THE FUNNEL STUDY — how the corpus delivers, and what we should give away

Synthesis, 2026-10-26. Sources: the eight funnel profile files in synthesis/funnel/ (96 profiles covering 95 unique vendors in the teardown corpus; one vendor, Trade Ideas, was torn down twice and is counted once), the Morgue Engine design (synthesis/DESIGN.md), and the engine's build state (morgue/README.md: Slice 1 the Map, Slice 2 the Tape, Slice 3 the Board are built; Slice 4 the Ledger scorecards and Slice 5 the Studies are designed, with the append-only log itself already running under every slice).

The question: the vendors in the morgue each funnel, transmit, present and curate a slice of market data. The engine already computes most of those slices on owned data and publishes to a public web terminal in plain English, alongside a separate chart-review ranking board (the Lector ranker), with every figure written to an append-only record before it is shown. What free public service should that stack offer, and how does it avoid becoming the thing the teardowns dissected?

1. THE DELIVERY TAXONOMY

Eight archetypes cover the corpus. Each vendor is assigned to its primary archetype; hybrids are noted, not double-counted.

Plus one non-product: TradingView Desktop is a shell with no unit of its own (95 unique vendors accounted for).

Archetype 1 — The Live Terminal (20). Unusual Whales, FlowAlgo, Cheddar Flow, BlackBoxStocks, InsiderFinance, Tradytics, Bullflow, Quant Data, FlashAlpha, SpotGamma, MenthorQ, GEXRadar, Tapeboard, TradeEcho, Skylit, OptionScout, Profit Builders, Nansen, Arkham, VolSignals. The largest archetype and the most uniform: every unit is context — what traded, where dealer exposure sits — and in the entire group no unit contains an entry, stop and target together. The differentiation is never the math (much of it is the same published GEX family); it is metering, transport (API/MCP/WebSocket), and how much of the free surface is real. FlashAlpha and Bullflow sell the same terminal as data; SpotGamma and MenthorQ sell it as levels.

Archetype 2 — The Score Board (15). SqueezeMetrics, Danelfin, TipRanks, VectorVest, Kavout, Simply Wall St, AInvest, Chaikin, Zen Ratings, Prospero, StockRover, Intellectia, TickerNerd, Leadr, AltIndex. One number per name, refreshed daily, sold as a ranking. The unit is an opinion with the reasoning compressed into sub-scores. Exactly one vendor in this archetype (TickerNerd) publishes its full current dataset free and stamps every performance figure simulated vs live; exactly one (VectorVest) pairs the rating with a stop price, which is why it is the most actionable of the fifteen.

Archetype 3 — The Chart Overlay (11). The Saty stack, LuxAlgo, MarketCipher, ChartPrime, AlgoPoint, GainzAlgo, VIP Indicators, TJR, the ICT 2023 indicators, CorrectionScope, ApexQuant. Delivery is a TradingView invite; the vendor never hosts the moment of use, keeps no record of any reading, and cannot be audited after the fact — the chart is gone when the bar closes. The Saty stack is the honest pole (free, open source, no claims); ApexQuant is the other (a fabricated demo, a cosmetic paywall, a win-rate slider default presented as a calculator input).

Archetype 4 — The Signal Call (10). Zeiierman, CoinCodeCap, Options Cartel, Optionality, GammaEdge, EzAlgo, Stock Moe, TradeAlgo, Quantum Algo, Trade Ideas (Holly). The only archetype whose unit is routinely a complete trade. Quality splits entirely on the record: CoinCodeCap publishes a complete, recomputable public ledger; Optionality's tracker is engineered against flattering itself; Quantum Algo's public ledger rows carry null entry/stop/target, so nothing it claims can be re-scored.

