dev → master: ephemeral UTAs + IBKR-as-truth scope correction + FRED tools#171
Merged
luokerenx4 merged 15 commits intomasterfrom May 8, 2026
Merged
dev → master: ephemeral UTAs + IBKR-as-truth scope correction + FRED tools#171luokerenx4 merged 15 commits intomasterfrom
luokerenx4 merged 15 commits intomasterfrom
Conversation
Add a sibling subsection to the rolling dev → master rules covering when to break from `dev` and run an isolated branch off `master` instead. Default stays "everything → dev"; the isolation case is narrow but recurring (refactors that touch shared types / interfaces every broker implements, multi-day schema work, etc.). Codifies what we just discussed on PR #168 close-out: branch from freshly-merged master, rebase against master during the refactor, PR straight to master skipping dev, and have dev absorb the merge afterward via `git merge origin/master`. Adds the matching decision rule for sessions starting *after* such a refactor lands so they don't accidentally branch off stale dev. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The local worktree dance (per-worktree pnpm install, data/ copying, port discipline) costs more than it saves on a project this size. Cloud Claude sessions are the right concurrency primitive here — each gets a clean sandbox and returns a PR without disturbing the local dev server / data directory. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
externalTrade models the user manually trading on the exchange app — opening a short (covered call's call leg, naked put, perp short) is a legitimate user action there. The previous guard rejected any SELL without an existing position, which blocked all short-open flows in fixture-based testing. Distinct from placeOrder / _applyFill — those still reject SELL-without-position because Alice shouldn't silently flip a SELL intent into a short open. The asymmetry is intentional: external events let the user do whatever a real exchange allows; Alice's own orders go through stricter intent verification. Found while running a covered call scenario through the simulator console — first concrete coverage gap surfaced by the manual fixture testing loop. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Mock simulator UTAs were registered through the same persistence path as real broker accounts (Alpaca/IBKR/Bybit) — config row in accounts.json + persistent commit history under data/trading/<id>/. That meant test residue from one session polluted cost-basis calculations in the next: 7 reconcileBalance ops on AAPL accumulated over multiple QA sessions, weighted-average-cost the long-since-stale $198 instead of the actual trade price. Add ephemeral as a first-class UTA flag: - utaConfigSchema: ephemeral?: boolean, gated by a refine check that rejects ephemeral: true on any non-mock-simulator preset (would silently destroy real broker history at next boot). - Boot path (main.ts): purgeEphemeralUTAs runs before UTAManager initializes anything. For each ephemeral entry it wipes data/trading/<id>/ and removes the row from accounts.json, then the manager initializes only the survivors. Each session starts ephemeral mocks from a clean slate. - POST /api/trading/config/uta: accepts ephemeral: true in the payload (passes through utaConfigSchema's refine). - DELETE /api/trading/config/uta/:id: when the deleted UTA was ephemeral, also wipes data/trading/<id>/. For real brokers delete-from-config still preserves commit history (the audit trail outlives the connection); only ephemeral test UTAs are fully erased. Closes the "fixture import/export and persistence" line item from the simulator-console v2 backlog by establishing the test-account lifecycle that fixture-based testing depends on. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Surfaced via simulator covered-call test (see fix log: 0e09127 + ea8e1d2 in same session). The reconcileBalance + git WAC pipeline that solves CCXT spot's faked avgCost generalized over-aggressively and now destroys correct avgCost reported by brokers that actually know it (Mock, Alpaca, IBKR). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`_reconcileWalletPositions` was hardcoding the synthesized
`reconcileBalance` op's price to `position.marketPrice`, ignoring
whatever avgCost the broker had already supplied. The mark-bootstrap
behavior was designed for CCXT spot — where the broker fakes
`avgCost = markPrice` because it has no order history — but it
generalized over-aggressively and destroyed correct cost basis on
brokers that DO observe real fill prices.
Concrete case (covered call surfaced 2026-05-07): MockBroker
externalTrade records BUY 100 AAPL @ $148.50 → broker.avgCost is
$148.50. Mark moves to $152. UTA pipeline ran:
recordReconcile({ markPrice: position.marketPrice }) // = $152
Result: avgCost overwritten to $152, unrealizedPnL = 0, the +$350
real gain disappeared from the UI.
