- Advanced
- 60 min
- 12 min read
After this lesson you can
- Work out your capstone score from levels and weights.
- Describe what a proficient (level 3) result looks like for each of the six skills.
- List the ten mandatory items and explain how to meet each one.
- Self-grade before the final review and compare with your mentor.
Read these first
The capstone is five days, 30 hours, on one product. This page is the rubric: how the work is scored and what you must show. Score yourself before the final review, then compare with your mentor.
Pass mark: 70 points out of 100 AND all ten mandatory items. Less is “not yet”: fix forward and ask for a second review.
How the points work
There are six skills. Their weights add up to 100. For each skill you get a level from 1 to 4, and each level is worth a share of the weight.
| Level | Name | Share of the weight |
|---|---|---|
| 1 | Missing | 0 percent |
| 2 | Developing | 50 percent |
| 3 | Proficient | 80 percent |
| 4 | Master | 100 percent |
Worked arithmetic. If the skill is worth 20 points:
- Level 1 gives 0 points.
- Level 2 gives 10 points.
- Level 3 gives 16 points.
- Level 4 gives 20 points.
For the whole capstone, add the six results. Example: OSS selection at level 3 (16 of 20), patch-fork at level 3 (16 of 20), IaC and GitOps at level 2 (10 of 20), catalogue at level 3 (12 of 15), ITIL at level 3 (12 of 15), and epic by hand at level 4 (10 of 10). Total: 16 + 16 + 10 + 12 + 12 + 10 = 76. That is above 70, so the points are enough, but the ten mandatory items must still all pass.
Because level 3 is worth 80 percent, a learner who is proficient everywhere scores 80, which is a clear pass. A learner at level 2 everywhere scores 50, which is not a pass.
The six skills
For each skill, level 3 is described as the program defines it. The other levels are described here in plain words, following the level names.
Skill 1: OSS selection (weight 20)
Proficient (level 3) looks like: MUST and SHOULD needs written first. Three or more candidates. Licence file read. Fit percent with the unmet MUST list. One rung chosen. ADR filled.
| Level | What it looks like | TrailMug example |
|---|---|---|
| 1 Missing | No written needs, no candidates, or no ADR | The tool was chosen first. There is no decision record. |
| 2 Developing | Some needs and candidates, but the licence file was not read or the fit has no unmet-MUST list | Two candidates, fit percentages written, but the licence was copied from a web page. |
| 3 Proficient | As above | Three candidates, MUST and SHOULD needs written first, licence files read, fit calculated by hand, one rung chosen, ADR 0001 complete with a human name. |
| 4 Master | Level 3, and the ADR has a clear tech-debt statement, an exit plan and a reasoned note about a trap (for example a network copyleft candidate with the top score) | The ADR explains why the highest-scoring tool was rejected for its licence, and gives a one-week exit plan. |
Skill 2: Patch-fork, or the alternative deliverable (weight 20)
Proficient (level 3) looks like: the upstream pinned with a tag and commit id. An update-bot custom manager for the pin file. A ledger with an owner and a removal condition for each patch. The pipeline re-applies the patches and runs tests. The image is stored in the registry.
If no project in your capstone needs patches, the program allows an alternative deliverable that shows the same discipline (for example, a vendored dependency with a pin, a ledger and a refresh job). Agree it with your mentor on Day 1.
| Level | What it looks like | TrailMug example |
|---|---|---|
| 1 Missing | No pin, no ledger | The code was copied from upstream with edits mixed in. |
| 2 Developing | A pin and a few patches, but no ledger or no automatic re-apply | Patches exist as files but a person applies them by hand. |
| 3 Proficient | As above | Pinned tag and commit id, bot configured, PATCHES.md with owner and removal condition, pipeline applies patches and tests, image in the registry. |
| 4 Master | Level 3, and one patch has been sent upstream or a patch that no longer applies was fixed and explained | An upstream pull request is open for the delivery-report patch, and the ledger says “drop when merged”. |
Skill 3: Infrastructure as code and GitOps (weight 20)
Proficient (level 3) looks like: Kustomize base and overlays. A GitOps application with prune and self-heal. Secret names only through an external secret. Terraform for one resource. A drift plan with a fix-forward merge request.
| Level | What it looks like | TrailMug example |
|---|---|---|
| 1 Missing | No overlays, or live changes by hand | The service was deployed with a command typed on a laptop. |
| 2 Developing | Overlays exist but the controller has no prune or self-heal, or a secret is partly inline | Staging and production overlays, but a password-like value appears in a file. |
| 3 Proficient | As above | Base and two overlays, controller with prune and self-heal, secret names only, one cloud resource in Terraform with the plan in the merge request, a drift plan with a worked fix-forward merge request. |
| 4 Master | Level 3, and the same digest is promoted from staging to production, with a gate added after a drill | A small policy check fails any overlay that uses latest. |
Skill 4: Catalogue and map (weight 15)
Proficient (level 3) looks like: product, service and CI entries using real schema fields. Interim carriers where a field is not yet supported. Edges resolve. Search answers are written down.
