Business :: Ideas :: recond
⪼ Made with 💜 by Polyglot.
- Śavvy AI :: recond – Automating Idea Extraction
- Spec Forge - Autopilot PRD Request (ChatGPT)
- Automating Idea Extraction
Below is a complete PRD produced in **Autopilot mode**, using your Design Document as the sole source of truth.
No questions are asked. All unresolved choices were resolved with defaults and logged as assumptions.
---
# Product Requirements Document (PRD)
**Product Name:** Recon
**System:** Semantic Reconstruction Engine
**Document Version:** 1.0
**Mode:** Autopilot
**Source of Truth:** User-provided Idea Pack
**Primary Output Artifact:** Markdown document
---
## 1. Product Thesis
Recon is a semantic reconstruction engine that observes complex, dynamic web-based information environments and reconstructs their meaning into a single, coherent, human-readable Markdown document.
Recon does not scrape, index, archive, or mirror layouts. It interprets intent, infers structure, and rebuilds meaning. The system succeeds when the output reads as if it were deliberately authored in that form from the beginning, requiring no reference to the source experience.
---
## 2. Core Design Principles
1. **Meaning First**
Conceptual understanding precedes representation. Layout and DOM order are non-authoritative.
2. **Inference Over Templates**
No site-specific rules, schemas, or hardcoded assumptions.
3. **Completeness Without Speculation**
Capture everything observable, infer structure only from evidence.
4. **Single Durable Artifact**
One output, Markdown, self-contained and context-independent.
5. **Narrative Coherence**
Output must read linearly and logically to a human reader.
---
## 3. Personas
### P-001 Knowledge Worker
* Role: Researcher, analyst, strategist
* Goal: Preserve and reuse complex web-based knowledge
* Pain: Information scattered across interactive interfaces
* Success: A clean document usable without revisiting the source
### P-002 Technical Archivist
* Role: Engineer, documentation lead
* Goal: Create versionable, diffable knowledge artifacts
* Pain: Lossy exports and layout-bound captures
* Success: Stable Markdown suitable for long-term storage
### P-003 AI-Augmented User
* Role: Power user feeding downstream AI systems
* Goal: Provide structured, semantically rich inputs
* Pain: Raw web captures lack coherence
* Success: High-signal documents usable as prompts or references
---
## 4. Input Scenarios
* Complex dashboards with layered UI components
* Long-form interactive reports with modals and overlays
* Visually guided narratives split across sections
* Data explanations embedded in charts and tooltips
Out of scope: static PDFs, plain HTML articles, simple linear pages.
---
## 5. User Journeys
### J-001 Reconstruct a Web Experience
User provides a web-based experience. Recon captures, interprets, reconstructs, and outputs a Markdown document.
### J-002 Validate Knowledge Artifact
User reviews the Markdown output as a standalone reference without visiting the source.
---
## 6. UX Surface Inventory
| Screen ID | Name | Purpose |
| --------- | ---------------- | ----------------------------------- |
| S-001 | Input Surface | Accept target web experience |
| S-002 | Processing State | Indicate capture and reconstruction |
| S-003 | Output Viewer | Display final Markdown document |
---
## 7. Behavior and Editing Model
* Input is read-only. Recon does not modify source content.
* Processing is deterministic given identical observable inputs.
* Output is immutable once generated.
* No post-generation editing within Recon.
---
## 8. Constraints and Anti-Features
### Constraints
* No assumptions beyond observable data
* No speculative inference
* No layout mimicry
* No site-specific heuristics
### Anti-Features
* No alternate output formats
* No partial exports
* No DOM mirroring
* No summarization-only mode
---
## 9. Success and Failure Criteria
### Success
* Output reads as intentionally authored
* All observable information is represented
* Logical hierarchy is preserved
* Document is usable without source access
### Failure
* Missing concepts
* Disordered narrative
* Layout-driven structure
* Context-dependent references
---
## 10. North Star Metric
**Percentage of reconstructed documents that require zero source revisits for comprehension.**
---
## 11. Epics
* E-001 [MUST] Comprehensive Capture Engine
* E-002 [MUST] Semantic Parsing and Inference
* E-003 [MUST] Structural Reassembly Engine
* E-004 [SHOULD] Redundancy Collapse and Nuance Preservation
* E-005 [COULD] Output Validation Heuristics
---
## 12. User Stories with Acceptance Criteria
### E-001 Comprehensive Capture Engine
* US-001 [MUST] As a system, I capture all visible and interactable elements.
