Skip to content

TODO.md

Latest commit

 

History

History
613 lines (533 loc) · 37.1 KB

TODO.md

File metadata and controls

613 lines (533 loc) · 37.1 KB

Optimitron TODO - 4B Votes on the 1% Treaty

This file is the working priority list. If Mike opens it cold, the next useful work should be visible near the top.

Old sprint plans, session journals, stale PR checklists, and migration notes are in git history. Recover with git show <previous-commit>:TODO.md only when they directly unblock the 4B-voter campaign.

North Star

  • Goal: verified majority of humanity voting YES on the 1% Treaty referendum: roughly 4B humans.
  • Propagation math: 32 doubling rounds with each voter recruiting two more humans reaches about 4.3B.
  • Primary public site: warondisease.org, the International Campaign to End War and Disease.
  • Treaty text host: 1percenttreaty.org.
  • App/proof engine: optimitron.com.
  • Task tree: every campaign task is under optimize-earth (OPTIMIZE_EARTH_ROOT_TASK_ID, re-exported by packages/web/src/lib/tasks/task-keys.ts).

Campaign Priority Order

Do not let lower items crowd out higher ones.

  1. Increase treaty vote conversion.
  2. Increase referral propagation: each voter gets two more humans to vote.
  3. Get organizations to endorse, embed, and recruit their own people.
  4. Register plaintiffs and connect the case framing to voting.
  5. Remind country leaders and treaty signers — they are paid by the citizenry to promote welfare and are late on a 30-second task.
  6. Improve discoverability and trust in people, organization, task, and evidence pages.
  7. Preserve Optimitron's broader governance OS as the proof layer, not as a competing homepage.

Current State

  • Campaign mode is active. War on Disease is the product until the treaty passes.
  • Managed data now owns task tree, triggers, referendums, reference data, and seed shim.
  • Treaty vote, referral attribution, campaign emails, organization endorsement, plaintiff damages, and the simple /treaty skim-and-sign page exist.
  • Bio-template /love page is shipped; dating registry deferred per 2026-05-19 CBA.
  • Generic commerce schema is now the intended money path for live checkout: CommerceOffer/variant/order/item/fulfillment/entitlement cover shirts, sponsorships, subscriptions, digital access, and dating-app benefits without a shirt-only table. Product/vendor catalog IDs belong in managed data, not env.
  • Dating foundation schema is additive and separate from the task system: dating profiles, photos, prompts, questions, likes/passes/intros, matches, conversations, date plans, blocks, and safety reports get dating-specific privacy/moderation rules. Reuse Task only for the mission-output part of a date via DatingDatePlan.campaignTaskId, e.g. flyers, QR posters, outreach, or meetup follow-up.
  • /humanity-v-government and /court still need to unify plaintiff registration, verdict voting, and treaty settlement.
  • Visual review includes email screenshots; preview DB drift and unexplained missing screenshots still waste review time.

Active Review - 2026-05-19: Money and 4B Votes

Repo audit finding: the campaign machinery mostly exists. /vote and /treaty mount the treaty flow; AuthForm now locks the email-login controls after a sign-in link is sent; /join creates or opens organization tools; organization pages expose a survey URL, email starter, website button, iframe, preview, manager referral URL, and grant calculator; /poster has referral QR printing; /donate and /fund exist; signatories rank humans and organizations by attributable signatures; agent-readable mirrors and sitemap coverage exist for the public campaign surfaces; tests cover the vote API, referral attribution, invite-token paths, org endorsement, public detail sitemaps, share/email templates, treaty vote clicks, login, and reminder flows.

Main gap: this is not yet packaged as a fundable distribution sprint. The highest-value money path is a concrete 30-day institutional distribution sprint: fund outreach to organizations with audiences, get them to endorse/embed/recruit, and report signatures, organizations, referrals, and conversion bottlenecks.

