Skip to content

docs(README): first-run expectations, trim a repeat, qualify priority claim - #6

Merged
windaddict merged 1 commit into
mainfrom
docs/readme-accepted-fixes
Aug 9, 2026
Merged

docs(README): first-run expectations, trim a repeat, qualify priority claim#6
windaddict merged 1 commit into
mainfrom
docs/readme-accepted-fixes

Conversation

@windaddict

@windaddict windaddict commented Aug 9, 2026

Copy link
Copy Markdown
Owner

Applies the low-risk fixes surfaced by a cross-provider Impasse review of the README.

Changes

  • "First run — what to expect" note under Requirements — the onboarding gap the review flagged:
    you need an account with the reviewer's provider (separate from the host, spends its tokens), the
    first send blocks for consent, rough cost/time, and what a review does to your files
    (reviewer is read-only; the host applies verified fixes to your working copy). Also answers the
    "does 'review this' edit my files?" ambiguity.
  • Trim a repeat — dropped a near-verbatim "reviewer never edits / critic ≠ editor" from the
    flowchart caption (already stated in the opening; it appeared 5×).

Deliberately not changed

Per prior operator rulings, the benefit-first hero, "decorrelation beats intelligence", and the
CLAR priority claim ("the term was introduced here" — stated plainly; the term was coined here)
stay as-is. The opening's confident framing is intentional, and the "host-directed, not enforced"
honesty sits prominently right below it.

On the new three-paper section

It was run through Impasse (handles the inconvenient Xiang asymmetry result candidly — 2 low/medium
findings, both about slogans slightly broader than the described evidence). Impasse can't fetch
papers, so both new arXiv citations were verified against their abstracts: Xiang et al.
(2607.21656)
and Kumar et al. (2606.15689) — titles, authors, and headline figures all match.
Exact in-body numbers (quoted prompt strings, the 0.066 vs 0.847 PR-vs-synthetic figure) still warrant
a PDF check before leaning on them.

Docs-only.

🤖 Generated with Claude Code

Applies the low-risk fixes from a cross-provider Impasse review of the README:

- Add a "First run — what to expect" note under Requirements: you need an
  account with the reviewer's provider (separate from the host), it spends
  real tokens, the first send blocks for consent, and what a review does to
  your files (reviewer read-only; the host applies verified fixes to your
  working copy). Closes the onboarding/edit-authorization gap the review flagged.
- Drop a near-verbatim repeat of "the reviewer never edits — critic and editor
  stay separate" from the flowchart caption (already stated in the opening).

Deliberately NOT changed: the benefit-first hero, "decorrelation beats
intelligence", and the CLAR priority claim (the term was introduced here — that
is stated plainly, not hedged). The opening's confident framing is intentional;
the "host-directed, not enforced" honesty sits prominently right below it.

The new three-paper section (Xiang et al., Kumar et al.) was itself run through
Impasse — it handles the inconvenient asymmetry result candidly — and both new
arXiv citations were verified against the source abstracts (titles, authors,
and headline figures match).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X8XaW2bqgYnp7oRVgfHXid
@windaddict
windaddict force-pushed the docs/readme-accepted-fixes branch from b809e22 to 30a6a91 Compare August 9, 2026 17:31
@windaddict
windaddict merged commit 3d7c2c3 into main Aug 9, 2026
2 checks passed
@windaddict
windaddict deleted the docs/readme-accepted-fixes branch August 9, 2026 17:33
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant