feat(rows): say it once, and only when you know it

CHANGELOG.md

@@ -2,6 +2,58 @@
 
 ## Unreleased
 
+## v0.32.0 — 2026-08-23 — say it once, and only when you know it
+
+**Row 3 was restating row 2 at greater length.** The flash reader
+carried the pinned notice's key and never compared against it, so at
+session start the same sentence arrived twice, once short and once
+long: `- 12✕5h left · 5.8%/win` over `- budget ~12✕5h left · even
+5.8%/win · heading ~85%`. The function's own comment already promised
+it "never echoes the pin's own sentence" — now it skips record one and
+starts at record two, and row 3 shows the highest-ranked notice the pin
+is *not* already carrying. A calm start is two rows, the way it was
+supposed to be.
+
+**A 7d projection now needs a day of this window's own evidence.**
+Minutes after a weekly rollover the learned walk was projecting last
+week's Tuesday onto a pool 2% spent — measured: a 2.4h-old window with
+2% used came back `red, dry in 95h`, painted `×` across slot 14
+onward, and put `heading ~100%` on the budget line. Every input for
+that verdict predates the reset, including the trailing-24h blend the
+walk opens with, which describes a day lying on the far side of it.
+`SEVEN_DAY_YOUNG_SECS` (86400) silences the learned walk, the linear
+at-risk pace, and the dry-cell fallback until the window is a day old
+— one gate, so the forecast notices, the `×` cells, the heading, the
+accuracy logger and both subcommands go quiet together. Badges still
+report real percentages, and a young window that is genuinely spent
+still goes red on its own 85%+: the guard mutes pace, never facts. The
+5h window is untouched.
+
+**Pace waits for a fraction of its window, not a fixed 15 minutes.**
+One gate served both strips: 900 seconds, which is 5% of a 5h window
+and 0.1% of a week. An hour into a fresh week, 2% of the pool over
+0.6% of the time rendered `3.0✕` in red. The gate is now length/20 —
+5h still waits exactly 15 minutes, 7d waits ~8.4h, the same fraction
+`seven_day_elapsed` has always called the noise floor.
+
+**The fold token counts windows, not cells: `...▯(✕12)`.** The old
+`...▯5h✕10` counted hollow cells the fold hid, while the budget line
+beside it priced `12✕5h left` — two arithmetics on one row, and the
+reader was left to arbitrate. The 34-cell grid spans 170h against a
+168h period, which is where the two differed. The token is now
+parenthesized, drops the redundant `5h` unit, and counts the 5h
+windows remaining to the true reset: the same number the budget line
+prices, from the same instant. Dry tails still read `...×(✕12)` — `×`
+the cell, `✕` the operator, one column apart and never the same mark.
+
+**The budget voice drops its sigil.** `!` and `+` mark a notice that
+interrupts — you are about to hit something, or there is capacity to
+take. The week's resting reading interrupts nothing, and a leading `-`
+on a dim line reads as a bullet, which made rows 2 and 3 look like a
+two-item list. The pin now reads `12✕5h left · 5.8%/win`. Dim is the
+mark; the sentence carries itself. `--check` and `--week` still label
+their long form with the word `budget`.
+
 ## v0.31.0 — 2026-08-20 — the operator is not a reading
 
 **The ledger row was printing two different X's and calling them both

DESIGN.md

@@ -82,25 +82,27 @@ own signal.
 ## The ledgers (row 2)
 
 ```
-5h ▃▄▮▯▯ 0.6x @04:00  7d ▅▁▂ ▃▅ˍ▃▅ ▃▃▁▂▁ ▮▯▯...▯5hx14 0.7x @Wed 09:00
+5h ▃▄▮▯▯ 0.6x @04:00  7d ▅▁▂ ▃▅ˍ▃▅ ▃▃▁▂▁ ▮▯▯...▯(x14) 0.7x @Wed 09:00
 ```
 
 One grammar, two scales. `5h` = this window as 5 hours; `7d` = the
 period as 34 five-hour slots, oldest left, a gap at each local midnight
 *in history only* — the run from `▮` on is contiguous. History draws in
 full; the live row folds the 7d future after two kept cells into
-`...▯5hx14` — the slots left to the reset, all alike, `×` red when the
-tail projects dry (the `week` report still draws every slot). Each strip ends
-with its pace (used ÷ elapsed; dim <1x, pressure ≥1x, hidden under 15 min)
-and the reset its right edge is — axis labels, not restated badges.
+`...▯(x14)` — the 5h windows left before the reset, all alike, `×` red
+when the tail projects dry (the `week` report still draws every slot). The
+count is the budget line's own, priced from the same instant. Each strip ends
+with its pace (used ÷ elapsed; dim <1x, pressure ≥1x, hidden until 5% of the
+window has run) and the reset its right edge is — axis labels, not restated
+badges.
 
 ```
 ▁▂▃▄▅▆▇█   burned; height = points that cell cost (▁ ≤2 … ▅ ≤11 … █ >20)
 ˍ          ran, negligible — a bar of height zero, on the baseline
 ░          unknown — no sample; never drawn as idle
 ▮          now
 ▯          ahead — the hollow of ▮
-...▯5hx14  the folded future: 14 more 5h slots to the reset (live 7d row)
+...▯(x14)  the folded future: 14 more 5h windows before the reset (live 7d row)
 ×          pace won't cover it (7d: learned forecast, linear when cold; 5h: linear)
 ```
 

README.md

@@ -91,7 +91,7 @@ Every component earns its place:
 | User tier | Neutral white-weight (MAX bold, PRO normal, dim otherwise) — identity, never a status color. Truncated display name. |
 | Quota | Integer percentages. The 5h badge always carries its reset time while a window is live — `5h[42%@14:30]` reads "42% used, resets at 14:30" — because on a 5h horizon the reset is the number you plan the current sitting around. Wall-clock, not a countdown, on purpose: Claude Code only re-renders the statusline on activity, so a relative "@1h38m" silently decays into a lie during idle gaps, while "@14:30" stays true in a frozen frame. (The 7d badge is hybrid: day-relative `@5d` while the reset is >= 24h out — decays one day per day, mild and narrow — switching to the same wall-clock `@04:00` inside the last day, where an `@6h`/`@<1h` countdown decayed by the hour exactly when pressure keeps the suffix visible.) When a window's utilization climbs between renders, a reverse-video `+N` token appears right after the badge for ~60s: `5h[44%@14:32]+2` means "you just burned 2%". A drop (window reset) stays quiet — the fresh low number is its own signal. **7d is forecast, not leveled**: a learned per-weekday burn profile (EWMA over your own usage history) plus your recent 24h burn project whether the quota outlasts the window — your heavy Tuesday counts more than a generic average. The verdict is color alone; under pressure the badge shows when relief arrives: `7d[44%@5d]` red means "at your pace, dry days before the reset 5 days from now"; inside the last day it reads `7d[92%@04:00]` — resets at 04:00. Cold start (<14 days history) falls back to window-average pacing. Recovery color when reset is imminent. **Model-scoped weekly quota**: when the usage API carries a per-model weekly limit (`limits[]`, `kind=weekly_scoped`) for the model your session is running, it renders right after the model+context block — `fabl5[1m][12%] fb[67%]` on a Fable 5 session, `op[33%]` on Opus — because the quota is a property of the model you're running, not of the account-wide 5h/7d cluster. It's a weekly number (same reset as the 7d badge), scoped to one model. Other models' scoped quotas stay hidden: only the limit constraining *this* session is signal. Supersedes the legacy `seven_day_opus`/`seven_day_sonnet` fields, which the API now sends as null. |
 | Extra usage | Monthly spend, limit, prepaid balance. `--extra auto` shows when quota runs out. |
-| Week row | **A row of its own, under the badges**: `5h ▅█▃▮▯ 0.9✕ @23:00  7d ▅▁▂ ▃▅ˍ▃▅ …▮▯▯...▯5h✕19 0.7✕ @Wed 09:00` — this sitting by the hour, the week as its 5h windows (day-gapped, the far future folded to a counted `...▯5h✕19`), height = what each cell burned, `▮` now, `×` where the pool runs dry, each strip ending with its pace and reset. Reconstructed from your own usage log; `auto` shows it once there is history to show. See [Week row](#week-row). |
+| Week row | **A row of its own, under the badges**: `5h ▅█▃▮▯ 0.9✕ @23:00  7d ▅▁▂ ▃▅ˍ▃▅ …▮▯▯...▯(✕19) 0.7✕ @Wed 09:00` — this sitting by the hour, the week as its 5h windows (day-gapped, the far future folded to a counted `...▯(✕19)`), height = what each cell burned, `▮` now, `×` where the pool runs dry, each strip ending with its pace and reset. Reconstructed from your own usage log; `auto` shows it once there is history to show. See [Week row](#week-row). |
 | Deadman | **Invisible until a switch is armed.** Surfaces [deadman](https://github.com/thevibeworks/deadman) — a dead man's switch that hands the session off when you stop responding. `[☠ armed 42m]` (dim) counts down to the auto-handoff; `[☠ warned 3m]` (yellow) means the phone warning went out; `[☠ due]` means the handoff fires imminently. Sits on the left lane next to the path — it describes this session's lifecycle, not a quota. One `command -v` when the tool is absent, one fast file read when present; nothing armed renders nothing. `--deadman off` disables it. |
 | Cache health | **Quiet until it bites.** Claude Code never re-renders an idle session, and while you work the prompt cache is always freshly ~1 TTL from expiry — so a proactive "expiring soon" isn't honestly observable, and `auto` spends no width on it. It speaks only when a rewrite actually happens: `≡!419k` the instant you resume onto a dead cache (idle longer than the TTL) or a mid-session prefix collapse — a 419k-token re-cache at ~20x the read rate (and the same burn on your 5h/7d quota on subscriptions). **Bold red past 200k** — the premium-band miss. `≡~` while a large prefix rebuilds. `--cache always` additionally keeps the freeze-safe deadline `≡@15:20` (last request + TTL; a past time in a frozen frame reads "expired at 15:20"). TTL defaults to 1h (claude.ai subscriber sessions) or 5m (API-key auth); an observed usage breakdown overrides it. The `≡` glyph (U+2261) reads as stacked cache layers — one terminal column, quiet and distinct. |
 