Cost-benefit gate for near-term work:

Proposed change Benefit Cost / risk Decision
Write a 30-day funder/distribution sprint packet (docs/funding-sprint.md first; public page only after copy review) Gives donors and partner orgs a concrete ask: budget, targets, proof links, and done conditions 2-4 hours, no schema, no new UI required Do now
Seed or curate the first outreach task queue in existing Task records Converts "get organizations" into accountable follow-up work using the current task model 0.5-1 day; avoid new outreach schema Do after the packet if outreach starts
Polish /join and organization tools from first 5 real outreach attempts Removes actual conversion friction at the organization step 1-3 hours per observed issue; screenshots and copy approval required Do only from observed friction
Add a cheap weekly metrics report from existing referral/org/vote tables Shows whether money bought votes, orgs, referrals, or nothing 2-6 hours; avoid a dashboard until reports are used Do as a script/export, not a product surface
Redesign /fund or expand prize/fund mechanics Could look more fundable, but funders need a concrete sprint first High copy/UI review cost; risks leading with mechanism instead of distribution Defer
Poster style selector, PDF export, OG variants Useful if physical distribution proves real 0.5-2 days plus UI screenshot review Defer until poster scans/signatures show demand
Dating registry and per-app dating templates Potential niche channel, but /love bio-template is already the cheap test Schema, moderation, and privacy cost Parked until /love attribution clears the threshold
Public-figure catalog Could create social proof, but attribution disputes are expensive New data policy, likely schema, manual source review Defer until org sprint has traction
Full analytics dashboard Useful later; premature if nobody reads the report 1-3 days and another surface to maintain Use SQL/export first

Do not start a new feature unless it helps the sprint get money, convert votes, convert referrals, convert organizations, or prove the quantified case to a specific funder/partner.

Backlog - Organization Join Task Scaling

  • Keep the first 0-200 researched priority organizations in managed data when each row deserves code review. Build the bulk org-task import script when the target list crosses 200 researched organizations, not before. The script should consume reviewed CSV/JSON, upsert organizations and Task rows by stable source refs/task keys, and reuse the same organization join template.
  • For 200-5K organization targets, store the roster outside git and import it through the script; do not add thousands of per-org rows to managed-seed-data.ts.
  • For 5K+ targets, import from an external organization registry such as Form 990, GuideStar, or Charity Navigator, then curate priority slices before assigning public join tasks.

Active Checkout Launch Gates - 2026-05-20

  • Deploy the commerce migration before enabling paid shirt checkout.
  • Run managed-data sync after the migration so the shirt offer, variants, and CustomCat fulfillment mappings exist in the target database.
  • Keep env limited to secrets/ops toggles: STRIPE_SECRET_KEY, STRIPE_WEBHOOK_SECRET, CUSTOMCAT_API_TOKEN, CUSTOMCAT_SANDBOX, SHIRT_COMMERCE_ENABLED, and R2 credentials. Do not put product or variant IDs back in env.
  • Verify CustomCat catalog SKUs against GET /catalog/952 using the live API token, then run one Stripe test-mode/sandbox CustomCat order end to end. Local verification on 2026-05-20 skipped because no local CUSTOMCAT_API_TOKEN was present.
  • Only flip production SHIRT_COMMERCE_ENABLED=true after Stripe Tax, R2 public artwork URLs, CustomCat sandbox submission, and webhook retry behavior are all verified.

Active Dating Foundation - 2026-05-20

  • The first dating implementation should stay MVP: opt-in profile, photos, prompts, match questions, discover list, like/pass/intro, mutual-match messages, block/report, and optional mission-date plan linked to a Task.
  • Do not turn normal dating mechanics into tasks. A like, intro, match, private message, block, or safety report is not campaign work and should not enter the task tree.
  • Use existing Task records only when a pair chooses a concrete campaign activity: hang flyers, distribute QR posters, invite people to vote, host a singles meetup, or verify completion evidence.
  • Before public launch: decide photo moderation policy, approximate-location display rules, minimum age/consent checks, DM reporting workflow, and whether dating profiles are visible only to opted-in dating users.
  • Backfill jurisdictionId across the 25 new models added in PR #86 (DatingProfile + 14 dating models, CommerceOffer/CommerceOrder + 6 commerce models, TaskFundingTarget + TaskFundingPledge + TaskFundingEvent). CLAUDE.md requires it on every model. Deferred from PR #86 because adding it mid-deployment would require a coordinated schema migration + all server helpers + managed-data resync. Own PR: schema-only diff, default-backfill to the campaign jurisdiction, then update creators in a follow-up.

