Level 11 · 15 levels · Builder · about 90 minutes
Level 11 — Assetize from the inside: evidence, versions, ASK, the tool map and how we build
By the end: Explain what Assetize is and who it serves, read a record's sources, rank and version, say why answers come from facts and not from a model at run time, send any customer question to the right tool, and ship a change to the site the way the team does.
IBM documentation and other sources flow into the Assetize evidence base of source-backed records. From it come the free tools and ASK, and three products: mxH for upgrades, mxR for shipping changes and the Data Workbench for loading data, each pointing at a customer Maximo.
Lesson 1
What Assetize is, and who it serves
Assetize is an evidence-grounded IBM Maximo knowledge base, plus the tools and products built on it. It reads IBM documentation, IBM support pages, fix lists and practitioner writing, turns them into short records that each state one fact with its sources, and publishes them at assetize.com.au. Everything else stands on that base.
IBM documentation and other sources flow into the Assetize evidence base of source-backed records. From it come the free tools and ASK, and three products: mxH for upgrades, mxR for shipping changes and the Data Workbench for loading data, each pointing at a customer Maximo.
The base is large and it keeps growing. When this level was written, on 1 November 2026, the corpus_stats tool answered with these figures. The live numbers are on the /method page.
corpus_stats, 1 November 2026
# Assetize corpus coverage
- Published records: 82797
- Unique source documents: 136578
- Versions represented: 1318
- Raw evidence items: 314268
Every published record is source-backed.| Who comes to Assetize | What they need | Where they start |
|---|---|---|
| A Maximo administrator planning an upgrade | What changes between their release and the target, and by when they must move | Compare versions, the Upgrade Desk, Support deadlines |
| A developer with an error on screen | What the BMXAA code means and how others fixed it | Error lookup, ASK |
| A customer with years of customisation | What the upgrade will break in their own scripts, screens and Java | mxH |
| A team shipping changes every week | A safe, tested path from DEV to PROD | mxR |
| A data team loading a spreadsheet | A load they can check, preview and prove | The Data Workbench and Workbench 3 |
| An IBM Business Partner | A way to scope and deliver assessments for clients | mxH for partners |
| Anyone using an AI assistant | Answers with sources instead of guesses | The MCP server |
The team works to a priority order, set in AGENTS.md. First, getting sources in, processed and published, with the truth kept intact. Second, the quality of ASK's answers. Third, how useful and current the records are to explore. Fourth, the website's clarity and working product paths. Last, internal plumbing, and only where it serves the first four.
The mistake: Describing Assetize as "an AI chatbot for Maximo". Do this instead: It is a base of source-backed records with tools on top. Every answer links back to the IBM page or practitioner source it came from.
Remember: Lead with what a customer can do today: compare two releases, look up an error, assess an upgrade, ship a change, load a sheet.
Lesson 2
The evidence model: sources, ranks and labels
Trust is the product. A Maximo team will act on an answer only if they can see where it came from. So every published record carries its sources, and every source carries a class that says how much weight it bears.
A pyramid of sources: IBM documentation and support at the top, recognised practitioners in the middle, general community posts at the bottom. A record card shows its sources with their ranks.
| Source class | Examples | How it is treated |
|---|---|---|
| IBM official | ibm.com documentation and support pages, IBM's own documentation sites on GitHub such as ibm-maximo-dev.github.io | The strongest evidence |
| IBM Business Partner content | Articles from IBM Business Partners | Published only after explicit approval, beyond a successful fetch |
| Reputable Maximo expert | A short, named list of practitioner sites | Published only after explicit approval. Labelled as practitioner, never as IBM |
| Community | community.ibm.com, user-group sites | A useful lead to check, carrying the least weight |
| Multi-source verified | Two or more approved sources that agree | Corroborated |
A candidate fact that has not passed review is AWAITING_VERIFICATION. It is private and is never returned by any public page, API or tool. Nothing is published on a hunch.
Here is a real record, as the get_record tool returns it. Read it from the bottom up: the source line tells you this fact is IBM's own words, then the scope line tells you which releases it covers.
get_record("testing-a-script")
# Testing a script
The scripting application now has a Test button to help test the script that you authored. ...
