trustworthy as your ability to say why it came out that way.
Governing part
None.
A worked example
From Jones's own carbon research. Two equations: atmospheric carbon C, a reservoir R that the atmosphere drains into, emissions E, and pre-industrial level C₀.
The mechanic
Do this on your own dissertation problem. It works for optimization as well as dynamics; most of this lab's problems are optimization, so there are examples of both below.
Deliverable
A one-page PDF: the pencil derivation, the spreadsheet or notebook link, the degenerate-case comparison table (pencil / spreadsheet / simulation, same inputs, three outputs), and the two sentences from step 4.
Turning it in
If you do it, do it as one week's arithmetic entry. Submit it to that week's WA assignment in Canvas, the same one you would have submitted anyway, with the one-page PDF attached and, in the text box, the pencil result in one line and whether the three rungs agreed.