- user_gear table (migration 008) + GET/PUT /api/gear (brewer keys
validated against the taxonomy, grinders deduped/trimmed)
- /gear page: silhouette multi-select for owned brewers (autosaves) and
a grinder list; nav gains Brewing → Equipment
- Brews page: with gear configured, the picker shows only owned brewers
(plus the edited brew's method) with a show-all toggle; grinder field
suggests your grinders; empty-gear state links to /gear
- Included in full backups, per-user export, and the OpenAPI spec
Co-Authored-By: Claude Fable 5 <[email protected]>
The extraction schema gains cultivarGroup (typica/bourbon/ethiopian/
hybrid, classified by the LLM from coffee genetics); when the named
variety isn't in the cultivar table, deriveFields now falls back to the
group's first-crack anchor (number still comes only from the reference
table) instead of leaving 1.4 blank and breaking every downstream
calculation. deriveFields exported + unit-tested.
Co-Authored-By: Claude Fable 5 <[email protected]>
Free-text recipe (bloom, pours, timings) distinct from tasting notes and
next-time notes: new brews.recipe column (migration 007), form textarea,
shown under the recipe summary in brew history, carried by Brew again,
included in backups and the OpenAPI spec.
Co-Authored-By: Claude Fable 5 <[email protected]>
Brewing:
- roasted_beans + brews tables; /beans (bean management with LLM URL
prefill) and /brews (silhouette brewer picker across immersion/
percolation/espresso, recipe fields, auto ratio, 0-10 rating, tasting
notes); bean remaining weight derived from logged brew doses
- Green inventory lot form also prefills from a product URL
Navigation/UX:
- Side nav is now generated from one definition in nav.js, grouped
Roasting / Brewing / account, consistent on every page
LLM:
- 'Ask the LLM' chat drawer on the planner (stateless /api/plan-chat)
grounded in the plan, computed ledger, learned pace, and a new
roaster-behavior profile aggregated from uploaded .alogs
(/api/roaster-profile: TP lag, phase RoR, median milestone temps)
- The profile also feeds roast reviews and the planner curve's fallback
milestone temps
API platform:
- User-generated bearer tokens (rpt_…) with account-page management;
token requests skip CSRF; hand-authored OpenAPI 3 spec at
/api/openapi.json rendered by self-hosted Swagger UI at /api-docs
- Full-database backup export/import (admin) + per-user data export
Co-Authored-By: Claude Fable 5 <[email protected]>
- New server/llm.js consolidates ModelRuntime + model selection: admin
setting (app_settings.llm_model) > LLM_MODEL/PREFILL_MODEL env > first
available; used by both prefill and roast evaluation
- GET/PUT /api/admin/llm lists configured models and stores the choice
(validated against the list; empty = auto; audited); admin page gains
an LLM section with a model picker
- All user-facing 'Pi agent' wording is now 'LLM'; no_model error message
no longer references the pi CLI
- /roasts table: Review column removed (review lives in the detail view)
Co-Authored-By: Claude Fable 5 <[email protected]>
- POST /api/roasts stores the original .alog verbatim (multiple per plan),
parses it server-side, and kicks off an async zero-tool Pi agent review
(grade, highlights, concerns, next-batch suggestions, plan-vs-actual)
- /roasts page: table of historical actual roasts with review summaries;
row detail renders the BT curve as SVG with the plan curve overlaid,
shows the full review, and offers original-.alog backup download,
re-evaluate, and delete
- Planner's Reference curve drawer gains a multi-file finished-roast
uploader that attaches to the open (synced) plan
- Failed reviews record the error on the row and are retryable via
POST /api/roasts/:id/evaluate (e.g. once a model is configured)
Co-Authored-By: Claude Fable 5 <[email protected]>
Replaces the flat additive batch-size correction with a per-user
multiplicative pace factor learned from each account's own logged
roasts (shared/learn.js, GET /api/machine-profile), plus a lot-scoped
"last refine" auto-suggestion. Also fixes the reference data being
framed as "your own roasts" on a now-multi-user product, and
reclassifies the drying/Maillard sanity checks as informational since
they're algebraically derived from the DTR check rather than
independent (yellow = 0.56 x first crack, not entered separately).
Bug fixes found by an adversarial Opus review of the first pass:
- Unicode minus sign (U+2212) broke duration parsing against the
app's own generated refine-suggestion text
- Printed sanity-checks table still showed a bare pass/fail glyph for
the now-informational drying/Maillard rows
- Printed time-ledger box didn't show the pace multiplication step,
so it stopped reconciling by hand once pace != 1
- Field 6.4 (manual batch correction) was double-counted: excluded
from the learned-pace fit but added back after the multiplication
- Learned pace had no outlier rejection or hard clamp
- computeLedger had no test coverage
- Batch-size help copy overstated what the pace factor models (it's
a single blanket ratio, not conditioned on batch weight)
A follow-up Opus pass also caught the per-user profile cache
surviving logout/account-switch in a shared browser; fixed by
sweeping it alongside the existing plan-draft cleanup.
Co-Authored-By: Claude Sonnet 5 <[email protected]>
Ports inventory management and SCA-style cupping scoring from
hope_roaster: lot tracking with audit-logged consumption, cupping
sessions with server-authoritative scoring and a live radar chart,
and links from the planner (draw-from-lot, open-cupping-session).
Also fixes the Blend/Single-origin toggle layout, replaces tooltips
with an in-context "why" teaching layer, and stages the planner UI
into pre-roast vs. post-roast phases.
- shared/ ports the worksheet's ledger math, time parsing, and reference tables
(cultivars/processes/roast-levels/machine bands) as browser-safe ESM, imported
by both the server and the browser so the arithmetic can't drift between them.
- server/prefill.js runs a zero-tool Pi Coding Agent SDK turn to extract page facts
from a bean product URL, then derives worksheet field IDs deterministically from
reference-data.js — the model never invents a first-crack anchor or a modifier.
- server/alog.js ports the Python alog_parser.py's format handling, including the
tokenizer-based Python-dict-literal-to-JSON conversion the real files need.
- public/ is the worksheet reproduced as a live HTML form: worksheet.css is a
verbatim copy of the paper worksheet's print CSS, print.js bakes values into
the print layout so the same DOM renders both on screen and on paper.
- Ledger math verified against both of the paper worksheet's worked examples;
alog parser verified against all 14 real logs in ref/roasts/.
Co-Authored-By: Claude Sonnet 5 <[email protected]>
snowspeederandClaude Sonnet 5