Hazel Work Dashboard
A copy of the Hazel Work Dashboard that Matthew runs on his laptop, published within a minute or two of each change to its notes, and every 15 minutes for the status from GitHub. Its localhost links work only there.
Updated 2026-10-16 00:15 MDT, by ~/localhost-index/update.py (PR titles and status from GitHub).
History: every change to this page, a folder for each year, month and day, also pushed to matthewhammer/hazel-work-notes (public). Published within a minute or two of each change to our notes, and every 15 minutes for the status from GitHub: matthewhammer.org/hazel/work-dashboard/. How we work with this page: agent-instructions.md. This is version 0.0.3: VERSIONS.md lists what each version has. Closed tickets: their own page. This page reloads itself when it changes (checked every 90 s while it is open).
Current ticket: ^ticket(127)
performance · from Matthew, 2026-10-15 🌙 long run 🪷 blocked on Matthew
Started Thu Oct 15, 11:44 PM MDT · elapsed
A report on the performance PRs to dev: an executive summary and a merge plan. Matthew, 2026-10-15 21:37: a report about the dashboard's set of performance PRs (https://matthewhammer.org/hazel-work-notes/#perf-prs: #2681, #2679, #2678, #2675, #2673, #2661, #2651, #2652, #2614, #2610, #2671, #2672, #2606, #2603, and the #2490 spike), giving an executive summary for Cyrus, or Alexander, or Andrew, or Matt K: for each PR, (1) why it is important (the savings in responsiveness, or memory, etc., with the measured numbers and where they come from), and (2) reasons not to merge it to dev ASAP, hopefully not too many, but honest and open, doing the work for them (conflicts with dev's head, tests and CI state, risks, what a reviewer should look at, what overlaps #2469). And a merge plan (Matthew, 21:38): an order, which can go together, what depends on what, and what each needs first. A draft in the performance series for his review; nothing posted to GitHub or published until he says. Tonight, after hours (🌙). 21:38, Matthew: include a dependency graph of the PRs as a picture (which PR builds on, or conflicts with, which), and repeat the relevant plots from earlier reports where they show a PR's savings (with their sources).
Now: Overnight summary (Matthew back about 06:30). LIVE: Dress Up 0.0.57 (42 flowing stream and brook, the swing swings its sitter; 128 shorts and swimsuits on posed legs; 129 the Pose tab first: standing, cross-legged meditating with closed eyes and hands in the lap, five dances; turning feet; 130 the hug: far arm behind; 131 the profile puddle). WAITING ON YOU (🪷, all local, nothing posted): ^report(16) draft, two drag fixes side by side (http://localhost:8131/series/016/; coalescing cuts a long drag's freeze 4.0 -> 2.2 s, 10 of 10 rounds; views in the worker do not help) [91]; ^report(17) draft, the performance PRs' executive summary and merge plan, fifteen with #2608, and all fifteen merged pass the full suite, 4,756 tests (http://localhost:8131/series/017/) [127]; the outline deep link ?outline=ux&ux=... (local 438812718b) [25]; the cell memo (page time -2/3, nothing visible moved: recommend not pushing) [24]; CI budgets as local branches off dev in ~/hazel-ab-coal: statics passes [13], a recursive-library step budget [14], byte budgets [15], and 19's finding (no call memo on dev; jsoo GC counters are zeros) [19]. Nothing is running now. Servers: 8000 dashboard, 8131 series, 8133 Dress Up, 8353 the outline link + cell memo build (#2635-based); 8311 #2597 (hazel-splicerefs-full) restarted at 23:45.
Work Queue (Active Tickets)
What we mean to do next, in order: a few tickets, not the backlog. We push a ticket onto the front or the back, and reorder as we go; a ticket leaves the queue when it closes. Each links to its full row in the backlog.
