Redesign audience reaction UI with FAB and popup grid - #9
Merged
Merged
Conversation
Replace the cramped 8-emoji horizontal pill bar (which overflowed on small phones) with a clean FAB + popup grid pattern: - Single floating action button on bottom-right, above nav controls - Tap to expand a frosted-glass popup with 4x2 emoji grid - Each reaction has proper touch targets with labels - Cooldown shown as an SVG ring around the FAB - Floating emoji rise from FAB when reactions come in - Auto-dismiss on selection or tap outside - Host view unchanged (count badges + floating emojis) Removes all old .reaction-btn CSS (including duplicate block) since mobile reactions are now fully inline-styled in the component. https://claude.ai/code/session_014T4yzapvSSZdPBJCJgsRkP
Solo mode changes:
- Server: when host triggers CHAOS in solo mode, immediately pick a
random twist and inject it instead of starting a 15s vote timer
with no audience to vote
- Host UI: hide AudienceReactionBar and PlotTwistVoting in solo mode
(no audience to show reactions from or voting UI to)
- Rename CHAOS button to TWIST in solo mode
- Update toast message for solo ("PLOT TWIST incoming!" vs audience voting)
Voting phase was already skipped server-side for solo mode
(teleprompter.service.ts goes straight to RESULTS).
https://claude.ai/code/session_014T4yzapvSSZdPBJCJgsRkP
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
jackson-jpeg
added a commit
that referenced
this pull request
Jul 29, 2026
…ount The cost section now carries the credit-tier table it was missing. One credit is one round (deductCreditOrReject), lib/credits.ts prices four tiers, so #9 is answerable on data for the first time: every tier covers the $0.053 inference cost, and the original "$0.05-$0.07 per round" estimate was right at the low end. The thin cell is Studio Head sold through Apple IAP — $0.070 a script against $0.053, about a 24% gross margin before hosting. It goes negative around 4000 output tokens, and scripts are ALREADY running 50-80% over their requested line count, so that is not a hypothetical. Web-first materially helps here: Stripe's ~3% against Apple's 30% is the difference between a 1.8x and a 1.3x margin on the tier a heavy host is most likely to buy. Also fixes a stale figure. The packet claimed scripts "came back at 45-64 lines" — carried over from an earlier run, when the current one produced 59-73. A hardcoded number went wrong inside a single regeneration, so the range is now computed from the data it describes. Same class of mistake as the ordering gate earlier today: a number written down beside the thing it measures instead of derived from it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
jackson-jpeg
added a commit
that referenced
this pull request
Jul 29, 2026
…aight man Jackson's rulings, Q1-Q4. Q1 — length. Target 30-38 lines, ceiling 2,600 tokens, over my proposed 35-45. Applied to BOTH branches of generateScript: the host UI sends scriptLength:'standard', so a real game takes the customization path, and capping only the default would have measured clean and shipped inert. The ENSEMBLE x1.25 multiplier is gone with it — it re-inflated standard to 3,250 on exactly the mode being capped, and cast size does not change tokens per line. Three fresh 8-player generations: 38/38/38 lines, none truncated. Output fell 39%; a round went $0.0531 -> $0.0407, and output's share of it 83% -> 56%. SOLO measured too rather than assumed, since every number behind 2,600 was ENSEMBLE. A ceiling reached mid-JSON is a parse failure, not a short script, so stop_reason === 'max_tokens' now throws with a message naming the ceiling instead of surfacing "Unexpected end of JSON input" from the parser. Q2 — DECISIONS #9 closed on measured margins. No repricing. Studio Head is web/Stripe only and gets re-run if it ever ships through Apple IAP; 1.7x on observed mean is not a business. Stripe modelled at 2.9% + $0.30, not a flat 3% — the fixed fee is 6% of a $5 purchase. Q3 — the straight man is no longer characters[0]. The prompt picks it. Q4 — logging kept. Two things found while doing the above, both caught by cross-checking a change against its instrument rather than either against itself: - A FIFTH franchise leak, in the live SOLO prompt: "If the setting is from a known show/universe, USE THOSE CHARACTERS". Every layer passed it — the source audit looks for franchise names and there is no name in that sentence, and layer 3 screens output that the instruction had asked for. A leak does not need a proper noun, it needs permission. New gate reads prompt text for permission-shaped instructions; verified by reintroducing this line and the old "Yoda talks like Yoda" and watching it go red on both. - The harness's script counter keyed on max_tokens >= 5000, calibrated to ENSEMBLE's old 10,000. At 2,600 it counted zero and failed three checks on a working product. Already latent: lightning asks for 2,048. The floor now sits between the populations and asserts its own separation at startup. Also: disable Vercel preview deploys for this branch. The builds fail on the missing Clerk publishable key, and the project cannot serve this product anyway — start is a custom Socket.IO server that Vercel's preset never runs. Suite 450/450, harness 49/49, tsc clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
jackson-jpeg
added a commit
that referenced
this pull request
Jul 30, 2026
plotupdate.md is written for an outsider with no context — what the product is, where it stands, what this session did, and the honest list of what is unverified. Writing it surfaced a real finding: DECISIONS #9's margin table still assumes $0.0407 per round. #14 raised the line budget the same week, recorded the new cost, and never recomputed the tiers. Re-measured at $0.0560; every tier still clears on the web, but Studio Head via Apple IAP is 1.25x. The entry's own tripwire was set to fire on a payment-rail change while the thing that moved was cost — a guard watching one input while the other moves is not a guard. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Redesigned the audience reaction interface from a horizontal button bar to a modern floating action button (FAB) with an expandable popup grid. This improves mobile UX by reducing screen clutter and providing better visual feedback. Also added solo mode support for plot twist voting.
Key Changes
Audience Reaction UI Redesign
Host View (Unchanged)
Solo Mode Plot Twist Support
Code Quality
MOTIONanimation presets from@/lib/animationsfor consistent spring/easingisSoloModeprop toHostPerformingcomponent for mode detectionImplementation Details
AnimatePresencefor smooth enter/exit animationsdelay: i * 0.03for a cascading effectstrokeDasharrayfor smooth ring animationbackdrop-filterwith webkit fallback for iOS supporthttps://claude.ai/code/session_014T4yzapvSSZdPBJCJgsRkP