Active Handoff - 2026-05-13

  • Codex hook cleanup: Mike prefers deleting repo-local .codex hooks instead of condensing them. The intended change is to remove .codex/hooks.json and .codex/hooks/*; keep .codex/agents/* and .codex/config.toml MCP config. The hooks were mostly Claude-Code guardrails and add friction/background risk for Codex.
  • Adaptive email rendering: the bug was a process-wide renderSurface global leaking between async email renders and normal web renders. The safer shape is React context via EmailRenderSurface, plus a regression test proving an async email render does not affect a web render. Watch the "use client" boundary: if components/adaptive/index.tsx or ParameterValue.tsx must be client-side for web pages, also verify /dev/email/* still renders server-side.
  • Verification already run for the adaptive/email fix: pnpm --filter @optimitron/web exec tsc --noEmit; focused Vitest suite covering adaptive components, ParameterValue.email, share footer, monthly digest, post-vote share email, first-conversion email, magic-link host dispatch, and share message; direct renderPreviewBodyHtml for all six email previews. The dev /dev/email/*?raw=1&full=1 route timed out while copy:preview had the Next dev server overloaded, so re-check it after the server is calm/restarted.
  • Parallel-agent boundary: packages/web/src/app/employees/page.tsx, packages/web/src/components/tasks/PresidentManagementSystemSection.tsx, and packages/web/src/lib/tasks/overdue.ts were already staged by another agent. Do not unstage, revert, or fold them into unrelated commits.
  • Process note: another copy:preview run drove the Next dev child above 10 GB private memory. I stopped only the copy:preview worker chain; the shared 3001 dev server stayed up and root responded afterward.

P0 - Conversion and Email Safety

  • Login page: excess space between slider (CTA / framing element) and the submit button pushes the submit below the fold. Reduce vertical spacing so the entire form is visible above the fold on common mobile viewports without scrolling. Audit gap-* / mt-* / py-* on the AuthForm container.

  • Server-side sign-in-link send rate limit is a defense, not the next product bottleneck. Client-side repeat-send spam is already mitigated by AuthForm's hasSubmitted state hiding the email/Google/demo controls after success. Add a server-side limiter only if Resend/auth logs still show repeat sends after the UI fix, or before a high-volume outreach push. Avoid an in-memory limiter as a false guarantee on serverless.

  • Add a min-font-size validation pass — emails first, then web. We need automated detection so this doesn't recur. Two layers:

    • Email-specific Playwright test: render every email preview via the existing pnpm copy:preview / email:preview-md flow, walk every text node in the rendered HTML, fail if computed font-size < 14px UNLESS the node has an explicit data-allow- small="footnote" opt-out attribute. Reuses the existing contrast-audit.spec.ts pattern at packages/web/e2e/.
    • Web ESLint rule (faster, complementary): flag JSX className patterns containing text-xs, text-[N]px with N<14, and inline style={{ fontSize: "<14px" }}. Allow overrides via // eslint-disable-next-line min-font-size comments naming why. Lives in the existing ESLint config alongside the strict-mode rules.
    • Long-term: replace numeric fontSize with named tokens (body, lead, caption, footnote) in adaptive/email-styles.ts so the size policy lives in ONE place and surfaces use semantic intent. Token-based then the lint rule has a clean allowlist to enforce against.

  • Poster follow-ups after v1. /poster now exists with a logged-in referral QR, generic logged-out QR, compact copy affordance, letter/A4 print CSS, and the default treaty-editorial style. Remaining: style selector, OG image variants, optional PDF download, and share-template text selection if printed posters become a measured channel.

  • Apocalypse framing standardization follow-up. Canonical War on Disease site description exports exist, but source and generated copy still contain direct "apocalypse" phrasing in route metadata, referendum-site copy, managed task triggers, Grandma Kay seed text, and share templates. Finish only after the copy gate approves one standardized phrase.