Versions: ALL_VERSIONS | Type: GOTCHA | Area: AUTOMATION_SCRIPTS
Web: https://assetize.com.au/record/testing-a-script
## Sources
- [IBM_OFFICIAL] Testing a script
https://ibm-maximo-dev.github.io/maximo-autoscript-documentation/troubleshooting/testingscriptsThe site shows four kinds of content, and they never look the same. A source-backed record is one fact with an approved source, date, version scope and provenance. "What changed" is a factual summary of a sequence of reviewed records. An Assetize observation is a clearly labelled interpretation that connects verified records; it is never presented as IBM's statement. "Related by topic" is a search cue that suggests what to read next; it is not evidence of a link.
- Never make a fact up. When the evidence does not support an answer, the tools say so and say what would settle it.
- Practitioner material never inherits IBM authority, even when it sits next to IBM records on the same topic.
- D1, the database, is the source of truth. Vectorize, the search index built from it, is derived and is never treated as truth.
- Sample data is welcome in demos. Label it once, clearly, and never dress it up as real evidence.
- Corrections stay visible in the public record. The trail is never rewritten silently.
A pyramid of sources: IBM documentation and support at the top, recognised practitioners in the middle, general community posts at the bottom. A record card shows its sources with their ranks.
The mistake: A forum post and an IBM page say the same thing, so you write the record as "IBM says". Do this instead: Cite the IBM page as IBM, and the forum post as community. If only the forum post exists, the record carries the community label.
Lesson 3
Versions on every fact
Maximo is not one product. It is a family of releases with real differences: Maximo Asset Management 7.6.1 on WebSphere, and Maximo Application Suite 8 and 9 on Red Hat OpenShift, each with its own fix packs. A fact about 9.2 is not a fact about 7.6.1. So every record carries its version scope, and every tool answers for a version.
On the left, Maximo 7.6.1 runs as one application on WebSphere servers. On the right, Maximo Application Suite runs in containers on Red Hat OpenShift: a core for users and licences, and applications such as Manage, Monitor, Health, Predict and Visual Inspection, each as pods.
When a source is read, the version is taken from the page address first, then from its opening text, and a fact that genuinely applies to every release is marked ALL_VERSIONS on purpose. Each version also gets a family: Maximo 7.6, MAS 8 or MAS 9. Compare two records the tools returned this month:
| Record | Version scope | Source |
|---|---|---|
| Testing a script (the Test button in Automation Scripts) | ALL_VERSIONS | IBM, ibm-maximo-dev.github.io |
| Diagnostics and testing when moving from Java 17 to Java 25 | 9.2 | IBM Support page 7270577 |
| End of support for Maximo 7.6.1 | 7.6.1, 7.6.3, 7.6.7, 7.6.8 | IBM Support |
Dates are facts with versions too. The support_lifecycle tool returns each date with the IBM page it comes from. Regular base support for Maximo Asset Management 7.6.1.x ended on 30 September 2025. All MAS 8.x releases moved to Extended Support only on 1 May 2026. Dual Support for MAS with 7.6.1.x ends on 30 April 2027. MAS 9.0 moves to Extended Support on 1 July 2027.
A timeline: 30 September 2025 end of regular support for Maximo 7.6.1.x, 1 May 2026 MAS 8.x on Extended Support only, 30 April 2027 Dual Support ends, 1 July 2027 MAS 9.0 moves to Extended Support. A marker shows today.
The upgrade tools use five catalogue releases: 7.6.1, 8.11, 9.0, 9.1 and 9.2. The test stub and the script tools use the same five as lenses: 761, mas811, mas90, mas91 and mas92. Here is the test_automation_script_evidence tool checking a completion script against the 7.6.1 dictionary of MXAPIWODETAIL. It reads which attributes the script uses and says whether each one is evidenced on that release.
test_automation_script_evidence(source, "761")
# Static evidence check — MXAPIWODETAIL on Maximo 7.6.1
Attributes referenced: FAILURECODE, STATUS
Status: CLEAN
- FAILURECODE: EVIDENCED (object-structure) — FAILURECODE is in the Maximo 7.6.1 MXAPIWODETAIL dictionary.
- STATUS: EVIDENCED (object-structure) — STATUS is in the Maximo 7.6.1 MXAPIWODETAIL dictionary.Two doors. The 7.6.1 door accepts a maxauth user and password and an API key. The MAS 9 door accepts only the API key; the maxauth caller bounces with 401.
The mistake: Answering a customer from a record without checking its version scope against their release. Do this instead: Read the Versions line first. If the record's scope does not include their release, say so and look for the record that does.
