Origin, access model, eligibility, and substitution are reviewer diagnostics, not judging criteria. Authentication requirements are not penalized.
02 The two blind reviews
Two independent reviewers scored this project blind, from a sanitized evidence packet.
Scores are shown separately so the reasoning stays inspectable. A withheld score means the reviewers differed by more than two points.
Reviewer A (round 1)
confidence 81%
Leverage
9 /10
WebMCP is central: structured reports, live drafts, shared view state, and approval-gated submission address the reliability and auditability problems of agents operating dense ERP screens.
Evidence cited
Agent can query real ledger numbers without scraping or menu navigation.
draft_quote writes a live editable card and get_current_view reads the same draft.
submit_draft suspends for human approval before irreversible commit.
Execution
8 /10
The packet describes a broad but internally coherent ERP and a carefully designed draft-to-human-approval workflow; no video is submitted, so the score is execution claimed rather than observed.
Evidence cited
Live judge demo is reported alive with explicit credentials.
Description covers multiple ERP domains and a concrete draft/approval flow.
Public repository and hackathon-period provenance are stated.
Impact
9 /10
Taiwanese SMEs have a real need for reliable ERP operation and accounting accuracy; shared drafts and human approval directly reduce costly agent errors.
Evidence cited
Specific SME audience and business workflows.
Human retains final approval for posting to books.
Ledger, tax, inventory, and payroll use cases are consequential.
Creativity
8 /10
The agent-as-drafter/human-as-approver model is a thoughtful interaction pattern for high-stakes enterprise software, beyond simply exposing CRUD tools.
Evidence cited
Approval by structure and live shared draft editing.
Same-screen agent and human collaboration across ERP state.
Reviewer B (round 2)
confidence 87%
Leverage
8 /10
Agent and person operate on the same structured ERP draft and view, avoiding scraping and preserving a human approval boundary; this is a central product workflow, not a bolt-on.
Evidence cited
Frames show WebMCP ready with 12 registered tools and a trace of get_current_view, partner/product searches, query_report, and dashboard summary.
Frames show a quote draft rendered in the ERP, human edits, readback, and an Approve & create gate.
About text says submit_draft suspends to human approval and the agent reads the same draft back.
Execution
8 /10
The frame sheets show a credible end-to-end quote workflow and concrete tool traces, with draft editing, totals, non-submission, list view, and approval modal.
Evidence cited
Frames visibly show Blue Harbor Trading quote draft, line items, TWD 116,000 subtotal, Quotes & Orders list, and Create this quote? modal.
Frames show 12 WebMCP tools registered and named tool calls.
Screenshot packaging confirms ERP demo and WebMCP activity panel.
Impact
9 /10
SMEs need ERP access for sales, finance, tax, and operations, and safe agent assistance around irreversible accounting actions has substantial practical value.
Evidence cited
Product visibly includes sales, purchases, contacts/products, financial reports, e-invoice, VAT return, and withholding tax.
Quote workflow is customer-specific and human-approved.
Pitch targets Taiwanese SMEs and fragmented ERP access.
Creativity
8 /10
Shared-screen ERP drafting with tool-driven state and explicit human approval is a strong, disciplined application of WebMCP to dense business software.
Evidence cited
One draft, two editors behavior is visibly demonstrated.
Agent uses tools rather than clicking and final approval is restricted to a human.
03 Review highlights
Standouts across reviewers
Strong human-approval boundary for irreversible accounting actions.
Concrete shared-state ERP workflow.
Concrete named tool trace leading to a shared quote draft.
No submitted video frames could be visually verified.
Repository documentation distinguishes prior work from challenge work; exact new-scope boundary is not independently checked here.
Demo is presented through frame sheets rather than a transcript.
04 Evidence
What each artifact proves is labeled on the artifact itself. A Devpost page capture is packaging evidence, not proof the product runs;
video frames are evidence from the submitted demo, not live verification.
Contact sheets from the ?s video the team submitted. This is what reviewers were shown;
it demonstrates the product in motion but is not independent verification.
· watch the original
EX-V1 Sheet 1 of 3 · reported video evidenceEX-V2 Sheet 2 of 3 · reported video evidenceEX-V3 Sheet 3 of 3 · reported video evidence