* Given a web experience
* When capture runs
* Then all text, modals, tooltips, charts, and metadata are recorded
* US-002 [MUST] As a system, I do not assume source structure.
* Given any site
* When capturing
* Then no site-specific rules are applied
---
### E-002 Semantic Parsing and Inference
* US-003 [MUST] As a system, I identify conceptual groupings.
* Given captured elements
* When parsing
* Then related concepts are grouped by meaning
* US-004 [MUST] As a system, I infer artifact type.
* Given captured content
* When parsing
* Then the document type is inferred from evidence only
---
### E-003 Structural Reassembly Engine
* US-005 [MUST] As a system, I generate hierarchical sections.
* Given parsed concepts
* When assembling
* Then headings reflect logical structure
* US-006 [MUST] As a system, I preserve nuance without redundancy.
* Given overlapping content
* When assembling
* Then redundancy is collapsed without information loss
---
### E-004 Redundancy Collapse and Nuance Preservation
* US-007 [SHOULD] As a system, I merge repeated explanations.
* Given repeated concepts
* When assembling
* Then explanations are unified
---
### E-005 Output Validation Heuristics
* US-008 [COULD] As a system, I validate standalone readability.
* Given final Markdown
* When validated
* Then no external references are required
---
## 13. Traceability Map
| Story | Epic | Journey | Screen | Priority |
| ------ | ----- | ------- | ------ | -------- |
| US-001 | E-001 | J-001 | S-002 | MUST |
| US-002 | E-001 | J-001 | S-002 | MUST |
| US-003 | E-002 | J-001 | S-002 | MUST |
| US-004 | E-002 | J-001 | S-002 | MUST |
| US-005 | E-003 | J-001 | S-003 | MUST |
| US-006 | E-003 | J-001 | S-003 | MUST |
| US-007 | E-004 | J-002 | S-003 | SHOULD |
| US-008 | E-005 | J-002 | S-003 | COULD |
---
## 14. Lo-fi UI Mockups (ASCII)
### S-001 Input Surface
Purpose: Accept target web experience
States: empty, loading, error
+-----------------------------+ | Target Input | | [ URL or Source Context ] | | | | [ Start Reconstruction ] | +-----------------------------+
---
### S-002 Processing State
Purpose: Indicate capture and reconstruction
States: loading, error
+-----------------------------+ | Reconstructing Meaning... | | [ Capture ] [ Parse ] [ ✓ ] | +-----------------------------+
---
### S-003 Output Viewer
Purpose: Display Markdown output
States: success, error
+-----------------------------+
| Reconstructed Document |
|---|
| # Title |
| ## Section |
| ... |
| +-----------------------------+ |
---
## 15. Decision Log
| ID | Question | Winner | Confidence |
| ----- | ------------------- | ------------------------ | ---------- |
| D-001 | Output format | Markdown only | 0.95 |
| D-002 | Layout vs meaning | Meaning-first | 0.90 |
| D-003 | Site-specific rules | None | 0.92 |
| D-004 | Multiple outputs | Single artifact | 0.88 |
| D-005 | MVP scope | Core reconstruction only | 0.80 |
All decisions auto-committed.
---
## 16. Assumptions
* MVP timebox is 2 to 4 weeks
* Web-first system
* No compliance requirements for MVP
* Input sources are observable web experiences
* No native applications in MVP
* No paid data dependencies
---
> **This PRD is complete.**
> Copy this Markdown into Word, Google Docs, Notion, or directly into a coding model.