azcomputerguru/claudetools
1
0
Fork 0
Code Issues 58 Pull Requests Actions Packages Projects Releases Wiki Activity

sync: auto-sync from DESKTOP-0O8A1RL at 2026-05-13 10:19:52

Browse Source 
Author: Mike Swanson
Machine: DESKTOP-0O8A1RL
Timestamp: 2026-05-13 10:19:52
This commit is contained in:
Mike Swanson
2026-05-13 10:20:07 -07:00
parent 4828be10e2
commit c74e5bccbb
431 changed files with 250844 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

158
clients/grabb-durando/ai-demand-review/NICHOLS-case-analysis.md Normal file
View File

@@ -0,0 +1,158 @@
# Nichols Case Analysis — Folder Structure & Prompt Structuring
**Source:** NICHOLS, LUIS.rar (SMB case folder)
**Purpose:** Cross-reference real case folder structure with Robert's template; analyze demand letter construction to inform prompt design
---
## 1. Actual Folder Structure vs. Template
Robert's template uses 11 numbered folders. The real SMB case uses named folders with sub-folders:
| Real Folder | Maps To Template | Notes |
|---|---|---|
| `INTAKE/` | 01 Intake and Client Facts | Fee agreement, intake form, liens & costs summary |
| `ACCIDENT REPORT/` | 02 Liability Documents | TPD report + records request |
| `PHOTOS TAKEN BY CLIENT/` | 03 Photos and Videos | Vehicle damage photos |
| `DOCS FROM CLIENT/` | 01 (partial) | IDs, insurance cards, UC summary, transaction receipts |
| `MEDICAL RECORDS & BILLS/` | 05 + 06 combined | All provider R&B in one folder |
| `LIENS/Cigna/` | 07 Liens and LOPs | ERISA proof letters, Rawlings/Machinify correspondence |
| `WAGE LOSS/` | 08 Wage Loss | Wage loss form |
| `CORRESPONDENCE/State Farm-Def/` | 09 Adjuster Correspondence | Rep letters, SF responses |
| `CORRESPONDENCE/Geico-UM UIM/` | 09 (partial) | UIM carrier correspondence, DEC page, LOA |
| `SETTLEMENT DEMAND/` | 11 AI Output | Demand letter (this is OUTPUT, not input) |
| `IMPACT STATEMENT/` | Not in template | Client's written impact narrative — critical input |
| `MEDICAL REQUESTS/` | Not in template | Admin correspondence to obtain records — NOT useful to AI |
| `COSTS/` | Not in template | Paid receipts for record retrieval — NOT useful to AI |
| `NOTES/`, `NOTES.docx` | Not in template | Attorney case notes — high value |
| `Eval Sheet - Nichols.xlsx` | Not in template | Case facts, parties, insurance info, summary — high value |
**Critical finding:** The firm does not use Robert's 11-folder template in practice. The app must accommodate the real named-folder structure, not the template.
---
## 2. Document-to-Demand-Letter Cross-Reference
| Demand Letter Section | Source Documents |
|---|---|
| Opening — claim number, adjuster, settlement deadline | `Eval Sheet - Nichols.xlsx` (Cassandra Luque, claim 03-91P4-55L) |
| Total specials ($7,538.68) | `INTAKE/Liens & Medical Costs.pdf` |
| Liability narrative — police diagram, A.R.S. citation, driver statements | `ACCIDENT REPORT/TPD Report.pdf` |
| Property damage description | `PHOTOS TAKEN BY CLIENT/` (referenced visually) |
| Medical chronology (Oct 6 – Jan 7) | `MEDICAL RECORDS & BILLS/` — four providers |
| Specials breakdown by provider | `MEDICAL RECORDS & BILLS/` — itemized from bills |
| Pain/function/loss of enjoyment | `IMPACT STATEMENT/Impact Statement.pdf` |
| Wage loss narrative (3 weeks, career adjustment) | `WAGE LOSS/Wage Loss Form.pdf` + Impact Statement |
| Future damages (residual at discharge) | `MEDICAL RECORDS & BILLS/NWMC R&B.pdf` (PT discharge notes) |
| Causation / no pre-existing condition | `MEDICAL RECORDS & BILLS/` (PT notes: "no past medical history") |
| Arizona jury value framing | No source doc — attorney judgment / style |
| Settlement posture / deadline | No source doc — firm boilerplate |
**Documents NOT referenced in the demand letter:**
- `MEDICAL REQUESTS/` — admin only, never cited
- `COSTS/` — admin only, never cited
- `CORRESPONDENCE/` — rep letters and SF/Geico responses not cited in demand
- `LIENS/` — Cigna lien acknowledged but not yet resolved at time of demand
---
## 3. Key Observations for Prompt Design
### 3a. Document categories to INCLUDE in AI prompt
Send these to Claude, labeled by category:
1. **Eval Sheet / Intake Form** — parties, claim numbers, accident summary, injuries, insurance
2. **Police Report** — liability facts, diagram, citation, A.R.S. section, driver statements
3. **Medical Records & Bills** — full chronology, diagnoses, treatment progression, discharge notes, provider totals
4. **Impact Statement** — client narrative, functional limitations, wage loss story
5. **Wage Loss Form** — documented lost work and earnings
6. **Liens correspondence** — lien holder identity, amounts if known, ERISA status
7. **Vehicle photos** — Claude vision; describe damage for liability section
8. **Attorney notes** — attorney's own framing and emphasis points
