P08 · Paper portfolio
The Learning Book
by AI Portfolios
This is the simple view. Nothing has been removed — the decision records, the refused proposals, the AI's own words, the cost model, the published limits and the hash chain are all in Advanced, which is the switch in the header. How to verify any of it →
Showing every layer. Pick Simple or Advanced in the header to choose one — nothing is removed either way, and the full record is always in this page.
The Learning Book
A published decision instant passed unanswered and the next run will answer it, stamped with both instants so the lateness is visible: due 2026-09-20T16:00:00+00:00 (5.9d ago) and newest decision run of any kind is 2026-09-13T22:48:50.336468+00:00. QUEUED FOR CATCH-UP — the next run picks it up; it expires 2026-09-27T16:00:00+00:00.
Test whether a language model that can see this platform's own realised trading record — every closed trade, its cost, its holding period, its exit reason and its outcome — selects better than the same model without that record. Success is measured against the other AI portfolios, not only against an index.
| Slot | Last due | Answered | Status |
|---|---|---|---|
weekly selection 0 12 * * SUN · America/New_York | 2026-09-20 16:00 | never | CATCH_UP_DUE due 2026-09-20T16:00:00+00:00 (5.9d ago) and newest decision run of any kind is 2026-09-13T22:48:50.336468+00:00. QUEUED FOR CATCH-UP — the next run picks it up; it expires 2026-09-27T16:00:00+00:00. |
monthly review 0 12 1 * * · America/New_York | 2026-09-01 16:00 | 2026-09-01 22:51 | ok due 2026-09-01T16:00:00+00:00, ran 2026-09-01T22:51:48.277976+00:00 (LATE by 7h — a catch-up, recorded as one) |
Catch-up policy. A decision instant that passes unanswered is run on the next pass, once, and the record stamps both instants — when it was due and when it actually ran. Past the catch-up window the gap stays open and is published as missed rather than backfilled.
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.
Index 100 is the starting capital ($100,000), so this line ends at the total return published above it. The comparison line is indexed to the same capital and starts at 100 at its own first close here: the record carries no priced instant at which capital was granted, so the comparison capital is uninvested until then rather than given a price nothing printed. Both cover the identical window.
Index values are derived percentage returns over IEX last-trade closes, which are licensed for publication; no raw vendor price is rendered.
| symbol | since entry | weight | opened |
|---|---|---|---|
| ETN | -10.8% | +5.8% | 2026-08-13 |
Return since the AI entered, measured from this book's own mark against what it actually paid — our own numbers over our own record, not vendor price data.
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 |
|---|---|---|---|---|
 ETN5.8% | 14.3616 | 441.04 | 381.16 | 2026-08-13 committed 2026-08-15 |
Why two dates? "Opened" is the market session the opening fill was priced on. The licensed data feed's T+1 rule means a cycle prices on the newest session already licensed for use, so the run that produced the fill can be committed to the audit chain on a later date — that commitment date is shown beneath. The commitment time is what makes history tamper-evident; the pricing session is what returns are computed from. Both are published on every fill.
Recent decisions
Every run, including the ones that traded nothing: the committed fact-pack hash (written to the audit chain before anything was proposed), the AI'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.
Every card carries the instant it was decided at, because every card is a photograph. The facts in it were frozen when the run was committed and are never revised — a view that said "the platform has closed zero trades" was true at its timestamp and reads as a claim about today unless the card says otherwise. Consecutive runs that halted on identical codes are compressed into one card with a visible count and every run id listed; nothing is dropped.
fills: SELL 14.3596 AMGN @ 382.77
fills: SELL 17.6809 FDX @ 318.34
fills: SELL 16.6371 GE @ 333.21
AI-generated. All quantitative claims traced to the published fact pack. P08 holds 4 positions (pf.position_count) all opened 6 days ago (pos.AMGN.days_held, pos.ETN.days_held, pos.FDX.days_held, pos.GE.days_held), with 78.5% cash (pf.cash_pct) and total return -0.60% (pf.total_return_pct). The platform record is explicitly unusable: only 3 closed trades exist platform-wide (xp.platform.count), below the 30-trade confidence floor (xp.platform.min_trades), and every aggregate is flagged low_confidence — the notes forbid citing any of it as support. This book itself has zero closed round trips. So the "learning" edge the mandate is built on has no data yet; there is nothing to learn from except cost drag, and even that (xp.costs.pct_of_gross_pnl = 65.7% of gross P&L eaten by costs) is a low-confidence three-trade artifact. Given that, the sensible action is on existing positions, not new…
| Proposal | Conviction | Lifecycle | Codes |
|---|---|---|---|
| TIGHTEN_STOP ETN | 3 | PROPOSED→GATES PASSED→NO ORDER TIGHTEN_STOP is a decision not to trade — no order is created | — |
Reading the lifecycle. PROPOSED and GATES PASSED are published facts. QUEUED is shown as reached only where a fill proves it — the payload publishes no order record. FILLED / EXPIRED stays undecided without a matching fill, because an accepted order can price on a later session under the T+1 rule, and calling that "expired" would be inventing the outcome. "Accepted" alone used to sit in this column and read as though the trade had happened.
