FundPilot Assurance StudioEVIDENCE WORKSPACE · NO RELEASE AUTHORITY

01 / Overview

Interface build identity not collected. Packet revisions below identify dated evidence.

Loading checksummed dated evidence…
RELEASE READINESS · DATED EVIDENCEAI REVIEW + DETERMINISTIC VERIFICATION

Understand the risk.
Know the next action.

FundPilot connects a consequential software change to its source evidence, GitLab Duo review and independent checks. Investigate the claim. Propose the correction. Keep unresolved risk visible.

How AI is used: GitLab Duo supplied the recorded AI-powered code review. Every interaction in this workspace is deterministic; it runs no live language-model inference.

FULL-STUDIO OBSERVATION

EVIDENCE NOT LOADED

NEXT EVIDENCE-BASED ACTION

Selected packet · scope and evidence summary

NO OBSERVATION

Investigate one claim.
Trace every safeguard.

Choose retained evidence.

For maintainers of consequential decision software: inspect a changed claim, its unchanged evidence, the recorded GitLab Duo code review and deterministic checks. This workspace runs no live language-model inference. Start with the historical case; dated MR packets explain remaining blockers.

01

Investigate

Compare an unsafe eligibility label with the unchanged source and recorded AI finding.

02

Act

Define a correction, run packet checks and independently replay the source-bound receipt.

03

Decide

Inspect verified, failed and unknown conditions. Local proposals never grant release authority.

EVALUATION, NOT MARKETING

Can the evidence distinguish a safe next step?

Run a frozen, deterministic comparison against simple baselines. These are constructed evidence cases, not an independent user study, an AI benchmark or measured time savings.

Not run in this browser.

Method and limitations

RUNNABLE INVESTIGATION · BROWSER COMPUTATION

From a native finding to a checkable correction.

Connect an authentic GitLab Duo finding to its code change, a proposed corrective action, independently executed packet checks and a receipt another maintainer can replay. No fix, approval or remote write is performed here.

Public synthetic evidence only. Drafts stay in this page until reload; downloaded files are yours to retain privately.

RECORDED AI CONTRIBUTION

Inspect the original finding

Related returned code diff

                
HUMAN CORRECTIVE-ACTION PLAN
Define the change and its proof
Optional external correction / owner records
These are user-supplied references, not verified implementation or authorization.

              

              
Reproduce independently: node scripts/replay_workbench.cjs --receipt your-receipt.json. Checks inspect retained fields, counts and job observations; they do not rerun original JUnit, audits or the AI agent.

            

            
Reopen and verify an exported action receipt
Select the matching dated investigation above, then open a JSON receipt. Verification runs in this browser against the independently loaded, checksummed packet. The file is not uploaded. It proves consistency, not publisher identity, freshness or release permission.
No receipt opened.

            
          

        

        

          
FUNDPILOT · RELEASE READINESS DOSSIER
Know what blocks this release.

          
GitLab Duo supplied the recorded AI-powered review. This workspace checks retained evidence deterministically, with no live language-model inference or new agent execution.

          
Loading retained evidence…

          

          
Inspect all eight conditions ↓

          
Source-backed investigation · local proposals never authorize release
            

              
The evidence in one place
These are dated observations, not freshly executed tests or current production clearance. Each record links directly to its original source.

              

            

            
Write a human follow-up proposal
This session-only ledger records a proposed next step. It does not apply a correction, establish an independent reviewer verdict or supply owner approval.

            

          

          
All eight conditions · no checks omitted

          
Independent reproduction: node scripts/replay_dossier.cjs release-readiness-dossier.json. Hash consistency is not publisher authentication. Notes stay in memory until reload or Clear all; downloaded copies are yours to retain privately.

        

        

          

            

              
TRACE THE CHAIN

              
Where does confidence stop?

            

            SELECT A STAGE
          

          

          

          
        

        

          
ACTIONABLE READINESS

          
What still prevents a release?

          

          

            All states are scoped to the selected observation. No packet
            establishes current production health or human release
            authorization.
          

        

        
PROVENANCE NAVIGATOR · RECORDED HISTORY
Follow the recorded sequence.
Inspect each source record. Comment times, commit times and pipeline creation times are distinct observations—not invented completion or causal timestamps.

          
COMPARE OBSERVATIONS

          
What changed—and what did not?

          
          

        

        

          
AUTHENTIC AI CONTRIBUTION

          
GitLab Duo review, in context.

          

            Comments are authentic API observations. Diff positions can move
            after pushes. Neither a comment nor a finished session proves formal
            approval or the exact execution input revision.
          

          
          

          

            Textual code change—not a behavioral conclusion
            
            

          

          

            Security inventory and original artifact links
            

            

          

        

        

          
PUBLIC INTERACTIVE · SESSION-LOCAL DRAFT

          
Turn a blocker into a review request.

          
Selected native finding is attached when you save.

          

            Session memory only. Downloads are owner-managed copies. No account
            or server storage.
          

        

        
Local advanced capabilities — separate Python adapter required
These functions are not part of the public static deployment. The separate public funding demo remains available from the sidebar.

          
CONNECT TO THE PROTECTED APPLICATION

          
Decision Explorer

          

            This local adapter runs the original rules engine, not an LLM.
            Compare declared facts and amended public text; unknown and
            unsupported conditions remain unresolved. No eligibility
            certification.
          

          

            Install the separate local runtime to enable calculations. Static
            hosting preserves evidence exploration; it does not pretend to run
            the evaluator.
          

          

          

            
            

              

                
              

              

                
              

              

                
              

            

            
          

        

          

            No current analysis.
          

        

        

          
SYNTHETIC SIMULATION · NOT A LIVE REVIEW

          
What if this evidence changed?

          

            The unchanged canonical assessment runs locally. These controls
            never alter GitLab, production, or the real packet above.
          

          
          
No simulation executed.