| Next | Ticket | Area | Ticket |
|---|---|---|---|
| 🪷 Blocked on Matthew: waiting for his response | |||
| 1 | ^ticket(18) 🪷 blocked on Matthew | performance | Option B from ^report(5): evaluate each livelit's view in the worker, keep the last Html sent per livelit, and post patches to it, so view… |
| 2 | ^ticket(102) 🪷 blocked on Matthew | hazel | Reply to Alexander on the deep links PR (#2583; Matthew writes the reply by hand). Alexander (7h3kk1d) replied today at 11:00 to our… |
| 🌙 Long runs: for when Matthew is away from the keyboard | |||
| 3 | ^ticket(91) 🌙 long run 🪷 blocked on Matthew | performance | Drag stalls: how many changes does a drag process, and does coalescing slider input help? From ^ticket(90): on #2635's head a 1.5 s drag of… |
| 4 | ^ticket(25) 🌙 long run 🪷 blocked on Matthew | testing, UX | Code-driven tests of the outline modes: a URL parameter (or test hook) that opens a slide in UX mode with chosen GUI rows, so a script… |
| 5 | ^ticket(24) 🌙 long run 🪷 blocked on Matthew | performance | Open cells draw their livelit views without the view memo (live_html, 449 ms per change profiled in ^report(8)'s B with ^ticket(6)'s fix),… |
| 6 | ^ticket(127) current 🌙 long run 🪷 blocked on Matthew | performance | A report on the performance PRs to dev: an executive summary and a merge plan. Matthew, 2026-10-15 21:37: a report about the dashboard's… |
| 7 | ^ticket(13) 🌙 long run 🪷 blocked on Matthew | performance, CI | Statics counts on dev: the analyses-per-expression counter and Test_StaticsPasses from #2635 (bd90642654, da3251d3e5), with a small corpus… |
| 8 | ^ticket(14) 🌙 long run 🪷 blocked on Matthew | performance, CI | A budget on a slide: the evaluator's work after a figure move in the woods, where reuse failed before #2675. 2026-10-15T21:29: Matthew:… |
| 9 | ^ticket(15) 🌙 long run 🪷 blocked on Matthew | performance, CI | Byte budgets: elaboration sizes on dev, and the worker's message sizes for a fixed interaction. 2026-10-15T21:29: Matthew: after hours… |
| 10 | ^ticket(19) 🌙 long run 🪷 blocked on Matthew | performance, CI | Budgets harder to fool (^report(6)'s update, a count is not a time): count the call memo's own work (the nodes its keys put in form) and,… |
Localhost ports
| PR # and title | Repo and branch | Localhost |
|---|---|---|
| #2635 Livelits × modular editors: merge #2469 into the livelits branch | hazelgrove/hazelintegration/livelits-modular-editors | localhost:8341 Scenes slide |
| #2597 User-Defined Livelits, as in PLDI '21, and beyond. | hazelgrove/hazelintegration/livelits-splicerefs-full | localhost:8311 |
| #2635 Livelits × modular editors: merge #2469 into the livelits branch | hazelgrove/hazelperf/2635-names (into integration/livelits-modular-editors) | not served |
| #2675 Incremental evaluation reuses expressions around a recursive let | hazelgrove/hazelrecursive-let-reuse (into dev) | not served |
| #2676 The cursor inspector can show a long type whole, on several lines | hazelgrove/hazelcursor-inspector-full-types (into dev) | not served |
| — Hazel performance analysis (the report's sources) | matthewhammer/hazel-performance-analysis, checked out at ~/hazel-perf-reportmain (public: a push publishes, so drafts wait) | localhost:8131/series/ |
| — Messed Up Dress Up (the game) | matthewhammer/kids-choice-games, folder dress-up/, public with its history from 2026-10-09Its own Pages site deploys every push to main. Edit ~/kids-choice-games: ~/hazel/experiments/polygons-plain/dress-up.html is a symlink to it.main | not served |
Other PRs we are tracking
Not about performance: features, fixes, docs, CI.