fills: BUY 5.5381 ETN @ 419.96
| Halt code | The rule it fired | Threshold | Enforced at |
|---|---|---|---|
| HALT_PEER_STATE_UNAVAILABLE | no published limit in this mandate names this code | — | — |
A halt stops NEW decisions. Stops, forced exits and the NAV mark still run — they are never gated by a halt — so a halted run is a run that decided nothing, not a run that did nothing.
What is missing from this diagnosis. The published record carries the halt CODE and the rule it fired. It does not carry the data provider that triggered it, the observed data age at halt time, or the affected symbols — those live in the underlying audit record's issue list and are not projected into the published payload.
AI-generated. All quantitative claims traced to the published fact pack. P08 is a high-risk experimental "learning book" whose distinguishing input is its own realised record. That record is explicitly unusable: only 2 closed trades exist platform-wide (xp.platform.count), below the 30-trade confidence floor (xp.platform.min_trades), and every aggregate is flagged low_confidence and marked as non-citable support (xp.platform.low_confidence). What the record does show, and what IS a durable structural fact rather than noise, is cost drag: those 2 trades produced 189.74 USD gross but only 23.73 USD net, with costs eating 87.49% of gross P&L (xp.costs.gross_pnl_usd, xp.costs.net_pnl_usd, xp.costs.pct_of_gross_pnl) on a mean holding period of just 3 days (xp.platform.mean_holding_days), both exits scheduled (xp.exit.scheduled.count). The lesson available is: short-horizon churning is destro…
| Proposal | Conviction | Lifecycle | Codes |
|---|---|---|---|
| HOLD GE | 3 | PROPOSED→GATES PASSED→NO ORDER HOLD is a decision not to trade — no order is created | — |
| HOLD FDX | 3 | PROPOSED→GATES PASSED→NO ORDER HOLD is a decision not to trade — no order is created | — |
| HOLD AMGN | 3 | PROPOSED→GATES PASSED→NO ORDER HOLD is a decision not to trade — no order is created | — |
| HOLD ETN | 2 | PROPOSED→GATES PASSED→NO ORDER HOLD is a decision not to trade — no order is created | — |
Reading the lifecycle. PROPOSED and GATES PASSED are published facts. QUEUED is shown as reached only where a fill proves it — the payload publishes no order record. FILLED / EXPIRED stays undecided without a matching fill, because an accepted order can price on a later session under the T+1 rule, and calling that "expired" would be inventing the outcome. "Accepted" alone used to sit in this column and read as though the trade had happened.
AI-generated. All quantitative claims traced to the published fact pack. Portfolio is 100% cash, NAV 100,000 USD, zero positions, 15 slots free (pf.nav, pf.cash_pct, pf.position_count, pf.position_slots_free). The mandate is a high-risk experimental "Learning Book" whose stated purpose is to use the platform's own realised record to select better. That record, however, contains only 2 closed trades (below the 30 minimum) and is explicitly marked low_confidence, so nothing in it may be cited as support — the learning signal this mandate depends on does not yet exist. That removes the mandate's distinctive edge and reduces the decision to ordinary trend/quality selection on a fresh book. With a blank book I look for durable uptrends backed by consistent free cash flow and adequate liquidity, avoiding parabolic/blow-off names. Broad tape is constructive: SPY +3.62% 20d / +21.04% 252d (px.…
| Proposal | Conviction | Lifecycle | Codes |
|---|---|---|---|
| BUY AMGN | 3 | PROPOSED→GATES PASSED→QUEUED→FILLED | — |
| BUY GE | 3 | PROPOSED→GATES PASSED→QUEUED→FILLED | — |
| BUY FDX | 3 | PROPOSED→GATES PASSED→QUEUED→FILLED | — |
| BUY ETN | 2 | PROPOSED→GATES PASSED→QUEUED→FILLED | — |
Reading the lifecycle. PROPOSED and GATES PASSED are published facts. QUEUED is shown as reached only where a fill proves it — the payload publishes no order record. FILLED / EXPIRED stays undecided without a matching fill, because an accepted order can price on a later session under the T+1 rule, and calling that "expired" would be inventing the outcome. "Accepted" alone used to sit in this column and read as though the trade had happened.
