diff --git a/.gitignore b/.gitignore index 4d4680f..69917bb 100644 --- a/.gitignore +++ b/.gitignore @@ -8,3 +8,6 @@ # stays tracked; `git restore internal/webdist/dist` after local testing) /internal/webdist/dist/* !/internal/webdist/dist/index.html + +# internal working docs (specs, handoffs, mockups) stay local-only +/docs/ diff --git a/docs/m3-frontend-handoff.md b/docs/m3-frontend-handoff.md deleted file mode 100644 index 842cd1f..0000000 --- a/docs/m3-frontend-handoff.md +++ /dev/null @@ -1,77 +0,0 @@ -# Handoff: milestone 3 frontend (slices 2-12) - -You are implementing the spreadlab dashboard frontend. The design is fully -decided and the backend is ready; this segment is Vue/TypeScript only. -Repo root: grant-proposal/tool/ (paths below are relative to it). - -## State as of 2026-06-10 - -- `docs/ui-spec.md` is the NORMATIVE spec: read it end to end before any - code. Page layout, component tree, every interaction with - trigger/behavior/states, visual tokens for both themes, URL scheme, - accessibility checklist, responsive rules, and the slice order - (section 10). Slice 1 is done; you start at slice 2. -- `docs/mockup/rendered.html` is the approved visual reference. View it in - a browser (or headless, see Method) in light, dark (`?dark`), and at - phone width. Match its look; do not port its code. -- `POST /api/scenario` is live (slice 1): `{config, strategy}` in, - `{config, result, edges}` out, edges as deterministic `[from, to]` - node-index pairs. 400 with `{"error": "..."}` on invalid input. - Generated types in `web/src/types/` already include - `ScenarioRequest`/`ScenarioResponse` (edges arrive as `number[][]`). -- The current `web/src/App.vue` is the milestone 2 parity page. Slice 3 - replaces it; `ComparisonTable.vue` evolves into the collapsed - ResultsTable (spec 4 and 8). Keep its test current, do not orphan it. -- The Go side needs no changes in this segment. If one becomes necessary, - raise it first; remember `go generate ./...` + commit the regenerated - types (CI has a drift guard). - -## Method - -- Vue/TS is Justin's home turf: drive fast and mostly autonomously. No - micro-briefs needed; keep commits per slice with clear messages, push - after each slice (CI must stay green: type-check, lint, vitest, build, - plus the Go jobs). -- THE DESIGN BAR IS THE HARD PART. Justin rejected the first mockup as "a - very weak mishmash of elements"; the approved one is the bar: one card - system, real type scale, consistent spacing, both themes, mobile - first-class. Never show boxes-on-a-page, not even as an intermediate - state. Slice 3 is the make-or-break visual slice. -- Self-check visually before claiming a visual slice done: screenshot the - running app headlessly and compare against the mockup, e.g. - `chromium --headless --disable-gpu --screenshot=/tmp/app.png - --window-size=1366,900 http://localhost:5173` (also dark theme and - `--window-size=390,1400`). Read the PNGs; if it does not look like the - mockup, it is not done. -- Dev stack runs in the shared tmux session "spreadlab" (window 1 runs - ./dev.sh: API :8080, Vite :5173). Restart it there with C-c + - `./dev.sh` via tmux send-keys; never spawn invisible background tasks. -- Work through the accessibility checklist (spec section 8) as you build - each piece; slice 12 is the audit, not the first time you think about it. -- One shared `formatPct` (Math.round to whole percent) used everywhere; - the mockup says 83% where Go's printf said 82% for 99/120, and that - class of mismatch must not reach the UI. -- Descriptive variable names; no em dashes anywhere. - -## Small open choices (decide with Justin via AskUserQuestion, batched) - -- d3-force dependency: `d3-force` + `d3-quadtree` only (recommended) vs - full d3. The seeded random source (mulberry32, spec section 6) is - hand-written either way. -- Inter font: `@fontsource-variable/inter` (recommended) vs static weights. -- Panel ids: `crypto.randomUUID()` (recommended, zero deps) vs nanoid. -- Reach chart: hand-rolled SVG paths (recommended at this scale) vs a - chart library. - -Everything else is decided in the spec; do not re-open decided items -(rejected alternatives are listed in spec section 11). - -## First steps - -1. Read `docs/ui-spec.md` fully, view the mockup in both themes and at - phone width. -2. Batch the four open choices above into one AskUserQuestion round. -3. Slice 2 (store + URL + debounced parallel runs) with Vitest coverage: - it is pure logic, test it well; it carries every later slice. -4. Slice 3, compared pixel-wise against the mockup before you call it done. -5. Continue down spec section 10; each slice leaves the app shippable. diff --git a/docs/m3-implementation-handoff.md b/docs/m3-implementation-handoff.md deleted file mode 100644 index 52a8eef..0000000 --- a/docs/m3-implementation-handoff.md +++ /dev/null @@ -1,81 +0,0 @@ -# Handoff: milestone 3 implementation (from the UI design session) - -You wrote `docs/ui-design-brief.md` and asked for a design session. It ran on -2026-06-10, every decision made with Justin via AskUserQuestion, and it is -done: nothing UI-shaped is left open. This doc tells you what exists now, -what changed relative to your milestone 3 assumptions, and where to start. - -## What exists now - -- `docs/ui-spec.md` is the normative spec. Page layout with wireframe, - component tree, every interaction with trigger/behavior/states, API - requirements, visual tokens for both themes, a testable accessibility - checklist, 12 ordered implementation slices, and one-line rejected - alternatives. Build from it without re-deciding; if you disagree with - something, raise it with Justin, do not silently deviate. -- `docs/mockup/index.html` is the approved visual reference (v2; Justin - rejected v1 outright). It is a design artifact with stand-in layout and - edges, not app code. To view: bake a live `/api/comparison` response into - it (replace `__DATA__`, see `data.json` + `rendered.html` already there), - open in a browser, `?dark` for the dark theme, narrow window for mobile. - Match its look; do not port its code. - -## Deltas against what you probably assumed for milestone 3 - -- The dashboard is not the fixed three-strategy comparison. Panels are a - dynamic list (cap 6) of scenarios: one shared base `Config` plus sparse - per-panel overrides and a strategy. The classic trio is just the default, - loaded from a study preset module (`web/src/presets/deepfake-school.ts`) - that owns all study copy, so no research question is hardcoded. -- New endpoint needed, and it is slice 1, your Go territory: - `POST /api/scenario` takes `{config, strategy}`, returns - `{config, result, edges}` where `edges` is the deterministic topology as - node-index pairs. This closes the topology gap flagged in the brief. The - engine already has the graph; expose it. Same validation and 400-string - behavior as comparison. tygo regen, determinism test on edges. - `/api/comparison` stays as-is (parity table and external consumers). -- The frontend fires one `/api/scenario` per panel in parallel, debounced - 400 ms, atomic swap, keep-last-good on errors. No batch endpoint - (rejected, one-liner in the spec). -- The optimisation feature (your milestone 5) got its API shape reserved: - `POST /api/optimize` goal-seek, request/response sketched in spec - section 2. M3 ships no optimizer UI, only a layout that can host it. -- In scope for M3 beyond what the brief listed: export (JSON, per-node CSV, - PNG snapshot), both light and dark themes with a toggle, and mobile as a - first-class target (Justin expects first contact by phone). Animated - GIF/WebM export is explicitly post-M3. -- The accessibility criteria your locked decisions called UNDEFINED are now - defined: spec section 8 is the checklist, and slice 12 is the audit. -- Pinia and Vue Router: decided not needed. Module-scope composable store, - modal focus, one page. - -## Calls Justin made explicitly (do not relitigate) - -- Focus view is a modal overlay (recommendation was expand-in-place). -- Sharing is address-bar-only, no Share button, no toast - (`history.replaceState` after every successful run). -- Both themes ship, not light-only. -- Dynamic panels, export, and the optimizer hook all came from him - unprompted; they are wanted, not gold-plating. - -## Method notes - -- Your working agreements stand: teaching-paced TDD on Go (slice 1 gets a - micro-brief, failing test first, small annotated diffs, learning-log - commit message); Vue/TS driven fast