feat(web): /calc expression evaluator -- pipeline-ab card 2 Arm A (BASELINE)

[eksno]

EXPERIMENT PR (card 2: harder, separating) -- DO NOT MERGE. pipeline-ab Arm A (BASELINE).

Baseline feature layout: Opus hands + full mid-process review loop + George QA + review-reconcile.

Correctness-deep card (expression parser/evaluator) chosen to separate Composer-hands-vs-Opus-hands. Both arms build from ONE shared plan. Stays OPEN for eyeball evaluation against its live Railway preview.

---

^{Need help on this PR? Tag /codesmith with what you need. Autofix is disabled.}

[railway-app]

🚅 Deployed to the testing-pr-11 environment in testing

Service	Status	Web	Updated (UTC)
testing	✅ Success (View Logs)	Web	Jun 12, 2026 at 8:31 pm

[coderabbitai]

Warning

Review limit reached

@eksno, we couldn't start this review because you've reached your PR review rate limit.

More reviews will be available in 1 minute and 32 seconds. Learn how PR review limits work.

Your organization has run out of usage credits. Purchase more credits in the billing tab to continue.

⌛ How to resolve this issue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

We recommend that you space out your commits to avoid hitting the rate limit.

🚦 How do rate limits work?

CodeRabbit enforces hourly rate limits for each developer per organization.

Our paid plans include higher PR review limits than trial, open-source, and free plans. In all cases, reviews become available again over time. During sustained high-volume PR review activity, CodeRabbit may temporarily slow when the next review becomes available.

Please see our Fair Usage Limits Policy for further information.

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: d5593b03-f548-4621-b512-3319118aa442

📥 Commits

Reviewing files that changed from the base of the PR and between c20b967 and a5a83ea.

⛔ Files ignored due to path filters (1)

• web/package-lock.json is excluded by !**/package-lock.json

📒 Files selected for processing (25)

• docs/fixes/0002-eslint-svelte-ts-runes-module-parser-gap.md
• web/package.json
• web/src/lib/calc/calculate.test.ts
• web/src/lib/calc/calculate.ts
• web/src/lib/calc/evaluate.test.ts
• web/src/lib/calc/evaluate.ts
• web/src/lib/calc/format.test.ts
• web/src/lib/calc/format.ts
• web/src/lib/calc/index.ts
• web/src/lib/calc/parse.test.ts
• web/src/lib/calc/parse.ts
• web/src/lib/calc/tokenize.test.ts
• web/src/lib/calc/tokenize.ts
• web/src/lib/calc/types.ts
• web/src/lib/clipboard.test.ts
• web/src/lib/clipboard.ts
• web/src/lib/components/site-header.svelte
• web/src/lib/components/toaster.svelte
• web/src/lib/toast-core.test.ts
• web/src/lib/toast-core.ts
• web/src/lib/toast.svelte.ts
• web/src/routes/+layout.svelte
• web/src/routes/calc/+page.svelte
• web/src/routes/calc/+page.ts
• web/vitest.config.ts

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/calc-arm-a

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[eksno]

Final review (advisory -- not a merge gate)

calc Arm A (BASELINE) -- advisory red-team review

Worktree: /shared/repos/startino/.worktrees/calc-arm-a (web/)
Diff base: origin/alpha...HEAD (commit a5a83ea)
Status: ADVISORY -- does NOT block ship. tests/check/build pre-verified green.

Verdict: SHIP. The implementation is a faithful, high-fidelity realization of
calc-plan.md. Parser correctness is CLEAN across the pinned floor plus the
red-team edge cases I traced by hand. No act-worthy code fixes were found; the
only triage item is the docs/fixes/0002 mis-filing, judged keep-flagged (see
disposition). One real, reproducible defect exists OUTSIDE the feature's
acceptance gate: npm run lint (eslint half) fails on toast.svelte.ts -- this
is a pre-existing repo eslint-config gap, correctly diagnosed in 0002, with the
fix deferred because eslint.config.js is DO-NOT-TOUCH for this work.