@@ -255,7 +255,7 @@ right edge shared with line 1:
 
 ```text
 proj (main*)       +84/-14 8m $6.72 fabl5[1m][██░░42%] fb[66%] [MAX|@work] 5h[38%@23:00] 7d[39%]
-- 3✕5h left · 20%/win   5h ▅█▃▮▯ 0.9✕  7d ▅▁▂ ▃▅ˍ▃▅ ▃▃▁▂▁ ▅ˍ▂▁▁ ˍˍˍ▃▅ ˍˍˍ▅ ▆▆ˍ▂▮▯▯ 0.7✕ @Wed 09:00
+3✕5h left · 20%/win   5h ▅█▃▮▯ 0.9✕  7d ▅▁▂ ▃▅ˍ▃▅ ▃▃▁▂▁ ▅ˍ▂▁▁ ˍˍˍ▃▅ ˍˍˍ▅ ▆▆ˍ▂▮▯▯ 0.7✕ @Wed 09:00
 ```
 
 The 5h strip prints no reset: line 1's `5h[38%@23:00]` already carries
@@ -273,12 +273,12 @@ edge and the full sentence sits flush-left beneath them:
 ```text
 proj (main*)   fabl5[1m][██░░42%] fb[66%] [MAX|@work] 5h[38%@23:00] 7d[39%]
    5h ▅█▃▮▯ 0.9✕ @23:00  7d ▅▁▂ ▃▅ˍ▃▅ ▃▃▁▂▁ ▅ˍ▂▁▁ ˍˍˍ▃▅ ˍˍˍ▅ ▆▆ˍ▂▮▯▯ 0.7✕ @Wed 09:00
-   - budget ~3✕5h left · even 20%/win · heading ~52%
+   budget ~3✕5h left · even 20%/win · heading ~52%
 ```
 