and mostly autonomously. -- Justin's design bar is high. His words on the first mockup: "a very weak - mishmash of elements"; he wants a clean, responsive SPA that "looks - professional and feels amazing". The approved mockup is the bar: app - shell, type scale, one card system, real SVG icons, consistent spacing. - Never show him boxes-on-a-page, not even as an intermediate state demo. -- Dev stack runs in the shared tmux session "spreadlab" (`./dev.sh`, API on - :8080, Vite on :5173). Use it, do not spawn background tasks. -- One shared percent formatter everywhere; the 82-vs-83 rounding mismatch - already bit the mockup once (spec section 6). - -## First steps - -Read `docs/ui-spec.md` end to end, view the mockup in both themes and at -phone width, then propose the slice 1 plan (the `/api/scenario` endpoint) -and start the usual loop. Spec section 10 is the full M3 slice order; each -slice leaves the app shippable. diff --git a/docs/m4-handoff.md b/docs/m4-handoff.md deleted file mode 100644 index 4179dfe..0000000 --- a/docs/m4-handoff.md +++ /dev/null @@ -1,103 +0,0 @@ -# Handoff: milestone 4 (single binary + deploy) - -You are shipping spreadlab to Justin's server. Milestone 3 is done; this -segment turns the repo into one deployable artifact: a Go binary with the -built frontend embedded, a public container image, and a Portainer stack -behind Caddy. Repo root: grant-proposal/tool/ (paths below relative to it). - -## State as of 2026-06-10 - -- All 12 M3 slices are merged and CI is green: the dashboard, playback, - chart, controls, panel management, focus modal, tooltips, mobile pass, - export, and the accessibility audit (results in commit d9b3812). -- `docs/ui-spec.md` was amended in place where the implementation - deliberately deviates (playback pacing and trickle in 5.4, seed range in - 5.3, viewBox and layout forces in 6, `readingCaption` in 1). The spec is - normative again; do not "fix" the app back toward old numbers. -- Deploy shape is DECIDED in `grant-proposal/tool-handoff.md` (read it): - multi-stage Dockerfile (node build, Go build with go:embed, minimal - final image), a compose snippet for a Portainer stack joining the - existing Caddy network, and a GitHub Actions job pushing a public image - to ghcr.io on merge to main. No Docker in the dev loop: native - `./dev.sh` (tmux session "spreadlab", API :8080 + Vite :5173) stays. -- The UI already carries the "illustrative, not validated" framing - everywhere, including PNG exports; deploy adds no copy. - -## Method - -- This is Go/infra territory: Justin is learning Go from the diffs, so - the M1/M2 working agreements apply again (unlike the fast autonomous - Vue pace of M3). Micro-brief each new concept before the code: - go:embed, http.FileServer over an embedded FS, multi-stage builds, - GHCR auth and the packages:write permission. Small annotated diffs, - one slice per commit, push after each, CI stays green. -- Verify each slice for real: run the binary and curl both / and /api; - build the image once locally and run it; after the GHCR job lands, - pull the public image and run it cold. -- `go generate ./...` + commit if generated types ever change (CI drift - guard); nothing in this segment should need it. - -## The one design decision to settle first - -`//go:embed` requires the embedded directory to exist at compile time, -but `web/dist/` is gitignored, so a naive embed breaks `go test ./...` -for anyone (and CI) without a frontend build. Decide with Justin via -AskUserQuestion, batched with the open choices below. Recommended: keep -the embed in its own small package (e.g. `internal/webdist`) with a -committed one-line placeholder `dist/index.html` ("run `npm run build`"); -the Dockerfile overwrites it with the real build. Alternatives: build -tags (dev binary without embed) or making CI build the frontend before -every Go job. - -## Suggested slices (each leaves main shippable) - -1. **Serve the SPA from Go**: embed web/dist, serve it on / alongside - /api (single page, no client routes; unknown paths can 404). Local - proof: `npm run build`, `go run ./cmd/spreadlab`, one origin serving - both. The vite dev proxy keeps working unchanged. -2. **Dockerfile**: multi-stage (node build → Go build → minimal final - image), plus .dockerignore. Build and run locally once to verify; - Docker still stays out of the daily dev loop. -3. **GHCR workflow**: build and push ghcr.io/justinzeus/spreadlab on - push to main (permissions: packages: write), then make the package - public. Verify with a cold pull. -4. **Portainer stack**: compose snippet joining the external Caddy - network plus the Caddyfile entry, documented in the README (deploy - section). Justin applies it in Portainer; verify the public URL, - both themes, a shared link with panels and focus, and a PNG export - from the hosted app. - -## Small open choices (decide with Justin via AskUserQuestion, batched) - -- Embed strategy for web/dist (see above; placeholder file recommended). -- Final image base: scratch vs distroless vs alpine (distroless static - recommended: CA certs and tzdata without a shell). -- Image tags: latest + commit SHA on main (recommended) vs semver tags. -- Listen port and a /healthz endpoint for compose healthchecks - (recommended: keep :8080, add the trivial healthz handler). -- The public hostname, and whether the Caddyfile entry is managed by - hand on the server or checked into the repo as documentation. - -## Traps learned in M3 (do not rediscover) - -- npm ci on CI rejects a lockfile that was updated incrementally after - installs (missing optional deps like @emnapi/*); regenerate - package-lock.json from scratch (`rm -rf node_modules package-lock.json - && npm install`) whenever dependencies change. -- Prettier must never touch `web/src/types/` (generated; CI drift guard). - `web/.prettierignore` already enforces this; keep it. -- Headless theme screenshots: `web/public/__light.html` and - `__dark.html` (gitignored one-liners that set localStorage and - redirect; recreate if missing). Screenshot light, dark, and 390 px - wide before calling visual work done. -- Compound shell commands need every pipe segment allowlisted, not just - the chromium part. -- Playback feel is perceptual: if any slice touches animation, have - Justin look at it running in his browser early, not after polish. - -## First steps - -1. Read `grant-proposal/tool-handoff.md` (deploy shape, working - agreements) and skim README + .github/workflows/ci.yml. -2. Batch the open choices above into one AskUserQuestion round. -3. Slice 1 with its micro-brief on go:embed; continue down the list. diff --git a/docs/mockup/data.json b/docs/mockup/data.json deleted file mode 100644 index cbed859..0000000 --- a/docs/mockup/data.json +++ /dev/null @@ -1 +0,0 @@ -{"config":{"numStudents":120,"edgesPerNode":3,"triangleProb":0.45,"forwardProb":0.38,"numEducated":36,"origin":0,"graphSeed":17,"thresholdSeed":2,"educationSeed":1},"results":[{"strategy":"none","educated":[],"reachedAtRound":[0,4,4,2,3,3,5,2,3,4,4,4,1,7,5,1,5,6,5,5,5,2,3,7,3,4,2,4,4,3,4,4,5,6,2,5,1,5,-1,2,6,5,3,4,-1,5,4,5,4,4,3,5,3,-1,6,5,3,-1,4,5,-1,5,-1,5,2,-1,1,4,3,4,3,5,2,4,4,3,3,-1,5,3,-1,5,-1,-1,5,3,3,-1,6,3,5,5,-1,-1,6,2,5,5,8,5,-1,5,6,6,4,7,-1,4,4,-1,-1,5,3,-1,1,-1,3,3,-1,5],"numReached":99,"numRounds":9,"reachedPct":82.5},{"strategy":"random","educated":[2,4,6,7,8,10,21,23,24,27,28,33,37,39,40,43,48,50,53,56,61,62,63,68,74,79,80,85,87,89,90,97,101,102,111,117],"reachedAtRound":[0,4,4,2,3,3,5,2,3,-1,7,4,1,7,5,1,5,6,-1,6,5,2,-1,7,3,7,2,5,5,-1,-1,6,6,6,2,5,1,5,-1,2,6,-1,3,4,-1,-1,4,-1,4,-1,-1,5,-1,-1,6,5,3,-1,-1,-1,-1,-1,-1,-1,2,-1,1,-1,5,4,-1,5,2,-1,-1,-1,3,-1,-1,3,-1,5,-1,-1,6,3,-1,-1,6,3,7,-1,-1,-1,7,2,-1,7,8,5,-1,5,6,-1,-1,7,-1,4,4,-1,-1,5,-1,-1,1,-1,3,-1,-1,-1],"numReached":70,"numRounds":9,"reachedPct":58.333333333333336},{"strategy":"most-connected","educated":[1,2,3,4,5,6,7,8,9,10,11,12,13,15,16,19,20,21,22,23,24,25,26,27,28,30,31,32,33,35,37,38,39,45,47,77],"reachedAtRound":[0,-1,-1,-1,-1,-1,-1,2,-1,-1,-1,-1,1,-1,-1,1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,1,-1,-1,-1,-1,-1],"numReached":7,"numRounds":3,"reachedPct":5.833333333333333}]} \ No newline at end of file diff --git a/docs/mockup/index.html b/docs/mockup/index.html deleted file mode 100644 index 9b21301..0000000 --- a/docs/mockup/index.html +++ /dev/null @@ -1,445 +0,0 @@ - - - - - -spreadlab mockup v2 (design artifact, not app code) - - - - -
- - - spreadlab - - Illustrative model, not validated - - - -
- -
-
-