---

Lens 1 -- Spec conformance (grammar / semantics / error-taxonomy / formatting)

PASS. Hand-traced every anti-ambiguity item in plan section 9 against the code:

• ^ right-assoc (2^3^2 -> binary(^,2,binary(^,3,2))): parsePower recurses
  the RIGHT operand through parseUnary -> correct (parse.ts:74-84). Tested.
• Unary minus looser than ^ (-2^2 -> unary(-, binary(^,2,2))): parseUnary
  consumes - then descends to parsePower which consumes the whole 2^2 as
  operand -> correct. AST shape test asserts it (parse.test.ts:13-24).
• Exponent may be unary-negated (2^-2 = 0.25): RIGHT operand goes through
  parseUnary -> correct.
• No implicit multiplication: 2(3)/2 3 -> trailing-token Unexpected token
  at top-level (parse.ts:136-138). Correct.
• Number lexing (.5, 3., 1.5e3, 2E-2; reject bare . and 1e): sticky
  regex /(?:\d+\.?\d*|\.\d+)(?:[eE][+-]?\d+)?/y + the .-needs-trailing-digit
  guard (tokenize.ts:6,29). 1e matches just 1, leaves stray e ->
  Unexpected character "e". Correct and tested.
• Division-by-zero guard (incl 0/0) in evaluate (evaluate.ts:25-27); non-finite
  final result gated in calculate (calculate.ts:17-20). Correct.
• formatNumber: -0/0 -> '0'; round to 12 sig digits; exp branch for
  abs>=1e21 || abs<1e-6; trailing-zero normalization of the mantissa
  (format.ts). 0.1+0.2 -> '0.3', 1e21 -> '1e+21', 0.0000005 -> '5e-7'
  all correct.
• Error taxonomy: every pinned string is produced at the right site. Empty
  pre-check, unexpected char (lex), unexpected token / incomplete / mismatched
  (parse), division by zero (eval), not-finite (calculate), defensive
  Invalid expression / Empty expression. All asserted by calculate.test.ts
  and the unit suites.

SPEC GAP (informational, NOT a code defect): plan line 353 contains an aside --
"a bare ) with no opener is Unexpected token". In this parser a LEADING bare
) reaches parsePrimary at a primary position and yields Incomplete expression (parse.ts:117-118), not Unexpected token. This is an internal
inconsistency in the SPEC: the pinned, tested () case (plan line 356, test
calculate.test.ts:156-158) REQUIRES rparen at primary position to yield
Incomplete expression. The implementer correctly satisfied the pinned test;
the aside is unreachable without positional context that the grammar does not
carry. Leading bare ) is not in the test floor, so behavior is undefined by
the floor and the chosen message is defensible. No action -- changing it would
break the pinned () case. Both arms will share this; it is a measurement
non-issue for the A/B.

---

Lens 2 -- Standards (Svelte 5 runes, cn(), prettier, ASCII, imports)

PASS.

• Svelte 5 runes used correctly: $state, $derived, $effect (page +
  toaster + toast.svelte.ts). Module-scope runes live in .svelte.ts per
  convention. $props() in layout untouched-style.
• $effect cleanup return in toaster.svelte:6-8 is the canonical unmount-leak
  pattern.
• Intra-$lib imports use the .js extension and index.js barrels, matching
  the repo's existing convention.
• ASCII house style honored: -- (not em-dash) in copy and comments; straight
  quotes; tabs; single quotes; semicolons. No non-ASCII introduced.
• No cn() misuse -- conditional class arrays use Svelte's native
  class={[...]} (matches existing site-header.svelte style); inline strings
  elsewhere. Acceptable; plan said "use cn() IF combining conditional classes,
  otherwise inline" -- the array form is idiomatic here.
• prettier --check passes (pre-verified). eslint FAILS (see Quality / 0002).

