Inspiration

AI conversation exports often look like ordinary JSON archives, but evidentiary questions depend on structure that a flat transcript can erase: branches, roles, timestamps, missing values, tool and system events, attachment references, and duplicate families.

GenAI Evidence Workbench began from a practical need for an offline and reproducible way to preserve that structure. The Build Week challenge was to explain the workflow clearly and interactively without publishing live evidence, sensitive filenames, vendor-specific adapters, detector thresholds, or conclusions that the records cannot prove.

What it does

The workbench presents a four-stage evidence path:

Acquire — lock an abstract input and establish its byte-level hash. Restore — preserve branches, roles, timestamps, missing values, and event classes. Reconcile — account for duplicates, attachment references, and candidate relationships without treating them as proof of provenance or causation. Report — produce a reproducible manifest, data dictionary, and audit report.

The public Build Week experience uses fixed synthetic counts and abstract nodes. It contains no upload endpoint and cannot receive live evidence. A JP/EN switch presents the same narrative in both languages.

How we built it

The public experience is a responsive React and TypeScript site deployed with Codex Sites. It combines:

a narrative chapter structure; an accessible bilingual interface; a deterministic four-stage synthetic simulator; an original 2D forensic AI analyst and mascot; a public-safe explanation of the offline forensic workflow; a controlled-evaluation proposal for vendor-supplied synthetic or benchmark exports.

The regulated evidence workspace and public submission site are deliberately separated. The private production method remains offline and outside this submission.

How Codex accelerated the work

Codex was the primary development environment for the Build Week extension. It helped:

separate the public website from the regulated evidence workspace; translate forensic policy into product and disclosure requirements; design the narrative information architecture; implement and validate the bilingual React interface; build the deterministic synthetic simulator; generate and refine the original analyst and mascot; identify the risk of exposing proprietary extraction logic; replace that risk with an explicitly synthetic demonstration; prepare the deployment, README, demo video, narration, subtitles, and submission assets.

Human decisions controlled the anime direction, female guide character, evidence-preservation principles, public/private boundary, and controlled-evaluation positioning.

How GPT-5.6 was used

GPT-5.6 was used meaningfully through Codex during the Build Week extension to reason about the safety boundary, translate regulated forensic requirements into a testable public experience, implement the interface, review disclosure risks, and validate the project narrative and build.

GPT-5.6 is not called by the public demo at runtime. No live evidence is sent to GPT-5.6 or any hosted model. This is an intentional privacy and evidence-preservation decision.

Challenges

The hardest problem was not rendering JSON. It was deciding what a public demo must not reveal.

A realistic extractor could disclose vendor-specific adapters, thresholds, candidate-scoring logic, or evidence-derived details. We designed the simulator to teach the workflow using fictional counts and abstract nodes instead.

The second challenge was making hashes, branches, duplicates, safe stops, and audit lineage understandable without implying that a candidate match proves training, leakage, provenance, or causation.

Accomplishments

Created a complete public narrative for a technically sensitive workflow. Built a working bilingual simulator requiring no account, API key, or upload. Established a clear boundary between demonstration, controlled evaluation, and private production logic. Created a distinctive original visual identity that makes audit work approachable. Produced a reproducible, source-text-free testing path for judges.

What we learned

For evidence tooling, restraint is a product feature.

A compelling demo does not need live customer data or exposed extraction logic. It needs a clear chain of decisions, explicit limits, deterministic behavior, and an honest account of what every result can and cannot prove.

Prior work and Build Week work

The underlying private forensic workbench predates Build Week.

During Build Week, Codex and GPT-5.6 were used to create:

the complete public-safe website; the anime-inspired design system and original female analyst; the bilingual narrative interface; the deterministic synthetic simulator; the controlled-evaluation protocol; the public/private disclosure boundary; the Codex Sites deployment; the demo video, English narration, subtitles, and submission materials.

Timestamped Codex task history and the Build Week commit history document this extension.

What GEW does:

Offline, local-first, post-hoc reconstruction and verification of AI execution artifacts, state transitions, and evidentiary claims Independent corroboration of whether AI-reported outcomes (e.g. "confirmed," "succeeded," "verified") correspond to actual execution evidence Detection of state-integrity failures where output from terminated processes survives context compaction, loses its evidential qualifier, and is inherited as confirmed operational state Reproducible evidence paths with byte-identical deterministic verification (Python/Node dual implementation, SHA-256 bundle hashing)

What GEW does not do:

Real-time inline monitoring or request-path enforcement Cryptographic signing of live API traffic Policy-based allow/redact/block decisions on AI requests in transit

Complementary relationship:

Inline enforcement gateways (such as DeepInspect) sit in the request path between users and AI endpoints, producing cryptographically signed audit trails of every API call. GEW occupies a different layer: the failure class it addresses occurs inside the LLM's own state transitions, where corruption happens within the model's internal context rather than on the HTTP request path. No amount of request-level logging can detect this class of failure in principle. The two approaches are complementary, not competing.

