From 9d05cb94cb22f9010f9b3251458fa40182224620 Mon Sep 17 00:00:00 2001 From: dytlzl Date: Wed, 18 Nov 2026 22:55:50 +0900 Subject: [PATCH 1/3] docs: explain why proof-driven development helps review --- README.md | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) diff --git a/README.md b/README.md index 0e4f30b..8af8b79 100644 --- a/README.md +++ b/README.md @@ -2,12 +2,20 @@ **Proof-Driven Development for Go.** -When a change touches who can sign in, what can be authorized, or when a balance can be debited, examples alone leave cases unchecked. Write the rule as a contract on the production Go function. `pdd` runs Gobra to check it, runs Go tests for the surrounding code, and writes a receipt you can bring to code review. +ProofSteer checks decision rules in production Go code, runs tests for the surrounding behavior, and writes a receipt you can bring to code review. Start with one decision in an existing Go project. The core workflow works without an LLM account. [View live report examples](https://proofsteer-demo.cruskell.chatgpt.site) · [Try the login example](#try-it) · [Use it in your project](#use-it-in-your-project) · [Download pdd](https://github.com/cruskell/proofsteer-go/releases/tag/v0.4.0) +## Why proof-driven development? + +AI agents can write code quickly. Human review can become the bottleneck when someone has to trace each branch of an authorization or payment rule and check it against the intended behavior. Passing tests confirm the cases they exercise. + +Start by describing the rule in plain language. An agent can draft a Go contract, and a human reviewer approves the contract and its assumptions. After implementation, `pdd` runs Gobra to verify that the Go function satisfies the contract for every input it covers. Reviewers can focus on whether the rule matches the intent, whether its assumptions hold, and whether tests cover behavior outside the proof. + +In the wallet demo, a fixed balance of 100 still gives **16 cases** to check by hand. Combine two authentication states, two lock states, and four withdrawal amounts (0, 1, 100, and 101). Gobra takes over checking the withdrawal decision for those cases. Go and browser tests exercise the HTTP boundary and the withdrawal flow. [See the review comparison](https://proofsteer-demo.cruskell.chatgpt.site/en/#review-cost) + ## What a proof catches Suppose the requirement is **allow a login exactly when the account exists, the password matches, and the account is unlocked**. The specification lives beside the production code in a Go comment that Gobra reads. From 987bb61c45054204c6a0601d016f794155b80d94 Mon Sep 17 00:00:00 2001 From: dytlzl Date: Wed, 18 Nov 2026 22:57:30 +0900 Subject: [PATCH 2/3] docs: make the withdrawal example standalone --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 8af8b79..52cff15 100644 --- a/README.md +++ b/README.md @@ -14,7 +14,7 @@ AI agents can write code quickly. Human review can become the bottleneck when so Start by describing the rule in plain language. An agent can draft a Go contract, and a human reviewer approves the contract and its assumptions. After implementation, `pdd` runs Gobra to verify that the Go function satisfies the contract for every input it covers. Reviewers can focus on whether the rule matches the intent, whether its assumptions hold, and whether tests cover behavior outside the proof. -In the wallet demo, a fixed balance of 100 still gives **16 cases** to check by hand. Combine two authentication states, two lock states, and four withdrawal amounts (0, 1, 100, and 101). Gobra takes over checking the withdrawal decision for those cases. Go and browser tests exercise the HTTP boundary and the withdrawal flow. [See the review comparison](https://proofsteer-demo.cruskell.chatgpt.site/en/#review-cost) +Consider a withdrawal rule with a balance of 100. Two authentication states, two lock states, and four withdrawal amounts (0, 1, 100, and 101) give **16 cases**. Gobra can take over checking the decision against the contract for those cases. Go and browser tests exercise the HTTP boundary and the withdrawal flow. ## What a proof catches From 7296c59865d3fc956ad94d4d46ce0ca041f38b28 Mon Sep 17 00:00:00 2001 From: dytlzl Date: Wed, 18 Nov 2026 22:58:16 +0900 Subject: [PATCH 3/3] docs: compare review responsibilities inline --- README.md | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 52cff15..05da52d 100644 --- a/README.md +++ b/README.md @@ -12,9 +12,15 @@ Start with one decision in an existing Go project. The core workflow works witho AI agents can write code quickly. Human review can become the bottleneck when someone has to trace each branch of an authorization or payment rule and check it against the intended behavior. Passing tests confirm the cases they exercise. -Start by describing the rule in plain language. An agent can draft a Go contract, and a human reviewer approves the contract and its assumptions. After implementation, `pdd` runs Gobra to verify that the Go function satisfies the contract for every input it covers. Reviewers can focus on whether the rule matches the intent, whether its assumptions hold, and whether tests cover behavior outside the proof. +Start by describing the rule in plain language. An agent can draft a Go contract, and a human reviewer approves the contract and its assumptions. After implementation, `pdd` runs Gobra to verify that the Go function satisfies the contract for every input it covers. -Consider a withdrawal rule with a balance of 100. Two authentication states, two lock states, and four withdrawal amounts (0, 1, 100, and 101) give **16 cases**. Gobra can take over checking the decision against the contract for those cases. Go and browser tests exercise the HTTP boundary and the withdrawal flow. +Consider a withdrawal rule with a balance of 100. Two authentication states, two lock states, and four withdrawal amounts (0, 1, 100, and 101) give **16 cases**. + +| Review task | Manual code review | Proof-driven review | +| --- | --- | --- | +| Is this the intended rule, and are its assumptions sound? | Reviewer checks the specification | Reviewer checks the specification | +| Does the Go function make the right decision for these 16 inputs? | Reviewer traces the code | Gobra checks the contract | +| Do HTTP handling and the browser flow behave as expected? | Reviewer checks test coverage and results | Reviewer checks test coverage and results | ## What a proof catches