 Each strip ends with its **pace** (used ÷ elapsed: `0.7✕` is on track,
-`1.6✕` caps early — dim below 1✕, pressure-tinted from 1✕, hidden for
-the first 15 min of a window) and the **reset** its right edge stands
+`1.6✕` caps early — dim below 1✕, pressure-tinted from 1✕, hidden until
+5% of the window has run: 15 min for 5h, ~8h for the week) and the **reset** its right edge stands
 for (`@23:00` inside 24h, `@Wed 09:00` beyond) — axis labels for a
 timeline, not badges restated. When the row shows, the advisor's calm
 budget line shows with it (windows left, what even looks like, where
@@ -292,9 +292,11 @@ fire.
   left, the last a 3h stub), height = the 7d points that window burned,
   with a thin gap at each local midnight so days read as clusters — and a
   day that held five windows shows it — without a ruler. History draws in
-  full; the future folds: two hollow cells after `▮`, then `...▯5h✕28` —
-  28 more 5h slots to the reset, all alike (`×` red when the tail projects
-  dry). The `week` subcommand's wide ledger still draws every slot.
+  full; the future folds: two hollow cells after `▮`, then `...▯(✕28)` —
+  28 more 5h windows before the reset, all alike (`×` red when the tail
+  projects dry). That count is the budget line's own `~28✕5h left`, priced
+  from the same instant: one row, one arithmetic. The `week` subcommand's
+  wide ledger still draws every slot.
 
 | Cell | Meaning |
 |------|---------|
@@ -304,8 +306,8 @@ fire.
 | `▮` | the cell you are in now |
 | `▯` | a cell still ahead of you — the hollow of `▮`, an empty slot waiting |
 | `×` | a cell the pool will not cover at the current pace (7d: the learned forecast's dry point, linear when untrained; 5h: linear, the same projection as the badge) |
-| `...▯5h✕28` | the folded 7d future: 28 more 5h slots to the reset, one token instead of 28 hollow cells (`×` red when the tail ends dry) |
-| `✕` | not a cell — the multiplication sign, the row's one operator (`...▯5h✕28`, `0.7✕`, `- 19✕5h left`). Deliberately not `×` (U+00D7), which is already a reading: cells are the ink, the operator is punctuation, and `...×5h✕28` has to say both at once. One terminal column and no emoji fallback, so the row still meets line 1's edge; override with `MULT_GLYPH` (`╳` and `✖` look stronger but are ambiguous-width and emoji-presentation respectively) |
+| `...▯(✕28)` | the folded 7d future: 28 more 5h windows before the reset, one token instead of a run of hollow cells (`×` red when the tail ends dry). The count is windows-to-reset — what the budget line prices — not the number of cells the fold happened to hide |
+| `✕` | not a cell — the multiplication sign, the row's one operator (`...▯(✕28)`, `0.7✕`, `19✕5h left`). Deliberately not `×` (U+00D7), which is already a reading: cells are the ink, the operator is punctuation, and `...×(×28)` has to say both at once. One terminal column and no emoji fallback, so the row still meets line 1's edge; override with `MULT_GLYPH` (`╳` and `✖` look stronger but are ambiguous-width and emoji-presentation respectively) |
 
 Burn cells take their badge's pressure color; everything else is neutral,
 so the row never adds an alarm channel of its own. Both strips are
@@ -365,7 +367,7 @@ clauses about one window can never disagree.
 | `+ 5h ~40m left · 70% unused` | Said only when the *week* is stranding capacity — an unspent 5h window is otherwise headroom, not waste, since the 5h window is a rate limit and not a budget. |
 | `+ ~62% will expire · go heavier` | On pace to strand a large chunk of the subscription. Speaks only to an engaged, unsqueezed session. |
 | `+ work 5h[8%] free` | A sibling account in the same shared home is idle while this one is pinned. |
-| `- 19✕5h left · 1.1%/win` | The calm budget: runway, what even looks like, where you land (long form). |
+| `19✕5h left · 1.1%/win` | The calm budget: runway, what even looks like, where you land (long form). It wears no sigil — `!` and `+` interrupt, the week's resting reading does not. |
 
 `--notice off` keeps row 3 quiet; `--advisor off` silences both. `--check`
 and `--week` print the long form, since a terminal command has a whole
@@ -509,7 +511,7 @@ run. Setting only `CLAUDE_CACHE_DIR` keeps the legacy single-dir behavior.
 npm exec --yes bats -- t/
 ```
 
-418 tests across `t/statusline.bats` (406 statusline + integration) and
+423 tests across `t/statusline.bats` (411 statusline + integration) and
 `t/install.bats` (12 installer). CI runs on push and PR to `main`.
 
 ## Project Structure