| PR # and title | Branch → base | State | CI | Review | Merge | Updated | Note |
|---|---|---|---|---|---|---|---|
| #2676 The cursor inspector can show a long type whole, on several lines | cursor-inspector-full-types → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-10-12 | The type-expansion panel (issue #2674). Its fix for module signatures' parentheses is only on #2635: dev prints a signature as a labeled tuple. |
| #2583 Deep links: the URL can say which slide, which panel, and where the cursor is | deep-links → dev | open | ✓ 5 passed | ⚠️ changes requested | conflicts | 2026-10-15 | |
| #2613 Colors: every var() a stylesheet reads is one the theme defines | configuration-undefined-colors → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-09-29 | |
| #2605 Substitution: a let's definition sees the outer binding, not its own | substitution-let-def-scope → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-09-27 | |
| #2608 Leave the source's whitespace and comments out of elaborations | elab-no-secondary → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-09-30 | |
| #2592 Report each error once, not once per wrapper | error-dedup-wrappers → dev | draft | ✓ 5 passed | no review yet | mergeable | 2026-09-23 | |
| #2491 docs: research-paper implementation status report | docs/research-status-report → dev | draft | ✓ 7 passed | no review yet | mergeable | 2026-09-22 | Docs. |
| #2577 Serve the release web build from an Internet Computer canister | ic-canister → dev | draft | ✓ 7 passed | no review yet | conflicts | 2026-09-22 | |
| #2616 non-`dev` (`integration/livelits-splicerefs-full`): Remote refs: splices a Fumola instance reads and writes | remote-splices → integration/livelits-splicerefs-full | open | ✗ 2 failing: checks / Fast checks (failure), checks / Full tests (failure); ✓ 1 passed | no review yet | conflicts | 2026-10-04 | Remote splices, the base of #2650. On the back burner since 2026-10-05. |
| #2650 non-`dev` (`remote-splices`): Hazel's state and Fumola instances on an Internet Computer canister | ic-canister-fumola → remote-splices | open | ✗ 2 failing: checks / Fast checks (failure), checks / Full tests (failure); ✓ 1 passed | no review yet | mergeable | 2026-10-05 | The IC canister demo. On the back burner since 2026-10-05. |
| #2536 non-`dev` (`integration/modules-livelits`): experimental language integrations | experimental-lang-integration → integration/modules-livelits | draft | ✓ 5 passed | no review yet | mergeable | 2026-09-22 | Fumola-in-Hazel, shelved since 2026-09-16 (its issues go to Adapton/fumola). |
Repositories we deploy
Pushed to directly, without PRs. The version is the newest in each project's VERSIONS.md. Deployed means Pages has built the branch's latest commit; a repository without a Pages site runs where its links point.
| Repository and branch | Latest commit | Version | Deployed | Open PRs | Note |
|---|---|---|---|---|---|
matthewhammer/kids-choice-gamesmain | c058d1d Messed Up Dress Up 0.0.57: poses, flowing water, and a hug (2026-10-16) | Messed Up Dress Up 0.0.57 (2026-10-15) | ✓ deployed c058d1dthe game, live localhost:8133/dress-up.html 🎵 the tunes (MP3s) | none | The kids' toys, prototyped in plain JS with real players before the Hazel slides formulate them as user-defined livelits (#2635, #2597; ^report(1) on). Each version is tagged dress-up-X.Y.Z. |
matthewhammer/dress-up-picturesmain | 048e549 2026/10/15: Messed Up Dress Up 0.0.55 (2026-10-16) | — | ✓ deployed 048e549every picture, by day localhost:8133/screenshots/ | none | The Dress Up picture archive, apart from the game (^ticket(106)): each day, every background and chip menu at its defaults, as PNG + SVG + settings. Unchanged files are symlinks to the day they last changed. ~/du-shots/tools/archive-today.sh makes a day. |
matthewhammer/dress-up-musicmain | bce8822 days.json, for the music player (ticket 119) (2026-10-16) | — | ✓ deployed bce8822every tune, by day localhost:8133/music/ | none | The Dress Up music archive, apart from the game (^ticket(106)): each day, every tune at its defaults, one loop each, as MP3. Unchanged tunes (by fingerprint) are symlinks to the day they last changed. |
matthewhammer/hazel-work-notesmain | 5e00bb5 Publish 2026-10-16 00:00 MDT: index.html, closed.html (2026-10-16) | Hazel Work Dashboard 0.0.3 (2026-10-13) | ✓ deployed 5e00bb5the dashboard, published the dashboard, localhost:8000 its history, localhost:8000/notes/ | none | This page's code (dashboard/, run in place: ~/localhost-index links to it), its history and agent-instructions.md. PUBLIC since 2026-10-13: its Pages site serves a copy of this page (noindex), redirected to from matthewhammer.org/hazel/work-dashboard/. Cron commits and pushes a snapshot, data.json and the public copy when the page changes; the code is committed by hand. |
matthewhammer/hazel-performance-analysismain | 3262f63 ^report(15): a dated update, published 2026-10-15: not like for like (shape counts) (2026-10-16) | — | no Pages site the series, published drafts, localhost:8131/series/ | none | The performance analysis series, for outside readers (^report(1) on), with its history since 2026-10-09 and its own agent-instructions.md. Public since 2026-10-12. Published by copying a report to the homepage repo; a push here is public too, so drafts wait for review. |
Performance Analysis and Improvement
Tracked in hazelgrove/hazel#2609 (performance challenges). Write-ups: the performance analysis series (local sources).
