Rebuild the execution boundary demo and give it its own page - #130
Merged
Conversation
The demo said what CTRLRun decided and never said what it was deciding instead of. A reader arriving from an agent framework could not tell which part of the picture was their model and which part was this library, so the whole thing read as one more layer of the same kind. Three changes fix that. The explorer opens with three lanes -- the LLM decides what to do, CTRLRun decides whether it may run, the real system is where it becomes real -- and every outcome now pairs what the model believed against what the kernel did. Under the request sits the ladder of checks in order, marking which one stopped the action, so a refusal is attributable to a check rather than appearing as a verdict from nowhere. Each outcome carries the line the caller's own code would see and a link to the concept page behind it. Every scenario is now addressable. `?domain=&action=&situation=` is read on mount, the share row hands over the link, and `/try` hosts the same explorer as a page of its own -- the home page section stays where it is. The lost response section links into the case it describes rather than only naming it. A new home page section says where CTRLRun sits in plain English and hands off to the architecture for the internals, and the medical affairs build map opens rather than hiding behind a summary. The link checker was reporting `/try?situation=uncertain` as a missing page: a query string is not part of a path, and the strip is tested in both directions so a genuinely missing page carrying a query still fails.
Contributor
|
Preview deployment for your docs. Learn more about Mintlify Previews.
💡 Tip: Enable Automations to automatically generate PRs for you. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changed
The demo now says which part is the model and which part is CTRLRun. The explorer opens with three lanes — the LLM decides what to do, CTRLRun decides whether it may run, the real system is where it becomes real — and every outcome pairs what the LLM believed against what CTRLRun did. That pairing is the whole point: "the call errored, so it failed" next to "an error is not evidence of failure" explains the product faster than any paragraph on the page.
A refusal is now attributable to a check. Under the request sits the ladder — authority, policy, approval binding, effect reservation, outcome — with the check that stopped the action marked, and a note per check saying what it found. Each outcome also carries the line the caller's own code would see (
raise ApprovalMismatch(...)) and a link to the concept page behind it.Every scenario is addressable.
?domain=&action=&situation=is read on mount, a share row hands over the link, and/tryhosts the same explorer as a page of its own. The home page section stays exactly where it was. All 48 domains are still browsable; six quick chips sit above the picker for the common ones.A plain-English "how it works" section on the home page (
#how-it-works) says where CTRLRun sits, in four steps, and hands off to/docs/ARCHITECTUREfor the internals. It ends by naming what CTRLRun is not: it does not read prompts, choose models, rank tools, or judge the agent's reasoning.Cross-links between the two demos.
/tryillustrates the decisions and says so on the page;/docs/try-itruns the released wheel in the tab. Each now points at the other, and the difference is stated rather than implied.Medical affairs: the build map is
<details open>rather than hidden behind a summary.Link checker:
/try?situation=uncertainwas reported as a missing page. A query string is not part of a path. Stripped in_resolveand in the planned-page lookup, with a test asserting both directions — the working link resolves, and a genuinely missing page carrying a query still fails.Verified
pytest tests/— 4446 passed, 47 skipped.mint devand driven with Playwright: the deep link lands on the right scenario through a client-side navigation, the approve → execute → retry → duplicate flow runs, dark mode and 390px both hold, no console errors.completedoutcome no longer claims an approval was consumed on the path where none was required.