Fix: prefer `position.avgCost` when non-zero, fall back to
`marketPrice` only when the broker has nothing. Behavior matrix:
- 'broker' avgCostSource: filtered out before this loop, no-op
- 'wallet' + broker.avgCost > 0 (Mock externalTrade, future
CCXT-with-fetchMyTrades): bootstrap at broker.avgCost — fixes
the cost-basis-destruction bug
- 'wallet' + broker.avgCost == 0 or missing: bootstrap at
markPrice — preserves the CCXT-spot fallback path
- 'wallet' + CCXT spot today (avgCost == markPrice by
construction): bootstrap at avgCost == bootstrap at markPrice,
behavior unchanged
Verified end-to-end through the simulator covered-call scenario.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
FRED was completely broken on the SDK path. Three independent failures
in the credential chain had to misalign at the same time for the request
to fail without a clear error — and they all did:
- src/webui/routes/config.ts test-provider entry referenced a
nonexistent `fred` provider; the registry has `federal_reserve`.
- credential-map's single mapping table sent `fred_api_key` to both
paths, but the in-process opentypebb SDK auto-prefixes declared
credentials with the provider name (Provider constructor at
packages/opentypebb/src/core/provider/abstract/provider.ts:54-59),
so it actually expects `federal_reserve_api_key`.
- getFredApiKey only looked for `fred_api_key` / `api_key`, missing
the prefixed form even when the credential made it through.
On top of that, two data-quality bugs:
- fetchFredSeries defaulted to `sort_order=asc`, so any limit-only
query returned 1946-era observations (GDP `limit=5` first row was
1946-01-01, value `.`).
- fredRegionalApi had three independent errors in one helper: wrong
base URL (`/fred/geofred/...` 404s; GeoFRED is at `/geofred/...`),
wrong query param name (`series_group` should be `series_id`), and
wrong response-parse path (`data.data` should be `data.meta.data`).
Fixes:
- src/webui/routes/config.ts: test-provider fred entry now uses
`provider: 'federal_reserve'` + `credField: 'federal_reserve_api_key'`.
- src/domain/market-data/credential-map.ts: split into httpKeyMapping
(legacy Python sidecar HTTP path, `fred -> fred_api_key` — the 5
openbb-api clients still call buildCredentialsHeader and need this)
and sdkKeyMapping (in-process opentypebb, `fred -> federal_reserve_api_key`).
Public signatures unchanged.
- packages/opentypebb fred-helpers.ts: getFredApiKey prefers the
prefixed form with legacy fallbacks; fetchFredSeries defaults to
desc and reverses to ascending before returning so date-merge
consumers keep their shape; new GEOFRED_BASE constant + optional
base param on buildFredUrl; fredRegionalApi uses series_id and
parses data.meta.data.
- packages/opentypebb fred-regional standard model: schema description
reflects series_id semantics.
Tests — the structural piece. market-data had zero e2e specs while
trading has 14 (one per broker, real paper accounts). bbProvider specs
cover only the inner fetcher slice and miss the credential-mapping +
route-wiring layers where the community-reported "fred 配不下来"
failure modes actually live. Filling that gap now while the FRED
bug is fresh:
- src/domain/market-data/__tests__/credential-map.spec.ts (10 unit
cases, CI main suite): pins the two-table contract so a future
merge attempt fails loudly.
- src/domain/market-data/__test__/e2e/{setup,market-data.e2e.spec}.ts:
first market-data e2e harness. Mirrors the trading e2e pattern
(lazy-singleton TestApp + Hono app.request without listening port)
but lighter (no isPaper guard — read-only by construction). Five
FRED cases hit real webui routes, reproducing the actual
user-facing failure path. Future fmp/yfinance/oecd describe blocks
drop in cheaply.