An interim carrier is a documented stand-in used when the catalogue does not yet have a field for something. Edges are the links between entries; they resolve when each link points to an entry that exists.
| Level | What it looks like | TrailMug example |
|---|---|---|
| 1 Missing | No entries | The catalogue folder is empty. |
| 2 Developing | Entries exist but have missing fields or dead links | The returns service entry names a dependency that has no entry. |
| 3 Proficient | As above | Entries for the product, the returns service and its pipeline, all fields filled, every link resolves, three written questions with answers found by search. |
| 4 Master | Level 3, and the entries answer an “impact” question (what changes if this service goes down) | A short note lists the three entries that depend on the returns service and who owns them. |
Skill 5: ITIL change and incident (weight 15)
Proficient (level 3) looks like: a complete change record with a human risk decision. A rehearsal: flag off, then a forward fix. A blameless review with five whys. A service level target sheet.
| Level | What it looks like | TrailMug example |
|---|---|---|
| 1 Missing | No change record or no rehearsal | The incident was “fixed” by reverting. |
| 2 Developing | A change record without a risk decision, or a review that names a person | The five whys end with “the engineer forgot”. |
| 3 Proficient | As above | Change record with a named risk owner; rehearsal with a named commander, flag off and a forward fix; five whys ending at a process cause; one SLI and one SLO marked as a proposal. |
| 4 Master | Level 3, and the review adds a gate that is shown failing on the original change | A test is added that would have caught the duplicated email, and it is shown red on the old commit. |
Skill 6: Epic and roadmap by hand (weight 10)
Proficient (level 3) looks like: clear outcome, a journey, a repo list, testable done lines, five ordered milestones, and accepted risks with a name.
| Level | What it looks like | TrailMug example |
|---|---|---|
| 1 Missing | No epic, or it was written by AI | The epic text came from a chat tool. This also fails mandatory item M6. |
| 2 Developing | An epic with no measure or no milestones | A goal and a task list, no dates. |
| 3 Proficient | As above | One epic by hand with outcome, journey, repo list, testable done lines, five ordered milestones, and named risk owners. |
| 4 Master | Level 3, and others can retell the epic without notes | The mentor repeats the epic in one minute after reading it once. |
The ten mandatory items
These are pass or fail. All ten must pass. Tick each only when it is really done.
| Item | What it requires | How to meet it | Example of failing |
|---|---|---|---|
| M1 | No plaintext secrets | Names only in git; values in a vault | A token in a config file |
| M2 | No kubectl apply, patch or create |
Look with get, describe, logs, diff; change through merge requests |
A hand-typed apply |
| M3 | ADR present, status Accepted, one rung | The ADR has status “Accepted” and names rung (b), (c) or (d) | An ADR with status “Draft” |
| M4 | Licence checked and recorded | A licence-check record with SPDX id, file read, date, usage pattern, result | “MIT, I think” with no file |
| M5 | Fix forward, not revert | Every repair is a new change that moves forward | A revert commit as the repair |
| M6 | Epics hand-written | You wrote every epic and roadmap word | AI-written epic text |
| M7 | Planned means simulated, and I marked it | Any Planned feature is marked “simulated” | Claiming a Planned step is running |
| M8 | Human names on the ADR, the licence sign-off and the risk decision | Three named people (use real names in your own private repo) | “Approved by the team” |
| M9 | Digests only: no :latest, no local builds |
Images are referenced by digest; builds run in the pipeline | An image tagged latest |
| M10 | Mentor sign-off on all five checkpoints | The mentor signs after Day 1, 2, 3, 4 and 5 | Four of five signed |
The five checkpoints are one per day: (1) epic and roadmap by hand, OSS selection, ADR 0001; (2) the patch-fork or alternative; (3) Kustomize, the GitOps application, external secret, Terraform, drift test plan; (4) catalogue entries, change record, service level sheet; (5) incident rehearsal, blameless review, self-grade.
The 2030 Product Development Manager lens
- You decide: the epic, the rung, the risk owners, how you grade yourself, and when a mandatory item is truly met.