  • Other human-language candidates while we're sweeping copy:

    • "magic link" in user-facing error strings (/auth/signin/page.tsx:12 "That magic link is invalid or has expired." + AuthForm error fallback "Unable to send a magic link right now."). Non-tech users don't know what a magic link is. Use "sign-in link" or "email login link". Keep magic-link as the internal slug / file name — only the user-visible strings change.
    • KEEP (intentional Wishonia-voice satire — do NOT change): "Humanity Manager", "promoted to", "subordinate humanity managers", "Earth Optimization Services LLC", "Earth Optimization Points", "AEOSP / Authorized Earth Optimization Services Provider". These are the corporate-org-chart joke, not jargon.

P0 - Increase Treaty Vote Conversion

Keep /treaty boring and fast

  • Preserve the one-page skim-and-sign treaty flow: headline, treaty body, signature box, YES/NO. No stepper, slide split, competing Court CTA, or decorative explanation before the vote.
  • After the PR #75 managed referendum sync reaches production, regenerate and commit the treaty/h-v-g/join markdown snapshots so citation URLs reflect the fixed upstream manual refs.
  • Keep treaty copy parameter-backed. Do not hand-type 4B, 32 rounds, 122 apocalypses, trial multiplier, or eradication-timeline numbers where a ParameterValue or generated parameter exists.

Make the logged-in dashboard a command surface

  • Dashboard top priority: vote/verdict status, canonical share message, plaintiff status, assigned campaign tasks.
  • Remove duplicate embedded surfaces from the dashboard when a dedicated page exists. Presidents belong on /employees or its eventual rename; full treaty text belongs on /treaty; signatories belong on /signatories.
  • Add the Humanity Manager status panel only when it increases the next action: direct converts, overdue humans, overdue presidents, and one copy action.

Simplify task execution

  • Task-list rows should behave as one link to /tasks/<id>. Assignee names and avatars inside lists should not trap row clicks; assignee navigation belongs on the detail page.
  • /tasks page redundancy: the top section heading is "Humanity's Tasks" but the root row chip shows assignee "You" (because isAssignedToYou is true for any humanity-org-assigned task), then a separate "Your Tasks" section appears below. Hide the assignee chip on the Humanity's Tasks section (section heading already establishes the context) — the empty-state "Your Tasks" CTA below should also render via the same task-row component as the rest, not a one-off TreatyVoteCta box.
  • Task-table column hiding: when no task in a given task-list rendering has a value for a column (e.g. Military Budget on non-signer task sets), hide that column entirely rather than rendering empty cells. Today columns are static.
  • Impact inheritance for tasks without their own numbers: recommendation — show "—" (no inheritance) for tasks without an impact.selectedFrame. Reason: inheritance from parent creates double-counting when aggregating children (sum of N children "inheriting" parent X gives N×X, which is wrong); explicit per-task numbers force data-entry discipline. If the empty cell feels too bare visually, fall back to "shows parent estimate — this task has no estimate yet" with the parent's number, but mark it as inherited and exclude from aggregations. Per-task overrides take precedence over parent values.
  • On /tasks/[id], keep title, assignee/avatar, due date, primary action, markdown body, comments, complete/reassign controls, and admin disclosures. Remove duplicated metadata blocks.
  • Decide the logged-in task action label. Current "Claim Task" is bad. Working candidate: "Do this." Do not use "Take this on."
  • Reframe enum labels in the viewer state strip so users do not see raw "Claimed / In Progress / Completed / Verified" workflow labels.
  • Add one E2E regression that a signed-in user can open an assigned/private task from "Your Tasks" without hitting 404.

P0 - Increase Referral Propagation

Canonical share message and post-vote email

  • The post-vote email is a single forward-friendly share kit fired when a YES treaty vote is counted. No drip sequence. No generic reminder spam.
  • Canonical share-message wording now lives in packages/web/src/lib/share-message.ts. Keep dashboard, post-vote flow, monthly digest, and email footer aligned to that source.
  • First-conversion email stays one-time only per referrer. Do not notify on every conversion.

share-templates.ts is the source of truth for ALL reminder copy

  • packages/web/src/lib/tasks/share-templates.ts is the canonical voice-variant registry: ~26 named templates (Trump versions, office memo, performance review, polite reminder, etc.) keyed by recipientModes (leader | humanity | one_human | peer) with token-based interpolation.
  • Monthly-chain-digest and the shared email footer now read from share-templates.ts; referral-first-conversion inherits that footer. Remaining audit: post-vote-share still builds from share-message.ts, and task-comment-notification needs a final check for hand-rolled reminder copy.
  • Every email module that includes reminder/share copy should pull recipient-appropriate templates from share-templates.ts where a reusable template fits. Hand-rolled copy stays only when no template fits AND a new template would be too narrow to reuse.

