amosblomqvist
52c02533d1
Per-batch pipeline is now extract → spine → audit → merge (owner order, 2026-09-30), the spine inside the serialized audit section so it reads the graph exactly as the following audit will. Measured motive: 0 of 133 chains in the live corpus crossed a stretch boundary — no stage minted cross-stretch chains. - src/agent/subagents/spine.ts: SESSION/NOTES spine preambles + shared SPINE_CORE (span doctrine; extraction doctrines carried over), the submit_spine TypeBox schema, validateSpine (mechanical span law: every chain touches >=1 fresh endpoint AND >=1 existing digest endpoint; unknown ids, self-reference, and id collisions are loud), the digest prompt builder, and the runner (strong tier, retry-only). - extraction.ts: 'assembling' phase, spine stage in the per-need gate, usage/billing stage, assemble() inside withAuditLock for both front doors; first stretch of a source skips the call (empty digest, no spend); the audit sees COMBINED propositions (chunk + connectives), never chains; merger receives the combined findings. - core/graph.ts mergeFindings: chain endpoints resolve through the job-local map OR pass through verbatim as existing n_ ids (the spine's mixed namespace). - EXTRACTION_CORE: steps-of-reasoning selection bar (examples, notation glosses, and restatements that no argument needs are not steps). - Factory chunk budget 4096 -> 8192 (fewer seams, more local context). - healStringifiedArrays moved to the shared submit seam (submit-agent.ts). - Cost panel: Assembly row; usage records carry a spine stage (optional on legacy records). - Tests +5: prompt pins, span-law matrix, digest rendering, seeded end-to-end flow (spanning chain + connective, verbatim n_ endpoint, spine billed as its own stage), non-spanning refused + retried.
144 lines
6.9 KiB
TypeScript
144 lines
6.9 KiB
TypeScript
import path from "node:path";
|
|
import { fileURLToPath } from "node:url";
|
|
|
|
/** Repository root (this file is at <root>/config.ts). */
|
|
export const ROOT = path.dirname(fileURLToPath(import.meta.url));
|
|
|
|
/**
|
|
* THE data-location seam (the Electron contract, 2026-09-17): LH_DATA_DIR
|
|
* relocates every MUTABLE record — chat sessions, graph ledgers, the
|
|
* embedding model cache — when set. The desktop shell points it at
|
|
* userData so nothing is written inside the read-only .app bundle
|
|
* (macOS translocation guarantees the bundle is read-only anyway).
|
|
* Unset (plain web dev / repo work) it resolves to the repo's data/,
|
|
* byte-for-byte as before. Read-only assets (prompts) stay ROOT-anchored
|
|
* on purpose: they ship inside the app bundle by design.
|
|
*/
|
|
const DATA_ROOT = process.env.LH_DATA_DIR
|
|
? path.resolve(process.env.LH_DATA_DIR)
|
|
: path.join(ROOT, "data");
|
|
|
|
export const PATHS = {
|
|
root: ROOT,
|
|
/** Teacher system prompt — the agent's instructions (src/agent/). */
|
|
teacherPrompt: path.join(ROOT, "src", "agent", "system-prompt.md"),
|
|
data: DATA_ROOT,
|
|
/** Chat sessions (AI SDK UIMessage transcripts, one JSON file per chat). */
|
|
chats: path.join(DATA_ROOT, "chats"),
|
|
graph: path.join(DATA_ROOT, "graph"),
|
|
/** Practice problem sets (one JSON document per set — src/store/practice.ts).
|
|
* Bounded self-contained snapshots (propositions + chains frozen at
|
|
* creation), not append-only ledgers. */
|
|
practice: path.join(DATA_ROOT, "practice"),
|
|
} as const;
|
|
|
|
// ─── Models ────────────────────────────────────────────────────────────────
|
|
// BYOK ALL THE WAY DOWN (owner, 2026-09-22): there are NO factory default
|
|
// models. Every engine resolves from the user's model slots (settings.json
|
|
// models section: strong/cheap/vision) over per-role/ChatSettings overrides,
|
|
// and a chain that ends empty REFUSES honestly (503 models_not_configured).
|
|
// .env stays for secrets (API keys) only. What remains here are ENGINE
|
|
// CONSTRUCTION FACTS: the thinking-level scale and the default order.
|
|
|
|
export type ThinkingLevel = "off" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max";
|
|
|
|
/** Every agent role in the app. The conversation pair (prober/teacher)
|
|
* resolves chat.model → strong slot; background roles resolve their own
|
|
* override → tier slot (core/settings-resolution.ts ROLE_SLOT). */
|
|
export type ModelRole =
|
|
| "prober"
|
|
| "teacher"
|
|
| "researcher"
|
|
| "demoBuilder"
|
|
| "visualMaker"
|
|
| "pdfReader"
|
|
| "extractor"
|
|
| "spine"
|
|
| "auditor"
|
|
| "partitioner"
|
|
| "problemGenerator"
|
|
| "grader";
|
|
|
|
/**
|
|
* The LIVE graph store (src/store/graph.ts) — owned by the app's extraction
|
|
* pipeline. Line shapes: src/core/contracts.ts. CHAINS ARE THE STORED
|
|
* TRUTH (owner, 2026-09-30): edges are a derived view, never stored.
