GrantBack

The description states that GrantBack is a ChatGPT App and public Closeout Workbench. It uses GPT-5.6 to interpret fragmented post-award grant evidence and compile it into a source-verified candidate receivable package. This package includes:

• Resolution of the controlling document path;
• Verification of exact source spans, quotes, and hashes;
• Checks for required evidence and contradictions;
• Monetary calculations with deterministic Decimal arithmetic;
• Creation of a candidate receivable package for accountant review;
• Binding of approval to the exact source root, amount, recipient, subject, body, attachments, and idempotency key;
• Separation of provider acceptance, mail-server delivery, recipient observation, and payment;
• Blocking of downstream actions if one character changes in the source.

It is described as a strict TypeScript and pnpm workspace, with modules such as @grantback/compiler, @grantback/verifier, @grantback/evaluator, etc., that handle authority resolution, verification, evaluation, and package creation. The system also integrates with the OpenAI Apps SDK and ChatGPT Work.

Not evidenced: No actual customer data, revenue, or usage metrics are provided.

The author positions GrantBack as a solution to a pre-ledger gap in grant accounting — where systems typically begin after a receivable has been identified and entered into the ledger. GrantBack addresses the challenge of scattered evidence across multiple documents before an entry exists.

It claims to be a "typed obligation program" that resolves authority through a decision chain involving:

• GPT-5.6 proposal → restricted typed obligation IR;
• Source authority → controlling agreement and amendment path;
• Evidence → document hash, typed locator, exact quote, quote hash, and parser revision;
• Money → server-owned Decimal execution trace;
• Approval → source root, program, amount, recipient, notice, attachments, actor, nonce, and expiry;
• Action → provider acceptance, delivery, observation, and payment remain separate truth states.

The system is described as not owning arithmetic, authority resolution, approval, recipient selection, delivery truth, or payment truth — but rather binding these elements to the source evidence.

Inferred: The positioning implies a niche application in public grants and government-funded R&D, where precision and auditability are critical. However, no claims about market size, adoption, or competitive differentiation beyond its own self-description are made.

The description states that the author is Sergio, a Washington State CPA, who specializes in public grants and government-funded R&D. The system is built for use in post-award grant closeout workflows.

It targets environments where:

• Evidence is scattered across multiple documents;
• Accountants must determine which source controls;
• Required evidence must be complete;
• Exact amounts must be supported by sources;
• Actions remain authorized only if the evidence does not change.

Not evidenced: No specific customer personas, industries, or use cases beyond the synthetic examples are described. No indication of whether this is intended for large enterprises, small firms, or government agencies.

The description does not provide any information about pricing, monetization, or business model. It only describes the technical architecture and functionality of the system.

Not evidenced: No revenue streams, pricing tiers, or customer acquisition strategies are mentioned.

The project is built using:

• OpenAI GPT-5.6
• ChatGPT Work and Apps SDK
• Codex for development and auditing
• TypeScript, Node.js, pnpm, Zod, Decimal.js, Vitest, Vercel, Resend, WebCrypto, Ed25519, GitHub Actions

Modules include:

• @grantback/contracts
• @grantback/compiler
• @grantback/verifier
• @grantback/evaluator
• @grantback/pipeline
• @grantback/receivable
• @grantback/actions
• @grantback/replay
• @grantback/widget

The system is described as using deterministic code to verify evidence, perform calculations, and bind approvals. It also includes:

• A one-character source-change test that blocks downstream actions;
• A release-gate run with 271 of 271 automated tests passing;
• Zero findings in privacy, Gitleaks, or OSV scans.

Inferred: The system appears to be a prototype or proof-of-concept built for a hackathon. No production deployment details, scalability assumptions, or delivery mechanisms beyond the described synthetic case are provided.

The description states that:

• A public Closeout Workbench was built with 23 synthetic institution cases;
• The system demonstrated a final production MCP advertised 15 tools, and actively exercised the hero and live-model-replay paths;
• A fresh authenticated ChatGPT Work execution was captured against the final production release;
• The system reproduced USD 10,069.57 with exact Decimal arithmetic;
• It proved that a one-character source mutation blocks amount, package, approval reservation, and action;
• All 24 held-out evaluation cases passed (12 exact-calculation, 12 required-abstention);
• The final release scans reported zero public-data privacy findings, zero Gitleaks findings, and zero OSV findings.

Not evidenced: No real-world usage, customer feedback, or adoption metrics are provided. The system is described as synthetic and not connected to any actual customer evidence.

The description does not mention any competitors or direct market comparisons. It positions itself as solving a gap in grant accounting workflows, particularly where systems fail before ledger entries exist.

Not evidenced: No competitive landscape, pricing, or differentiation from existing tools is described.

• No real-world adoption: The system is described as synthetic and not connected to any actual customer evidence.
• Unverified claims: The description makes strong technical claims but lacks independent verification or third-party validation.
• Limited scope: It appears to be a hackathon project, not a production-ready product.
• Unclear business model: No monetization strategy or pricing information is provided.
• High technical complexity without traction: The system involves complex integration with AI models and financial systems, but no evidence of operational use.

Inferred: The risk of over-engineering for a niche problem without real-world validation is high. The project may be more of a proof-of-concept than a scalable product.

1. What is the actual use case or customer pain point that this system solves?
2. Has it been tested with real grant accountants or in live workflows?
3. How does it handle edge cases beyond the synthetic examples provided?
4. Is there any plan to integrate with existing accounting systems or ERP platforms?
5. What are the legal and compliance implications of binding approvals to source evidence?
6. Are there plans for monetization, and how will pricing be structured?
7. How does the system scale beyond the current synthetic test cases?