Humanity Manager status report

  • Replace direct-only monthly counts with transitive chain counts when the query is ready.
  • The dashboard version should expose copyable messages for overdue humans and presidents instead of motivational filler.

P1 - Organizations Endorse, Embed, and Recruit

  • Persist the organization grant/application workflow: request data, review status, and follow-up outreach. The current calculator/request framing is not enough for operational follow-through.
  • Keep organization attribution first-org-wins for ReferendumVote, matching referredByUserId. Later org links should not steal attribution.
  • Keep neutral partner/embed copy where full Wishonia voice would make adoption harder. Partner-safe is not the same as bland.
  • Adopt the "Authorized Earth Optimization Services Provider" framing for partner orgs in campaign-facing copy. Extends the corporate-promotion voice from the post-vote-share email: voters are Humanity Managers at Earth Optimization Services LLC; partner orgs are Authorized Earth Optimization Services Providers, each with a vendor-style certification badge they can display. Update /join to register orgs under this category. Per the neutral-partner-copy note above: keep the application form itself professional enough not to scare off serious nonprofits — AEOSP framing lives in campaign-facing pages, shared snippets, and the badge artifact, not the onboarding form.
  • For the next sprint, do not build new organization admin surfaces before outreach proves the existing tools are the blocker. Current org pages already provide the share URL, email starter, linked HTML starter, website button, iframe, preview, manager referral URL, and grant calculator.

P1 - Person/Org Conversion Surfaces (post-PR #81)

Roadmap from Mike's 2026-05-15 brainstorm. /people/[id] redesign lands in PR #81; the rest follow once that merges. Ship in this order for maximum reuse.

  • PR-A: Foundation-action seed tasks. Seed optimize-earth task tree with concrete user actions: donate $1, print poster, hang flyer, social-share. New task category campaign-action. Hours-prevented impact metric per task. One file: packages/db/src/managed-data/managed-seed-data.ts. Smallest, lowest risk; unlocks PR-B and PR-C.

  • PR-B: /orgs/[slug] task display. Mirror /people/[id] conversion-surface pattern onto org pages. Reuse SufferingPreventedMetric (extracted in #81). Wire getOrganizationTasks MCP to a page consumer. Primary CTA: ENDORSE (visitor not in org) vs SHARE (visitor's org already endorsed). Below-fold: org-completed tasks + member leaderboard.

  • PR-C: Add-org + assign-task UX. Backend primitives exist (createOrganization + createTask MCP). New surfaces: /orgs/[slug]/admin/tasks for org admin self-assignment; /admin/assign-task (superuser, Mike-only) for cross-org. Gate behind superuser role until proper moderation. Audience: org admin who joined via /join, wants to coordinate their members.

  • PR-D: Hand-curated public-figure catalog. Seed Person rows for top 50-100 public figures (scientists, politicians, celebrities). Each row: displayName, handle (mlk, einstein, gates), 50-word deadpan-Wishonia bio, 1-3 attributed campaign-relevant actions with documented public-statement sources, impact-estimate (DALYs averted, methodology-cited from parameters-calculations-citations.ts), isPublicFigure: true flag (new bool on Person). /people/<famous-slug> renders with "Public-figure record" eyebrow. Visitors can co-sign the figure's position.

  • PR-E: AI-cataloged public-figure activity (deferred). Scale PR-D with NER on Wikipedia/news/govt records → impact-attribution model → manual-review pipeline. Scaffolding only after PR-D's manual seed proves the UX works. Not on the campaign critical path.

Out-of-scope for this roadmap: peer-task-assignment on /people/[id] (Mike deferred 2026-05-15; needs whitelist/opt-in moderation design — separate plan).