How a non‑consensual deepfake spreads through a school

-

One simulated year group, 120 students. Educate the same 30% of them, but change who: - the fake reaches 82% of the school with no program, - 58% educating at random, and - 6% educating the best‑connected students.

- Illustrative model, not validated -
- -
- -
-

Scenarios

- - - -
- -
- -
- Forwarded the fake - Educated to refuse - Not reached - Origin -
- -
- - - - -
- Round 9 of 9 - 1× -
- -
-
-

Reach over time

- -
-
-

World

-
- -
- 38% -
-
- -
- 36 · 30% -
-
- Advanced -
- Students120 - Friends per student3 - Clique tendency0.45 - First poster#0 - Friendship network17 - Who resists2 -
-
-
-
- - -
- - - - diff --git a/docs/mockup/rendered.html b/docs/mockup/rendered.html deleted file mode 100644 index 30a1420..0000000 --- a/docs/mockup/rendered.html +++ /dev/null @@ -1,445 +0,0 @@ - - - - - -spreadlab mockup v2 (design artifact, not app code) - - - - -
- - - spreadlab - - Illustrative model, not validated - - - -
- -
-
-

How a non‑consensual deepfake spreads through a school

-

One simulated year group, 120 students. Educate the same 30% of them, but change who: - the fake reaches 82% of the school with no program, - 58% educating at random, and - 6% educating the best‑connected students.