Remember: No record found is a reason to search further and record the gap, never proof that something does not exist in Maximo.
Lesson 4
AI at build time, facts at run time: ASK and the MCP server
The house rule for every Assetize product is short: AI at build time only, with no model call in production logic. At build time AI helps read sources, extract claims and shape records, and every record is reviewed and stored with its sources. At run time, when a customer asks, the answer comes from typed facts and ranked records that already exist.
At build time, AI helps read sources and shape records, and every record is checked and stored with its sources. At run time, the answer is built from typed facts and ranked records that already exist; no AI model decides it.
Why so strict? Three reasons. The same question gets the same answer every time, so a customer can rely on it and a test can check it. Every sentence can be traced to a record and a source, because nothing was invented on the fly. And it is fast and cheap: the answer is on screen in the time a few database reads take.
ASK (/ask) is how a person asks a question. Its primary lane, Ask v2, answers from typed, versioned claims and a version-aware, trust-aware rerank of the published search, with no model call. Each claim in the answer carries an [R#] marker back to its record. When the published evidence cannot support an answer, ASK says so instead of guessing, and the question is logged as a gap for the corpus to fill. Where model-written prose appears, it is added after the facts are delivered and never decides them.
| A typical chatbot | ASK |
|---|---|
|
|
MCP, the Model Context Protocol, is an open standard that lets an AI assistant (Claude, Copilot, Cursor and others) call tools on a server. The Assetize MCP server lives at https://assetize.com.au/mcp. It is public and needs no login. It offers 19 tools and a set of resources, so an assistant can look things up in the records instead of guessing.
| Tool | Answers |
|---|---|
| ask_maximo | A cited answer to a practical question, or a decline |
| search_records, get_record | Keyword search over the records; one full record with every source |
| lookup_error_code | The records that explain a BMXAA or IJ code |
| compare_maximo_versions, upgrade_plan, upgrade_checklist | What a move between two releases means, as a report, a plan and a phased checklist |
| support_lifecycle, lifecycle_dates | End-of-support and Extended Support dates, with the IBM page |
| deprecation_register, fixes_gained, maximo_changelog, change_watch | What was removed or renamed, what fixes a move brings, what changed in 9.x |
| list_gotchas, behaviour_history, records_by_version, maximo_mobile | Known traps, how a behaviour evolved, everything at one release, Maximo Mobile |
| test_automation_script_evidence, corpus_stats | A static check of a WORKORDER script against one release; coverage figures |
Smoke-test the MCP server: one JSON-RPC message per POST
curl -s https://assetize.com.au/mcp \
-H 'Content-Type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'The public limits are per IP address: ask_maximo answers to how-to and incident questions are 10 a minute and 100 a day; the other tools are 30 a minute and 500 a day. Daily allowances reset at 00:00 UTC, and a refusal says when to try again. A rate limit is just a 429, and you know from Level 9 how good clients handle one.
The mistake: Adding a model call to an API route to "polish" an answer before it is returned. Do this instead: Keep the answer built from facts and records. If wording needs work, improve the record or the template at build time, where it can be reviewed and tested.
Lesson 5
The tool map: which tool answers which customer question
A customer never asks for a tool by name. They ask a question. Your job is to hear the question and send them to the one page that answers it. Learn this map and you can help anyone on a call.
Customer questions on the left connect to Assetize tools on the right: what changes in the upgrade to Compare and the Upgrade Desk, what an error means to Error lookup, what will break to mxH, how to ship safely to mxR, how to load a sheet to the Workbench, and anything else to ASK.
