You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Retargeted (website#498): the NSG connector deliverable is superseded by NEMAR compute (#493, nemarOrg/nemar-cli#2021). website#498 removes the dataset page Compute dialog's link to this issue; that change is live once it reaches production. The other deliverables were closed out or overtaken on 2026-08-03, see the comments.
Wire Vis / Compute placeholder UI to open a "coming soon - track progress on issue #X" modal linking to the follow-up issues (Compute: superseded, see above)
DNS cutover plan in .context/cutover_plan.md: switch www.nemar.org to Cloudflare Pages, keep legacy.nemar.org for 90 days, redirect map for old /dataexplorer/... URLs
Out of scope
Implementing the viewer or NSG connector (these become their own epics)
Acceptance
Two architecture docs reviewed and merged
Three follow-up issues created and linked
Lighthouse a11y >= 95, perf >= 90, SEO >= 95 on landing + Discover + dataset detail
Cutover doc has a concrete date + rollback procedure
Closing: the phase's purpose was to design the deferred live features and plan the cutover. Both were overtaken by actually building and shipping them.
DNS cutover — DONE. The apex cutover landed 2026-07-29 (Apex cutover: ww2 becomes nemar.org, legacy moves to ww1, legacy URLs must keep resolving #190); nemar.org, www, ww2, and app.nemar.org all serve this build, with ww2 deliberately kept alive so persistently-cached 301s land on a working page. The host model is documented in AGENTS.md rather than a one-off .context/cutover_plan.md.
Viewer architecture doc + 'coming soon' placeholder — superseded by the real thing. The WebGL EEG viewer is built and shipped (src/lib/eeg-viewer/: traces, montages, topo, DSP, filters, events, all unit-tested), served from zarr.nemar.org. A design doc for a feature already in production is not worth writing.
Accessibility, theme, and performance passes — done incrementally across the epic instead of as one phase-end sweep (7953145 design-review a11y fixes, 27ff388 AA contrast, cdac663 dialog a11y, and the cold-paint perf epic Epic: Cold-paint perf + cache hygiene for dataset detail #66, closed).
The one deliverable with no successor is the NSG connector architecture doc; there is no compute-launch flow today and no open work depending on that document. If NSG compute gets picked up, it should start as its own epic against the current architecture, not from this issue.
Open follow-ups that came out of this area are tracked on their own: #208 (viewer zarr fetches on unmount), #212 (staging app-host counterpart).
Reopening, narrowed. I closed this a few minutes ago as done-or-superseded, which was wrong on one count: the live product links here.ActionBar.astro:215 renders "Direct compute access is coming soon... tracked on #6" inside the Compute dialog on every dataset page, and that link is live on production right now. Closing it points real visitors at a closed issue.
Remaining scope is exactly one thing: the NSG compute connector. One-click submission from a dataset page (Worker proxy with NSG OAuth, launch flow, status polling). The Compute dialog today is a good manual fallback (NSG portal instructions, a nemar-cli recipe, direct file access), not the connector.
Everything else this issue asked for is done or was overtaken, and is out of scope here:
Activity
design: Compute button -> 'Compute instructions' with 3-route modal
Phase 2 polish: 6 design-review fixes (fonts, a11y, tree layout, keyw…
neuromechanist commentedon Aug 3, 2026
Closing: the phase's purpose was to design the deferred live features and plan the cutover. Both were overtaken by actually building and shipping them.
nemar.org,www,ww2, andapp.nemar.orgall serve this build, with ww2 deliberately kept alive so persistently-cached 301s land on a working page. The host model is documented in AGENTS.md rather than a one-off.context/cutover_plan.md.src/lib/eeg-viewer/: traces, montages, topo, DSP, filters, events, all unit-tested), served from zarr.nemar.org. A design doc for a feature already in production is not worth writing.7953145design-review a11y fixes,27ff388AA contrast,cdac663dialog a11y, and the cold-paint perf epic Epic: Cold-paint perf + cache hygiene for dataset detail #66, closed).The one deliverable with no successor is the NSG connector architecture doc; there is no compute-launch flow today and no open work depending on that document. If NSG compute gets picked up, it should start as its own epic against the current architecture, not from this issue.
Open follow-ups that came out of this area are tracked on their own: #208 (viewer zarr fetches on unmount), #212 (staging app-host counterpart).
neuromechanist commentedon Aug 3, 2026
Reopening, narrowed. I closed this a few minutes ago as done-or-superseded, which was wrong on one count: the live product links here.
ActionBar.astro:215renders "Direct compute access is coming soon... tracked on #6" inside the Compute dialog on every dataset page, and that link is live on production right now. Closing it points real visitors at a closed issue.Remaining scope is exactly one thing: the NSG compute connector. One-click submission from a dataset page (Worker proxy with NSG OAuth, launch flow, status polling). The Compute dialog today is a good manual fallback (NSG portal instructions, a nemar-cli recipe, direct file access), not the connector.
Everything else this issue asked for is done or was overtaken, and is out of scope here:
.context/cutover_plan.md.7953145,27ff388,cdac663, plus perf PRs perf: edge-cache via caches.default in middleware #14, Phase 1 (#65): bump edge-cache namespace to v2 #67, Epic #66: Cold-paint perf + cache hygiene for dataset detail #70, #74: Tree totals from summary.totals.bytes + 4s summary timeout on partials #75, Lazy chunked dir tree + drop catalog-row SSR fetch #79, Speed up Zarr viewer startup #113 and the cold-paint epic Epic: Cold-paint perf + cache hygiene for dataset detail #66). No formal Lighthouse run against the stated 95/90/95 thresholds was ever done; if that gate still matters it should be its own issue with a repeatable command, not a leftover acceptance criterion here.So: keep open until either the NSG connector is built, or the Compute dialog stops pointing here.