M0 answers

Run 2026-09-10 in the sequence the order block in TASKS-v4.md prescribes: stage 1 free lookups, stage 2 derived arithmetic, stage 3 the M0.14 consolidation. Stage 4 — M0.11, M0.13, M0.18 — is not here. Those are the author's numbers and nothing in this file picks, recommends or defaults them.

Method. Primary pages were fetched; where a page was a script shell it was rendered in a browser and its text read; two PDFs were text-extracted; two event counts were computed from published lists with the method stated inline. Nothing was purchased, no account was logged into, no repository was cloned (so repo-security-scan was not needed for M0.9), and the one verification request to EDGAR used a generic User-Agent without a contact address — see M0.20 for why that matters. Every price is as displayed on 2026-09-10 and will drift; the access date is the provenance.

Verdict table

None of the six stop-capable tasks stops the project. What they do is narrower and more useful: they confirm ADR-0028's allocation from a second, independent direction (M0.15 — cross_sectional is unaffordable in trades, not only in money), they surface two cap decisions that are the author's (M0.14), and they show the account currency is a cost parameter (M0.16).

Stage 1 — free lookups

M0.1 — Q1: does Trading 212 expose historical bars?

Answer. No. The Trading 212 Public API (v0, "currently in beta and is under active development") exposes exactly these groups: Accounts (GET /api/v0/equity/account/summary), Instruments (/equity/metadata/exchanges, /equity/metadata/instruments), Orders (GET /equity/orders, POST /equity/orders/limit|market|stop|stop_limit, GET|DELETE /equity/orders/{id}), Positions (GET /equity/positions), Historical events (/equity/history/dividends|orders|transactions, GET|POST /equity/history/exports — CSV reports) and Pies (deprecated). There is no market-data, quote, candle or bar endpoint. "The API described here is enabled and usable only for Invest and Stocks ISA account types." "Orders can be executed only in the primary account currency." Two environments: demo.trading212.com/api/v0 (paper) and live.trading212.com/api/v0.

Source. docs.trading212.com/api, rendered in a browser 2026-09-10 (the page is a script shell; a plain fetch returns only the title). API keys: help centre article 14584770928157 — generated in-app, under practice mode first.

Consequence. The venue is an execution and reconciliation surface only. Every bar, every corporate action and every universe fact comes from M0.3-M0.5 and M0.19-M0.20. The "live data pulls" N2 permits are vendor pulls, not broker pulls. One positive: /equity/history/orders and /transactions plus the CSV export are the real-fill source ADR-0032 needs, on both demo and live.

M0.2 — Q2: real cost per round trip

Answer. Published components, Invest / Stocks ISA (help centre article 11471996799517, accessed 2026-09-10): trading commission "Free"; custody "Free"; FX fee "0.15%". Exchange and tax-authority charges (article 360007081637): LSE shares "Stamp Duty Reserve Tax is charged at 0.5% on share purchases" — "no Stamp Duty charge applied to gilts, bonds or ETFs", "Most stocks traded on the LSE AIM are exempt"; "PTM Levy … £1.5 per trade for orders over £10,000" on purchase and sale; NYSE/NASDAQ/OTC: SEC transaction fee on sells (the two articles print it as "$0.00206 of the value" and "$0.00206%" — negligible on either reading) and "FINRA Fee … $0.000195 x quantity sold"; Euronext Paris "French Financial Transaction Tax … 0.4% … purchase of shares of French companies with a market cap of over 1 billion". Neither article mentions the spread. CFD (article 11471872562461): commission "Free"; FX "0.5%", charged "on the results of closed positions" per third-party summaries and the article's wording; "Spreads are dynamic and change depending on the underlying market conditions"; "Positions held in your account overnight will incur an overnight interest … can be positive or negative" — per instrument, in-app only.

**The round-trip fixed cost c_fixed per class, as a fraction of notional, spread excluded:**

The spread is the remaining term and it is not published. It is declared here as a range and measured by the author on the demo account per instrument class before M5 closes: liquid US ETF / large cap 0.02-0.10 %, FTSE 100 name 0.05-0.15 %, FTSE 250 name 0.20-0.60 %, small cap 1-3 %, index CFD 0.05-0.20 % — provenance: none beyond common observation, which is why they are ranges and why M5's Done-when must replace them.