### 3b. Document categories to EXCLUDE from AI prompt
These add noise, inflate token count, and contain no case-analysis value:
- `MEDICAL REQUESTS/` — letters to providers requesting records; content is boilerplate
- `COSTS/` — paid receipts for CIOX/ChartSwap; dollar amounts already in Liens & Costs summary
- `CORRESPONDENCE/` rep letters — standard form letters; no case facts
- `SETTLEMENT DEMAND/` — this is the OUTPUT, never feed back as input
- `Thumbs.db` — Windows thumbnail cache, always skip
- Duplicate files (e.g., "Northwest Urgent Care R&B (dup).pdf")
### 3c. Demand letter style observations
Robert's demand letter is significantly more aggressive and structured than a generic PI demand. The prompt must capture this:
- **No opening dollar demand** — demands a written offer from the adjuster instead
- **Trial-ready framing throughout** — "the first liability exhibit for the jury will be..."
- **Preemptive defense rebuttal** — addresses "minor impact / soft tissue" objection directly
- **Arizona-specific statutory language** — A.R.S. citations, jury instruction framing
- **Human story leads** — client's age, occupation, and how injury affects livelihood
- **Specials presented as gross billed**, with exclusions clearly noted
- **Future damages discussed without a dollar figure** — flags residual findings without fabricating a number
- **Eggshell plaintiff doctrine addressed explicitly**
- **Specific instructions to the adjuster** on how to format their response
### 3d. Prompt restructuring recommendations
**Robert's current prompt** asks for 10 items in one pass. Recommend splitting into two stages:
**Stage 1 — Fact extraction (per document category):**
```
For each uploaded document, extract:
- Dates and timeline entries
- Provider names and diagnoses
- Dollar amounts (bills, liens, lost wages)
- Liability facts (parties, citations, statements)
- Client-reported symptoms and limitations
- Missing items (providers referenced but no records received)
Output as structured JSON — not prose.
```
**Stage 2 — Demand letter generation (using Stage 1 output):**
```
Using the extracted facts, draft a settlement demand letter in the style of
Grabb & Durando, P.C.:
- No opening dollar demand; request a written offer by [deadline]
- Lead with liability — cite the police report, citation, A.R.S. section, driver statements
- Frame property damage using photo descriptions
- Include a full medical chronology in date order, by provider
- Calculate gross billed specials by provider, total them, note exclusions
- Human story section: client's age, occupation, how injury affects earning ability
- Address the anticipated "soft tissue / minor impact" defense preemptively
- Reference Arizona jury value and jury instruction framing
- Specify what the adjuster's response must include
- Future damages: flag residual findings without fabricating a dollar figure
- Eggshell plaintiff doctrine if pre-existing susceptibility is plausible
```
**Why two stages:** Stage 1 with structured JSON output gives the app an auditable fact base. Staff can review the extracted facts before the demand is drafted. If a bill amount or date is wrong, they catch it at Stage 1, not buried in a paragraph.
---
## 4. App Design Implications
1. **Folder mapping UI needed** — since the real folder names don't match the template, the upload UI should let staff map their folders to the standard categories (or auto-detect by name)
2. **Exclusion logic** — app should auto-exclude MEDICAL REQUESTS, COSTS, Thumbs.db, and duplicates
3. **Eval Sheet parsing** — the Excel file is high-value structured data; parse it separately and inject as JSON context (don't treat as a document to summarize)
4. **Photo handling** — Claude vision for vehicle damage photos; describe damage for liability section
5. **Specials tally** — Stage 1 should output a provider-by-provider specials table that staff verifies before Stage 2 runs
6. **Tracked changes** — firm keeps `.tracked.docx` versions; app should save both clean and tracked output
7. **Lien status** — Cigna/Rawlings lien was still unresolved at demand time; app should flag "lien unresolved" as a checklist item
---
## 5. Updated Document Category Mapping (for app)
| App Category Label | Real Folder(s) | Priority |
|---|---|---|
| Intake / Case Facts | `INTAKE/`, `Eval Sheet`, `DOCS FROM CLIENT/` | High |
| Liability / Police Report | `ACCIDENT REPORT/` | High |
| Photos | `PHOTOS TAKEN BY CLIENT/` | High (vision) |
| Medical Records & Bills | `MEDICAL RECORDS & BILLS/` | High |
| Impact Statement | `IMPACT STATEMENT/` | High |
| Wage Loss | `WAGE LOSS/` | High |
| Liens & LOPs | `LIENS/` | Medium |
| Adjuster Correspondence | `CORRESPONDENCE/State Farm-Def/`, `CORRESPONDENCE/Geico-UM UIM/` | Low |
| Attorney Notes | `NOTES/`, `NOTES.docx` | Medium |
| SKIP | `MEDICAL REQUESTS/`, `COSTS/`, `SETTLEMENT DEMAND/`, Thumbs.db | Exclude |