Self-Study Guide¶

Page Maps¶

graph LR
  family["Reproducible Research"]
  program["Deep Dive DVC"]
  section["Orientation and Study Practice"]
  page["Self-Study Guide"]
  outcome["Independent module completion"]

  family --> program --> section --> page
  page -.supports.-> outcome

flowchart LR
  question["opening question"] --> lesson["core lesson"]
  lesson --> prediction["written prediction"]
  prediction --> investigation["worked or executable investigation"]
  investigation --> practice["cumulative exercises"]
  practice --> review["answer comparison"]
  review --> decision["module decision"]

This guide replaces the coordination an instructor would normally provide: what to read, when to run evidence, how to use answers, what to record, and how to decide whether to continue.

It is designed for learners completing the course alone, returning after missed classes, or studying asynchronously with a review partner.

Prepare one study workspace¶

Keep generated material under the repository's artifacts/ boundary:

artifacts/learning/deep-dive-dvc/
├── questions.md
├── predictions.md
├── module-decisions.md
├── review-packets/
└── command-logs/

Do not place learning logs beside course source files. Generated audit routes already write to artifacts/audit/reproducible-research/deep-dive-dvc/.

At the start of a session, record:

date:
module:
question:
time available:
last accepted capability:
known blocker:

This takes less than a minute and makes interrupted study resumable.

Use one complete module rhythm¶

Frame the question¶

Read the module index and write the question you expect the module to answer. If you cannot state it, use the index's symptom or route table rather than browsing every page.

Read core lessons in order¶

The early page defines language used by later pages. On the first pass, do not jump directly to the most technical file.

For each core lesson, keep two notes:

rule:
failure if violated:

If the rule cannot be stated without copying a paragraph, explain it to an imagined reviewer and revise.

Predict before observing¶

Before every worked example or audit, predict:

which declaration or authority controls the decision;
which stage, candidate, or artifact should be accepted or rejected;
which result value or absence will matter;
what the command cannot establish.

Prediction turns execution into a test of your model. Without it, output can feel persuasive merely because it is detailed.

Work the investigation¶

Follow the evidence in its natural time order:

initial declaration or baseline;
controlled mutation;
pre-repair status or comparison receipt;
ordinary execution;
semantic result or artifact presence;
repaired declaration or policy;
replay of the original challenge;
convergence or final integrity receipt.

Preserve the original failure. Do not overwrite it with the repaired output.

Complete all ten exercises¶

Exercises within a module are cumulative. They usually build one packet rather than ten unrelated answers.

Use this sequence:

flowchart TD
  attempt["attempt without answers"] --> evidence["produce requested artifact"]
  evidence --> check["run acceptance check"]
  check --> compare["read matching model answer"]
  compare --> difference{"reasoning differs?"}
  difference -- yes --> revise["revise and explain why"]
  difference -- no --> retain["retain original"]
  revise --> reflection["record one transferable rule"]
  retain --> reflection

Do not grade by matching wording. Grade whether your artifact:

names the governing claim;
cites the right evidence layers;
distinguishes observation from inference;
makes a decision;
states a limit.

Close with a module decision¶

Append to module-decisions.md:

Module:
Capability accepted:
Evidence route:
Failure I can now diagnose:
Remaining uncertainty:
Next module:

If capability is not accepted, record the narrow reason and repeat the relevant lesson or exercise. Do not restart the whole module automatically.

Choose a session size¶

Time available	Honest scope
20 minutes	one core section plus a written rule and failure mode
45 minutes	one core page plus one evidence prediction
90 minutes	two core pages or one worked investigation
2–3 hours	complete exercise packet and answer review
half day	full technical module only if prerequisites are already stable

Avoid starting an executable investigation when there is no time to preserve and interpret its receipts. Reading it conceptually is better than rushing through commands and calling execution proof.

Recover after a missed class¶

Use three artifacts:

module index completion standard;
worked investigation;
final cumulative exercise and matching answer.

Try the final exercise first. Its difficulty reveals where to return:

Difficulty	Return to
terms are unfamiliar	glossary and earliest core lesson
cannot predict command behavior	declaration or state model lesson
can predict but cannot interpret receipt	evidence-reading lesson
can interpret but cannot decide	boundary or policy lesson
decision is sound but proof packet is incomplete	exercises

This recovery route replaces chronological rereading with capability diagnosis.

Use answer pages without losing the work¶

Open an answer only after saving your attempt. Compare:

Did both answers use the same authority?
Did both preserve the relevant before-state?
Did both join tool receipt to semantic meaning?
Did either claim more than the evidence supports?
Is the model answer showing one defensible design or a required invariant?

When your answer differs but remains defensible, add a short justification. The goal is reviewable judgment, not compliance with one implementation.

Study with a review partner¶

Exchange artifacts, not conclusions. The reviewer should receive:

the claim;
declaration or baseline;
mutation;
raw receipt;
semantic assertion;
decision and limit.

The reviewer then asks one challenge question. Good challenges attack an authority, hidden influence, comparison boundary, recovery assumption, or consumer contract.

Do not let the author explain missing evidence orally before the reviewer attempts the route. The handoff itself is part of the test.

Handle blocked commands¶

Classify the block:

Block	Continue with
missing tool or platform dependency	platform setup and conceptual prediction
network or credential boundary	inspect local evidence; mark remote claim untested
test too expensive for current session	run focused audit; schedule full verification
generated state conflicts with local work	stop mutation; preserve status; use isolated audit
receipt differs from page	inspect structured fields and record version context

Never replace a blocked claim with an unqualified pass. Record UNTESTED and the exact missing authority or capability.

Weekly review¶

At the end of each study week, answer:

Which capability can I now demonstrate without the page?
Which claim still depends on a warm workspace or oral explanation?
Which receipt can another reviewer interpret?
Which failure would I recognize earlier now?
What is the earliest unsupported boundary in the course map?

Use the answer to select the next session. Page count is not progress if a prerequisite claim remains unsupported.

Self-study completion standard¶

Independent study is working when you can:

resume from your evidence packet after an interruption;
choose a bounded session without instructor scheduling;
predict before running;
use answers to audit reasoning rather than copy outcomes;
classify blocked evidence honestly;
decide readiness by capability rather than pages viewed.