v0.next — complete product pass: history nav, usage transparency, long-session perf

[lroolle]

The complete pass

v0.next is not a patchwork release. It's one coherent product pass across history navigation, usage transparency, and long-session performance. Four issues, one milestone, one quality bar.

The bar: ship the smallest thing that fully solves one user problem. Then the next. No dangling threads when tying them off is cheap. Tests, docs, and polish are part of "done", not optional extras. When a real fix is within reach, don't ship a workaround.

The work

• feat(web): per-turn token breakdown with cost attribution (surpass ccusage) #2 — Per-turn token breakdown with cost attribution (surpass ccusage) — P0 crown jewel
• bug(web): hash/deep-link navigation silently fails inside collapsed load-earlier sections #3 — Hash/deep-link navigation into collapsed load-earlier sections auto-expands — P1 afternoon fix
• perf(web): profile-driven rewrite of the long-session render path #4 — Long-session loading performance: profile-driven rewrite — P1, gated on feat(web): per-turn token breakdown with cost attribution (surpass ccusage) #2
• feat(web): left-side time-machine vertical timeline nav rail #5 — Left-side time-machine vertical timeline nav rail — P2, can slip to v0.next+1

Why these four, and why together

Each has its own shipping contract, but they share a story:

• feat(web): per-turn token breakdown with cost attribution (surpass ccusage) #2 answers the feat: multi-provider support (Claude Code + Codex) #1 user question right now: "where did my quota go?" It's the reason the others matter.
• bug(web): hash/deep-link navigation silently fails inside collapsed load-earlier sections #3 is a small UX bug that becomes glaring the moment deep links get shared from the breakdown in feat(web): per-turn token breakdown with cost attribution (surpass ccusage) #2. Ship it alongside, not after.
• perf(web): profile-driven rewrite of the long-session render path #4 is the performance work that turns feat(web): per-turn token breakdown with cost attribution (surpass ccusage) #2 from "interesting for medium sessions" into "usable for 2000-message monsters where people actually lose track of cost." Gated on feat(web): per-turn token breakdown with cost attribution (surpass ccusage) #2 so we measure what matters, not what feels slow in the abstract.
• feat(web): left-side time-machine vertical timeline nav rail #5 is the visible proof that the release wasn't a patch. P2 because its absence doesn't hurt anyone; its presence is what makes the release unmistakably done.

Evidence summary

Current-state citations live in each sub-issue. High-level snapshot:

• Token usage: internal/parser/types.go:63-78, internal/parser/session.go:42-52, internal/web/templates.go:470-488 — session-level aggregate only, per-message usage thrown away during parse.
• Load-earlier navigation: internal/web/templates.go:617, internal/web/templates.go:5487-5506, internal/web/server.go:303 — hash handler has no branch for hidden ranges.
• Render path: internal/web/server.go:283, internal/web/templates.go:547-562 — full parse + full render on every request; progressive loading is CSS-hide, not real lazy render.
• Timeline rail: doesn't exist today. internal/web/templates.go:317-327 is the outline sidebar, not a time axis.

Prior art for the token work: ccusage. Table-only CLI, no per-turn breakdown, no web UI, several accuracy bugs around sidechains and streaming partials (#913, #938, #934, #933, #935). ccx's tree model and web UI are the natural leapfrog surface.

Ships when

Milestone ships when #2, #3, #4 are green. #5 can slip to v0.next+1 without holding the release.

Every sub-issue ships with:

• Tests covering the new logic (not I/O plumbing)
• CHANGELOG entry
• README / screenshot update where UX-visible
• No dangling ?all=1 workaround where a real fix exists

Not in this milestone

• Real-time live-tail cost during active sessions (needs streaming parse — separate issue)
• Session replay / playback animation
• MCP server mode
• Full-text cross-session search index
• Any auth / multi-user mode

We are not boiling the ocean. We are finishing a release.