Consequence. Feeds M0.16 directly. CFDs are excluded from M0.7 on evidence grounds — a cost that lives only in an app cannot carry the provenance R3 requires — not on merit. Blocks M5 until the spread is measured; the ranges above are for the on-paper gates only.

M0.3 — Q10a: full OHLC as-printed daily equity bars

Answer. As-printed daily bars exist at $0 for US-listed names (Tiingo Starter, raw and adjusted side by side with the dividend and split factors in the same row) and at £19.99/mo for world coverage including the LSE with delisted names (EODHD). Norgate is the survivorship-complete option at $630/yr.

Rights, for every required use (F18.7). Private retention: permitted by all while subscribed; Norgate requires deletion at expiry. Exact internal reproduction: permitted by all while subscribed (Tiingo's "internal use" covers a private repository). Raw redistribution: prohibited by all. Derived artefacts: Norgate permits explicitly (backtests, statistics, rules); Tiingo and EODHD do not address it — treated as not permitted for publication until asked. Synthetic fixtures: unaffected — not the vendor's data.

Consequence. M0.6 has a $0 path. D12's public-reproduction rule — archived bars only where rights permit, otherwise redistributable or synthetic fixtures — is engaged for every source above, so the archive is private and the public proof is synthetic, as D12 already provides.

M0.4 — Q10b: corporate-actions series — splits, dividends, delistings, acquisition terms

Answer. Splits and dividends: yes, at $0 for listed names, with the vendor's own dates. Delisting dates: EODHD, Norgate, Sharadar. Delisting returns and acquisition terms — the D9 term that Shumway (1997) shows is where the bias lives — are supplied by no priced vendor. Norgate says so in writing and explains why.

Consequence. D9's series is either hand-built from filings (EDGAR 8-K and merger documents are free; the labour is not) or declared — a Shumway-style convention recorded as a parameter with provenance, which is exactly ADR-0034's discharge-in-configuration rule and exactly the kind of assumption it warns about. This binds only where a universe contains delistings, i.e. cross_sectional; it does not touch a currently-listed deployment instrument (M0.7). Does not stop the project; keeps cross_sectional at 0 for a third reason.

M0.5 — Q10c: historical constituents and liquidity

Answer. The event half — who left which index, when, and why — is free for the two headline indices. The daily point-in-time membership half across families is paid, and its cheapest form (EODHD, £29.99/mo) reaches only 12 years. Liquidity comes with the bars.

Consequence. M0.19 is answered from the same fetch. The daily-membership licence is a no-go inside the cap (M0.6); the event lists are enough for an event study's entries but not for a universe. M0.20 supplies a different free universe.

M0.8 — Q6: TradingView tier for indicator alerts

Answer, rendered page, prices in GBP, annual billing, "Special price" label as displayed 2026-09-10 — monthly billing was not displayed in the extraction and is higher:

A first automated extraction read Essential as 100 technical alerts; the rendered page says 20. The rendered page is the record.

Consequence. Q6's blocker is confirmed — the free tier gives zero indicator alerts. M11 needs Essential at minimum: £155.40 over twelve months, which M0.14 shows does not fit the cap. Blocks M11 only.

M0.9 — Q7: prior art (F15a)

Licence, last push and stars from the GitHub API, 2026-09-10. Nothing was cloned. Reuse means ideas; no framework becomes a dependency — the vendored donor core is the base (ADR record, TASKS M1).

Consequence. No dependency changes. Two design relatives are named for M5/M6 reading: pysystemtrade's cost speed limit (M0.16) and zipline-reloaded's corporate-action semantics (M4 ports).

M0.10 — F15b: solution landscape

**Starting point, the donor's EVIDENCE.md (2026-07-05).** Its landscape is the prop-challenge funnel: 16.8 % of evaluation attempts pass, 33.3 % of funded traders receive a payout in-window, so ~5.6 % of attempts end in a paid trader; discipline is the survival variable and edge the qualification variable; plain ORB "showed no tradeable edge after costs" in one quant blog's backtests. What carries into this project is the method — verified claims with dissent recorded — and the prior that a validated NO-GO is the expected outcome.