Performance PRs
Against #2609; the last three (#2606, #2603, #2490) are from before it.
| PR # and title | Branch → base | State | CI | Review | Merge | Updated |
|---|---|---|---|---|---|---|
| #2681 A value known to have a type is not walked again for it | ascriptions-has-type → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-10-13 |
| #2679 An ascription pushed through a value gives the value back when it already has the type | ascriptions-copy-on-change → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-10-12 |
| #2678 CI fails when an evaluation does more work than its budget | ci-perf-budgets → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-10-12 |
| #2675 Incremental evaluation reuses expressions around a recursive let | recursive-let-reuse → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-10-12 |
| #2673 Incremental evaluation recognizes an unchanged expression by its Merkle hash | reuse-check-merkle-hash → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-10-13 |
| #2661 Types get a Merkle hash; the evaluator asks is_consistent once per pair | typ-hash-consistent → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-10-10 |
| #2651 Statics: a module binding analyzes its definition once, not twice | statics-module-def-once → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-10-05 |
| #2652 Eval worker in Firefox: start it at load, and do not time out a reply that is queued | eval-worker-start-at-load → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-10-06 |
| #2614 Load text about 2.5× faster: whole tokens, one re-grout | parser-perf → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-09-29 |
| #2610 Parse the load fallback with the segmented parser, projectors kept | parse-fallback-segmented → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-09-30 |
| #2671 MapPiece keeps unchanged pieces as the same objects | mappiece-keep-identity → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-10-10 |
| #2672 Measuring the font no longer scrolls the editor to the caret | font-metrics-keep-scroll → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-10-10 |
| #2608 Leave the source's whitespace and comments out of elaborations | elab-no-secondary → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-09-30 |
| #2606 Keep the incremental cache in the eval worker | eval-worker-held-cache → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-09-30 |
| #2603 Statics: a let is analyzed once unless it refers to itself, plus three smaller speedups | dev-statics-perf → dev | open | ✓ 5 passed | no review yet | mergeable | 2026-09-26 |
| #2490 SPIKE: compile the Hazel evaluator with wasm_of_ocaml and benchmark it | wasm-eval-bench-spike → dev | draft | ✓ 7 passed | no review yet | conflicts | 2026-09-22 |
The integration branches also carry these fixes:
| PR # and title | Branch → base | State | CI | Review | Merge | Updated |
|---|---|---|---|---|---|---|
| #2635 Livelits × modular editors: merge #2469 into the livelits branch | integration/livelits-modular-editors → integration/livelits-splicerefs-full | draft | ✗ 2 failing: checks / Fast checks (failure), checks / Full tests (failure); ✓ 1 passed | no review yet | conflicts | 2026-10-15 |
| #2597 User-Defined Livelits, as in PLDI '21, and beyond. | integration/livelits-splicerefs-full → integration/livelits-splicerefs | open | ✓ 5 passed | no review yet | mergeable | 2026-10-15 |
Full Backlog (All Tickets)
All our open tickets: important and not done; the few we mean to do next are in the work queue. Numbered in our own sequence from #1; a number is never reused. A closed ticket moves to the closed tickets, with how it ended.
