One thing blocks the shape of this
The keynote goal issue says the talk is September 2026. The planning notes from July say the talk is October, with preparation starting in September. Those cannot both be true, and the difference is an entire month of work. Everything below assumes October. If it is September, development is already over and this page is the wrong shape — say so and it gets rewritten rather than patched.
Why September goes to open.gismo
Three efforts are in good health. The report builder released a version with a real efficacy section on 27 August, taking it from six displays to twenty-six. The chart library has thirteen chart types and a legacy backlog that finally moved. The R package has a rebuilt census and thirteen widget bindings.
The talk is about none of those. The summer deliverable, stated in July, was open.gismo — a full end-to-end environment for safety and RBQM and anything else. That is the claim the keynote makes, and open.gismo is the thinnest thing in the programme: one merged pull request in the last fortnight, a v0.2 that shipped an architecture the maintainer had said he disagreed with, and a design discussion meant to settle that which has been open since 12 July.
So the month is not "continue the queue". It is: make the headline claim true enough to demonstrate, and stop doing anything else.
The demo that matters already has its design. The August data-loading session measured the problem and recommended a path: put the folder on disk as the vendor sent it, run one command, and get back a single document that prices every gap in terms of the charts it turns off. That is the moment where a stranger's data becomes charts — and it is the most interesting five minutes available in a talk about this kind of tool.
The month, and who does what
Most of it is not you. The parts that are, are cheap and they unblock the rest.
Week 1 — mostly yours, and the cheapest week of the month
@jwildfire
- Write down what version one contains. A morning. Everything else in this area is downstream of it, and until it exists the honest answer to "is it ready" is "ready for what".
- Answer the five decision pages that survived triage, and pick a branch-protection tier — one choice between three options, with the tooling already merged and waiting.
- Sign off the two harness pull requests that touch guardrail files. They are 13 and 11 days old.
- Bind the demo study repository to a goal. It is a one-line edit inside the approval carve-out, and it is the reason a broken weekly pipeline went unnoticed for three weeks — a repository in no goal cannot be picked up by any session.
- Plug in a backup drive.
Weeks 2–3 — the build, almost entirely agents
agents
The data-loading path in open.gismo, to the design the August session already produced and measured. Read a folder as the vendor sent it; report what is usable and what is not; price every gap in the charts it turns off. This is implementation rather than discovery, which is why three weeks is credible.
Alongside it, the demo study needs to be reliably green from a cold start rather than green last Tuesday — it is the thing a version one would be judged on, and it failed three consecutive weekly rebuilds in silence before that was fixed.
Week 4 — not a build week
@jwildfire + agents
Rehearsal. Running the demo on the real machine, from a cold start, repeatedly.
The reason to protect this week is specific rather than general. Every serious defect this programme found in August was something that looked fine: four checks that never started and so raised no red mark, a page claiming requirement coverage it did not have, a merge that combined two sections into one number and raised no conflict. Those are exactly the failures that survive a quick look and die in front of an audience.
If the month slips, slip week three, not week four.
What not to build, and why that is the harder half
Not the statistical analysis plan
Designed on 27 August as D0030, and it is good work — most of the plan already exists dispersed across twenty-six displays, and a display turns out to render as an empty shell from its own spec, which makes it cheaper than it looks. It is also v0.4, it serves no part of the talk, and it would eat a fortnight.
Not the review layer
Designed as D0029. It is the largest capability gap in the portfolio measured against competing platforms, and that is an argument for doing it, not for doing it now.
Not the laptop migration
Already decided; recorded here so it stays decided.
Four, each with a recommendation
Is the talk in September or October?
The goal issue and the July planning notes disagree. This is not a preference — it is a fact one of us has wrong, and it decides whether September is a build month or already past.
Does September go to open.gismo, or to finishing what is already strong?
- open.gismo. Makes the headline claim demonstrable. Costs the momentum of three efforts that are currently going well.
- Consolidate. Ship another report-builder version and more chart types. Safer, and leaves the talk claiming an end-to-end environment while demonstrating its parts.
Are the SAP and the review layer genuinely deferred?
Where is the hard stop, and is it real?
What was waiting on you, after triage
Checked on 31 August rather than relayed from the queue.
| Item | Verdict |
|---|---|
| D0028 — the widget contract | Answered by the work. The two widgets shipped on 27 August using named table arguments — exactly the shape the page recommended. Nothing left to decide. |
| D0024 — the laptop move | Mostly overtaken. Nine of eleven questions hang off a migration that was called off. Two findings underneath are still live and were re-verified today. |
| D0026 — clinical priorities | Half answered by you. The ranking was acted on the night it was written. The version-one question stands, and it is week one above. |
| D0025 — CRAN | Advanced. Both pull requests are open upstream since 27 August. Currently waiting on the other maintainers, not on you; the CRAN submission is still yours. |
| D0022, D0027, D0029, D0030 | Genuinely open. Branch protections is the most urgent and the most understated — only one of four main branches is protected, and it is not the one holding the merge policy. |
| gsm.safety #68 and #69 | One release train, not two decisions. #69 contains #68 entirely. Both still carry a CI defect fixed on dev and never carried across. |
| obot.agent #198 and #273 | Genuinely yours. Both touch guardrail files, so the carve-out requires your sign-off. 13 and 11 days old. |
Eight decision pages and four release candidates reduce to roughly seven real items, most of them a single choice each.