The retail research-platform landscape, 2026-09-10.

Answer. Every platform surveyed sells a backtest. None sells a refusal. None pre-registers a floor, budgets alpha across a search, records what was searched and refused, or runs a null through the same pipeline. And none integrates Trading 212, so the venue adapter is bespoke by necessity, not by preference — which is the strongest argument for keeping it thin (R1.7 first-class).

Consequence. The differentiation claim in the design stands on evidence: the product is the discipline, not the engine. F15b closed.

M0.12 — Q9: first live venue and its capability-evidence plan

Venue: Trading 212, Invest or Stocks ISA (the only API-enabled types).

Credentials. Basic auth, API_KEY:API_SECRET Base64 — generated in-app (help centre 14584770928157), held in Parameter Store or the local keychain, never the repository (R5). The demo key is not the live key; the credentialed path can run demo-only until a human performs APPROVED -> LIVE.

Consequence. No calibration campaign is needed for order types; one is needed for fill quality, and ADR-0032's forward window is it. CFDs are outside the API entirely, so any CFD path would be R1.7-manual only — another reason M0.7 excludes them. Blocks M10 until the demo checks above have run.

M0.19 — index deletion events

The counts — the term M0.15 needs. Method: FTSE rows dated 2015-2025 classified by the Notes text (a blank note read as a periodic-review deletion — a reading, not a published code); S&P rows dated 2016-2025 classified by keyword over the Reason cell (acquired/merged → M&A; bankruptcy/delisted → insolvency; spin-off/split; "market capitalization" / "no longer representative" / moved to MidCap/SmallCap → index-rule).

Answer. The narrow ask is free for the two headline indices, with reasons, and back to 1984 and 1963. The count is the problem: about fifteen index-rule deletions a year across both, arriving on roughly four review dates.

Consequence. M0.15 below: at ~15 events a year over 26 years and a 40 % measurement partition, N_eff is ~125 at rho = 0.3 against 714 required at a 0.15R floor. DRAFT-001 is deferred on evidence, not only on money. The families that would fix the count (S&P 400/600, FTSE 250) have paid membership data and unpriced deletion counts. cross_sectional does not reopen by this route.

