Files
learning-garden/sources/psb-gemma-2026-04-14.md

94 lines
2.9 KiB
Markdown

---
pageType: source
id: source.psb-gemma-2026-04-14
title: psb-gemma-2026-04-14
sourceType: local-file
sourcePath: /home/topher/.openclaw/workspace-psb-gemma/memory/2026-04-14.md
ingestedAt: 2026-05-03T01:15:05.946Z
updatedAt: 2026-05-03T01:15:05.946Z
status: active
growth: seed
---
# psb-gemma-2026-04-14
## Source
- Type: `local-file`
- Path: `/home/topher/.openclaw/workspace-psb-gemma/memory/2026-04-14.md`
- Bytes: 2321
- Updated: 2026-05-03T01:15:05.946Z
## Content
````text
# 2026-04-14 - Tuesday
## Bootstrap Trimmed Today
Total bootstrap reduced from ~32KB to ~15.8KB:
- BREWERY_LAYOUT.md → memory/brewery-layout.md (on-demand)
- SESSION-STATE.md → memory/session-state.md (on-demand)
- HEARTBEAT.md → memory/heartbeat.md (on-demand)
- price_list.md → memory/price-list.md (on-demand)
- MEMORY.md trimmed from 9KB to 1.8KB (distilled to identity + critical rules only)
Everything still exists — just not reinjected every session. Read from memory/ when needed.
Also: `contextInjection: "continuation-skip"` applied to agents.defaults (skips bootstrap reinjection on normal continuation turns, still rebuilds after compaction/heartbeat).
## How to Undo
**contextInjection:**
```json
// Remove from agents.defaults in openclaw.json:
"agents": { "defaults": { "contextInjection": "continuation-skip" } }
// → revert by removing line or setting to "always"
```
**Bootstrap files:** Move back from memory/ to workspace root.
---
## Corrections Received Today
- Failed to read context files at session start → logged to shared mistakes
- Session-per-message overhead → identified contextInjection as likely cause
- Repeating consecutive replies → logged
- Self-improving skill missing mistakes.md → created it
## Active Fermentations
- HotForTeacher
- 70megOhm
- Seltzer Base (Orange TILT, targeting 10-12% ABV)
## BSMX Parser — Testing Phase
**Goal:** Reverse-engineer BeerSmith BSMX formula so we can fill in the order spreadsheet autonomously (later this summer).
**Current status:** Hybrid parser works — PDF for absolute grain weights, BSMX for percentages/water/yeast/mash/fermentation. Skill saved at `skills/bsmx/SKILL.md`.
**Next phase (when time permits):** With 10+ recipe pairs (BSMX + PDF), try to derive BeerSmith's internal grain calculation:
```
grain_lb = f(batch_size, efficiency, grain_yield)
```
**Files:**
- `skills/bsmx/parser.py` — main hybrid parser
- `skills/bsmx/SKILL.md` — skill doc
- `skills/bsmx/notes.md` — working notes
- `recipes/70_Meg_Omega_2_2024.bsmx.txt` — test file with known PDF
## Order Spreadsheet Context
Topher wants to upload BeerSmith recipes → get clean data → auto-fill the order spreadsheet. We're in testing/validation phase now. When the parser is reliable, it'll go into production workflow.
````
## Notes
<!-- openclaw:human:start -->
<!-- openclaw:human:end -->
## Related
<!-- openclaw:wiki:related:start -->
- No related pages yet.
<!-- openclaw:wiki:related:end -->