Shared component reuse: all five PRs share SufferingPreventedMetric (extracted in #81). No new design primitive.

Risks: public-figure attribution disputes (use only documented public statements; label "historical position, not active endorsement"); spam on /admin/assign-task (superuser-gated); org-task display competing with conversion CTA (keep task list below the fold).

Durable summary lives here; no loose .claude/plans/campaign-impact-attribution-roadmap.md file should be kept unless a fresh plan review starts.

P1 - Plaintiffs and Court Framing

/humanity-v-government plaintiff-first rework

  • Primary action becomes plaintiff registration. Hero CTA: "Name your dead" or the strongest approved variant, not "Support the settlement."
  • Show a running plaintiff count near the hero. Named plaintiffs are harder to ignore than an anonymous vote total.
  • Build the public death/plaintiff stream: a constantly updating, moderated list of humans submitted by loved ones who recently died, shown as wrongful-death claims against the governments that kept buying apocalypse capacity instead of medical progress. Use represented people / CourtCaseParty data first; do not scrape or invent deaths for v1. Landing-page placement: below the primary treaty vote/referral action, not above it. Show the most recent named person, a ranked/recent list, and a CTA to "Name your dead" / register the estate. Tie the legal theory to the 1900-counterfactual: disease deaths after the model cutoff and aging deaths after the aging cutoff are prima facie government-caused because a rational treaty could have redirected military growth to clinical trials and public goods. Reuse parameter-backed apocalypse / cutoff / damages claims; no hardcoded 122-apocalypse or 1950/1990 copy.
  • Add the missing counterfactual sentence: damages are what humanity would have had if governments had signed the 1% Treaty in 1900, freezing military spending growth and redirecting surplus to clinical trials and public goods.
  • Drop secondary hero CTAs to /vote and external evidence. Demote them below the plaintiff action.
  • Collapse "usual defenses" into a disclosure. Remove decorative case-caption repetition. Move DamagesSensitivityCalculator next to the damages/vote context instead of burying it.

/court as the operational Court surface

  • Build /court around the case caption, plaintiff/juror count, defendant status, settlement progress, and one treaty/verdict CTA.
  • Seed or sync the Humanity v. Government CourtCase row with claims, harms, evidence, parties, and the 1% Treaty settlement remedy.
  • Add the 193 governments as respondent parties and drive status from treaty signature/ratification state.
  • Surface implicit class membership on the dashboard before vote: "You are a potential plaintiff. Render your verdict to formalize your claim."

Represented people and estates

  • Plaintiff-registration aspect-ratio handling is no longer P0. SquarePhotoCropper is wired into user upload surfaces, and Grandma Kay now uses /img/grandma-headshot.jpg. Add managed-data validation for non-square seeded person images only when another seed image regresses or when plaintiff registration becomes the active sprint.

  • Reframe memorial/deceased-person registration as filing a wrongful-death claim for the estate, with descendants as beneficiaries.

  • Add pre-search before creating represented people: canonicalized display name

    • birth date + death date, then offer "join as co-next-of-kin" on match.

  • Avoid schema work unless duplicate Person rows become a real operational problem. Optional later: indexed Person.canonicalKey.

  • Earth Optimization Day stays separate until the case page is coherent. MVP: /earth-optimization-day, countdown/RSVP, existing verdict/treaty widgets, and isEarthOptimizationDayWindow() before seasonal CTA swaps.

P1 - Remind Leaders and Treaty Signers

  • Keep /employees as the president-accountability surface for now. Consider /presidents rename later if it improves comprehension.
  • Dashboard should link to the president surface instead of embedding the whole management table.
  • The government-side task wording is "Get 193 heads of government to sign." Avoid vague "get governments to adopt the treaty."
  • Monthly status and dashboard panels should identify overdue presidents and provide copyable reminder language.
  • Internal leader/signature tasks should stay under managed data, not ad hoc seed fragments.

P1 - Discoverability and Trust

Sitemap and evidence paths

  • Split sitemap files by entity type when tasks/people/orgs approach the 500-row cap.
  • Keep 1percenttreaty.org as a separate shareable treaty domain. Do not collapse it into warondisease.org/treaty.