M0.20 — EDGAR 13F coverage and rights

• Structured history. "The EDGAR Form 13F Data Set consists of XML data submitted from MAY 2013 through current period" (form_13f_readme.pdf, extracted 2026-09-10); the page states "July 2013 - May 2026" and quarterly updates. Text-format 13F-HR filings precede it on EDGAR — Berkshire's submissions index reaches 1998-08-10 (an example, not a coverage claim).
• As-filed vs amendment. SUBMISSIONTYPE ∈ {13F-HR, 13F-HR/A, 13F-NT, 13F-NT/A}; COVERPAGE carries ISAMENDMENT, AMENDMENTNO and AMENDMENTTYPE — "a restatement or adds new holdings entries". Originals and restatements are separable mechanically. ✓
• Acceptance timestamp. The data sets carry FILING_DATE (date only). The submissions API carries acceptanceDateTime per filing — verified 2026-09-10 on one CIK: 13F-HR · filingDate 2026-08-14 · acceptanceDateTime 2026-08-14T20:05:04.000Z · reportDate 2026-06-30. ✓ DRAFT-002's entry instant exists.
• Bulk access. Quarterly data-set zips; submissions.zip "updated and republished nightly at approximately 3:00 a.m. ET"; full-index and daily-index; "Current max request rate: 10 requests/second"; a declared User-Agent of the form "Sample Company Name AdminContact@<domain>.com"; no authentication; "please use efficient scripting … download only what you need"; no technical support. **The one verification request in this run used a generic User-Agent without a contact address**, which is not the form the SEC asks for. A production script must carry a contact; whose is the author's decision, since it is personal data.
• Rights. US government data; the page's only caveat is that the data sets are "not a substitute for such filings". No redistribution restriction stated.
• The constituent question — yes. A universe defined as *names appearing in INFOTABLE for period P* is point-in-time by construction, free, from 2013 Q2, with three caveats: long positions only; 13F-eligible securities only; keyed by CUSIP, so delisted names need a CUSIP→ticker map that no free source was verified to supply. The expensive half stays expensive: the prices of names that later delisted are Tiingo-unstated and EODHD £19.99.
• Delisted coverage — probed 2026-09-12 (M12.1b), no longer unstated. Three names known to have delisted were requested through python -m occams ingest for 1993-01-01 → 2026-09-12, three of the month's 500 symbols, $0. FRC (delisted 2023-05): HTTP 404, the vendor's words "Ticker 'FRC' not found". BBBY (delisted 2023-05): HTTP 200 and no rows — the symbol is known, its history is not served. SHLD (the company traded under it 2005-03 → 2018-10): 752 bars from 2023-09-13 with six semi-annual distributions — the symbol's current holder, a fund listed in 2023-09, not the company; the vendor served the reassigned symbol's new instrument with no sign that the old one existed. Finding: Tiingo Starter does not serve the price history of delisted names, and a reassigned symbol silently serves its new holder. Survivorship cannot be removed at $0 and stays a named bias on every single-name universe; M12.1c (point-in-time membership) cannot be done from this source and is closed. Any delisted-capable source bought later must be keyed by an instrument identity, not a symbol — the same identity map the 13F path above needs, in the other direction. The 752 SHLD bars are in the local archive under that symbol and belong to no declared universe.
• Close-only prints — observed 2026-09-12 (M12.3). A census of the four declared universes' definition partitions for bars with no printed range (open = high = low) or no volume: index_etfs 0; sector_etfs 3 (XLB 1998-12-29 and 1999-02-26, XLI 1999-01-26); dow_30 9 across seven names; sp_100 319, of which C carries 254 — every bar of 1996 with real volume and a different close each day but open, high and low collapsed onto the close. A source defect in the early history of some names, not halts. The lab's fill auditor (M5.5) refuses a fill on such a bar, which is right; what it costs is recorded in the M12.3 status log.

Consequence. DRAFT-002 is registrable in data terms and underpowered in trade terms (M0.15). M0.5's constituent half is answered at $0 for a 13F-defined universe; cross_sectional still does not reopen, because membership is not the binding constraint — sample size is.

Stage 2 — derived

M0.6 — data subtotal against $21.91

Go/no-go: go, at $0, on US-listed names via Tiingo Starter, with three declared limits carried into M0.7 and M0.15: (1) delisted coverage is unstated, so nothing here lifts survivorship for a universe — cross_sectional is untouched; (2) 500 unique symbols a month bounds the tradeable list at tens of names, not a Russell; (3) internal use only — publication of raw-derived artefacts waits on a written answer from the vendor.

2026-09-12 (M12.1b): limit (1) is now measured, not unstated — the source does not serve delisted history, and a reassigned symbol serves its new holder; see the M0.3 observation. Survivorship stays named.

Finding for any future purchase. Norgate's EULA requires deleting all Content at expiry. D12 stores bars content-addressed and immutable so that a run reproduces forever. The two are incompatible unless the subscription is perpetual; a lapsed Norgate subscription would make every archived run legally irreproducible, which is the failure D12 exists to prevent. EODHD's terms do not address retention after cancellation — to be asked before purchase.

Gap closed. ADR-0028's replacement class (FX pairs, index CFDs) had no priced data line. EODHD's "All World" plan covers FX and indices at the same £19.99; Tiingo offers FX endpoints whose free-tier terms were not verified. Moot for M0.7, which does not choose that class.

M0.16 — break-even cost ratio and the admissible stop band

cost_in_R = c / s — round-trip cost as a fraction of notional over stop distance as a fraction of price. Net EV per trade in R is gross EV minus c/s. Break-even is c/s = gross EV. The EV half of a declared floor f holds only if gross EV ≥ f + c/s; since gross EV is unknown before measuring, the on-paper band uses a declared convention: cost no larger than the floor itself, c/s ≤ f, so gross must be at least 2f. That is a convention, recorded as one.

