What's new.

Nav previously returned null for logged-out users — meaning visitors landing on /guides, /coming-soon, /whats-new, /analyze/example, or /deck/example had no header at all and no way back to /. Nav now renders a minimal public variant (logo + Sign in + Try it free) for unauthed users on every page except / (custom inline nav), /preview (internal), /login, and /auth/* (where it would be redundant). Replaces the standalone PublicShowcaseHeader component, which is removed. Plus: homepage logo now resizes responsively (h-6 on mobile, h-10 on sm+) — previously fixed at 40px which crowded the 60px mobile nav band.

Two surfaces previously missing internal links from /. (1) New Hard-Earned section on the homepage between the benefits grid and the closing CTA — pulls the two most recent published guides server-side and shows them as cards with title, excerpt, and a Read → link. Cross-linking from the highest-authority page = compounding SEO. Section auto-hides if there are no published guides yet. (2) Site-wide footer in app/layout.tsx — three columns (Read / Account / About) covering Hard-Earned, Coming soon, What's new, sign-in, contact, terms, privacy. Fills the gap that previously left /coming-soon and /whats-new only reachable by URL or sitemap.

Migrated key <img> tags to next/image — Crystal avatars on /results, /speak, /analyze/example, VoiceRecorder; brand Logo; guide featured images on /guides and /guides/[slug]. Adds automatic responsive sizing, lazy-loading, and LCP optimization. Allowed Supabase Storage host in next.config images.remotePatterns. Separately, /admin is now server-side gated: previously it was 'use client' and the role check happened in a useEffect, which briefly exposed the page chrome to non-admins (no data leaked, every API endpoint enforces role server-side, but the tab structure flashed). Page is now a tiny server component that checks role before sending HTML and redirects non-admins to /. AdminClient holds all the original interactive logic, defense-in-depth role check kept inside.

/analyze now auto-saves your transcript to localStorage on every keystroke (cleared on submit), so a browser crash or accidental refresh no longer drops what you typed. /deck/[id] step 2 (clarifying answers, edited storyline tiles, free-text revision, keep-original toggles) auto-saves to localStorage keyed per deck and restores on reload — plus a beforeunload warning if you close the tab with unsaved edits. Drafts only restore for decks not yet submitted, so a stale draft can't shadow a real decision. /whats-new and /coming-soon also opened to logged-out visitors (same family as the /guides public-routes fix).

Phase 1 audit fallout. (1) /guides + /guides/[slug] now load for logged-out visitors — they were unintentionally walled by middleware, killing organic traffic to the Hard-Earned essays. (2) Removed /api/debug-prompt, an old debugging endpoint that leaked system-prompt fragments + Supabase URL to any logged-in user. (3) Homepage textarea and guide CTAs now route to /analyze/example instead of /analyze, so logged-out users see the showcase before hitting the login wall (matches the orange Get-feedback button fix). (4) Added <h1>s to /analyze/example and /login for screen-reader navigation and document outline.

Homepage "Get feedback" CTA now lands on /analyze/example — a public, server-rendered showcase of a full Ms. Crystal critique (verdict, scorecard, So-What callout, word-by-word table, stop-using list, suggested rewrite). Renders the same components as /results, so what visitors see in the example is exactly what they'll get on their own pitch. Zero token cost, no agent calls. "Try it on your own" CTA at the bottom routes to /analyze for users who want to commit. Sister page to /deck/example.

New /guides section with public index + per-guide pages, and a Guides tab in /admin for writing them. Rich-text editor (TipTap) with bold, headings, lists, links, blockquotes, code, horizontal rules, and inline image upload to a public Supabase Storage bucket. Featured image, excerpt, slug auto-generated from title (editable), Save draft / Publish workflow, and a Preview link that lets you check drafts at /guides/[slug]?preview={id} before going live. Public pages render with Tailwind Typography plus brand-color overrides; per-guide SEO metadata + OpenGraph image; sitemap auto-includes published guides. Drafts return 404 to the public.

Lior can now rename every feedback block and (in text mode) swap stop_using ↔ rewrite directly from the Prompt tab — no engineering ping needed. New rendering-prompt builder assembles the system prompt from a block manifest + DB-saved label/order config; the CSS class contract stays locked. Each block shows a tiny color-correct preview swatch so the block being renamed is visually obvious. New Test button runs the draft prompt through Claude with a canned-but-editable sample, then validates the rendered HTML against a structural manifest (required classes, required children, valid tone modifiers, no rogue <style>/<script>, blocks-in-saved-order). Result is a per-block pass/fail checklist with the raw HTML available on demand. Default config produces byte-identical output to production — invariant covered by sanity tests.

After Ms. Crystal finishes, a What you said accordion appears above the feedback. It stays closed by default so the feedback is still the hero, and one click reveals the full transcript Deepgram heard. Every word/phrase Ms. Crystal flagged in the word-by-word table is highlighted inline in the same red/yellow/green pill, so you can see exactly where the buzzwords landed in your own speech. Voice sessions now also save the transcript, so you can come back to it later from /history.

Copy now writes both rich HTML and plain text to your clipboard — paste into Google Docs, Notion, or Gmail and the verdict box, scorecard, tables, and color-coded pills come through with formatting intact. New Download PDF button generates a branded report (logo + Ms. Crystal avatar + date + your context + the full feedback) ready to email or archive. Page-numbered footer.

Per-page metadata so each route gets its own title and snippet in Google. Dynamic OpenGraph image so links shared on Slack, Twitter, and LinkedIn render with a real preview instead of a blank card. robots.txt + sitemap.xml so crawlers index the right pages and skip the auth-gated ones. JSON-LD on the homepage so Google understands what MakeAPoint is. Crystal portrait now uses Next.js Image for automatic responsive sizing and faster loads.

