The content plan
Written by a person. Last read by a person on 2026-09-08, 20 days ago. Its facts were checked by the eval suite on 2026-09-28.
You are asking what the money actually produces, you doubt that more modules can fix a problem more modules created, or you want to see a content library sized against evidence rather than against a catalog.
Scenario-based sample. Halden Systems is invented, and so is every figure about it.
Bottom line. Halden Systems is not building 140 modules. We audited the 140 we have, 96 of them are being retired, and the library that replaces them is 79 pieces of which almost none is a course. Nothing runs longer than 15 minutes and most of it takes under 4.
Why the format changes and the volume falls
The evidence rules out doing more of what we did. 78 percent can follow our documented steps and 31 percent can say what those steps do, so the gap is not coverage. 29 percent rated our previous training useful, which is this function's own report card. 34 percent think more training would help, and they are right to doubt it.
A library of 140 modules is what produced those numbers. Replacing it with a larger one would be buying more of a metric that is already full.
The audit
Every one of the 140 was read against a single question: does somebody who works through this come out able to predict what the system will do. 96 could not pass it.
| What we found | Units | What happens |
|---|---|---|
| Teaches the steps and not the understanding, mostly screen recordings | 41 | Retired |
| Covers a tool we no longer use | 24 | Retired |
| Duplicates another unit, usually a regional copy | 19 | Retired |
| Compliance content that belongs with the policy, not here | 12 | Moved to the policy owner |
| Reference that is correct and still needed | 22 | Migrated unchanged |
| Right subject, wrong form | 22 | Rewritten |
That leaves 44 surviving units. 35 new pieces are written to fill what the audit exposed, which is a library of 79.
The 19 duplicates are worth naming separately. They are the $410,000 of regional spend showing up as content: 9 sites each solving the same problem, and none of them wrong to.
What a piece of content is now
Nothing here is a course and nothing has a completion certificate. Each piece is read at the moment somebody needs it, and the measure is whether they then do the thing, not whether they reached the end.
| Form | Length | When it is used | Count |
|---|---|---|---|
| Job aid | 2 to 4 min | At the keyboard, mid-task, with the thing open | 31 |
| Worked example on a real repository | 10 to 15 min | Once, early, with a checkpoint that proves it | 18 |
| Concept explanation | 6 to 8 min | When somebody asks why, not how | 14 |
| Decision guide | 2 to 3 min | When the situation has to be identified before acting | 9 |
| Wrongness log entry | 1 to 2 min | After the assistant was confidently wrong | 7 and growing |
Casual in register and exact in content. Written the way a colleague explains something at a desk, which is the register 61 percent of people said they would answer a question in.
One thing in the program is not in this table and matters more than any of it. The facilitated first change is 45 minutes, once, with a local expert, on a real task the person actually has. That is the intervention the Dublin pilot ran, and it is what moved days to a first merged change from 41 to 9.
The 79, by subject
Grouped by the system somebody is working in, because that is how the need arrives.
| Subject | Pieces | The piece that carries the group |
|---|---|---|
| Repositories and version control | 18 | What a branch actually is, and why yours diverged |
| Docs-as-code, previews and publishing | 14 | Why the preview looks right and the build does not |
| Issue tracking and planning | 9 | Which field the automation reads |
| Pipelines, and reading a failed build | 8 | The 6 lines of a build log that matter |
| Design tokens in a repository | 6 | Changing a color without changing 40 files |
| Support runbooks | 7 | Writing one somebody else can run |
| The assistant, and judging its output | 11 | Which answers need checking |
| Access, permissions, and what cannot be undone | 6 | What you can try safely, and what you cannot |
The last group is 6 pieces against a fear held by 64 percent of the population, and it is the smallest group that carries the most weight. Somebody who knows which actions are reversible stops needing permission to try.
What this costs to keep true
79 pieces against 4 engineering systems that ship continuously. At the rate those systems change, about 12 percent of the library is wrong at any moment, and a central team of 2.5 cannot chase 4 release trains.
The teams that own each system own the pieces about their system, the way they already carry release notes. The enablement function holds the standard and edits against it. That split is in the strategy and it is the reason the authoring line in the budget halves over 3 years.
What we are not making
No onboarding curriculum. A new joiner meets the same job aids as everybody else, plus the facilitated first change. A separate track for new people is a second library to keep true.
No video. Recordings teach the steps and skip the thinking, which is how the 47-point gap was made. Where motion genuinely helps, it is a 20-second clip inside a job aid rather than a recording somebody sits through.
No certificates of completion, except for the customer and partner certification, which is a different product for a different audience and is priced in the revenue case.
No content for the outer 1,850 in year 1. They have an assistant and no repository work, so the 11 pieces on judging its output are theirs and the rest are not. Writing for them before we have fixed the inner population would be starting a fifth rollout.