What's next

The next step is a controlled evaluation using synthetic or benchmark exports supplied by an AI vendor.

Inputs would be provided through a local read-only path, hashed before interpretation, processed offline inside a scoped workspace, and returned as source-text-free audit results.

No vendor partnership or endorsement is currently claimed.

Post–Build Week update: Evidence-bounded conversational analysis

After Build Week, the private offline workbench was extended with an evidence-bounded conversational analysis layer.

Using a real, read-only 15.6 GiB X archive, the local runner parsed 202,457 canonical records and created a separate confidential analysis corpus containing 123,986 text-searchable records. Two clean corpus builds produced the same byte-identical database, while the original archive and verified evidence artifacts remained unchanged.

The model is never treated as evidence. It can access records only through Workbench-managed deterministic retrieval methods for exact text, dates, record types, IDs, threads, related records, provenance, and record verification.

Every verifiable answer is written to a claim ledger and checked by deterministic code against record existence, content hashes, timestamps, source-archive identity, retrieval receipts, and the complete provenance chain. Fabricated IDs, unsupported assertions, zero-result claims, source-hash mismatches, and claims about records that were never retrieved are rejected.

Research selections are persisted as ordered record-ID sets rather than being held only in model context, so follow-up references remain stable across conversations and context compaction.

This is a post–Build Week private research extension, not part of the judged public demo. The public experience remains synthetic, upload-free, and does not send live evidence to a hosted model.

Built With

Share this project:

Updates

posted an update

Existing inline enforcement gateways such as DeepInspect record every AI request on the wire and produce cryptographically signed audit trails. However, the failure class documented by GEW occurs inside the LLM's own state transitions: output from a terminated process survives context compaction, loses its evidential qualifier, and is inherited as confirmed operational state by a later session. Because the corruption happens within the model's internal state rather than on the HTTP request path, no amount of request-level logging can detect it in principle. GEW occupies the complementary layer: independent, post-hoc reconstruction and verification of AI execution artifacts, state transitions, and evidentiary claims that inline gateways cannot observe.

Log in or sign up for Devpost to join the conversation.

posted an update

GenAI Evidence Workbench now has a recorded working demonstration using real incident data.

The demo shows the system operating end to end, including local source intake, integrity verification, evidence analysis, evidence graph construction, contradiction handling, verification, and proof-bundle generation.

This is the working implementation, not a mockup or conceptual demonstration.

The recording and technical record are maintained in a private GitHub repository. Access can be provided to reviewers and potential collaborators under controlled access.

Private repository: https://github.com/hiroki-tamba-research/mediator-ai

Log in or sign up for Devpost to join the conversation.

posted an update

Recorded product walkthrough now live.

The video shows GenAI Evidence Workbench running as a working product, not a concept mockup.

It demonstrates the evidence workflow for reproducible GenAI evaluation, including source integrity, cryptographic hashes, execution records, evidence relationships, verification, provenance, chain of custody, and reproducible proof artifacts.

The core principle is simple: an AI system reporting that it completed a task is not, by itself, evidence that the task was actually completed.

GenAI Evidence Workbench externalizes that evidence so evaluation results can be reconstructed and independently verified.

https://www.youtube.com/watch?v=mkwNUgt_ZI4

Selectable subtitles are available in English, Japanese, French, and German.

Log in or sign up for Devpost to join the conversation.

posted an update

The GenAI Evidence Workbench is now complete.

I have also obtained and analyzed primary-source data related to the July 2026 OpenAI–Hugging Face incident. Based on that analysis, I have proposed an independent corroboration process through UK AISI’s Inspect AI repository:

UKGovernmentBEIS/inspect_ai#4821

The proposal asks whether independently retained records can corroborate, contradict, or otherwise clarify specific evidentiary gaps identified in the hash-fixed dataset.

This is a real-world application of the same evidence-preservation and verification workflow implemented in the GenAI Evidence Workbench.

Log in or sign up for Devpost to join the conversation.

posted an update

Post-submission validation note — Compaction integrity

During continued validation of long-running AI-agent workflows, I identified a compaction-related integrity issue that warrants separate reproducible review.