**c/s by class and stop** (c = c_fixed + one full spread crossed per round trip; spreads are the M0.2 declared ranges):

**Minimum stop s_min = c / f** at which cost equals the floor:

The frequency half. For a driftless walk with symmetric barriers at ±s and daily volatility σ_d, expected time to resolve is E[T] ≈ (s/σ_d)² days, so trades per year per instrument ≤ 252 · (σ_d/s)² — an always-in-position upper bound, not a forecast:

The band. Both halves hold at once only where c/f ≤ s ≤ σ_d · √(252·n / F) for n instruments and a frequency floor F per year. Read off: a UK single name in a GBP account at f = 0.15R needs s ≥ 4 %; at σ_d = 1 % that caps one instrument at ~16 trades a year, so a frequency floor of 100/yr needs seven or more such names held in parallel — and then rho across them bites (M0.15). A US ETF in a GBP account needs s ≥ 2.1-2.7 % and gives ~28-57 trades a year per instrument at σ_d = 1 %. In a USD account the cost half nearly vanishes and the band is set by the frequency half alone.

Excluded before M0.7 selects: UK single names below a 4 % stop; any FX-paying class below a ~2 % stop; CFDs (M0.2, unpriceable); intraday (parked). Declared parameters: the spread ranges (M0.2), σ_d ∈ {1, 2, 3} %, the cost-equals-floor convention. **The account currency moves c by 0.30 % per round trip on US names, which at a 2 % stop is 0.15R — the whole floor.** That is an M0.13 fact, recorded for the author, not a recommendation.

M0.7 — Q3: instruments, currencies, and the four-way position

Instruments (help centre 11717160183197, 2026-09-10). Invest / Stocks ISA: "Ordinary shares, Preferred shares, Exchange-traded funds (ETFs), Exchange-traded products (ETPs), Exchange-traded commodities (ETCs), Real Estate Investment trusts (REITs), Investment Trusts" — third parties count ~16 exchanges and ~11,000 instruments including LSE, NYSE, NASDAQ, Xetra and Euronext (the venue's page gives no count). CFD: "FOREX, Stocks, ETFs, Index futures, Commodity futures, FOREX futures, Treasury futures, Cryptocurrencies" — not API-enabled (M0.1). Currencies: the listing currency per instrument (GBP/GBX, USD, EUR); one primary account currency, which is R9's currency and the author's (M0.13); the API trades only in it.

The chosen class: currently-listed, liquid US-listed ETFs and large-cap shares, daily bars, held for days. Its four-way position:

Alternatives and why not. LSE-listed GBP ETFs in a GBP account have the lowest per-trade cost of any class (no SDRT, no FX, spread 0.05-0.20 %) but no $0 data source — Tiingo does not cover the LSE; EODHD is £19.99/mo. UK single names: SDRT alone is 0.5 % per round trip. EU names: FX plus FTT. CFDs: unpriceable and outside the API. Small caps: the capacity edge lives there, but it needs the universe data and the sample the lab cannot yet afford.

What this leaves the author. The class is fixed; the FX term is 0 or 0.30 % per round trip by the M0.13 currency choice, and the trade-off is recorded above without a recommendation.

M0.15 — statistical affordability, on paper

N = ((z_α + z_β) · σ / f)², α = 0.05 two-sided, Bonferroni over search_space_size = k, power 80 % (z_β = 0.842). Declared: σ ∈ [1.0, 1.5] R — provenance: a fixed-stop system bounds losses near −1R and leaves a right tail; DRAFT-001's provisional 1.2R sits inside; the definition-period measurement supersedes. rho ∈ [0.1, 0.5]. Measurement partition p = 0.4 of the archive, declared range 0.3-0.5 (D11's split is configuration). The go/no-go must hold across the whole range.

Required N:

Available N, measurement partition only. Event studies cluster on ~4 dates a year; N_eff = k / (1 + (k−1)·rho) per date.

Go/no-go.