Copy and citation quality

  • Sweep public copy for startup-bro/system-architecture filler: off-ramp, enforcement stack, incentive layer, coordination mechanism, primitive, substrate, fundamentally, and similar.
  • Replace abstractions with concrete user action, villain, number, or outcome.
  • Keep a lightweight email/template validation test: render templates with fixtures, cap length, reject banned phrases, assert required tokens, and enforce one primary CTA unless the share footer is intentionally part of the action.

Visual review and preview workflow

  • Add a missing-screenshot banner to latest.html, auto-screenshot changed routes after preview READY, and write review pages under packages/web/output/playwright/pr-watch/.
  • Investigate Neon branch-per-preview or preview-scoped managed-data sync.
  • Add /dev/email index over EMAIL_PREVIEWS.
  • Extract large GitHub Actions inline JavaScript blocks when touched again.
  • Preview deploy smoke test (would have caught the 2026-05-14 stale /plaintiffs regression: Prisma error rendering on the PR #79 preview while production was healthy — preview DB out of sync with main):
    • One CI step that fires AFTER Vercel reports deployment_status: ready, hits ~5-8 critical routes against the preview URL with the _vercel_share bypass token, and asserts: HTTP 200 + body does NOT contain "Something went wrong" or "Application error" + the expected <h1> is present.
    • Routes to cover: /, /treaty, /plaintiffs, /tasks, /humanity-v-government, /employees, /court, /people.
    • Target: under 30 seconds total. NOT the full Playwright suite — that's slow + redundant with the existing local-build e2e + visual review.
    • Catches: preview DB drift, env var mismatches, Edge-vs-Node runtime errors, missing migration sync, generic 500s from Vercel.
    • Won't catch: internal logic bugs or UI regressions (already covered by existing CI against the local build).
    • Stash the bypass token in the GitHub Preview environment alongside VERCEL_TOKEN.
  • Production smoke test after deploy — same shape as the preview smoke test above, pointed at production after every prod deploy. Same routes, same assertions (200 + no error-boundary text + expected <h1>). Catches the bug class where a query works against demo/seed data but explodes against real production data shape (NULL relationships, large row counts, edge-case user states). Run as a GitHub Actions step on deployment_status: production_ready for the Vercel deploy.
  • Anonymized prod-DB fork for preview (parked) — Neon branching + a one-shot anonymization step (hash emails, redact non-public-figure display names, null out signatures, keep schema shape + row counts). Solves the "preview data ≠ production data" problem properly. Engineering cost: ~1-2 days + recurring maintenance as new sensitive columns land. Privacy is load-bearing — even with Vercel preview auth protecting against the public, team members + invited reviewers can see real data. Parked until campaign launch makes prod state diverse enough to bite regularly.

P2 - Preserve the Governance OS as Proof Layer

  • Optimitron supports the campaign. Do not rebuild the generic Optimitron homepage, feature archive, demo surfaces, or platform narrative while the vote funnel is the bottleneck.
  • Keep managed data as the source for semi-permanent app records. Missing from a manifest must not imply delete; only explicitly retired managed records should be soft-deleted.
  • Do not introduce a second task model. Personal/private, org-assigned, treaty-invite, and agent-proposed work all remain Task rows with scoped ownership/visibility.
  • Outreach stays on Task, TaskCommunicationEndpoint, TaskCommunication, TaskComment, ReferralInvitation, ShareAttempt, and EmailLog. Do not add special outreach models without a real path that the existing model cannot cover.
  • allowsUserSubtasks schema work is parked. Existing schema is enough until public subtask creation UI is immediate.
  • Funding split: retail donations fund campaign operations; chain treasuries are a separate prize-pool track after institutional-host signal. Do not divert Stripe/Endaoment charitable donations into the prize contract.