Verified end-to-end against the live FRED API: GDP $31,856B (2026-01-01),
UNRATE 4.3%, California per-capita personal income $85,518.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…e identity Phase 3 of the IBKR-as-truth refactor (commit 4bcf2f4) inserted a normalization step that rewrote CCXT's wire symbol "BTC/USDT:USDT" into "BTC-PERP" before storing it as Contract.localSymbol. That violated the actual architecture: aliceId = "{utaId}|{nativeKey}" nativeKey = broker.getNativeKey(contract) aliceId is the system-level uniqueness primitive. nativeKey is whatever the broker considers its primary key — IBKR uses conId, CCXT uses unified wire symbol, Alpaca uses ticker, etc. We never normalize across brokers; each broker chose its identity scheme for product reasons that shouldn't be erased. The :settle suffix in CCXT's wire format encodes margin variant. Two distinct perp products (USDC-margined vs USDT-margined ETH) collapsed to the same canonical "ETH-PERP", which propagated through getNativeKey and broke aliceId uniqueness within a single Bybit UTA. What "IBKR-as-truth" actually means stays: - Phase 2: buildContract / assertContract — structural validation - Phase 4: strict SecType union — taxonomy Each broker's localSymbol carrier remains broker-native. Changes: - ccxt-contracts.ts: drop canonicalLocalSymbol + ccxtExpiryToCanonical; marketToContract emits localSymbol = market.symbol (wire format). KEEP this session's FUT/OPT field derivation (orthogonal to identity) - CcxtBroker.ts: drop canonical import; searchContracts indexes via direct markets[localSymbol] lookup; resolveNativeKey uses single direct lookup. KEEP this session's resolveContractSync defensive guards (orthogonal to identity) - contract-ext.ts: doc rewritten to describe per-broker uniqueness schemes (IBKR conId, CCXT wire, Alpaca symbol, Mock) - Tests: spec assertions and 5 e2e files updated to wire format. The `secType === 'CRYPTO_PERP'` heuristic added during this session stays — it's a more semantic check than localSymbol.includes() Persisted-state impact: 16 commit.json aliceId entries on bybit-main + okx-test (from today's Phase 3 window) become orphan and re-bootstrap via reconcileBalance at current markPrice on next getPositions. Today's activity was exploratory — no real cost-basis history lost. Out of scope (filed in plan as follow-up): IBKR getNativeKey should return String(conId) instead of localSymbol; non-CCXT brokers' nativeKey schemes deserve a similar audit. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Surfaced during the Phase-3 revert (afddd41) when articulating per-broker uniqueness schemes. IBKR's symbol/localSymbol aren't reliably unique — conId is the actual primary key. Audit + fix across IBKR / Alpaca / LeverUp / Bybit-direct brokers. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The FRED HTTP route + executor path were repaired in a86612c, but nothing in src/tool/ wraps them — the SDKEconomyClient class has been exported all along while no factory ever instantiated it, so AI agents have never been able to read FRED data. Closes that gap and pins the provider name at the tool boundary so the AI never has to know about the federal_reserve provider naming. Three Vercel AI tools, all routed through EconomyClientLike with provider: 'federal_reserve' baked in: - economyFredSearch — keyword → series_id discovery - economyFredSeries — observations for one or more series_ids, comma-separated for date-merged multi-series - economyFredRegional — state-level cross-section for a regional series (e.g. WIPCPI for per-capita income by state) Wiring follows the equity/crypto pattern: branch on backend, both SDKEconomyClient and OpenBBEconomyClient satisfy EconomyClientLike, register once via toolCenter under the 'economy' namespace. Tests at src/tool/economy.spec.ts mirror trading.spec.ts — mock the client surface, verify the three things that silently break when a domain function becomes an AI tool: 1. Schema validation (zod): rejects missing required, rejects non-positive limit, rejects non-integer limit 2. Passthrough: args land at the client unchanged + provider is pinned + optional fields omitted (not sent as undefined) 3. Errors propagate (no swallowing) These run in the CI main suite (`pnpm test`) — unlike the live e2e specs which only run under `pnpm test:e2e`. So a future schema mismatch (LLM passes string where number expected, etc.) fails fast in CI, not in the field. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
… dead URLs
Audited the 8 entries in ALL_PROVIDERS and the parallel test-provider
table at webui/routes/config.ts:156-165. Several were stale or pointing
at services we don't actually implement on the SDK backend.