| The customer asks | Send them to | Route |
|---|---|---|
| "What changes if we move from 7.6.1 to 9.2?" | Compare versions, then the Upgrade Desk for the plan and checklist | /compare, /upgrade |
| "When do we have to be off this release?" | Support deadlines, each date with its IBM page | /upgrade/deadlines |
| "Is something we use being removed?" | The deprecation register | /upgrade/deprecations |
| "What does error BMXAA4129E mean?" | Error lookup | /error |
| "How are these objects related?" | The Relationship Explorer | /data-model |
| "What will the upgrade break in OUR customisations?" | mxH, the upgrade assessment of their own system | /mxh |
| "Can we fix that finding in one place?" | The Workbench 2.0 Studio: open an mxH finding, fix it in the right editor, send the evidence back | /studio |
| "Will this screen change look right?" | The UI Playground | /ui/playground |
| "Can we design or check a workflow?" | The Workflow Designer | /workflow |
| "How do I test my script without a server?" | The Automation Script Test Harness | /test-stubs/automation-scripts |
| "How do we ship changes safely, DEV to PROD?" | mxR | /mxr |
| "How do we load this spreadsheet and prove it landed?" | The Data Workbench and Workbench 3 | /workbench, /workbench3 |
| "What changed in 9.x that matters?" | Change Watch, most severe first | /change-watch |
| "Just answer my question." | ASK, or their own AI through the MCP server | /ask, /connect |
Customer questions on the left connect to Assetize tools on the right: what changes in the upgrade to Compare and the Upgrade Desk, what an error means to Error lookup, what will break to mxH, how to ship safely to mxR, how to load a sheet to the Workbench, and anything else to ASK.
Two questions sound alike and need different tools. "What changes between 7.6.1 and 9.2?" is about IBM's releases, and Compare answers it for everyone. "What will the upgrade break for us?" is about the customer's own scripts, screens and Java, and only mxH, run against their configuration, answers it. Level 12 covers mxH in depth.
Every tool is listed in one directory at /tools, with its status, its docs and the evidence it runs on. When you are unsure, start there.
The mistake: Sending every question to ASK because it can answer anything. Do this instead: ASK is the right start for an open question. For a known job (a date, an error code, a comparison, an assessment) the dedicated tool gives a fuller, more structured answer.
Lesson 6
How the site is built: the stack and the one registry
Assetize is one Git repository, managed with pnpm, written in TypeScript. The site is a React single-page app (Vite, Tailwind CSS, framer-motion for motion). Cloudflare Pages serves it. Cloudflare Workers run the API, the MCP server, the scheduled jobs (crons) and the server-side rendering that gives search engines a real page for every route. Records live in Cloudflare D1, a managed serverless database with SQLite's SQL semantics. The search index lives in Vectorize and is always rebuilt from D1, never the other way round.
- Source fetched
- Claim extracted (AI at build time)
- Reviewed, versioned, ranked
- Published to D1
- Indexed in Vectorize
- Served by Workers to the site, ASK and MCP
How a fact reaches a customer
Every tool on the site is registered once, in artifacts/assetize/src/lib/site-map.ts. That one file feeds the header, the footer, the breadcrumbs, the section landing pages, the home page bands, the /tools directory, the generated docs pages, the tool strip at the foot of each tool page, the tour, and the Worker's server-rendered pages. So what a search engine sees matches what a visitor sees. A tool is a site-map item with a product block: an id from PRODUCT_IDS, a group, a status, an action, a result, its steps and its docs link.
Six cards, each with a picture: KISS, DRY (one source feeding many places), YAGNI, single responsibility, fail closed (a gate that shuts when unsure) and idempotent (pressing twice gives one result).
One site-map item (shortened)
{ href: '/mxr', label: 'mxR release pipeline',
desc: 'CI/CD/CT for Maximo Manage: every change is a work order, raised from your Git diff, deployed one site at a time and tested by artefact type.',
...
product: { id: 'mxr', group: ..., status: ..., action: ..., result: ..., docs: '/mxr/docs', steps: [...] } }Tests guard the registry. site-map.test.ts fails when a route is not served or a product is not registered. tool-data-versions.test.ts fails when the manifest of which evidence each tool runs on is out of date. You run them all with pnpm --filter @workspace/assetize test:lib before you push.
Your first day at the keyboard
pnpm install
pnpm --filter @workspace/assetize test:lib # the library tests, registry included
pnpm exec tsc -p tsconfig.json --noEmit # type check, from artifacts/assetize
PORT=5000 BASE_PATH=/ pnpm --filter @workspace/assetize run dev # the site on localhost:5000Two pull requests side by side. The small one, twelve lines, is reviewed and merged in minutes. The big one, two thousand lines, hides a bug that slips through and breaks later.
The mistake: Adding a new tool by hard-coding a link in the footer and another in the directory. Do this instead: Add one site-map item with its product block. Every list on the site picks it up, and the tests check it is served.
Remember: One registry, many readers. If you are about to write a second list of tools, stop and read site-map.ts.