---

Lens 3 -- Quality (parser correctness, theme, leaks, SSR, dead code)

PARSER CORRECTNESS: CLEAN. Hand-traced beyond the floor:

• (2+3)4 -> Unexpected token (no implicit mult). Correct.
• - alone / 2+ -> Incomplete expression. Correct.
• 2**3 -> RHS parseUnary->parsePower->parsePrimary sees * -> Unexpected
  token. Correct.
• ((1+2)) nested + whitespace -> 3. Correct.
• Right-assoc + unary-exponent interaction (2^-3 = 0.125) correct.
• No off-by-one in the sticky-regex cursor advance (i += text.length).
• Division-by-zero short-circuits BEFORE producing Infinity/NaN, giving the
  specific message; overflow (1e308*10) reaches the calculate-level not-finite
  gate. The two paths are correctly separated.

THEME: No hardcoded hex/rgb anywhere in new files. Toaster + page use only
token utilities (bg-card, text-card-foreground, border-border, text-destructive,
text-muted-foreground, text-foreground, bg-muted). Reads correctly in both
modes by construction.

LEAKS: Toast auto-dismiss timers tracked in a Map keyed by id; schedule()
clears a prior timer for the same id (idempotent re-schedule); clearAll() on
viewport unmount via $effect cleanup. The leak probe (toast-core.test.ts)
asserts clearAll prevents a pending callback from firing and pending() returns
to 0. Solid. The viewport is mounted ONCE globally, so clearAll fires only on
app teardown -- acceptable and matches plan intent.

SSR / PRERENDER: calculate('') is pure and SSR-safe; input starts empty ->
empty-state render. copyToClipboard SSR-guarded (typeof navigator + clipboard +
writeText checks). No window/document/navigator at module scope or in render --
only in event handlers. /calc/+page.ts restates prerender = true. Clean.

DEAD CODE: none material. evaluate.ts inner switch has no default and the
function relies on TS exhaustiveness of the op literal union to return on all
paths -- svelte-check is green, so the exhaustiveness holds; not dead code.

DOCS/FIXES/0002: see disposition. Diagnosis verified reproducible
(npx eslint src/lib/toast.svelte.ts -> Parsing error: Unexpected token interface, exit 1). Root cause (eslint-plugin-svelte **/*.svelte.ts glob
lacks the tseslint sub-parser) is correct. The fix-entry is arguably mis-filed
per "Log fixes only" (it records a DEFERRED gap, Commit: n/a) -- flagged, NOT
removed (removal would destroy a correct, reproducible, high-value diagnosis;
the fixes-format itself anticipates n/a for no-code-change entries).

---

Act-worthy fixes applied

NONE. No cheap-clearly-correct behavior-preserving code defect was found. The
code is spec-conformant and the gate is green; touching anything would be
churn, not improvement.

Left (judgment calls)

1. docs/fixes/0002 mis-filing -- KEPT, flagged. Removal is destructive (loses a
   correct reproducible diagnosis), not "clearly-correct cleanup". The real
   follow-up is the one-line eslint.config.js widening, which is DO-NOT-TOUCH
   for this feature work.
2. Leading bare ) yields Incomplete expression vs the spec aside's
   Unexpected token -- spec internal inconsistency; the implementer correctly
   satisfied the pinned () test instead. No action.
3. npm run lint (eslint) currently fails on toast.svelte.ts -- pre-existing
   repo config gap, OUTSIDE this feature's acceptance gate (test/check/build/
   format). Surfaced in the PR body as a known follow-up.

[eksno]

calc Arm A -- review disposition (PR-ready)

ADVISORY review. Does NOT block ship. tests/check/build verified green
(pre-existing) -- not re-run since no code fix was applied.

Outcome: SHIP. No code changes applied.

The Arm A (baseline) implementation faithfully realizes calc-plan.md. Parser
correctness is clean across the pinned test floor and the additional red-team
edge cases hand-traced during review. No cheap-clearly-correct code fix was
warranted, so the working tree is unchanged and no review commit was made.