- You delegate to agents (where the starter kit allows): building the patch-fork pipeline, writing overlays, drafting catalogue entries from your decisions, running checks.
- The brief you would write: “For each mandatory item M1 to M10 in this repository, say met or not met, cite the file and line, and quote the text. Do not fix anything.”
- You verify by: opening each cited file yourself and writing your own level for each skill before reading the agent’s report.
- Disciplines involved: strategy (epic), engineering (pipelines), security (M1, M2, M9), legal (M4), finance (the tech-debt statement), operations (drift), service management (change and incident).
- On the agent ladder: the capstone is orchestrate and govern in miniature. Several agent sessions may work at once on different days’ pieces, under written rules, with a mentor acting as the gate.
Working with AI agents
- Where allowed: build, format, check. Where the task does not say AI is allowed, it is not.
- Never: epics, the self-grade, the oral review.
- Disclose: one line per use, kept in your capstone repository.
- Planned features must be simulated and marked so. An agent that quietly “makes it work” for a Planned feature has made a false claim.
Common mistakes
- Chasing points and missing a mandatory item. A pass needs both.
- Scoring yourself level 4 everywhere. Your mentor will compare.
- Skipping the unmet MUST list. It is a level 3 requirement.
- Leaving an ADR in “Draft”. M3 needs Accepted.
- Claiming Planned as running. M7.
- Four checkpoints signed, not five. M10.
Practice
- Write your own five-line plan for Day 1, with the product of each hour.
- Score a sample capstone: use the TrailMug examples for level 3 in every skill and add up the points. (Answer: 16 + 16 + 16 + 12 + 12 + 8 = 80.)
- Make a card with M1 to M10 and tick it after each day.
- After the capstone, write what you would do differently in a three-line note to your mentor.
Checklist
- I know the weights: 20, 20, 20, 15, 15, 10.
- I know the level shares: 0, 50, 80, 100 percent.
- I know the pass mark: 70 points and all ten mandatory items.
- I can say how to meet each of M1 to M10.
- I will self-grade before the final review.
Where to go next
Read the Track B weekly checklist and capstone plan and the Master Developer assessment overview. For how the six Dojo Readiness areas line up with the rubrics, see the readiness map.
Go deeper
Further learning
Free or low-cost material from other publishers. We opened every link; check the cost label before you start.
ADR community (adr.github.io)
Architectural Decision Records (ADRs) (opens the publisher's site in a new tab)
- Free
- Docs
- Intermediate
A community-maintained home for architectural decision records: short files that record a decision, the options considered and the reason. It explains the idea, lists templates and links to examples. Read it before you write your first decision record, so you can capture decisions in git in a form that a colleague, or an AI agent, can read later.
Mend Renovate
Renovate documentation (opens the publisher's site in a new tab)
- Free
- Docs
- Intermediate
The official docs for Renovate, a bot that opens merge requests when dependencies have new versions. Read the introduction and the configuration basics. It supports keeping a patched fork close to upstream.
Harbor project
Harbor documentation (opens the publisher's site in a new tab)
- Free
- Docs
- Intermediate
The official docs for Harbor, an open source container registry with scanning and signing. Use it to learn what a registry does for a team. Read the overview before the install pages.
Argo Project
Argo CD documentation (opens the publisher's site in a new tab)
- Free
- Docs
- Intermediate
The official docs for Argo CD, a tool that keeps a Kubernetes cluster equal to what is in git. Start with the overview and the getting started guide. It is reference material to dip into.
HashiCorp
What is Terraform (opens the publisher's site in a new tab)
- Free
- Docs
- Beginner
A short introduction to describing cloud resources as code, planning a change and applying it. It explains the plan step that this program asks you to put in the merge request. Read it first, then try a tutorial.
Google
Google SRE Book: Postmortem culture (opens the publisher's site in a new tab)
- Free
- Book
- Intermediate
Explains how to learn from failure without blame and how to make reviews useful. It pairs well with the five-whys method. Read it before you write your first review.
SOPS project
SOPS documentation (opens the publisher's site in a new tab)
- Free
- Docs
- Intermediate
Explains how to encrypt the values in a file so the file can sit safely in git. It helps you understand encrypted secrets next to the rule of never committing plaintext. Read the overview and the quick start.
GitHub
Open Source Guides: The legal side of open source (opens the publisher's site in a new tab)
- Free
- Article
- Beginner
Explains in plain language why licences matter and what happens when code has none. It supports the rule that code with no licence file is not open source. A short, readable guide.