Removed (no SDK provider; Test would 500 with "Provider not found"):
- nasdaq: Nasdaq Data Link is not registered in
packages/opentypebb/src/providers/. The pre-existing comment at
domain/market-data/equity/symbol-index.ts:13 already flagged it
("intrinio, nasdaq, tradier 暂未接入") but the form still offered
it. Test button was always going to fail.
- tradingeconomics: same problem — no SDK provider. The TE Python
upstream is also paid-only with no free tier worth advertising.
Refreshed (URL pointed at retired or marketing pages):
- bls hint: registrationapps.bls.gov/bls_registration → data.bls.gov/
registrationEngine/. The old host is what the BLS site used to redirect
through; OpenBB upstream's instructions field already points at the
new path, we just hadn't followed.
- econdb hint: "Optional" → "Required (free signup)". Anonymous API
calls now return 401 (verified live), the free-but-keyless era is
over. Status quo would have users seeing inexplicable 401s for
EconDB-routed economy queries.
- eia hint: eia.gov/opendata → eia.gov/opendata/register.php. Direct
registration link instead of the index page.
- fred hint: fredaccount.stlouisfed.org/apikeys → the standard "fred.
stlouisfed.org → My Account → API Keys" flow. The deep link works
for already-signed-in users but bounces unauthenticated visitors;
matches the OpenBB upstream onboarding text now.
Widened (descriptions undersold what the provider actually returns):
- intrinio: "Options snapshots, equity data." was misleading — Intrinio
fetchers cover equity quotes/historical/info, financials (income/balance/
cash flow), ratios, calendars, news, ETFs. Now: "Equities, ETFs,
fundamentals, news, options snapshots."
- fmp: added "commodity" — FMP's CommoditySpotPrice path is wired and
tested (fmp.bbProvider.spec.ts has gold/silver/brent), it was just
missing from the user-visible blurb.
No behavior change for working providers; users with stale nasdaq /
tradingeconomics keys in their config still keep them on disk (the
schema entries in core/config.ts and the credential-map tables are
untouched), they just disappear from the form. Re-introduce when the
underlying SDK providers actually exist.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
EIA went silently broken in three independent places. Surfaced when a new EIA API key got plugged into the same /test-provider button that caught the FRED bug, except this provider's failure mode was layered. 1. providers/eia/index.ts declared `credentials: ['eia_api_key']` — the already-prefixed form. The Provider constructor at abstract/provider.ts:54-59 unconditionally prepends the provider name to declared credentials, so the runtime credential field became `eia_eia_api_key` (double prefix). Symmetric to the FRED bug from a86612c, just from over-correcting instead of from a provider-name mismatch. Fix: declare `['api_key']` like every other provider in the package. 2. Both EIA fetcher models built the `sort` query parameter as JSON.stringify([{column, direction}]). EIA Open Data API v2 takes PHP-style bracket syntax (`sort[0][column]=period&sort[0][direction]=desc`) — the JSON form is silently rejected with 403 on every endpoint that uses sort. Verified live: same key + URL, swap the param shape, request goes from 403 to 200. 3. EIA serialises numeric observations as strings ("75.10") with null only for missing periods. Both fetchers passed `obs.value` straight through to the Zod schema, which expected number — so even after fixes 1+2 every row failed validation with "Expected number, received string". Fix: typeof check + parseFloat, NaN-skip the malformed ones. Updated the local response interface to admit string|number|null instead of lying about it being number. Tests: - src/domain/market-data/__test__/e2e/market-data.e2e.spec.ts: added an EIA describe block mirroring the FRED structure. Three live cases (test-provider button + STEO + petroleum status), each pinned to catch the specific bug it was originally hidden behind. Note: EIA endpoints sit under /commodity/* (OpenBB upstream classification), not /economy/*, and need explicit ?provider=eia because the commodity asset class default is yfinance. Verified live with the user's key: - STEO crude oil price: 120 monthly rows, latest 2027-12 = $68 (forecast=true flag working — observed rows go through 2026-04ish, forecast extends ~2 years out) - Crude oil stocks weekly: 260 rows, latest 2026-05-01 = 457,182 thousand barrels (~457M, sane US commercial inventory level) - Refinery utilization: 90.1%, typical seasonal high Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ools
Two new AI tools under the economy namespace, completing the FRED + EIA
macro surface for AI agents:
- economyEnergyOutlook — EIA STEO monthly observations + forecasts
(oil/gas price, production, consumption)
- economyPetroleumStatus — EIA Weekly Petroleum Status Report
(crude/gasoline/distillate stocks, refinery
utilisation, US crude production)
EIA endpoints route under /commodity/* on the wire (OpenBB upstream
classifies oil/gas as commodity data) but conceptually they're macro,
so they belong with FRED in the economy tool namespace from the AI's
perspective. createEconomyTools now takes both economyClient (FRED) and
commodityClient (EIA) — the asymmetry between AI namespace and HTTP
route is documented in the file header.
