docs(roadmap): the order 0.11.0 is worked in, and why it is a dependency rather than a preference

[stephrobert]

Thirteen issues remain in 0.11.0 and two of them move the numbers the other eleven quote. The order is written down because it is not obvious and is expensive to get wrong.

What it says

1. Trust the measurement — The behaviour axis is not reproducible: two identical runs disagree on one operation, and a regeneration can demote three #398 (behaviour returns 313 and 314 on two identical runs) and Three axis percentages published in the documentation and in #390 are wrong, and nothing in the repository refuses a hand-written measured number #406 (three published axis percentages are wrong, and nothing refuses a measured number written by hand). Working parity before these means aiming at a moving target and republishing wrong figures.
2. The question that can cancel most of the rest — The shape axis stands at 14 %, not 100 %, and it is the weakest of the seven — nobody had ever read it #407: 292 of shape's 318 zeros are unrecorded, and this repository already holds 619 recorded exchanges in corpus/ that do not feed shapes/. Whether they can is answerable offline, without an account. Going back to three accounts for answers already committed would be the worst possible order.
3. The image catalogue chain — The image catalogue cannot answer a device mapping until a snapshot backs it: filled, no state of the pack satisfies shapes:check, score.sh and terraform.sh at once #389, then The emulated image catalogue answers no BlockDeviceMappings and no PermissionsToLaunch #383, then A Vm answers no UserData, no BlockDeviceMappings and no Tags, where the cloud answers all three #378, as one model rather than three patches.
4. Parity — Fifteen Scaleway and twelve Exoscale address operations are driven by nobody, and IPAM is the least reached of all #414 first because it is cheapest, Machines are proven ten ways on Outscale and one hundred and thirty-four times less on Scaleway #409 last because it is 134 gaps.
5. Documentation — Three documents describe a version of this project that shipped two releases ago, and the README hero promises the one client the emulator deliberately refuses #403 deliberately last, so it describes the release's final state instead of being corrected twice.

The cut, if the release must ship sooner, is named too: after wave 3. That is the maintainer's decision, and writing it down is not the same as taking it.

One thing the page now says about itself

The section ends by stating that the rest of roadmap.md is out of date and #403 rewrites it — it still presents feint replay, coverage --observed and fault injection as work to come, and all three shipped.

That warning sits before the first stale line rather than after. A roadmap presenting shipped work as future work is worse than no roadmap: it is wrong in the direction that undersells the project, and a reader who catches it stops trusting the pages that are right.

The milestone description carries the same order, for whoever is working rather than reading.

Related

Refs #403, which rewrites the rest of this page.