| Ticket | Area | Text | From |
|---|---|---|---|
^ticket(13) | performance, CI queued 🌙 long run 🪷 blocked on Matthew | Statics counts on dev: the analyses-per-expression counter and Test_StaticsPasses from #2635 (bd90642654, da3251d3e5), with a small corpus subset in the quick suite. 2026-10-15T21:29: Matthew: after hours tonight if time is left (🌙), as local branches against dev for his review. | ^report(6) |
^ticket(14) | performance, CI queued 🌙 long run 🪷 blocked on Matthew | A budget on a slide: the evaluator's work after a figure move in the woods, where reuse failed before #2675. 2026-10-15T21:29: Matthew: after hours tonight if time is left (🌙), as local branches against dev for his review. | ^report(6) |
^ticket(15) | performance, CI queued 🌙 long run 🪷 blocked on Matthew | Byte budgets: elaboration sizes on dev, and the worker's message sizes for a fixed interaction. 2026-10-15T21:29: Matthew: after hours tonight if time is left (🌙), as local branches against dev for his review. | ^report(6) |
^ticket(16) | CI | Ask the maintainers to require the test jobs for merging into dev: today no check is required, so a red budget is a signal, not a gate. | ^report(6) |
^ticket(17) | performance, CI | A timing report that informs and doesn't fail: base and head on one runner, repeated, phase by phase, perhaps from #2161's harness (David's call). | ^report(6) |
^ticket(18) | performance queued 🪷 blocked on Matthew | Option B from ^report(5): evaluate each livelit's view in the worker, keep the last Html sent per livelit, and post patches to it, so view evaluation leaves the main thread. Needs the view runner (ViewCmdRunner) in the worker, a patch format, an element on the page that owns each livelit's DOM and applies patches, and handlers dispatched by id. Keep the page's own path for an interaction's immediate redraw, or every drag waits a round trip. Its condition is now met (2026-10-12, see ^ticket(10)): evaluating the view is about 69% of the main thread in a woods figure move, and C would save about 2%. Weigh it against making the view itself cheaper first (^ticket(1), ^ticket(2)). PROTOTYPED 2026-10-14 (local branches, nothing pushed; work notes http://matthewhammer.org/hazel-work-notes/work-notes/2026/10/14/2026-10-14T2120-0600.html): MvuShape and ViewCmdRunner moved to the language library (61d031fd93, no change in behavior); the worker runs view commands to Html, reusing an unchanged view per projector (fast_equal), and whole programs fold their views in too (experiments on experiment/2635-views-in-worker-all, ede920b07e). Full Body: the main thread about 540 ms (UX mode) / 220 ms (whole program) less per change, the redraw no later; Kids' Choice: release about 45 ms later (144 → 187–196 ms). Waits for Matthew's design calls: which livelits' views move (all, or the costly ones), and the on-screen ids in the request. 2026-10-15T18:52: tagged 🪷 blocked on Matthew: its next step is his call (see above). | ^report(5) |
^ticket(19) | performance, CI queued 🌙 long run 🪷 blocked on Matthew | Budgets harder to fool (^report(6)'s update, a count is not a time): count the call memo's own work (the nodes its keys put in form) and, if js_of_ocaml can report it, memory allocated, beside the evaluator's steps. 2026-10-15T21:29: Matthew: after hours tonight if time is left (🌙), as local branches against dev for his review. | ^report(6) |
^ticket(24) | performance queued 🌙 long run 🪷 blocked on Matthew | Open cells draw their livelit views without the view memo (live_html, 449 ms per change profiled in ^report(8)'s B with ^ticket(6)'s fix), where the whole-program editor uses memo_view (163 ms in C). Give cells the memo. (Corrected 2026-10-12: B's 0.48 s of of_segment was the autosave printing text, not the cells' code views.) 2026-10-15T21:29: Matthew: tonight after hours (🌙), local only. | ^report(8) |
^ticket(25) | testing, UX queued 🌙 long run 🪷 blocked on Matthew | Code-driven tests of the outline modes: a URL parameter (or test hook) that opens a slide in UX mode with chosen GUI rows, so a script reaches ^report(8)'s B without clicking through the outline. Matthew: "let's consider that too with our UX feature". 2026-10-15T21:29: Matthew: tonight after hours (🌙), local only. | ^report(8) |
^ticket(43) | dress up | Messed Up Dress Up: postcards. Save the friend and their scene as one picture, with a caption to write on it, ready to send. | Matthew, 2026-10-13, in chat (ideas from ^ticket(40)'s doorways) |
^ticket(54) | performance, #2469 | Offer ^ticket(7)'s three ports to #2469, from the local branch, with ^report(13)'s numbers. ^report(13)'s What next, step 1: "Offer the three ports to #2469, from the local branch, with this report's numbers." The branch: port-2469-t7 in ~/hazel-port-2469, three commits on #2469's head ^report(13)'s numbers for them; #2469 is Andrew's PR, so a progress note, no @-mention, no action asked; Matthew approves the post before it goes. | Matthew, 2026-10-14 (^report(13), What next, step 1) |