New users on /analyze now get a richer welcome card that previews the 6 sections of feedback they're about to receive (the same orange pill flow used in admin). After analysis on /results, a What's next? card surfaces /speak and My Projects so the voice mode stops being hidden in the nav. Card has a Don't show again link.

First tab in /admin is now a Help page — orientation card explaining the editorial-vs-locked split, six "How do I…" cards that link to the right tab (tune Ms. Crystal, A/B test, ratings, replies, sessions), and a callout listing when to ping Jonathan. Lands on the same surface anyone opens admin to.

A small "About this prompt" panel at the top of the Prompt tab spells out: tone and persona are yours, output structure and HTML are handled by the page. Different content per Text and Voice mode, since they have different fixed structures.

Ms. Crystal (Clear)'s voice analysis now lands in the same look as text feedback — a verdict box at the top, green pills for keywords to keep, color-coded So What status (Landed / Partial / Missing), and a Filler-word watch callout where every filler is highlighted in red. Same five sections, much easier to scan.

Every voice take you record on /speak is now saved to My Projects with the context (audience, occasion, CTA), the date, and the feedback Ms. Crystal (Clear) gave you. Each retry is its own row so you can compare takes. Your transcript is still never saved — you don't read what you said.

The admin Prompt tab now has a Text mode / Voice mode toggle. Voice mode shows the same editor + version history as text, scoped to the /speak prompt — so the voice character can be tuned without touching the database. Reference files and A/B variants stay text-only for now.

New /speak page lets you record up to 5 minutes of your pitch and Ms. Crystal (Clear) listens. She returns the strongest keywords from what you actually said, the takeaways that landed, the So Whats (or the missing ones), and a Filler-word watch — "um," "like," "you know" — pinned to where they cost you. The transcript itself stays server-side; you never have to read what you said. Up to 3 takes per session.

New /contact page with a four-field form (name, email, topic, message) where anyone can reach out without needing an email client. Submissions land in the admin tab "Talk to me" with unread badges, mark-as-read / mark-as-responded actions, and a quick mailto reply link. Honeypot + per-IP rate limit handle basic spam.

Feedback now renders with the full design system applied — hero verdict pull-quote, color-coded scorecard cells, alternating rows in the word-by-word table, and proper Stop Using / Rewrite blocks. Sections lead with greens before reds. Truncated outputs are fixed with a 2x token budget.

The results page is now on the same design system as the rest of the app — brand orange buttons (was dark gray), tokenized colors, primitive components, and the new mobile-friendly sticky bottom bar. All feedback rendering, follow-up flow, ratings, and admin A/B tabs work the same.

The landing page now leads with a tighter hero + side quote card, a scrolling ticker of Ms. Crystal one-liners, three principles, a real sample feedback panel showing what the product actually returns, a four-step how-it-works, the creator section, a dark closing CTA, and a real footer. All built on the design system primitives.

Every page is now on the new design system — brand orange CTAs, underlined form fields, tighter typography rhythm, and consistent component primitives. /analyze, /history, /login, /coming-soon, the legal pages, and the nav all got the same treatment.

The Feedback tab badge in admin now only lights up when there are unread mentions — not whenever any pending feedback exists. Pending status is still visible inside the tab; the indicator is now signal, not noise.

A new avatar with your initials sits in the top-right of the nav. Click it to see your name, email, role, and sign out. The same identity block now anchors the top of the mobile menu.

The home page leads with a concrete claim — sharpen your message in 30 seconds — and explains what the tool does and who it's for. The CTA is now the brand orange, and Lior's bio reads in first person.

Each feedback item in the admin tab now has a conversation thread. Tag teammates to bring them into the discussion — they'll see an unread badge in the nav and a New pill on the feedback card. Threads with unread mentions sort to the top of the list.

Fixed the auth callback so it handles both Supabase redirect formats — invite emails now route correctly to the password-setup screen instead of bouncing users to a blank login form.

The admin Users tab now supports per-user actions: change role, suspend, reactivate, resend invite, and delete. Last-admin and self-action guards prevent lockout.

Body keeps the system font stack (clean, modern, fast). Poppins is loaded for design moments — page titles use Poppins 300 in orange, buttons use Poppins 700. Body color updated to true black.

Introduced a Button component matching the Figma spec (orange/gray/teal variants, four sizes) and updated the brand orange to #EC6526 site-wide.

Added a Download my data link on the My Projects page footer, next to Delete my account. Sends a prefilled support email so we can export your transcripts, feedback, and account metadata on request.

Added placeholder Terms of Service and Privacy Policy pages so the footer links go somewhere useful. Full versions are coming before public launch.

Retired the public /roadmap page. The roadmap now lives only in the admin tab, with click-to-expand rows that show notes, original messages, admin notes, and full metadata.

The changelog now lives at /whats-new, reachable from a bell icon in the top nav. A small orange dot signals when there's something new to read.

Refreshed the makeapoint.ai logo across the landing page header and the in-app navigation.

Updated the browser tab icon to match the MakeAPoint logo: the orange 'a' circle.

Test alternative prompts side-by-side. Enable Variant B or C in the Prompt tab, then run an analysis to see comparison tabs on the results page. Admin-only feature for evaluating prompt performance.

Store and restore the last 5 system prompt versions. Add optional notes when saving. Access via the Prompt tab.

Improved readability with editorial spacing: larger body text (16px), better line height (1.8), clearer heading hierarchy, and more breathing room between paragraphs.

Replaced the explicit 'Ask' button with an inline send icon. Press Enter to submit, Shift+Enter for newlines.

Sticky progress header on scroll, expand/collapse all notes button, dynamic last-updated date, and empty state for community suggestions.

Replaced word-by-word typing animation with a clean 300ms fade-in. Feels more professional and doesn't delay content access.