|
|
*/
|
|
export const LEDGER_FILES = {
|
|
nodes: path.join(PATHS.graph, "nodes.jsonl"),
|
|
chains: path.join(PATHS.graph, "chains.jsonl"),
|
|
sessions: path.join(PATHS.graph, "sessions.jsonl"),
|
|
/** Per-message extraction coverage (contracts/graph.ts CoverageEntry;
|
|
* the checkpoint's replacement for chat extractions — notes keep
|
|
* sessions.jsonl, whole-file as they are). */
|
|
coverage: path.join(PATHS.graph, "coverage.jsonl"),
|
|
} as const;
|
|
|
|
/**
|
|
* Notes-import bookkeeping (NOT a ledger): one JSON map, src-id → NotesSource
|
|
* record (contracts/graph.ts) — the file path + import time + the run's usage
|
|
* log. Read-modify-write like a chat file, not append-only like the ledgers.
|
|
*/
|
|
export const SOURCES_FILE = path.join(PATHS.graph, "sources.json");
|
|
/** The layout ledger (read-modify-write like sources.json): nodeId →
|
|
* {x,y} resting positions from the incremental settle (core/layout.ts).
|
|
* Carries an updatedAt stamp for future live-physics conflict handling. */
|
|
export const POSITIONS_FILE = path.join(PATHS.graph, "positions.json");
|
|
|
|
/**
|
|
* Graph render settings (the settings-panel controls): one small JSON blob,
|
|
* read-modify-write. Shared shape/sanitizer: src/core/render-settings.ts.
|
|
*/
|
|
export const RENDER_SETTINGS_FILE = path.join(PATHS.graph, "render-settings.json");
|
|
|
|
/**
|
|
* THE USER SETTINGS RECORD (settings-storage wave, 2026-09-19): one
|
|
* versioned, DELTA-shaped JSON blob at the data-dir seam holding every
|
|
* non-secret product preference — general, agent (background
|
|
* model/thinking overrides + tool toggles + prompt-override switches),
|
|
* appearance, graph. Explicit user choices only: ABSENT keys resolve to the
|
|
* factory defaults compiled into src/core/settings.ts (FACTORY_SETTINGS), so
|
|
* a software update that adds a knob (new tool, new role, new appearance
|
|
* field) needs no migration — old records simply lack the new keys. The
|
|
* version field is a PRIVATE schema counter for breaking shape changes
|
|
* only (field renamed/meaning changed), unrelated to the app release
|
|
* version. Secrets (API keys, login tokens) NEVER live here — auth.json /
|
|
* .env.local keep that role. (The `chat` default-settings section died
|
|
* with per-session models, owner 2026-09-28 — sessions own their model
|
|
* from birth, stamped from the strong slot.)
|
|
*/
|
|
export const SETTINGS_FILE = path.join(DATA_ROOT, "settings.json");
|
|
|
|
/**
|
|
* User prompt-override files (settings-storage wave, 2026-09-19): the
|
|
* per-segment system-prompt overrides mirroring the SOURCE layout
|
|
* (src/agent/prompts/) — the shared base as prompts/system-prompt.md,
|
|
* fragments as prompts/teacher/<file> / prompts/prober/<file>. File
|
|
* PRESENCE alone is not the override: settings.json's agent.prompts map
|
|
* (segmentPath → boolean) is the explicit switch, so a user can keep a
|
|
* customized file on disk but run the source default to compare. The
|
|
* segment list itself is ALWAYS derived from source at runtime (the
|
|
* loader in src/store/prompt-overrides.ts) — code updates that add,
|
|
* remove, or reorder fragments adapt automatically; an override keyed to
|
|
* a vanished segment stops composing but stays on disk.
|
|
*/
|
|
export const PROMPTS_OVERRIDE_DIR = path.join(DATA_ROOT, "prompts");
|
|
|
|
/**
|
|
* THE AUTH/KEY STORE (auth-keys wave, 2026-09-19): BYOK provider keys in a
|
|
* flat {VAR: value} JSON map — the SAME file the Electron app has always
|
|
* hand-edited (owner decision: option A, zero migration). Desktop main.cjs
|
|
* passes LH_AUTH_FILE pointing at userData/auth.json (mode 0600); in dev
|
|
* (no Electron) the default is the repo-root auth.json (gitignored, same
|
|
* pattern as .env.local — a worktree may symlink it to the master copy).
|
|
* The SERVER reads/writes this file directly at call time (no restart to
|
|
* change a key); it reflects the values into process.env at boot with
|
|
* FILL-GAP precedence — real environment variables (incl. .env.local in
|
|
* dev) always win (src/server/keys.ts). Hand-edits still work: the file is
|
|
* re-read from disk on every API access. Secrets NEVER go in settings.json
|
|
* — this file keeps that role, alongside .env.local.
|
|
*/
|
|
export const AUTH_FILE = process.env.LH_AUTH_FILE
|
|
? path.resolve(process.env.LH_AUTH_FILE)
|
|
: path.join(ROOT, "auth.json");
|