Lesson 7
House rules: one page one intent, voice, protected parts, merge, deploy, fix
A handful of rules shape every change to the site. Each one exists because something went wrong without it, and most have a test that fails when the rule is broken. Do not edit a guard test to make it pass; fix the page.
| Rule | What it means | The test that guards it |
|---|---|---|
| One page, one intent | The H1 states the page's job. The first thing under it is the tool or the answer, with the main action above the fold. Background never comes first. A child route renders its own content. The server title and H1 match the app's H1 and tab title | one-page-one-intent.test.ts |
| Full width | Pages use the whole screen. Never cap a page in a narrow centred column; a readable width for paragraphs of prose is fine | full-width.test.ts |
| Docs template | Every product docs page uses the /mxh/docs template: sections on the left, "On this page" on the right, Ctrl K search, from 1024px up | docs-template.test.ts |
| One registry | Every tool is one site-map.ts item | site-map.test.ts |
| Protected parts | The header menu and the home page Change Watch panel change only with the owner's approval in the current conversation | protected-surfaces.test.ts |
Six cards, each with a picture: KISS, DRY (one source feeding many places), YAGNI, single responsibility, fail closed (a gate that shuts when unsure) and idempotent (pressing twice gives one result).
The house voice: lead with the solution. Say what the product does, show it working, and give the next step. Short sentences. Write "your Maximo" and "your system". Avoid corporate filler such as "surface" and the other banned words in CLAUDE.md. On anything a customer sees, never headline what Assetize has not done or cannot do; internal records (docs, evidence packs, test reports, pull request descriptions) keep the full, truthful status. A demo shows the whole picture on sample data, labelled once.
A Git graph: the main line, a branch splitting off, two commits on it, a pull request with tests and a review tick, then the branch merging back into main.
Merge and deploy are yours. GitHub Actions is switched off for everything (it cost about $100 in one cycle), and its workflow files sit in .github/workflows-disabled/, where they stay. The checks you run locally are the gate. When work is ready and those checks pass, you merge the pull request yourself and deploy production yourself from a clean copy of current main:
Deploy (needs the Cloudflare token and account id in your environment, never in the repo)
git fetch origin main && git checkout origin/main
bash scripts/deploy-production.shThe script builds the site, bundles the Worker, promotes the build to the pages branch, then deploys Pages and the assetize, discovery and watchdog Workers. Afterwards you check two things: that production is the "ad_hoc" upload you just made, and that the routes you changed answer 200 on assetize.com.au.
A pull request card: the title, a diff with red removed lines and green added lines, checks turning from running to passed, and a reviewer approving.
Any failure you find, you fix. When a check, test, build, deploy or probe fails in front of you, you reproduce it, fix the cause, run the check again and report what you fixed, even when your change did not cause it and it sits outside your area. "It fails on main too" is a finding, not a reason to stop. You leave a failure alone only when the fix needs the owner's approval (a protected part, a destructive production step) or a permission you do not have; then you say exactly what fails, why, and the fix you would apply.
- Never expose secrets, credentials or customer data.
- Never knowingly corrupt evidence or its provenance.
- Never run a destructive production step without understanding its effect.
- Never replace current main with stale work.
- Never change a protected part without the owner's approval.
Remember: The final rule: protect the truth, protect customer data, protect production from destructive mistakes. Then build boldly.
Check yourself: the Level 11 quiz
18 questions. Pass mark 70%.
- Look at the lit part. What is it, and why does it matter? (picture: Assetize in one picture)
- The website header; it holds the menu
- The base of source-backed records; every tool reads from it
- A list of customers
- The billing system
- Which picture lights the strongest kind of evidence?
- Top rank lit
- Middle rank lit
- Bottom rank lit
- A named Maximo expert's blog explains a fix. Which rank is lit for it, and how is the record labelled? (picture: Ranks of evidence)
- As IBM fact, because it is accurate
- As practitioner evidence, never as IBM
- It is ignored
- As community rumour
- A candidate fact is AWAITING_VERIFICATION. Where can a customer see it?
- On the record page with a warning
- Nowhere: it is private and never returned by any public page, API or tool
- Only in ASK
- Only in the MCP server
- The published evidence does not support an answer. What does ASK do?
- Guesses the most likely answer
- Declines, and the question is logged as a gap to fill
- Makes up a source
- Answers from a model's memory
- A record reads "Versions: 9.2". A customer runs 9.1. What do you do?
