From 62d7ffaa967c3198df66e424b893d9c99d29c0a1 Mon Sep 17 00:00:00 2001 From: "John M. P. Knox" Date: Fri, 31 Jul 2026 12:18:41 -0500 Subject: [PATCH] docs(README): lead with the benefit, not the mechanism MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The opening buried the value under jargon — an abstract "independent second opinion" tagline, then "reference implementation of CLAR," then a flowchart — before a lay reader learned why they'd care (mastermind feedback). Rework the opening to sell the benefit first: - Hero: "Impasse gets you better results from your AI — the first time." - "Why a second AI?" explainer (blind spots) follows the promise. - Reframe the reviewer as an "independent advocate for one thing: the quality of the result you ship" — advocate for the outcome, not the operator's position, and not merely against the host AI. - Move CLAR, the flowchart, and the pre-release caveats below the pitch. Copy reviewed cross-provider with Impasse (dogfood). First pass caught the draft overclaiming ("can't", "only", "don't match" — softened to accurate claims). A marketing-calibrated pass cleared the headline as defensible and caught a real accuracy bug — verify/reconcile were attributed to the reviewer when the host performs them (a core invariant); fixed. Voice lint clean. Co-Authored-By: Claude Opus 4.8 (1M context) Claude-Session: https://claude.ai/code/session_01X8XaW2bqgYnp7oRVgfHXid --- README.md | 29 ++++++++++++++++++----------- 1 file changed, 18 insertions(+), 11 deletions(-) diff --git a/README.md b/README.md index 8c0e5c9..d744488 100644 --- a/README.md +++ b/README.md @@ -1,16 +1,23 @@ # Impasse -> **An independent second opinion for any high-stakes call — a decision, an essay, a research -> claim, a dataset, or code — from a cross-provider AI whose blind spots are *less likely to match* your own.** - -*Impasse is the reference implementation of **Cross-Lab Adversarial Review (CLAR)** — defined -below, and written up in [the CLAR essay](https://www.movingavg.com/essays/cross-lab-adversarial-review.html).* - -The independent **reviewer never edits your work** — it argues, with evidence. Keeping the critic -away from the pen is the point: fixes get applied by the host — the AI agent you're already working in (Claude Code or Codex), which -drives Impasse — or by you, never by the model that's supposed to be checking you. And unlike a plain code reviewer, it doesn't hand you -a raw list to triage — it **verifies each finding, reconciles the two models, and escalates only -the genuine disagreement.** You get the verified problems, plus the call(s) that are actually yours. +> **Impasse gets you better results from your AI — the first time. It brings in a second AI, from a +> different company, to pressure-test the work and catch the problems your AI missed, so the result +> you commit to is one you can trust.** + +Why a *second* AI? The one that did the work brings the same blind spots to the check that it brought +to the work, so a sloppy assumption or a missed edge case can sail straight through. A **rival model +from a different provider** is less likely to share your AI's blind spots. Think of it as an +**independent advocate for one thing: the quality of the result you ship.** The second AI challenges +the work with evidence; then Impasse checks each objection against your actual work, fixes the +problems that hold up, and brings you the few calls that are genuinely yours to make. + +The reviewer **never edits your work** — the critic and the editor stay separate: fixes are applied by +the host you're already working in (Claude Code or Codex, which drives Impasse), or by you, never by +the model that's supposed to be checking you. And unlike a plain code reviewer, it doesn't hand you a +raw list to triage. + +*Impasse is the reference implementation of **Cross-Lab Adversarial Review (CLAR)** — defined below, +and written up in [the CLAR essay](https://www.movingavg.com/essays/cross-lab-adversarial-review.html).* ```mermaid flowchart TB