- Illustrative model, not validated -
- -
- -
-

Scenarios

- - - -
- -
- -
- Forwarded the fake - Educated to refuse - Not reached - Origin -
- -
- - - - -
- Round 9 of 9 - 1× -
- -
-
-

Reach over time

- -
-
-

World

-
- -
- 38% -
-
- -
- 36 · 30% -
-
- Advanced -
- Students120 - Friends per student3 - Clique tendency0.45 - First poster#0 - Friendship network17 - Who resists2 -
-
-
-
- - -
- - - - diff --git a/docs/ui-design-brief.md b/docs/ui-design-brief.md deleted file mode 100644 index dd0ad3e..0000000 --- a/docs/ui-design-brief.md +++ /dev/null @@ -1,96 +0,0 @@ -# UI/UX design session brief - -You are running a design session for spreadlab's milestone 3 dashboard. The -deliverable is a written spec, `docs/ui-spec.md`, that an implementation -session can execute without re-deciding anything. Do NOT write application -code in this session. - -## How to run this session - -- Walk through the UX with the user top-down: audience and core message - first, then page layout, then each interaction, then states and edge - cases. One topic at a time. -- Use AskUserQuestion aggressively for every meaningful choice, with a - recommended option first. Use ASCII mockups in option previews when - comparing layouts. -- The user is a developer (Vue/TS background), not a designer; explain - design trade-offs the way you would explain code trade-offs. -- No em dashes in anything you write. - -## What spreadlab is - -A self-hosted dashboard running an agent-based model live: a non-consensual -deepfake spreads through a simulated school year group (120 students), and -an education program changes who forwards it. The headline result: the same -education budget reaches 82% / 58% / 6% of the school depending on who gets -educated (no program / random 30% / best-connected 30%). The product goal -is making that lever visceral. Always framed "illustrative, not validated". - -Audience, in order: grant-pitch audience and course graders (first contact, -must land in seconds), curious public on the hosted demo, the author. - -## What already exists (look at it) - -Paths below are given from the course workspace root -(/home/jv/src/uni/ai-in-open-society); the repo itself is -grant-proposal/tool/. - -- Run ./dev.sh from grant-proposal/tool/ and open http://localhost:5173 to - see the current parity page (a static table of the three scenarios). -- The Python prototype's rendered output defines the inherited visual - language: grant-proposal/pitch/figure/out/spread.png and spread.gif. - Read the PNG; it is the look to inherit. Teal donut ring = educated, - rose = forwarded the fake, haloed node = origin, slate = unreached; - colourblind-safe via shape, not colour alone. Modern-dashboard look, - Tailwind-ish palette (see constants in - grant-proposal/pitch/figure/spread_figure.py). -- API: GET /api/config/default, POST /api/comparison (full Config in, all - three strategies' Results out). Result.reachedAtRound gives per-node - activation rounds, exactly what an animation needs. Types in - web/src/types/ are generated from Go structs. -- KNOWN GAP: the API does not yet expose graph topology (edges), which any - network view needs. The spec should state what the frontend needs from a - new/extended endpoint; the implementation session will add it. - -## Locked decisions the spec must respect (from grant-proposal/tool-handoff.md) - -- Layout is computed in the frontend with a force simulation (d3-force or - equivalent) seeded from the config: a shared URL must render the same - picture. -- Config is the whole truth and goes in the URL: shareable, reproducible. -- Vue 3 + TypeScript, lean and conventional; Pinia/Router only when the - design actually requires them (note it in the spec if so). -- No shape redefinition in the frontend: all data types come generated - from Go. -- Quality bar includes accessibility, and the concrete criteria are - UNDEFINED so far. Defining them is part of this design session (keyboard - operation, reduced-motion behaviour for the animation, contrast, - screen-reader treatment of results at minimum). - -## Questions the spec must answer - -1. Page composition: prototype-style three panels side by side, or one - network view with a strategy switcher, or something better? What is - above the fold at first contact? -2. Which config levers get visible controls first, which live in an - "advanced" area (seeds?), and what control types (sliders, steppers)? -3. Animation: autoplay or user-initiated? Play/pause/scrub/speed? What does - round-by-round progression look like; what happens at reduced motion? -4. Metrics: what numbers/curves accompany the picture, and how do they - update while animating? -5. Sharing UX: how does config-in-URL surface to the user (copy button, - live-updating address bar)? -6. States: loading, API unreachable, invalid config (the API returns 400 - with an error string), tiny screens. -7. How is "illustrative, not validated" communicated without undermining - the demo? -8. Accessibility acceptance criteria (the concrete, testable list). - -## Output format for docs/ui-spec.md - -Page layout (with a final ASCII wireframe), component tree with -responsibilities, every interaction specced (trigger, behaviour, states), -data requirements from the API (including the topology gap), visual tokens -(palette, shape encodings), accessibility criteria as a checklist, and an -ordered implementation slice list for milestone 3. Decisions only, no -options left open; record rejected alternatives in one line each at the end. diff --git a/docs/ui-spec.md b/docs/ui-spec.md deleted file mode 100644 index 6fef963..0000000 --- a/docs/ui-spec.md +++ /dev/null @@ -1,601 +0,0 @@ -# spreadlab UI specification (milestone 3) - -Status: decided. This spec is the output of the design session run from -`docs/ui-design-brief.md`. An implementation session should be able to build -the dashboard from this document without re-deciding anything. Rejected -alternatives are listed at the end, one line each. - -The normative visual reference is the approved mockup in `docs/mockup/` -(`index.html` is the source; build `rendered.html` by replacing `__DATA__` -with a live `/api/comparison` response). The mockup is a design artifact: -its network layout and edges are stand-ins, its components are static. Match -its look (spacing, type scale, card system, both themes), not its code. - ---- - -## 1. Product shape - -A single-page Vue 3 + TypeScript dashboard. One page, no routes. The page -shows a dynamic list of simulation scenarios ("panels") that share a base -config; each panel may override any config field. The default experience is -the grant story: three panels (no program, educate 30% at random, educate -the best-connected 30%) over the default config, animated once on load. - -Audience order: grant-pitch audience and graders (message must land in -seconds, without interaction), curious public on the hosted demo (often on -a phone), the author. - -### Study presets, no hardcoded research question - -All study-specific copy and the initial scenario set live in a preset -module, not in components: `web/src/presets/deepfake-school.ts`, typed as -`StudyPreset`: - -```ts -interface StudyPreset { - headline: string // hero h1 - narrative: string // hero paragraph, may contain {0} {1} {2}... - // placeholders, replaced by each panel's - // reached% (same formatter as the panels) - disclaimerShort: string // badge text: "Illustrative model, not validated" - disclaimerLong: string // About popover body (2-4 sentences) - readingCaption: string // one quiet line under the legend explaining - // how to read the dots (added 2026-06-10) - toneThresholdPct: number // reached% <= threshold renders "good" (teal), - // above renders "bad" (rose); 30 for this study - base: Config // from generated types, the default world - panels: PanelSpec[] // initial panels (labels + overrides) -} -``` - -The app imports exactly one preset. A different research question is a new -preset file plus a changed import, zero component changes. `StudyPreset` -and `PanelSpec` are frontend-only types (they describe presentation, not -simulation data, so they do not violate the no-shape-redefinition rule; -`Config`, `Result`, `Strategy` stay imported from the generated types). - -### Panel model - -```ts -interface PanelSpec { - id: string // stable client id (nanoid-style) - label: string // user editable - strategy: Strategy // from generated types - overrides: Partial // sparse diff against base -} -``` - -Effective config for a panel = `{ ...base, ...overrides }`. Panels that do -not override `numStudents`, `edgesPerNode`, `triangleProb`, or `graphSeed` -share the base graph and therefore the identical layout; differences then -read as the experiment, not as noise. Maximum 6 panels; the Add button -disables at the cap with a tooltip "Maximum 6 scenarios". - ---- - -## 2. Data requirements from the API - -The API currently exposes `GET /api/config/default` and -`POST /api/comparison`. Neither returns graph topology, and comparison is -locked to the three fixed strategies of one config. Milestone 3 needs one -new endpoint; the implementation session adds it Go-side and regenerates -types with tygo. - -### New: `POST /api/scenario` - -Request body (Go struct, tygo-generated for the frontend): - -```jsonc -{ "config": { /* Config */ }, "strategy": "random" } -``` - -Response: - -```jsonc -{ - "config": { /* echoed effective Config */ }, - "result": { /* Result, as today */ }, - "edges": [[0, 5], [0, 12], ...] // undirected, node-index pairs, - // deterministic from the config's - // graph fields -} -``` - -- Validation and 400-with-error-string behavior identical to comparison. -- `edges` is small (roughly `numStudents * edgesPerNode` pairs) and is - always included; the frontend deduplicates layout work by hashing the - graph-shaping fields (`numStudents`, `edgesPerNode`, `triangleProb`, - `graphSeed`), so panels sharing a graph compute one layout. -- The frontend issues one request per panel, in parallel - (`Promise.all`), debounced 400 ms after the last input change, and swaps - all results in atomically when every request has settled. -- `GET /api/config/default` keeps seeding the preset's `base` fallback. -- `POST /api/comparison` remains for the parity table and API consumers. - -### Reserved: `POST /api/optimize` (post-M3, design hook only) - -Goal-seek endpoint: body `{ base: Config, strategy, goal: { maxReachedPct, -optimize: "numEducated" } }`, response: best config plus the candidate -sweep (config + reachedPct per candidate) so the UI can show its work as a -curve. Milestone 3 ships no UI for this; the controls card layout simply -leaves room for one extra button row, nothing more. - ---- - -## 3. Page layout - -Desktop wireframe (mobile differences in section 9): - -``` -┌──────────────────────────────────────────────────────────────────────┐ -│ ◆ spreadlab (ⓘ Illustrative model, not validated) ☾ ⌥src │ app bar, sticky -├──────────────────────────────────────────────────────────────────────┤ -│ How a non-consensual deepfake spreads through a school │ hero (from preset) -│ One simulated year group, 120 students. Educate the same 30%, │ -│ but change who: 83% ... 58% ... 6% ... │ live numbers -│ │ -│ SCENARIOS [↓ Export] [+ Add] │ toolbar -│ ┌──────────────────┬──────────────────┬──────────────────┐ │ -│ │ ▪ No program ⋮ │ ▪ Educate rand ⋮ │ ▪ Educate conn ⋮ │ │ panel cards -│ │ 83% │ 58% │ 6% │ │ -│ │ 99/120 · 9r · 0e │ 70/120 · 9r · 36e│ 7/120 · 3r · 36e │ │ -│ │ [override chips] │ [override chips] │ [override chips] │ │ -│ │ (network SVG) │ (network SVG) │ (network SVG) │ │ -│ └──────────────────┴──────────────────┴──────────────────┘ │ -│ ● Forwarded ◎ Educated · Not reached ◉ Origin │ legend -│ │ -│ ( ↻ ⏮ ▶ ⏭ ──●──·──·──·── Round 9 of 9 1× ) │ player pill, centered -│ │ -│ ┌─ REACH OVER TIME ────────────────┐ ┌─ WORLD ───────────────────┐ │ -│ │ (overlaid curves, playhead) │ │ Chance to forward ──● │ │ -│ │ │ │ Education budget ──● │ │ -│ │ │ │ ▾ Advanced (6 fields) │ │ -│ └──────────────────────────────────┘ └───────────────────────────┘ │ -│ ▸ Data table (collapsed) │ -│ Illustrative (not validated output) · toy model spreadlab · │ footer -└──────────────────────────────────────────────────────────────────────┘ -``` - -Above the fold at 1366x768: app bar, hero, toolbar, the full panel row, -legend, player. Chart and controls may start below the fold; the story -must not. - ---- - -## 4. Component tree - -``` -App -├── AppBar -│ ├── Wordmark (logo glyph + "spreadlab") -│ ├── DisclaimerBadge (opens AboutPopover; hidden on mobile) -│ ├── ThemeToggle -│ └── SourceLink (repo icon link) -├── HeroHeadline (preset headline + narrative with live %) -│ └── DisclaimerNote (mobile only: same popover trigger) -├── MobileScoreStrip (mobile only, sticky) -├── ScenarioToolbar ("Scenarios", ExportMenu, AddScenarioButton) -├── ErrorBanner (failed refresh / API unreachable) -├── PanelGrid -│ └── ScenarioPanel (xN) -│ ├── PanelHeader (accent swatch, label, kebab PanelMenu) -│ ├── PanelStats (big %, meta row, override chips) -│ └── NetworkView (SVG: edges, nodes; emits hover/tap) -├── LegendRow -├── PlayerBar (replay, step back, play/pause, step fwd, -│ RoundScrubber, round counter, speed pill) -├── ReachChart (overlaid cumulative curves + playhead) -├── ControlsCard -│ ├── LeverSlider (x2) (forwardProb, numEducated) -│ └── AdvancedFields (collapsible: students, edges/node, -│ clique tendency, origin, 3 seeds w/ dice) -├── ResultsTable (collapsible; accessible data table, -│ evolved from existing ComparisonTable) -├── FocusModal (enlarged ScenarioPanel + PlayerBar dock) -├── PanelEditorPopover (per-panel config diff editor) -├── NodeTooltip (single instance, positioned per hover) -└── FooterDisclaimer - -composables (module-scope singletons, no Pinia): -├── useSimStore state: base, panels[], resultsByPanelId, edgesByGraphHash, -│ round, playing, speed, focusPanelId, hoveredNode, -│ runState (idle|running|error), errorMessage -│ actions: setBaseField, add/update/remove/duplicatePanel, -│ run (debounced), syncToUrl/fromUrl -├── useLayout seeded d3-force layout per graph hash (cached) -└── usePlayback rAF round timer honoring speed + reduced motion -``` - -State lives in `useSimStore` (a plain `reactive` module singleton). No -Pinia, no Vue Router: one page, one store, modal focus is store state. - ---- - -## 5. Interactions, one by one - -Every interaction lists trigger, behavior, and states. - -### 5.1 Initial load -- Trigger: page open. -- Behavior: parse URL state (section 7); if absent, use the preset. Fetch - `/api/scenario` per panel. While loading, panels render as skeleton - cards (surface block, shimmer, fixed height) so layout does not jump. - When all results land: compute layouts, render round 0, then autoplay - (5.4). Hero percentages fill in with the results. -- States: success (autoplay), API unreachable (centered message in the - panel area: "Could not reach the spreadlab API at /api", retry - button, nothing else dimmed since nothing rendered yet), invalid URL - state (fall back to preset, show ErrorBanner "Link contained an invalid - configuration, showing defaults"). - -### 5.2 Changing a base lever (sliders, advanced fields) -- Trigger: input on any ControlsCard field. -- Behavior: update store immediately (slider thumb moves), debounce 400 ms - after the last change, then re-run all panels in parallel. While - running: panels dim to 60% opacity with a small spinner in the toolbar; - interaction stays enabled. On success: results swap in atomically, - animation resets to the final round (no autoplay re-trigger), URL - updates via `history.replaceState`. Hero numbers update. -- States: success; network failure (keep last good results undimmed, - ErrorBanner "Could not update: " with a Retry button); 400 - invalid config (keep last good results, attach the API error string - inline under the controls card header in rose, mark the offending field - if identifiable by name match, no banner). - -### 5.3 Seed rerolls -- Trigger: dice button next to a seed field (graph seed "Friendship - network", threshold seed "Who resists", education seed "Random picks"). -- Behavior: set that seed to a fresh random value in 0-9999 (amended - 2026-06-10: a full uint32 made fields and URLs unreadable), then as 5.2. - Rerolling - the graph seed invalidates the layout cache for affected panels; the new - layout fades in (200 ms opacity crossfade, skipped under reduced - motion). A text button "Reroll world" rerolls all three seeds at once. -- States: as 5.2. - -### 5.4 Playback -- Trigger: autoplay once after first successful load; afterwards only via - player controls. -- Behavior (amended 2026-06-10 after live review): rounds advance every - 1000 ms at 1x (speed pill cycles 0.5x, 1x, 2x). Within a round, each - node reached that round gets its own deterministic appearance moment - (full-avalanche hash of the node index) spread across the first 95% of - the interval; its unreached dot stays put until that exact frame, then - the rose dot fades and scales in (0.5 to 1.0) over 30% of the interval. - A round therefore reads as a continuous trickle, never a cohort blink. - The panel percentage, reached count, mobile strip, and focus modal - count via the same predicate, so the numbers tick up live exactly as - dots appear. Educated rings are visible from round 0 (they exist before - the spread). Panels whose run ended before the global max round hold - their final state. At the end, playback stops on the final round (no - loop), play button becomes replay affordance. The hero keeps the final - outcomes (its sentence claims outcomes, not playback state). -- Player controls: replay (jump to round 0 and play), step back, play / - pause toggle, step forward, scrubber (one tick per round, drag or click - to seek; seeking pauses), round counter "Round 4 of 9" (global max), - speed pill. -- Keyboard (when focus is inside the player, plus global except in text - inputs): Space toggles play/pause, ArrowLeft/ArrowRight step one round, - Home/End jump to first/final round. The scrubber is a native - `input[type=range]` so arrow keys work on it natively; - `aria-valuetext="Round 4 of 9"`. -- Reduced motion (`prefers-reduced-motion: reduce`): no autoplay, page - loads at the final round; play still works but rounds swap discretely - (no tween, no scale); the 1000 ms cadence becomes 1200 ms at 1x. -- States: round state is global and drives every panel, the chart - playhead, and the modal simultaneously. Round is not stored in the URL. - -### 5.5 Node hover / tap -- Trigger: pointerenter on a node (desktop), tap (touch). SVG circles get - a transparent 12 px-radius hit halo so 4 px dots are tappable. -- Behavior: NodeTooltip appears near the pointer: "Student 17", state in - this panel ("Forwarded in round 3" / "Educated, refused" / "Not - reached"; "Origin, posted it" for the origin), and "5 friends" (degree, - computed from edges). The same student gets a 2 px ink outline ring in - every panel simultaneously (the cross-panel echo: the same kid, reached - in one world, safe in another). -- Dismiss: pointerleave, tap elsewhere, Esc. No click-to-pin in M3. -- States: tooltips are pointer-only; the equivalent information is - available non-visually through the ResultsTable (per-panel aggregate) - and is acceptable per the accessibility decision (8). - -### 5.6 Add scenario -- Trigger: "+ Add scenario" button in the toolbar. -- Behavior: append a panel cloned from base (`strategy: "none"`, empty - overrides, label "Scenario N", next free accent color), run it, and - open its PanelEditorPopover anchored to the new panel. At 6 panels the - button disables with tooltip "Maximum 6 scenarios". -- States: while its first run is in flight the new panel shows the - skeleton card. - -### 5.7 Panel menu and editor -- Trigger: kebab menu on a panel header: Edit, Rename, Duplicate, Remove. -- Behavior: - - Edit opens PanelEditorPopover: strategy select (none / random / - most-connected) plus every Config field as a row showing the base - value; typing a different value creates an override (row highlights, - a reset icon appears per overridden field). Overrides render as chips - on the panel card ("numEducated · 60"). Apply follows the same - debounced-run path as 5.2 but re-runs only that panel. - - Rename: inline edit of the label in the popover header. - - Duplicate: copy panel with " copy" suffix, next accent. - - Remove: deletes immediately (no confirm; the action is cheap to - redo and the URL history keeps the old state one paste away). The - last remaining panel cannot be removed (menu item