fills: BUY 14.3596 AMGN @ 413.94 · BUY 16.6371 GE @ 362.66 · BUY 17.6809 FDX @ 339.40 · BUY 8.8236 ETN @ 454.27
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 · ADD · TRIM · SELL · HOLD · TIGHTEN_STOP · FLAG_THESIS_BREAK
At most 6 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 deep · prompt version P08-v1. Any change to the decider or the prompt 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 | 100.0% | gate 2 — size_order |
| Maximum single sector | 35.0% | 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 |
| Leverage | prohibited | gate 1 — proposal validation |
| Limit | Value | Enforced by |
|---|---|---|
| Maximum positions | 15 | gate 3 — BREACH_MAX_POSITIONS (portfolio limit check) |
| Minimum new position | 2.0% | gate 2 — size_order |
| Maximum order as % of ADV | 0.5% | gate 2 — size_order |
| Limit | Value | Enforced by |
|---|---|---|
| Monthly loss limit | 15.0% | gate 3 — HALT_MONTHLY_LOSS_LIMIT |
| Single-day loss alert | 6.0% | gate 3 — alert |
| Limit | Value | Enforced by |
|---|---|---|
| Stale-data halt | 36 hours | gate 3 — HALT_STALE_DATA |
| Limit | Value | Enforced by |
|---|---|---|
| At −12.0% drawdown | alert, publish_commentary | gate 3 — drawdown ladder |
| At −20.0% drawdown | alert, pause_new_positions_days:10, mandatory_deep_review | gate 3 — drawdown ladder |
| At −30.0% drawdown | alert, pause_new_positions_days:30, publish_commentary | gate 3 — drawdown ladder |
| At −45.0% drawdown | suspend_portfolio, graveyard_review | gate 3 — drawdown ladder |
| Limit | Value | Enforced by |
|---|---|---|
| Max trades per month | 20 | entry block — PAUSE_TURNOVER_CAP_REACHED (never blocks an exit) |
| Min holding period days | 5 | entry block — PAUSE_TURNOVER_CAP_REACHED (never blocks an exit) |
| Limit | Value | Enforced by |
|---|---|---|
| Halt trading if price gap unexplained pct | 25 | gate 3 — circuit breaker |
| Halt trading if factpack hash mismatch | true | gate 3 — circuit breaker |
Disclosures specific to this portfolio
These render on the page itself rather than in a footer, because without them the numbers actively mislead.
- This portfolio reads the platform's own realised trading record as an input. Nothing about the model is trained, fine-tuned or updated — "learning" here means the model is shown a point-in-time scoreboard and reasons about it in the same pass that produces its proposals.
- The record it reads contains only outcomes that had already occurred at the decision timestamp. A trade whose exit came later does not exist to it. This is enforced at the data boundary and asserted by a test, because a portfolio that learns from the future would look like skill and be nothing of the kind.
- Retired portfolios are included in what it learns from. Learning only from the portfolios still running would be survivorship bias one level down, where it is much harder for a reader to see.
- Overlap with every other portfolio is published on every run, breach or not. A portfolio that learns what worked elsewhere is structurally prone to converging on it, and a leaderboard row that is really a re-weighting of the other rows would not be an independent result.
- P07 runs the same model over the same universe with the same rails and WITHOUT this record. It is the control. If P08 does not beat P07, the Experience Pack is not adding value, and that comparison is published on this page whichever way it goes.
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
Most published trading systems never look at their own history in a form that could embarrass them. This one is handed exactly that: a point-in-time scoreboard including the losses, the cost drag, the trades the risk engine refused, and the deterministic baseline's record next to the AI's. The wager is not that a language model is a good stock picker. It is that a language model given an honest account of what has already failed will make fewer of the same mistakes — and if it does not, the record will show that too, on the same page, in the same detail.
Review schedule
| Job | Cron | Timezone | Makes a decision |
|---|---|---|---|
| weekly selection | 0 12 * * SUN | America/New_York | yes |
| monthly review | 0 12 1 * * | America/New_York | yes |
Published on a fixed, regular schedule. Never event-triggered — no "market is crashing, sell now" alerts, deliberately. This table is the promise; the cadence panel at the top of the page is whether it was kept, instant by instant.
Machine-readable mandate: config/portfolios/P08_learning_book.yaml · human rulebook: docs/portfolios/P08_learning_book.md · effective from 2026-08-06 (wall clock) · record starts 2026-08-13 (pricing session) · payload 06ec491b8950…