EXECUTE NOW

Target: $ARGUMENTS

Parse immediately: - If target is empty: evaluate all scenarios in test/scenarios/ - If target is a scenario name: evaluate only that scenario - If target is "compare": evaluate all and compare against a previous run if one exists

Execute these steps:

1. Discover scenario files

ls test/scenarios/*.md

Read each scenario file. Parse the frontmatter for frequency and the step list for Source paths and Hops values.

2. Measure fixed instruction bytes

For each step marked Fixed/Variable: fixed or Fixed/Variable: mixed, read the file referenced in Source and measure its byte count:

wc -c < {source-path}

Common fixed sources you'll encounter: - CLAUDE.md — always-loaded context (count once, not per-step) - kb/notes/COLLECTION.md — writing conventions for the notes collection - kb/instructions/cp-skill-connect/SKILL.md — connection skill body - kb/instructions/cp-skill-ingest/SKILL.md — ingestion skill body - kb/instructions/cp-skill-snapshot-web/SKILL.md — snapshot skill body - kb/sources/types/source-review.md — source review type template - the scenario fixture template under test/scenarios/types/

Important: Count CLAUDE.md bytes once per scenario (it's always loaded, not a per-step cost). Count other fixed sources once per scenario even if referenced in multiple steps (they stay in context after first load).

3. Estimate variable instruction bytes

For steps marked Fixed/Variable: variable, use these configurable defaults:

Variable element	Default estimate
Search results (note count)	3 notes
Average note size	2,000 bytes
Link-follow reads	2 notes
Area index reads	1 index at ~3,000 bytes
External URL content	5,000 bytes

If $ARGUMENTS includes override estimates (e.g. "notes=5 notesize=3000"), use those instead.

4. Count hops per scenario

Parse the Hops field from each step. Sum: - Fixed hops: steps where the hop count is a single number (e.g. "1") - Variable hops: steps where the hop count is a range (e.g. "2-4") — use the midpoint - Total hops: fixed + variable

For escalation steps, count separately and weight at 0.1 (estimated 10% occurrence).

5. Produce the cost table

Output format:

## Scenario Cost Report

Generated: {date}
Variable estimates: {defaults or overrides used}

### Per-Scenario Breakdown

#### {Scenario name} (frequency: {freq})

| Step | Hops | Fixed bytes | Variable bytes | Source |
|------|------|-------------|----------------|--------|
| 1. {name} | 0 | 5,251 | — | CLAUDE.md |
| 2. {name} | 3 | — | ~6,000 | search results |
| ... | | | | |
| **Subtotal (common path)** | **N** | **X** | **~Y** | |
| E1. {name} | ... | ... | ... | ... |
| **Subtotal (escalation, ×0.1)** | **N** | **X** | **~Y** | |

{Repeat for each scenario}

### Summary Table

| Scenario | Freq | Weight | Hops (common) | Hops (w/ esc.) | Bytes (fixed) | Bytes (total est.) |
|----------|------|--------|---------------|----------------|---------------|-------------------|
| Write a note | common | 1.0 | N | N | X | ~Y |
| Ingest a source | occasional | 0.3 | N | N | X | ~Y |
| Answer a question | common | 1.0 | N | N | X | ~Y |
| Respond to a change | occasional | 0.3 | N | N | X | ~Y |

### Weighted Totals

| Metric | Value |
|--------|-------|
| Weighted hops (common path) | N |
| Weighted hops (with escalation) | N |
| Weighted fixed bytes | X |
| Weighted total bytes | ~Y |

### Observations

{Note any architectural implications:
- Which scenario dominates the weighted cost?
- Where are the biggest variable cost drivers?
- What architectural changes would reduce costs?
- How does the current design compare to alternatives (e.g. inlining types saved N bytes at step X)?}

6. Verify plausibility

After producing the table, sanity-check key numbers: - CLAUDE.md should be in the range of 4,000-7,000 bytes - COLLECTION.md files should be in the range of 4,000-8,000 bytes - Skill files (connect, ingest) are typically 10,000-20,000 bytes - If any number looks implausible, re-read the source file and re-measure

START NOW. Read all scenario files, measure all referenced source files, produce the cost table.