Durable Guardrails

  • Read the relevant AGENTS.md before package edits.
  • Prisma schema or exported @optimitron/db type changes require explicit human approval.
  • Library packages must not import Prisma client or runtime DB code; use import type for cross-package type imports.
  • @optimitron/optimizer remains domain-agnostic: predictor, outcome, variable, measurement, effect size.
  • Never write tests that only assert mocks were called. Test shipped behavior or a real regression boundary.
  • Never merge PRs. When checks are green and valid review complaints are handled, report ready for human review/merge.
  • Plan files (under ~/.gstack/projects/<slug>/) MUST include a Cost-Benefit Matrix section before /autoplan final-gate review. Enforced by .claude/hooks/enforce-cba-table-on-plan-files.mjs.
  • New-feature plans MUST acknowledge existing routes/branches/commits matching the feature noun before drafting. Enforced by .claude/hooks/enforce-feature-preexistence-check-on-autoplan.mjs.

Parked Unless They Directly Unblock 4B

Dating registry — deferred until /love proves attribution

  • Person.isOpenToDating + /love/registry browse + email-bridge messaging — full plan reviewed 2026-05-19 by /autoplan, deferred per CBA verdict. /love bio-template page already shipped at packages/web/src/app/love/page.tsx (commit ba27f38d).
  • Kill threshold: ship only if /love bio-template channel produces ≥X treaty signatures attributable to dating-bio referrals per 30 days.
  • Wishonia-voice copy expansion (per-app Tinder/Hinge/Bumble templates, follow-up scripts, QR date-card variant) also deferred until /love instrumentation proves per-app variants matter.

Internationalization — centralize copy into a single message catalog

  • All user-facing copy currently lives inline in .tsx components, email modules, share-message.ts, share-templates.ts, managed task descriptions, routes.ts metadata, and external manual pages. Voice review (Wishonia rules, qa-editorial, voice-critic) currently greps prose in source.

  • Strategically correct migration target: a single messages/en.json (or packages/web/src/messages/en/*.json split by surface) feeding a typed key catalog, so the same prose can be translated to other locales without forking React components.

  • Cost vs. benefit today: thousands of strings to extract, voice review becomes harder (key-in-JSON harder to scan than prose-in-JSX), and the campaign is English-only until the 1% Treaty has English-speaker traction.

  • Trigger to start: either (a) English campaign demonstrates a verified- voter trajectory that justifies non-English rollout, or (b) a specific non-English country is targeted for launch. Until either happens, every hour spent on i18n plumbing is an hour not spent on the current English conversion funnel.

  • When started: lift share-templates.ts first (already token-based and the closest shape to an i18n catalog), then email modules, then component copy. Don't try to migrate everything in one PR.

  • Multi-agent/service-account architecture plans; AP2 / ACP / x402 payments.

  • Optimitron root rewrite and /features archive.

  • Donate-to-fund-task marketplace, Stripe Connect disbursement, WISH airdrop, VOTE-for-task-completion, monthly distributions, DAO-governed funding.

  • DIH migration, generic referendum/commission/EV-calculator work outside the campaign path, and broad email file renames.

  • Adopt @openai/codex-sdk for Codex dispatches. Current dispatches go through Bash → codex exec '...' < NUL > log 2>&1 with the enforce-codex-protocol hook gating substantive work. The SDK gives us streamed agent events, parallel threads with stable IDs, and turn/steer / turn/interrupt mid-run — solving the wrapper-opacity and stdin-hang failure modes today's session demonstrated. Setup is npm install @openai/codex-sdk in a dispatcher script; was wrongly estimated as "couple hours of work" before — actual cost is small. Implement when we hit the first failure mode direct exec can't cover (mid-turn steering, parallel agent coordination). Sources: https://www.npmjs.com/package/@openai/codex-sdk, https://developers.openai.com/codex/sdk

  • Single black-and-white style migration. Scope: ~560 brutal-* / BrutalCard references + most of packages/web/src/components/ui/ legacy decorative shapes (brutal-card, arcade-tag, game-cta, comparison-card, featured-info-card, icon-card, item-card, nav-item-card, numbered-step-card, rarity-badge, spending-bar, stat-bar, stat-card). Migrate to retroui primitives + semantic markup matching the editorial treaty style. Phased: (a) inventory by page surface, (b) opportunistic migrate when touching a file (current policy — keep this until usage drops), (c) once brutal-* references drop below ~50, do a final sweep + delete the legacy files. Eventually assign to a long-running Codex agent. Admin / game-demo / email markup may keep specialized styling.