Triage: open issues, planned fixes & performance findings

Full pass over the repository: every open issue read, baseline verified (install / lint / build / test all green on Node 22, 100% coverage threshold enforced), fixes prepared as minimal draft PRs, and a lightweight performance assessment of the transform hot path.

## Open issues

| Issue | Title | Category | Addressed-by-PR |
|---|---|---|---|
| #1 | Add tests | fixed-in-this-wave (close-candidate) | #103 |
| #3 | Better outline usage with asynciterator in docs | fixed-in-this-wave | #105 |
| #5 | add reset method | fixed-in-this-wave | #104 |

Notes:

- **#1** predates the merged test PR #6; the suite has existed since then and CI enforces 100% statement/branch coverage, so this was arguably already done. #103 adds regression tests for the previously untested cross-chunk escape states (the source even carried a `// Current error is that escapes are not being maintained accross chunks` note). Suggest closing #1 when #103 lands.
- **Not previously filed, fixed in this wave:** that known cross-chunk escape bug had three concrete failure modes, each silently dropping or corrupting output — a string literal closing in a later chunk, a chunk ending with a complete `\\` escape pair, and an object continuing past an escaped chunk boundary. All fixed in #103 and verified with a 2000-trial randomized round-trip fuzz (old code fails 1957/2000 trials, fixed code 0/2000).

## Performance findings

Microbenchmark: 4.8 MiB of realistic telemetry-style concatenated JSON, 16 KiB chunks, Node 22, median of 9 interleaved runs (2-core box shared with a live server — absolute numbers noisy, relative deltas stable across runs):

| variant | throughput | status |
|---|---|---|
| current scanner (post-#103) | ~51 MiB/s | baseline |
| `charCodeAt` integer compares | ~59 MiB/s (+14–23% across runs) | **PR #106** (stacked on #103) |
| `indexOf` string-literal skip + backslash-run parity | ~91 MiB/s (~1.9x) | prototype only, fuzz-clean, not PR'd |

- The hot path is the per-character scan; replacing `str[i]` single-character string comparisons with `charCodeAt` integer comparisons was the cheapest win (#106 — control flow unchanged).
- The bigger win is skipping string literals with `indexOf('"')` and deciding escapedness by counting the backslash run before each candidate quote (parity is equivalent to the current pair-stepping walk, including at chunk boundaries). The prototype passes the same 2000-trial fuzz, but it restructures the scanner and has subtle edge cases (empty chunk while mid-escape, runs truncated by chunk starts), so it deserves deliberate review rather than a drive-by change. Happy to turn it into a follow-up PR on request.
- Everything else is already lean: output is produced by `slice`/concat only, no regexes, no sync IO, no per-item allocations beyond the emitted strings.

## Needs your decision

- **Stray top-level `}` handling** — a `}` outside any object drives `nesting` negative, after which the *next* object's opening `{` no longer registers (nesting goes −1 → 0) and that object is silently lost. Options: clamp (`if (nesting > 0)`, treating a stray `}` like inter-object noise) or document as garbage-in/garbage-out. Either way changes behaviour for malformed input, so it was not fixed in this wave.
- **Top-level JSON arrays / primitives** — `[{...},{...}]` happens to emit the inner objects (brackets are skipped like whitespace), but a top-level array of primitives emits nothing. #105 documents objects-only as a limitation; decide whether array support is in scope.
- **`indexOf` scanner (~1.9x)** — see performance findings; want it as a follow-up PR stacked on #106?
- **Dependency staleness** — 10 dependabot PRs are open (some from 2022) and GitHub reports 58 vulnerabilities on the default branch, all in dev/tooling deps (jest 27, TypeScript 4.1, semantic-release 17 era). Probably worth one consolidated tooling-refresh pass instead of merging the piecemeal bumps.

Review timing: prepared with Claude; @jeswr will personally review before it progresses — expect active review Wed-Fri.

Review this triage issue alongside PRs #103–#106 and the repository's install, lint, build, and test commands. The work is not a single newcomer-sized change: completion requires maintainer decisions on malformed input, top-level arrays, the proposed scanner optimization, and dependency updates, followed by focused PRs and preserved coverage and benchmark results.

Written by the indexing model from the issue text.