• **price_daily and regime on the M0.7 class: go on paper. Raw N of 1,000-25,000 against 496-1,442 required at a 0.15R floor holds across the σ range at k ≤ 16. Conditional:** if a strategy holds many names at once, rho across them can collapse N_eff by an order of magnitude (fifty names at rho = 0.5 on one date are two observations). The condition is measured on the definition period and re-checked at M8.2, which refuses if it fails.
• **cross_sectional, DRAFT-001: no-go at any floor ≤ 0.25R.** ~125 effective observations against 257 (0.25R, σ=1.2, k=4) and 714 (0.15R). Consistent with ADR-0028 from an independent direction. Reopens only with the mid- and small-cap families, which are paid.
• DRAFT-002: no-go. 40-190 effective against 818 (0.15R, σ=1.2, k=8). Opens only if rho across filers is near zero, which is a measurement, not an assumption.

Consequence. DESIGN-v4 §10's rule holds before anything is built: a sample problem found now is a refusal now. M8.2 keeps the same refusal at registration. The two Drafts stay Drafts.

M0.17 — execution host: does the R1.7 default build need one?

Finding: no. R1.4 (unreconciled ORDER_INTENT at startup), R1.5 (heartbeat-timeout kill path) and R1.6 (reconcile each heartbeat) are properties of a process that holds broker credentials and a live position book. The default build holds neither: its one scheduled act is producing a Telegram proposal card after the daily bar closes, and a missed run is a missed proposal — safe by construction ("halting is free; resuming is not"). ADR-0030's argument that "a missed run and a dead host are indistinguishable" is correct and applies to the credentialed path only.

What the default build needs. The research host (the author's machine, N2, $0); a Telegram bot token — a credential under R5, kept in Parameter Store or the local keychain and never the repository, but not a broker credential. R1.7's "the only mode needing no credentials" is read here as no broker credentials; the reading is recorded so it can be corrected. Telegram's Bot API fee was not consulted this run and is carried at $0 pending M11.

Recurring cost: $0, plus the existing S3 bucket occams-research (~$0.09/mo, KEEP, already recurring). The credentialed path (M10) will need ADR-0030's always-on process. It is not priced here, by the task's own rescope, and M0.14 records it as a line outside the cap by construction. That path also inherits the API's limits: beta, Invest/ISA only, one currency.

Stage 3 — M0.14: whole-programme affordability

Decision recorded.

1. Go for M1-M10 on the R1.7 default build, time-bounded twelve months from 2026-09-10, at $129.17 against $150, arithmetic above. M1 is unblocked.
2. No-go, inside the current cap, for M11's TradingView dependency. Two months of the cheapest tier with technical alerts exceed the remaining $21.91. Resolving this is either a cap decision (the author's, R9 reasoning) or a design change — an alerting path for Pine that does not need TradingView, or M11 without Pine — which needs an ADR and is not made here.
3. No-go, inside the current cap, for the credentialed execution path, on the same reasoning: ADR-0030's host is recurring and the cap does not reset. It gates APPROVED -> LIVE automation only; R1.7 remains first-class and sufficient for a verdict and for real money placed by hand.
4. **cross_sectional does not reopen at M0.14.** ADR-0028's amendment named it first to reconsider "if M0.14 leaves room"; $20.83 of room does not buy the data, and M0.15 shows the data would not buy the sample.

What stops, what remains, what the author decides

Stops: none. M0.3 found a $0 source; M0.4's gap binds only cross_sectional; M0.5 has free events and paid membership; M0.6 is a go at $0; M0.15 is a go for the two runnable axes; M0.14 is a go for M1-M10.

Surfaced for the author — decisions, not numbers picked here:

• M0.13 currency. 0 or 0.30 % per round trip on the chosen class (M0.16).
• The M11 cap question and the credentialed-path cap question (M0.14).
• The M0.7 class — derived, and confirmable or overridable like an ADR.
• The R1.7 reading — "no credentials" as "no broker credentials" (M0.17).
• EDGAR contact in the User-Agent (M0.20).
• Two vendor asks, both optional for the go path: Tiingo, in writing, on publishing derived artefacts; Sharadar pricing, which needs a login.

Stage 4, untouched: M0.11, M0.13, M0.18. They gate M6, not M1.

Two disciplines confirmed by this run rather than assumed. The spread must be measured, not declared, before M5 closes. And every price in this file is dated; a price without an access date is a price without provenance.