- Use it anyway; versions are close
- Say this record covers 9.2, and look for the record that covers 9.1
- Change the record to 9.1
- Tell them to upgrade
- On this support clock, which date is already behind today's marker? (picture: The support clock)
- 30 September 2025: end of regular support for Maximo 7.6.1.x
- 30 April 2027: Dual Support ends
- 1 July 2027: MAS 9.0 moves to Extended Support
- None of them
- In this picture, where does the run-time answer come from? (picture: AI at build time, facts at run time)
- A language model writing a fresh answer
- Typed facts and ranked records that already exist; no model decides it
- A person on the help desk
- A random record
- Why does Assetize keep model calls out of run-time answers?
- Models are not allowed on Cloudflare
- So the same question gets the same, traceable answer, fast, with every claim tied to a source
- To save disk space
- Because customers dislike AI
- What is the Assetize MCP server for?
- Billing customers
- Letting AI assistants call Assetize tools, such as search_records and upgrade_plan, and get evidence instead of guessing
- Hosting videos
- Sending email
- A customer pastes BMXAA4129E into a chat. Where do you send them first? (picture: Which tool answers which question)
- The Workflow Designer
- Error lookup
- mxR
- The UI Playground
- A customer asks "when must we be off 7.6.1?" Which tool does the lit line lead to? (picture: Which tool answers which question)
- Support deadlines (/upgrade/deadlines)
- The UI Playground
- The Data Workbench
- mxR
- "What will the 9.2 upgrade break in our own scripts and screens?" Which tool? (picture: Which tool answers which question)
- Compare versions
- mxH, the assessment of their own system
- Change Watch
- Support deadlines
- You need to add a new tool to the site. Where do you register it? (picture: Six principles)
- In the footer, the header and the directory, one by one
- Once, as an item with a product block in site-map.ts
- In a spreadsheet the owner keeps
- In the database
- What does "one page, one intent" require at the top of a tool page?
- A long introduction to the background
- The H1 states the job, and the tool or answer comes right under it
- Three paragraphs on what the tool cannot do
- A list of every other tool
- Which change needs the owner's approval in the current conversation?
- Fixing a typo on a record page
- Moving the header menu breakpoint or changing the home page Change Watch panel
- Adding a test
- Improving a docs page
- Your local checks pass and the pull request is ready. What happens next at Assetize? (picture: Branch, commit, pull request, merge)
- Wait for GitHub Actions to deploy it
- You merge it yourself, deploy from current main with scripts/deploy-production.sh, and check production is the ad_hoc upload and your routes answer 200
- Email the owner and wait
- Leave it open for a week
- A check fails on your pull request. Your change did not cause it, and it fails on main too. What do you do? (picture: A pull request up close)
- Ignore it: not your change
- Reproduce it, fix the cause, run the check again and report what you fixed
- Edit the test so it passes
- Switch the check off
Remember it as a picture
- Assetize: A base of source-backed Maximo records, with free tools, ASK, mxH, mxR and the Data Workbench on top.
- IBM official: IBM documentation and support pages: the strongest evidence a record can carry.
- Practitioner evidence: Approved named experts. Strong, and always under its own label, never as IBM.
- Community: Forums and user groups: a lead to check, carrying the least weight.
- Versions on every fact: Every record and every date belongs to a release. Read the Versions line first.
- One name, many releases: 7.6.1 on WebSphere, MAS 8 and 9 on OpenShift. A fact without its version could belong to either.
- AI at build time: AI helps shape records at build time. At run time, answers come from typed facts and ranked records.
- ASK and MCP: Cited answers with [R#] markers, or a decline. The same tools reach AI assistants through assetize.com.au/mcp.
- Compare and the Upgrade Desk: "What changes between two releases?" A cited report, a staged plan and a checklist.
- The Workbench: "How do we load this sheet?" Map, check, preview, load and reconcile.
- site-map.ts: The one registry: header, footer, directory, docs, tour and the Worker's pages all read it.
- Fail closed: When unsure, the gate shuts: no guessed facts, no guessed stub answers, no guard test edited to pass.
- Merge and deploy yourself: Local checks pass, you merge, you run scripts/deploy-production.sh, you prove the deploy.
- Any failure you find, you fix: Reproduce, fix the cause, rerun, report. Even when you did not cause it.
Where these facts come from
Try it on the real tools
How did Level 11 go?
Rate the level and say what worked, what did not, what you liked and what you did not. Send it now, or with your full report.