disabled). -- States: editor validation mirrors the API (400 text shown inline in the - popover); a panel whose run failed shows its last good result plus a - small rose dot on the kebab. - -### 5.8 Focus modal -- Trigger: click anywhere on a panel's network area (cursor: zoom-in), or - "Focus" in the panel menu. -- Behavior: modal overlay (scrim `rgba(15,23,42,.5)`), panel rendered - large (max 90 vw / 85 vh), same SVG and round state, with the PlayerBar - docked at the modal footer (same store state, both bars stay in sync; - the background one is inert under the scrim). Close via X button, Esc, - or scrim click. `focusPanelId` is part of URL state, so a shared link - can open pre-focused. -- States: focus trap inside the modal; on close, focus returns to the - panel that opened it. `aria-modal="true"`, labelled by the panel label. - -### 5.9 Export -- Trigger: "Export" menu in the toolbar: "Data (JSON)", "Per-node CSV", - "Picture (PNG)". -- Behavior: - - JSON: downloads `spreadlab-export.json`: `{ base, panels, results }` - exactly as held in the store (configs echoed from the API). - - CSV: `spreadlab-nodes.csv`, columns `panel,label,node,educated, - reachedAtRound` (one row per panel x node; -1 kept as -1). - - PNG: renders the current view at the current round to a canvas - (serialize each panel SVG, draw side by side with panel labels and - percentages, current theme colors, legend row, disclaimer footer - line) and downloads `spreadlab.png` at 2x resolution. -- States: menu items disable while a run is in flight (export captures - settled state only). Animated WebM/GIF export is post-M3 (rejected - list). - -### 5.10 Sharing -- Trigger: none. The address bar is the share UI. -- Behavior: after every successful run and on panel/focus changes, the - full state serializes into the URL query via `history.replaceState` - (never `pushState`, no history spam). Copying the URL at any time - reproduces the exact view: same seeds, same layout (seeded force), same - panels, same focus. No Share button, no toast. - -### 5.11 Theme -- Trigger: ThemeToggle in the app bar. -- Behavior: default follows `prefers-color-scheme`; toggle overrides and - persists to `localStorage` (`spreadlab-theme`). Theme is not URL state. - Both themes use the token table in section 6; node semantics are - identical in both. - -### 5.12 About / disclaimer -- Trigger: DisclaimerBadge (app bar, desktop) or DisclaimerNote line under - the hero (mobile). -- Behavior: small popover, `disclaimerLong` from the preset: what the - model is (a seeded agent-based toy simulation with parameters chosen - for illustration), what it is not (a validated prediction of real - schools), and a link to the proposal/repo. The italic footer line - ("Illustrative (not validated output)") is always present and is - included in PNG exports. The headline itself carries no caveat. - ---- - -## 6. Visual tokens - -Inherited from the prototype figure (`grant-proposal/pitch/figure/ -spread_figure.py`) and extended for dark mode. Implement as CSS custom -properties on `:root` / `.dark`. - -| Token | Light | Dark | Use | -|---|---|---|---| -| `--bg` | `#F8FAFC` | `#0B1120` | page background | -| `--surface` | `#FFFFFF` | `#111A2E` | cards, app bar | -| `--border` | `#E2E8F0` | `#1E293B` | card borders | -| `--border-soft` | `#EEF2F7` | `#1A2438` | inner dividers, chart grid | -| `--edge` | `#E6E9EF` | `#243349` | network edges | -| `--unreached` | `#D8DFE9` | `#3B4A61` | unreached node fill | -| `--spread` | `#F43F5E` | `#FB7185` | forwarded nodes, bad % | -| `--edu` | `#14B8A6` | `#2DD4BF` | educated rings, good %, slider accents | -| `--ink` | `#0F172A` | `#F1F5F9` | headings, primary buttons | -| `--ink-2` | `#334155` | `#CBD5E1` | body, labels | -| `--ink-3` | `#64748B` | `#94A3B8` | secondary text (min for informative text) | -| `--ink-4` | `#94A3B8` | `#64748B` | decorative/duplicated text only (fails AA for body sizes; never sole carrier of information) | - -Panel accent palette (categorical, deliberately distinct from the rose and -teal node semantics; used for the header swatch, chart line, and mobile -chip): `#6366F1`, `#F59E0B`, `#0EA5E9`, `#8B5CF6`, `#EC4899`, `#84CC16`, -assigned by panel position, recycled on remove. - -Node encodings (shape carries meaning, color reinforces it; colorblind-safe -by shape alone): -- Forwarded: filled disc, `--spread`, r 4. -- Educated, not reached: donut, `--surface` fill, 2.2 px `--edu` stroke. -- Educated but reached (origin educated edge case): rose fill inside the - teal donut. -- Not reached: small disc, `--unreached`, r 3. -- Origin: rose disc with a 1.4 px rose halo ring at r 7.5. - -Type: `Inter, system-ui, sans-serif` (self-host Inter via fontsource; -system-ui fallback). Hero h1 30/700/-0.022em (22 on mobile). Panel -percentage 40/700 tabular. Section headings 13/600 uppercase tracked. -Body 15, secondary 13. All numerals `font-variant-numeric: tabular-nums`. - -Surfaces: radius 16 (cards) / 10 (buttons, fields) / 999 (pills, player). -Shadows: resting `0 1px 2px rgba(15,23,42,.05), 0 4px 16px -8px -rgba(15,23,42,.08)`; lifted (hover, modal, mobile player) deeper per -mockup. Spacing on an 8 px-ish scale as in the mockup; page max-width -1240 px, padding 28 px. - -Number formatting: one shared `formatPct` (round to whole percent) used by -hero, panels, strip, chart labels, and table. No other rounding anywhere. - -Network rendering: inline SVG (not canvas), viewBox 380x230 (amended -2026-06-10: taller than the mockup's 190 for breathing room). Layout via -d3-force (`forceLink`, `forceManyBody`, collision, plus anisotropic -`forceX`/`forceY` pulls so the cloud settles into the wide card shape -organically and the fit stays uniform-scale; axis stretching reads as -line patterns), run synchronously for a fixed 300 ticks with -`simulation.randomSource(mulberry32(graphSeed))` and seeded initial -positions, so identical configs give identical pictures on every machine -(the shared-URL guarantee). Layout results cached per graph hash. - ---- - -## 7. URL state scheme - -Query parameters, human-readable, no encoded blob: - -- Base config: each field by its JSON name, only when it differs from the - preset base: `?forwardProb=0.5&graphSeed=99`. -- Panels: one `panel=` param per panel (repeated params), value is - `label~strategy~overrides` with overrides as comma-separated - `field:value` pairs, URL-encoded: - `panel=No%20program~none~` - `panel=Big%20budget~random~numEducated:60`. - When the panel set equals the preset's, all `panel=` params are omitted. -- Focus: `focus=` when the modal is open. -- Unknown params are ignored; malformed state falls back to the preset - with the ErrorBanner notice (5.1). - ---- - -## 8. Accessibility acceptance criteria - -Testable checklist; "both themes" means verified in light and dark. - -- [ ] Every interactive control (toolbar buttons, kebab