Wiring:
- CommodityClientLike: added getPetroleumStatus / getEnergyOutlook
declarations (both SDK + OpenBB-API client classes already implement
them, just weren't on the interface). OpenBBCommodityClient gets the
same `as unknown as CommodityClientLike` cast we use for the legacy
HTTP path elsewhere — its untyped Promise<Record<string, unknown>[]>
doesn't satisfy the typed contract, but the legacy path is being
retired anyway.
- main.ts: createEconomyTools(economyClient, commodityClient).
Tests:
- economy.spec.ts: split mocks into makeMockEconomyClient +
makeMockCommodityClient. Added 9 cases covering schema (zod enum
rejects unknown category, missing category), passthrough (provider
pinned to 'eia', date range forwarded), error propagation, and
client-isolation (EIA tools must not touch economyClient and vice
versa). The cross-client isolation tests exist because the file
composes two clients — easy to wire the wrong one when adding a tool.
Verified live with the user's EIA key:
- economyEnergyOutlook crude_oil_price: 120 monthly rows, last
observed 2026-05 = $112/Barrel, forecast tapers to $68 by 2027-12
- economyEnergyOutlook natural_gas_price: 120 rows, 2027-12 = $4.14/MMBtu
- economyPetroleumStatus crude_oil_stocks: 260 weekly rows, latest
2026-05-01 = 457M barrels (typical commercial inventory level)
- economyPetroleumStatus refinery_utilization: 90.1%, gasoline
stocks 219.8M barrels — both consistent with seasonal norms
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
7 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Two distinct streams landed on dev since the last cut to master:
economy*toolset. UI provider-key form audited and pruned of dead links / unimplemented providers.Per-session contributions
2026-05-08 — Macro data layer (FRED + EIA) end-to-end + UI cleanup
fix(market-data): repair FRED end-to-end + add market-data e2e harness(a86612c)fredprovider (registry hasfederal_reserve); credential-map sentfred_api_keyto both paths but in-process opentypebb auto-prefixes; etc. Adds a market-data e2e harness for future regressions.feat(tool): expose FRED via three economy tools (search/series/regional)(0c7865e)SDKEconomyClient(which had been exported but never instantiated by a factory). Three Vercel AI tools route throughEconomyClientLikewithprovider: 'federal_reserve'pinned at the tool boundary. Schema-validation tests (economy.spec.ts) catch zod / passthrough / error-propagation regressions in CI.fix(ui): MarketDataPage hints — drop unimplemented providers, refresh dead URLs(86233d6)nasdaq+tradingeconomics(no SDK provider registered — Test button always 500'd). Refreshedbls(registrationapps.bls.gov was retired; now points at data.bls.gov/registrationEngine/),econdb("Optional" → "Required (free signup)" — anonymous API now 401),eia(direct register.php link),fred(the /apikeys deep link bounces unauthenticated users; standard fred.stlouisfed.org → My Account flow now). Widenedintrinioandfmpdescriptions to match what the providers actually return.fix(eia): repair credential prefix + sort syntax + value coercion(1182dd9)credentials: ['eia_api_key']was already-prefixed and got double-prefixed by the Provider constructor toeia_eia_api_key, blocking credential propagation (mirror of the FRED bug, opposite direction); (2) both EIA fetchers built thesortquery parameter asJSON.stringify(...), but EIA Open Data API v2 takes PHP-style brackets and silently 403s on JSON; (3) EIA serialises numeric observations as strings and the fetchers passed the string straight through to the Zod schema. All three fixed.feat(tool): expose EIA energy outlook + petroleum status as economy tools(b50f054)economyEnergyOutlook(STEO, with observed + forecast rows) andeconomyPetroleumStatus(weekly Petroleum Status Report). EIA endpoints route under/commodity/*upstream (OpenBB classifies oil/gas as commodity data) but conceptually they're macro, so the tool layer surfaces them in theeconomynamespace alongside FRED.createEconomyToolsnow takes botheconomyClient(FRED) andcommodityClient(EIA); the AI / wire asymmetry is documented in the file header.2026-05-07 — Trading: ephemeral UTAs, three bug fixes, Phase 3 revert
