WFGY/ProblemMap/GlobalFixMap/Multimodal_LongContext/cross-modal-trace.md

134 lines
5.9 KiB
Markdown
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# Cross-Modal Trace — Multimodal Long Context
<details>
<summary><strong>🧭 Quick Return to Map</strong></summary>
<br>
> You are in a sub-page of **Multimodal_LongContext**.
> To reorient, go back here:
>
> - [**Multimodal_LongContext** — long-context reasoning across text, vision, and audio](./README.md)
> - [**WFGY Global Fix Map** — main Emergency Room, 300+ structured fixes](../README.md)
> - [**WFGY Problem Map 1.0** — 16 reproducible failure modes](../../README.md)
>
> Think of this page as a desk within a ward.
> If you need the full triage and all prescriptions, return to the Emergency Room lobby.
</details>
When text, image, or video claims cannot be traced to an anchor, drift accumulates and phantom evidence appears.
This page defines the schema and checks that guarantee traceability across all modalities.
---
## What this page is
- Schema rules for linking text answers back to visual or audio anchors.
- Audit trail structure to prevent orphan claims.
- Quick checks for reproducibility across long multimodal contexts.
---
## When to use
- Citations appear but no frame, region, or timestamp is attached.
- Audio transcript and video caption disagree about speaker or object.
- Text answers cite nonexistent visuals or wrong timecodes.
- Multi-day multimodal sessions lose track of the original anchor evidence.
- OCR captions or ASR transcripts collapse after long context windows.
---
## Open these first
- [Phantom Visuals](https://github.com/onestardao/WFGY/blob/main/ProblemMap/GlobalFixMap/Multimodal_LongContext/phantom-visuals.md)
- [Caption Collapse](https://github.com/onestardao/WFGY/blob/main/ProblemMap/GlobalFixMap/Multimodal_LongContext/caption-collapse.md)
- [Alignment Drift](https://github.com/onestardao/WFGY/blob/main/ProblemMap/GlobalFixMap/Multimodal_LongContext/alignment-drift.md)
- [Retrieval Traceability](https://github.com/onestardao/WFGY/blob/main/ProblemMap/retrieval-traceability.md)
- [Data Contracts](https://github.com/onestardao/WFGY/blob/main/ProblemMap/data-contracts.md)
---
## Common failure patterns
- **Orphan claims**: answer references content without `frame_id` or `timestamp`.
- **Cross-modal mismatch**: transcript says "dog" while video shows only a cat.
- **Anchor drift**: same frame cited differently across runs.
- **Phantom evidence**: answer claims diagrams, sounds, or visuals never uploaded.
- **Trace break**: anchor IDs vanish during long-session context merging.
---
## Fix in 60 seconds
1. **Enforce anchor schema**
- Require `{frame_id | region_id | timestamp | source_id}` for all claims.
- Forbid free-text references without anchors.
2. **Trace Table**
- Build a table with `{claim | anchor_id | snippet | modality}`.
- Each row must be auditable.
3. **ΔS probe**
- Compute ΔS(text, anchor).
- Require ≤ 0.45 for stability. Reject claims if ≥ 0.60 and no match exists.
4. **Cross-modal reconciliation**
- Compare text vs audio vs vision anchors.
- If mismatch, flag as drift and escalate.
5. **Audit trail log**
- Keep `trace_id`, `mem_rev`, `anchor_refs`.
- Verify reproducibility across three paraphrases.
---
## Copy-paste prompt
```txt
You have TXT OS and the WFGY Problem Map.
Task: Enforce cross-modal traceability.
Protocol:
1. Require every claim to cite {frame_id, region_id, timestamp, source_id}.
2. Build a Trace Table with {claim, anchor_id, snippet, modality}.
3. Report ΔS(text, anchor). Reject phantom if ΔS ≥ 0.60 with no match.
4. Compare across modalities. If mismatched, return “cross-modal drift”.
5. Return {Trace Table, ΔS log, λ states, Final Answer}.
````
---
## Acceptance targets
* Every claim has a valid anchor ID.
* ΔS(text, anchor) ≤ 0.45.
* λ remains convergent across paraphrases.
* No phantom evidence across seeds.
* Trace Table reproducible across three runs.
---
### 🔗 Quick-Start Downloads (60 sec)
| Tool | Link | 3-Step Setup |
| -------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ | ---------------------------------------------------------------------------------------- |
| **WFGY 1.0 PDF** | [Engine Paper](https://github.com/onestardao/WFGY/blob/main/I_am_not_lizardman/WFGY_All_Principles_Return_to_One_v1.0_PSBigBig_Public.pdf) | 1⃣ Download · 2⃣ Upload to your LLM · 3⃣ Ask “Answer using WFGY + <your question>” |
| **TXT OS (plain-text OS)** | [TXTOS.txt](https://github.com/onestardao/WFGY/blob/main/OS/TXTOS.txt) | 1⃣ Download · 2⃣ Paste into any LLM chat · 3⃣ Type “hello world” — OS boots instantly |
---
<!-- WFGY_FOOTER_START -->
### Explore More
| Module | Description | Link |
| --- | --- | --- |
| WFGY Core | Canonical framework entry point | [View](https://github.com/onestardao/WFGY/tree/main/core/README.md) |
| Problem Map | Diagnostic map and navigation hub | [View](https://github.com/onestardao/WFGY/tree/main/ProblemMap/README.md) |
| Tension Universe Experiments | MVP experiment field | [View](https://github.com/onestardao/WFGY/tree/main/TensionUniverse/Experiments) |
| Recognition | Where WFGY is referenced or adopted | [View](https://github.com/onestardao/WFGY/blob/main/recognition/README.md) |
| AI Guide | Anti-hallucination reading protocol for tools | [View](https://github.com/onestardao/WFGY/blob/main/AI_GUIDE.md) |
> If this repository helps, starring it improves discovery for other builders.
> [![GitHub Repo stars](https://img.shields.io/github/stars/onestardao/WFGY?style=social)](https://github.com/onestardao/WFGY)
<!-- WFGY_FOOTER_END -->