menus, popover - fields, sliders, player controls, scrubber, collapsibles, modal, - theme toggle) is reachable and operable with keyboard only, in a - logical DOM order, with no trap outside the modal. -- [ ] Visible focus indicator on every focusable element: 2 px outline in - `--ink` with 2 px offset, both themes. -- [ ] Player keyboard map works as specced (Space, ArrowLeft/Right, - Home/End); scrubber is a native range input announcing - "Round x of y" via `aria-valuetext`. -- [ ] `prefers-reduced-motion: reduce` disables autoplay, tweening, hover - lift, and crossfades; stepping and scrubbing remain fully usable. -- [ ] All informative text meets WCAG 2.1 AA contrast (4.5:1, or 3:1 at - 18 px+/bold) against its background, both themes; `--ink-4` is never - the sole carrier of information. -- [ ] Node state is never encoded by color alone (shape encodings of - section 6 verified with a deuteranopia simulator on both themes). -- [ ] Network SVGs are `aria-hidden="true"`; each panel is a `region` - whose accessible name is its label and whose text content (big - percentage + meta row) carries the result. -- [ ] ResultsTable renders the live per-panel numbers as a real `` - with `
` headers (evolved from the existing - ComparisonTable), collapsed behind a disclosure reachable by - keyboard. -- [ ] One `aria-live="polite"` region announces results after each - successful run ("Updated: No program 83 percent, ..."), on pause, - on scrub release, and at playback end ("Final: ..."); it does not - announce every round during playback. -- [ ] FocusModal: `role="dialog"` `aria-modal="true"`, labelled by the - panel name, focus moves in on open, is trapped, Esc closes, focus - returns to the opener. -- [ ] Popovers (About, panel editor, export menu) follow the same - open/trap/restore rules as menus or dialogs per their role. -- [ ] ErrorBanner and inline 400 errors use `role="alert"`. -- [ ] Touch targets are at least 44x44 px on mobile (player buttons, - kebabs, chips, strip items); node hit halos at least 24 px. -- [ ] ``, landmarks: `header` (app bar), `main`, `footer`. -- [ ] Zoom to 200% at 1366 px wide loses no content or functionality - (panels reflow to the mobile stack). - ---- - -## 9. Responsive behavior - -Breakpoint: 760 px (one breakpoint only). - -Mobile (< 760 px): -- App bar keeps wordmark + theme + source; DisclaimerBadge hides, replaced - by the DisclaimerNote line under the hero. -- MobileScoreStrip appears: sticky under the app bar, horizontally - scrollable pill chips, one per panel (short label + colored %); tapping - a chip smooth-scrolls to its panel (instant under reduced motion). -- PanelGrid stacks single-column; toolbar stays one row (Export collapses - to an icon button if needed). -- PlayerBar fixes to the bottom edge (12 px inset, lifted shadow); step - buttons and the speed pill hide, leaving replay, play/pause, scrubber, - round counter. Page bottom padding clears it. -- Lower grid stacks: chart, then controls. Control rows go single-column - (label above slider, value right-aligned). -- FocusModal becomes a full-screen sheet. -- No horizontal scrolling at 320 px width; the strip is the only - horizontally scrollable element. - -Desktop >= 760 px matches the wireframe; the panel grid is -`repeat(3, 1fr)` for up to 3 panels, wraps to two rows of up to 3 beyond. - ---- - -## 10. Implementation slices (ordered) - -Each slice leaves the app shippable. - -1. **API topology**: add `POST /api/scenario` (config + strategy in, - result + edges out) in Go, tygo regen, tests for edge determinism. -2. **Store + URL**: `useSimStore` with preset loading, URL parse/serialize, - debounced parallel runs, atomic swap, keep-last-good error states. -3. **Static dashboard**: app bar, hero with live numbers, theme system and - tokens, panel cards with seeded d3-force NetworkView at final state, - legend, footer. (The page now beats the parity page; ComparisonTable - becomes the collapsed ResultsTable.) -4. **Playback**: usePlayback, PlayerBar, round-state rendering, autoplay - once, reduced-motion path, keyboard map. -5. **Reach chart**: overlaid cumulative curves, accents, playhead synced - to round. -6. **Controls**: lever sliders, advanced fields, seed rerolls, inline 400 - handling, ErrorBanner. -7. **Panel management**: add/duplicate/rename/remove, PanelEditorPopover - with override chips, 6-panel cap. -8. **Focus modal** with docked player and URL focus state. -9. **Node tooltip** with cross-panel echo and touch hit halos. -10. **Mobile pass**: score strip, fixed player, stacking, touch targets. -11. **Export**: JSON, CSV, PNG snapshot. -12. **Accessibility audit**: walk the section 8 checklist, fix, record - results in the PR description. - -Post-M3 (specced hooks, not built): `POST /api/optimize` goal-seek UI, -animated WebM/GIF export, additional study presets. - ---- - -## 11. Rejected alternatives - -- Single large network with strategy tabs: hides the comparison behind - clicks; the side-by-side comparison is the message. -- Fixed three-strategy panels only: cannot express budget or world - variations the dynamic-panel model needs. -- Fully independent config per panel: breaks layout comparability and - bloats URL and editor; overrides express the same power against a shared - base. -- Hardcoded study copy: blocks reuse for other research questions; preset - file chosen instead. -- Expand-in-place or routed focus view: modal chosen by user; routing - would pull in Vue Router for one interaction. -- Share button with clipboard toast: user chose address-bar-only sharing. -- Explicit Run button (and hybrid auto/manual): kills the - cause-and-effect immediacy that makes the lever visceral. -- Looping autoplay: hostile motion while reading; play-once chosen. -- Per-panel players: panels drift out of sync and the comparison dies. -- Per-panel sparklines and numbers-only metrics: hard to compare; one - overlaid chart chosen. -- Budget sweep / optimizer in M3: doubles scope; reserved as API hook. -- Animated GIF/WebM export in M3: as much work as the whole animation - feature; PNG + live share URL cover the need. -- Dark-only or light-only theme: both ship, prototype renders exist for - both and the toggle is cheap. -- Canvas/WebGL network rendering: SVG is ample at 120 nodes x 6 panels - and is easier to theme, hit-test, and export. -- Batch `/api/scenarios` endpoint: per-panel `POST /api/scenario` with - `Promise.all` is simpler and lets panels rerun independently. -- Base64 state blob in URL: readable repeated params chosen for a FOSS - demo; blob survives only as a fallback idea if URLs ever exceed limits. -- Fully focusable network nodes for screen readers: 120 tab stops x N - panels is worse than the table; aria-hidden SVG + live region + table - chosen. -- Pinia and Vue Router: not required by this design; module-scope - composable store and modal focus suffice.