fix(mock): allow externalTrade SELL to open short positions(0e09127)externalTrade(user-on-exchange semantics) now allows SELL-to-open;placeOrder/_applyFillretain the strict guard for Alice-initiated orders.feat(uta): ephemeral test UTA lifecycle (boot purge + manual destroy)(ea8e1d2)ephemeral?: booleanflag onutaConfigSchema, gated to themock-simulatorpreset. Boot path purges all ephemeral entries (config row +data/trading/<id>/) before UTAManager init.DELETE /api/trading/config/uta/:idalso wipes data dir for ephemeral UTAs. Closes the structural gap that let test residue pollute cost-basis across sessions.fix(uta): wallet bootstrap respects broker-reported avgCost(ab51371)_reconcileWalletPositionswas hardcoding the synthesizedreconcileBalanceop's price toposition.marketPrice, destroying broker-reported avgCost on first sight (Mock externalTrade, Alpaca, IBKR). Now prefersposition.avgCostwhen non-zero, falls back tomarketPriceonly when the broker has nothing — CCXT spot path unchanged (broker fakes avgCost = mark anyway), Mock/Alpaca/IBKR newly respected.revert(ccxt): drop canonical localSymbol; wire format is broker-native identity(afddd41)4bcf2f4) inserted a normalization step that rewrote CCXT's wire symbolBTC/USDT:USDT→BTC-PERPbefore storing it asContract.localSymbol. That collapsed the:settlesuffix and broke aliceId uniqueness within a single Bybit UTA (USDC-margined vs USDT-margined ETH perp became indistinguishable).Contracttype (secType union + per-secType fields) is canonical.Contract.localSymbolis broker-defined per broker — IBKR usesconId, CCXT uses unified wire symbol, Alpaca uses ticker. The system uniqueness primitive isaliceId = "{utaId}|{nativeKey}"wherenativeKeycomes from each broker'sgetNativeKey. We never normalize across brokers.buildContractvalidation funnel, strict SecType union) are taxonomy work — kept. Phase 3 was identity work — reverted.2026-05-07 — Workflow / CLAUDE.md
docs(claude): drop worktree convention; point parallel work at cloud(b6f7acc)pnpm install,data/copying, port juggling); cloud Claude sessions are the right concurrency primitive for projects with stateful working trees.TODO bookkeeping
docs(todo): cost-basis bootstrap overwrites broker avgCost(f294217)docs(todo): drop cost-basis bootstrap entry — fixed in ab51371(0239694)docs(todo): IBKR getNativeKey may collide on option-chain fanout(15fdba8)conIdis the actual primary key. Audit needed for IBKR + Alpaca + LeverUp + Bybit-direct getNativeKey implementations.Full commit log
Test plan
npx tsc --noEmitclean (root + ui workspaces)pnpm vitest run— 75 files / 1463 tests passing (was 1454; +9 EIA tool spec cases)pnpm vitest run --config vitest.e2e.config.ts src/domain/trading/__test__/e2e/— 12 files / 64 trading-e2e tests passing, 23 skipped (live-broker preconditions)pnpm vitest run --config vitest.e2e.config.ts src/domain/market-data/__test__/e2e/— FRED + EIA both green against live APIs (5 FRED + 3 EIA cases)getPositions. Today's window was exploratory testing, no real cost-basis history lost.economyFredSearch/economyFredSeries/economyFredRegional/economyEnergyOutlook/economyPetroleumStatusshould return real data without credential errors.🤖 Generated with Claude Code