Intraday AI Lab (EXPERIMENTAL)
Test, in public and with realistic costs, whether an LLM given only delayed intraday data can produce positive expectancy over a session horizon, flat by the close every single day.
Track record
Fewer than 60 daily observations. Annualised return, volatility, Sharpe, Sortino and Calmar are withheld: annualising a short sample produces numbers that look precise and are not.
Our own marks from our own fills (nothing here is vendor price data). Deliberately monochrome: colour that codes "good" and "bad" is an editorial thumb on the scale.
Weights are our own work product, computed at the newest audit-chain NAV mark — never re-derived later, never estimated from cost basis. Logos identify the holdings and imply no endorsement (operator decision, session 11); sector tints are structure, not judgement.
| Symbol | Quantity | Avg cost | Stop | Opened |
|---|---|---|---|---|
 AMZN15.1% | 55.0944 | 272.99 | 266.81 | 2026-08-06 |
 MSFT15.0% | 30.0084 | 499.43 | 489.86 | 2026-08-06 |
Recent decisions
Every run, including the ones that traded nothing: the committed fact-pack hash (written to the audit chain before the model was called), the model's published view, each proposal with the rule engine's verdict, and the fills that resulted. Rejected proposals stay up with their reason codes — a reader who can only see the trades that happened cannot judge the discipline that refused the ones that didn't.
This is the portfolio's first decision, starting with 100% cash (pf.cash_pct). The mandate is experimental intraday, flat-by-close every day, operating on 15-minute delayed data with full spreads and fees. The fact pack shows end-of-day data for 2026-08-06, post-close. Recent 5-day momentum shows strong dispersion: AMZN +15.61% (px.AMZN.ret_5d), NVDA +12.28% (px.NVDA.ret_5d), MSFT +10.81% (px.MSFT.ret_5d), META +9.44% (px.META.ret_5d), AVGO +8.44% (px.AVGO.ret_5d), and GOOGL +7.22% (px.GOOGL.ret_5d) all show strong upward moves. The broad indices (SPY +3.62%, QQQ +4.55% per px.SPY.ret_5d, px.QQQ.ret_5d) also rose but less dramatically. Over 20 days the picture diverges: MSFT is exceptional at +30.05% (px.MSFT.ret_20d), AMZN +10.21% (px.AMZN.ret_20d), but many others are negative or flat. TSLA is notably weak at -21.40% over 20d (px.TSLA.ret_20d) and -28.20% over 60d (px.TSLA.ret_60d). …
| Proposal | Conviction | Outcome | Codes |
|---|---|---|---|
| BUY MSFT | 4 | accepted | — |
| BUY AMZN | 3 | accepted | — |
fills: BUY 30.0084 MSFT @ 499.43 · BUY 55.0944 AMZN @ 272.99
What the AI is allowed to do
This portfolio's AI ranks candidates and explains its reasoning. It proposes direction and conviction only. Every price, quantity, weight and risk limit is computed by deterministic code the model cannot see or influence.
BUY · SELL · HOLD · TIGHTEN_STOP · FLATTEN
At most 2 proposals per run, each carrying a direction, a conviction from 1 to 5, a rationale, and references to the facts it used.
- set a price
- set a quantity or share count
- set a weight or position size
- set or move a risk limit
- execute anything
Model tier fast · prompt version P06-v1. Any change to either is recorded as a MODEL_CHANGE event in the audit chain — without it the record would be a chimera of two systems presented as one.
Enforced limits
Every limit below names the code path that enforces it. This is not decoration: a limit that is published but never fires is worse than no limit, because a reader has no way to tell the difference. A test walks every mandate, drives the engine into the state each declared control claims to protect against, and fails the build if nothing stops it.
| Limit | Value | Enforced by |
|---|---|---|
| Maximum gross exposure | 45.0% | gate 2 — size_order |
| Maximum single sector | 100.0% | gate 2 — size_order |
| Limit | Value | Enforced by |
|---|---|---|
| Shorting | prohibited | gate 1 — proposal validation |
| Derivatives | prohibited | gate 1 — proposal validation |
| Margin | prohibited | gate 1 — proposal validation |
| Limit | Value | Enforced by |
|---|---|---|
| Minimum new position | 3.0% | gate 2 — size_order |
| Maximum order as % of ADV | 0.05% | gate 2 — size_order |
| Limit | Value | Enforced by |
|---|---|---|
| Daily loss limit | 1.5% | gate 3 — HALT_DAILY_LOSS_LIMIT (halts and flattens) |
| Weekly loss limit | 4.0% | gate 3 — HALT_WEEKLY_LOSS_LIMIT |
| Monthly loss limit | 8.0% | gate 3 — HALT_MONTHLY_LOSS_LIMIT |
| Limit | Value | Enforced by |
|---|---|---|
| Stale-data halt | 45 minutes | gate 3 — HALT_STALE_DATA |
| Limit | Value | Enforced by |
|---|---|---|
| At −5.0% drawdown | alert, scale_risk_budget:0.6 | gate 3 — drawdown ladder |
| At −10.0% drawdown | alert, scale_risk_budget:0.35, publish_commentary | gate 3 — drawdown ladder |
| At −20.0% drawdown | suspend_portfolio, publish_post_mortem | gate 3 — drawdown ladder |
| At −30.0% drawdown | graveyard | gate 3 — drawdown ladder |
| Limit | Value | Enforced by |
|---|---|---|
| Max round trips per day | 6 | entry block — PAUSE_TURNOVER_CAP_REACHED (never blocks an exit) |
| Max round trips per week | 20 | entry block — PAUSE_TURNOVER_CAP_REACHED (never blocks an exit) |
| Limit | Value | Enforced by |
|---|---|---|
| Halt trading if price gap unexplained pct | 10 | gate 3 — circuit breaker |
| Halt trading if vix above | 40 | gate 3 — circuit breaker |
| Halt trading if market wide circuit breaker | true | gate 3 — circuit breaker |
| Halt trading if factpack hash mismatch | true | gate 3 — circuit breaker |
| Halt trading first n minutes of session | 15 | gate 3 — circuit breaker |
Execution
Fills are struck on a VWAP window that begins strictly after the decision timestamp plus a latency delay. No fill can use information from the bar in which the decision was made, so look-ahead is arithmetically impossible rather than merely avoided.
Philosophy
Intraday trading is where costs dominate and where an information disadvantage is largest. The model here sees data that is at minimum 15 minutes stale, pays the full spread on both sides, and pays every regulatory fee. If it still makes money, that is interesting. If it does not, that is a useful public result and the portfolio stays visible forever.
Review schedule
| Job | Cron | Timezone | Makes a decision |
|---|---|---|---|
| d0945 | 45 9 * * MON-FRI | America/New_York | yes |
| d1100 | 0 11 * * MON-FRI | America/New_York | yes |
| d1300 | 0 13 * * MON-FRI | America/New_York | yes |
| d1430 | 30 14 * * MON-FRI | America/New_York | yes |
| d1545 | 45 15 * * MON-FRI | America/New_York | yes |
| eod reconcile | 30 16 * * MON-FRI | America/New_York | no |
Published on a fixed, regular schedule. Never event-triggered — no "market is crashing, sell now" alerts, deliberately.
Machine-readable mandate: config/portfolios/P06_intraday_ai_lab.yaml · human rulebook: docs/portfolios/P06_intraday_ai_lab.md · effective from 2026-08-06 · payload b618c290568f…