I have created a private repository containing the controlled investigation and have provided OpenAI with direct access for review.

This work does not modify the submitted GenAI Evidence Workbench build. It extends the same underlying principle: persisted or summarized state should not be treated as verified evidence without checking it against the underlying artifacts and execution state.

Private review repository: https://github.com/hiroki-tamba-research/mediator-ai

Log in or sign up for Devpost to join the conversation.

posted an update

Expanding from evidence analysis to rights-aware AI infrastructure

GenAI Evidence Workbench was initially developed to reconstruct reproducible evidence paths across exported AI conversations, including timestamps, model metadata, related records, and cross-conversation recurrence.

The same evidentiary architecture may support a broader challenge facing creative industries: enabling authorized AI use while protecting creators, performers, publishers, studios, and other rights holders.

A rights-aware deployment could connect:

provenance and chain-of-custody records authorization and licensing conditions model and tool usage metadata versioned evidence packages attribution and remuneration records investigation workflows for disputed or unauthorized use

This creates a bridge between two goals that are often treated as opposites:

protecting intellectual property and enabling legitimate AI innovation.

Rather than relying only on post-hoc detection, the system could help establish verifiable evidence before, during, and after licensed AI interactions.

Potential applications include film, animation, publishing, voice performance, character licensing, and other creative sectors where trusted collaboration between rights holders and AI vendors requires transparent and auditable infrastructure.

The project remains focused on evidence reconstruction and provenance. Any future rights-management deployment would require formal authorization, controlled access, legal oversight, and cooperation with the relevant rights holders.

Log in or sign up for Devpost to join the conversation.

posted an update

GenAI Evidence Workbench has now been archived as a versioned research artifact with a permanent DOI:

https://doi.org/10.5281/zenodo.21421751

The archived v0.1.0 release includes:

citation metadata SHA-256 integrity manifest technical documentation security notes a versioned release package a companion PDF describing the system and validation workflow

Additional validation confirmed successful offline processing of a 15.6 GB exported archive on a standard consumer laptop, without cloud infrastructure.

The public Devpost page documents the project, methodology, and demonstration:

https://devpost.com/software/genai-evidence-workbench

The implementation repository will remain private. Access, evaluation, licensing, and deployment will be handled through direct authorization by Hiroki Tamba.

This preserves the distinction between the publicly documented research contribution and the non-public implementation layer, while maintaining a permanent, citable, and integrity-verifiable record of the project as submitted.

Log in or sign up for Devpost to join the conversation.

posted an update

Codex development workflow during Build Week

Codex was used continuously throughout Build Week to implement, test, debug, document, and extend the working application across multiple export formats and reporting workflows.

This iterative development process supported rapid validation and refinement of both the technical pipeline and the product experience.

Log in or sign up for Devpost to join the conversation.

posted an update

Public capability brief released

The public capability brief for GenAI Evidence Workbench is now available on GitHub.

The repository documents the evidence model, validation scope, operating boundaries, and reproducibility principles while intentionally excluding proprietary implementation details, production detection logic, thresholds, and raw datasets.

GitHub: https://github.com/hiroki-tamba-research/GenAI-Evidence-Workbench

Future updates will continue to focus on independent AI evidence engineering, reproducibility, and responsible AI audit methodologies.

Log in or sign up for Devpost to join the conversation.

posted an update

One interesting observation after submission:

The current implementation successfully analyzed a 15 GB real-world AI conversation archive on a standard laptop.

This suggests that reproducible AI evidence analysis can remain practical without specialized hardware or cloud infrastructure.

Log in or sign up for Devpost to join the conversation.

posted an update

GenAI Evidence Workbench has evolved from a research prototype into a reproducible workflow for inspecting, verifying, and exporting AI-generated evidence.

Recent improvements include:

• deterministic evidence exports • clearer provenance and verification records • reproducible forensic reports • vendor-neutral evaluation workflows • improved documentation and deployment

The project is designed around a simple principle: plausible AI output is not sufficient. Claims should remain traceable to observable evidence, execution history, and reproducible artifacts.

I also published a live deployment and expanded the project documentation so evaluators and developers can inspect the workflow directly.

Live app: https://genai-evidence-workbench.hiro4ever.chatgpt.site/

GitHub: https://github.com/hiroki-tamba-research/GenAI-Evidence-Workbench

The next phase will focus on long-horizon agent evidence, including interruption, compaction, resume behavior, and trajectory-level verification.

Log in or sign up for Devpost to join the conversation.