feat: add sampler.py — session value scorer (closes #17) #119

Rockachopa · 2026-04-15T03:03:36Z

Rockachopa commented

2026-04-15 03:03:36 +00:00

What

Build session_sampler that scores and ranks 20k+ sessions by harvest value, so the harvester processes the best ones first.

Scoring strategy

Recency: last 7d=3pts, last 30d=2pts, older=1pt
Length: >50 msgs=3pts, >20=2pts, <20=1pt
Repo uniqueness: first session for a repo=5pts, otherwise=1pt
Outcome: failure=3pts (most to learn), success=2pts, unknown=1pt
Tool calls: >10 invocations=2pts (complex sessions)

CLI

python3 sampler.py --count 100                          # Top 100
python3 sampler.py --repo the-nexus --count 20          # Per-repo
python3 sampler.py --since 2026-04-01                   # Date filter
python3 sampler.py --count 50 --min-score 8             # High-value only
python3 sampler.py --format paths                       # Pipe to harvester

Implementation

Fast scanning: reads only first line + last 8KB of each JSONL file
Processes 20k sessions in seconds (no full parse)
Extracts repos from tool call arguments
Detects failure signals from content keywords
3 output formats: table (human), json (machine), paths (pipeable)
--top-percent mode for percentage-based selection

Acceptance criteria

Scores all 20k sessions (fast scan, not full parse)
Output matches spec scoring weights
Filters by repo, date range, score threshold

Closes #17

## What Build session_sampler that scores and ranks 20k+ sessions by harvest value, so the harvester processes the best ones first. ## Scoring strategy - **Recency**: last 7d=3pts, last 30d=2pts, older=1pt - **Length**: >50 msgs=3pts, >20=2pts, <20=1pt - **Repo uniqueness**: first session for a repo=5pts, otherwise=1pt - **Outcome**: failure=3pts (most to learn), success=2pts, unknown=1pt - **Tool calls**: >10 invocations=2pts (complex sessions) ## CLI ``` python3 sampler.py --count 100 # Top 100 python3 sampler.py --repo the-nexus --count 20 # Per-repo python3 sampler.py --since 2026-04-01 # Date filter python3 sampler.py --count 50 --min-score 8 # High-value only python3 sampler.py --format paths # Pipe to harvester ``` ## Implementation - **Fast scanning**: reads only first line + last 8KB of each JSONL file - Processes 20k sessions in seconds (no full parse) - Extracts repos from tool call arguments - Detects failure signals from content keywords - 3 output formats: table (human), json (machine), paths (pipeable) - `--top-percent` mode for percentage-based selection ## Acceptance criteria - [x] Scores all 20k sessions (fast scan, not full parse) - [x] Output matches spec scoring weights - [x] Filters by repo, date range, score threshold Closes #17

Rockachopa added 1 commit 2026-04-15 03:03:36 +00:00

feat: add sampler.py — session value scorer (#17 ) 55797c8a3e

Timmy approved these changes 2026-04-15 04:13:27 +00:00

Timmy left a comment

Feature implementation reviewed - looks solid.

Scope: 1 file(s) changed (353+ / 0-)

Suggestions

Feature PR without tests. Consider adding test coverage.
Found 10 print/console.log statements - verify these are not leftover debugging.

Feature implementation reviewed - looks solid. **Scope**: 1 file(s) changed (353+ / 0-) ### Suggestions - Feature PR without tests. Consider adding test coverage. - Found 10 print/console.log statements - verify these are not leftover debugging.

Rockachopa commented

2026-04-16 01:54:16 +00:00

Closing: unmergeable due to conflicts or branch protection. Reopen if needed.

Rockachopa closed this pull request

2026-04-16 01:54:18 +00:00

Rockachopa commented

2026-04-16 01:54:28 +00:00

Closing: unmergeable due to conflicts or branch protection. Reopen if needed.