^ticket(91) | performance queued 🌙 long run 🪷 blocked on Matthew | Drag stalls: how many changes does a drag process, and does coalescing slider input help? From ^ticket(90): on #2635's head a 1.5 s drag of Full Body's arm length stalls the page about 4 s (whole program), and each processed change costs about 0.9 s of main thread, so the stall is mostly how many changes a drag triggers. Steps: (1) count, per drag: the slider's input events, the eval requests the page posts to the worker and the replies, and the heavy frames (gaps over 50 ms); (2) find where a slider's input becomes an edit, and whether changes already coalesce (the newest value only) while one is in flight; (3) if not, try coalescing on a local branch: while a change is in flight, keep only the newest value and process it when the change lands; (4) measure with drag_stall.py (C and B), release to redraw, and Kids' Choice, against the head; (5) then decide ^ticket(18) with both in hand. Local only until Matthew says. FINDINGS 2026-10-15: a transient input (a drag's preview) sends a quiet repaint action, and the page recalculates for each one, walking the whole program in EvalResult.calculate's reuse check (Calc.set -> fast_equal_with_lexemes, ~40% of a drag's main thread, profiled). Coalescing built (local perf/2635-coalesce-drag-repaint, 3eb4489124; counters in a MEASURE commit): only 1.5-3 of a drag's ~13 inputs coalesce. Measured in alternating sessions, whole program, drag stall: head 4.1 s, prototype (^ticket(90)) 2.6, coalescing 2.8, head 3.2, coalescing 4.2: the drift between sessions is larger than any build's effect, so neither fix is shown. Next (Matthew to say): an interleaved A/B, both builds served at once, alternating drag by drag, 10+ each; and try skipping the whole-program walk for a preview. Report: a new report on drags, once the A/B is in (the next free number: ^report(15) is now ^ticket(120)'s draft). 2026-10-15T18:52: tagged 🪷 blocked on Matthew: its next step is his call (see above). 2026-10-15T21:29: Matthew: yes, the interleaved A/B, tonight after hours (🌙): head against each prototype (coalescing, views in the worker), both served at once, alternating drag by drag, 10+ each. | Matthew, 2026-10-15 |
^ticket(102) | hazel queued 🪷 blocked on Matthew | Reply to Alexander on the deep links PR (#2583; Matthew writes the reply by hand). Alexander (7h3kk1d) replied today at 11:00 to our comment of 10-14 linking the plan page; he couldn't comment on that page, so he left his notes on the PR. His two points on the plan's sketch: (1) name `target`'s constructors `Documentation` and `Scratchpad`, not `Slide` and `Share`, and use records instead of tuples where there are several fields (`Scratchpad({name, share})`); (2) the `request` record allows a `position` without a `target`: he'd like a type that rules it out. Fact for the reply: on the branch as pushed (09-22), `?caret=` without `?slide=` puts the caret in whatever opens (the reader's last slide or scratchpad), so such a link depends on the reader's saved state. That supports nesting the position inside the place, e.g. `{place: option({target, position: option(position)}), panel: option(panel_request)}`. A panel-only link stays meaningful. Then revise the plan, and later the branch, to match whatever is agreed. He liked the rest. Also worth an answer: where he should comment on plans next time (the PR itself, or the PR description). 2026-10-15T15:59: tagged 🪷 blocked on Matthew (his reply, written by hand), at the end of the queue. | Matthew, 2026-10-15 |
^ticket(127) | performance current 🌙 long run 🪷 blocked on Matthew | A report on the performance PRs to dev: an executive summary and a merge plan. Matthew, 2026-10-15 21:37: a report about the dashboard's set of performance PRs (https://matthewhammer.org/hazel-work-notes/#perf-prs: #2681, #2679, #2678, #2675, #2673, #2661, #2651, #2652, #2614, #2610, #2671, #2672, #2606, #2603, and the #2490 spike), giving an executive summary for Cyrus, or Alexander, or Andrew, or Matt K: for each PR, (1) why it is important (the savings in responsiveness, or memory, etc., with the measured numbers and where they come from), and (2) reasons not to merge it to dev ASAP, hopefully not too many, but honest and open, doing the work for them (conflicts with dev's head, tests and CI state, risks, what a reviewer should look at, what overlaps #2469). And a merge plan (Matthew, 21:38): an order, which can go together, what depends on what, and what each needs first. A draft in the performance series for his review; nothing posted to GitHub or published until he says. Tonight, after hours (🌙). 21:38, Matthew: include a dependency graph of the PRs as a picture (which PR builds on, or conflicts with, which), and repeat the relevant plots from earlier reports where they show a PR's savings (with their sources). | Matthew, 2026-10-15 |