Findings -> disposition

#	Lens	Finding	Disposition
1	Quality/infra	npm run lint (eslint half) fails on web/src/lib/toast.svelte.ts: Parsing error: Unexpected token interface. Reproduced (npx eslint exit 1). Pre-existing repo eslint-config gap; eslint.config.js **/*.svelte.ts glob lacks the tseslint sub-parser.	LEFT. Outside this feature's acceptance gate (test/check/build/format). Fix is a one-line eslint.config.js widening -- DO-NOT-TOUCH here. Surface in PR body as known follow-up.
2	Quality/docs	docs/fixes/0002 documents a DEFERRED, un-fixed gap (Commit: n/a); arguably mis-filed per CLAUDE.md "Log fixes only".	LEFT (kept-flagged). Removal rejected: the entry is a correct, reproducible, high-value root-cause diagnosis; deleting it destroys debugging knowledge the fixes-log exists to preserve, and the fixes-format itself anticipates n/a for no-code-change entries. Not a "clearly-correct cleanup" -- it is a destructive judgment call. Flagged in PR instead.
3	Spec	Leading bare ) yields Incomplete expression, contradicting the plan's informal aside (line 353) that a bare ) is Unexpected token.	LEFT. SPEC internal inconsistency: the pinned, tested () case (line 356) REQUIRES rparen-at-primary -> Incomplete expression. Implementer correctly satisfied the pinned test. Not floor-tested; chosen message defensible. Shared by both arms -- A/B-neutral.

docs/fixes/0002 decision (explicitly requested)

KEPT, FLAGGED. Did NOT remove it. Rationale: the task offered removal as an
OPTIONAL cleanup ("if you judge it a cheap clearly-correct cleanup"). I judge it
is NOT -- it is destructive (the entry holds a verified, reproducible diagnosis
of a live lint failure with the exact one-line fix), and the fixes-format
explicitly accommodates Commit: n/a. The mis-filing is borderline, not
clearly-correct. Higher-value disposition: keep the diagnosis, surface the lint
gap + the one-line eslint.config.js follow-up in the PR body.

PR body -- suggested known-issues / follow-ups block

Known follow-up (out of scope for this feature; does not block):
npm run lint (eslint half) fails on src/lib/toast.svelte.ts with
Parsing error: Unexpected token interface. This is a pre-existing repo
eslint-config gap -- eslint-plugin-svelte's **/*.svelte.ts glob is not
wired to the typescript-eslint sub-parser, so the first module-scope-runes
file in the repo trips it. Fix is a one-line widening of
web/eslint.config.js (files: ['**/*.svelte', '**/*.svelte.ts', '**/*.svelte.js']), deferred because that file is DO-NOT-TOUCH for the
/calc work. Diagnosis recorded in docs/fixes/0002. The feature's own gate
(test/check/build/prettier) is green.

Gate after fix

n/a -- no fix applied; pre-existing green gate stands.

[eksno]

Known issues from QA (George -- baseline arm exploratory pass)

Triaged unblock-with-note (kept visible for the operator; not auto-fixed, so the two experiment arms stay comparable). Note: the parser/evaluator core is CORRECT on all edge cases -- these are UX/rendering issues a code review does not surface:

1. Result rendered TWICE (worst) -- every valid expression shows a large bold = N live result AND a smaller = N badge directly below it; looks like a rendering glitch. NOTE: this is a SHARED-PLAN flaw -- the plan specified both a live-result element and a result badge -- so it is present in BOTH experiment arms; only this arm's QA surfaced it.
2. Error rendered twice -- inline red text + a red badge below, for every error state.
3. Long history rows -- a long expression eats the row width and pushes the Copy button / result off.
4. Mobile @375px -- the "Compute" button label disappears (icon only); purpose unclear.
5. Compute button has no keyboard focus ring -- every other control does.
6. (Confirmed OK) history cap of 10 evicts correctly; 0^0=1 (JS semantics); 9^9^9 correctly errors as non-finite rather than showing Infinity.

These are the baseline arm's QA yield -- intentionally left visible as part of the experiment artifact.