Archetype 5 — The Automation Shell (16). Option Alpha, Capitalise, Composer, TrendSpider, WunderTrading, TradingKit, Quantix, AlgoBars, Alertsify, Tickeron, SummitIQ (automation in framing, advisory as shipped), AlgoAlpha (Echo copy), and the sold bots (M-Algo, Guardian FTMO, GridScalper EA, Zerith MACD Grid). The unit acts, so interpretation risk is replaced by construction risk: the best of these (TradingKit, AlgoBars, Option Alpha) ship their limitations in writing; the worst (Quantix) let a marketplace headline survive a census that refutes it.

Archetype 6 — The Reference Library (5). Edgeful, Market Chameleon, LevelFields, OptionStrat, Finviz. The customer arrives with a question; the unit is a base rate, a filtered list, or a fully priced structure. Edgeful is the cleanest claim surface in the whole corpus — frequencies printed with sample sizes, no platform win rate anywhere. OptionStrat's unit (exact legs, breakevens, max loss) is the most constructively actionable unit that is not a call.

Archetype 7 — The Research Letter (4). Citrini, Kobeissi, Schwar, and the 15.87 brief (a free method PDF funnelling to a paid daily desk). Human curation is the product; the portfolio is the unit that matters. Kobeissi is the archetype's proof that a self-published record can still be checkable: six annual ledgers whose totals recompute exactly under the stated methodology. Schwar shows the failure mode — a record split across two eras that cannot be chained, disclosed openly and still unusable as one number.

Archetype 8 — The Open Framework (13). ContestTrade, AutoHedge, Vibe-Trading, TradingAgents, OpenAlice, AI-Fund, AI-Trader, the Alpaca Trading Agent, Circuit Framework, Crypto Claude Desk, Helm Agents, Hypergrok Desk, Oft3r Desk. No funnel at all — distribution is GitHub, cost is the user's own API keys. Two sub-species matter: honesty machinery as the design centre (Vibe-Trading's grounding gate, AI-Trader's server re-pricing of agent-claimed fills, TradingKit-style frozen forward tests over in Archetype 5), and prose pretending to be machinery (AutoHedge ships a synthetic trade history at prices the asset never traded; Helm's run dated 2023 analyses with current data).

Cross-cutting findings.

• Nobody sells a decision except the callers and the bots. Of 95 vendors, the units that arrive trade-complete are the Signal Call archetype, the Automation Shell, and two outliers (Holly inside Trade Ideas, the 15.87 worksheet). Everything else — including every terminal and every score — leaves the last, hardest step to the customer while charging as if it didn't.
• Cadence is the most lied-about field. GEXRadar markets a 2-second refresh its shipped client never configures (fastest GEX poll: 15 s). Alertsify's alerts are sampled about every two minutes server-side. AInvest marketing claims real-time day-trading use its own knowledge base refutes. MenthorQ's two sources give two different end-of-day publication times for the same levels.
• The free surface predicts the trust posture. Vendors with a genuinely usable free surface (Bullflow, FlashAlpha, Tapeboard, OptionScout, TickerNerd, SqueezeMetrics, the 15.87 brief) are disproportionately the ones with clean or checkable records. Vendors whose free surface is a trial clock or a fabricated demo are disproportionately the ones with fabricated records.

2. HOW THE BEST MAKE IT ACTIONABLE

2.1 The unit grammar

Across the corpus, actionability is a property of the unit's fields, not of the vendor's confidence. A unit a stranger can act on — or deliberately reject — in under a minute carries:

1. A named event, not a naked number. Leadr's trigger taxonomy (prior-day high, opening-range break, resistance break, support reclaim) and TradeEcho's Discovery ledger (first-seen vs latest vs peak score) both refuse to show a score without the thing that happened. A score alone (Archetypes 2 and 3) is homework.
2. The price map in the same unit. Entry or trigger level, invalidation, target(s). CoinCodeCap's card is the corpus standard: pair, direction, leverage, an entry zone, two to three targets with R:R, a pre-set stop that trails after the first target, and a short rationale — formatted so a bot can parse it and a human can check it. Zeiierman's Discord unit (entry, TP1, TP2, SL) and Holly (entry, stop, target, hold discipline, a defined re-entry rule) are the same grammar. OptionScout's trade card reaches it from the other direction: a priced structure with breakevens, max profit/loss and probability of profit.
3. The base rate attached. Tapeboard prints its squeeze odds — including a negative median outcome — next to the score. Edgeful prints the sample size next to every frequency. A unit that says how often this has worked, on how many cases, converts marketing into arithmetic.
4. The record line. The strongest units carry their own history: what this exact signal type did over its last N logged instances. Almost nobody does this. It is the single largest open space in the corpus — and it is the Morgue Engine's Ledger by design.
5. Its biases on its label. SqueezeMetrics flags its outcome labels as outcome labels; Bullflow discloses the OI vintage in the field; FlowAlgo's buys-only intake is the counterexample — a structural bias disclosed only in teardown footnotes while every alert reads as a buy.

2.2 Push vs pull discipline

• Push is for events with a clock on them. A called trade, a Golden Sweep, a level break: value decays in minutes, so interruption is the service (Signal Call archetype; terminal alerts). Push without a clock is spam with a subscription fee.
• Pull is for state. Scores, maps, base rates and records are checked when the customer is deciding, not when the vendor is publishing. The Score Board vendors that push daily score changes are manufacturing urgency their unit doesn't contain.
• The honest hybrid is a digest with a fixed time. TradeEcho's Daily Echo Brief and OptionScout's 09:00 ET gamma briefing are point-in-time artefacts a non-member can read: one push a day, at a stated time, whose contents are frozen as published. That freeze is what makes them auditable later.
• Automation is push taken to its conclusion — and the corpus shows it only earns trust when the guardrails ship in the unit's terms (AlgoBars makes R-without-a-stop a compile error; the Alpaca Trading Agent's exits bypass its own exposure gates by design and says so).

2.3 Cadence honesty

The cadence printed on the unit must be the cadence the system runs: stated delay on delayed data (Bullflow's labelled 15-minute GEX pages), stated vintage on stale inputs (Prospero prints its data lags — dark-pool prints up to 24 hours late, 13F flow at least six weeks stale — on the signal pages themselves), stated sampling on alerts (Alertsify's two-minute check, disclosed on its own page). Every cadence claim the audits refuted was refuted by the vendor's own client, docs or knowledge base — the lie is never necessary, which is why it reads as a character tell.

2.4 The record of outcomes

The corpus contains exactly four record designs worth copying:

• Frozen daily records (Tapeboard): scores locked into dated archives, not re-scored when the formula changed from v5 to v6.
• Complete public ledgers (CoinCodeCap, Kobeissi): every call, losers printed, exclusions enumerated, totals that recompute. CoinCodeCap even published a dated correction replacing its own higher earlier figures with a full recount.
• Anti-flattery scoring (Optionality): unpriced closes excluded from every rate, a dead band around zero kept out of the win rate, peaks computed per cost-basis stretch so averaging down cannot inflate them, closed cycles that cannot be reopened.
• Frozen forward tests (TradingKit's Incubation): the backtest's source, settings and engine are frozen and paper-traded forward on closed candles, append-only — a backtest that has to survive the future instead of describing the past.

And one discipline that ties them together: settle at fixed horizons with a pre-declared rule (forward returns at 1h / 1d / 5d / 21d — the Ledger's design), never at the best moment inside the window.

2.5 Design prohibitions (the anti-patterns, named)

These are banned in anything we publish, by construction where possible:

• P-1 · No Peak Scoring. BlackBoxStocks scores options alerts by percent-to-post-alert-high; Trade Ideas' public records score "profit potential" to the session's subsequent high or low; Bullflow's marketing quotes winners at peakReturn. Maximum favourable excursion is not a result. Rule: outcomes are forward returns at fixed horizons from the logged timestamp.
• P-2 · No Buys-Only Completeness. FlowAlgo ingests only ask-side fills, so its tape cannot show a bearish print, and sells the result as "the flow." Rule: any intake bias is printed on the unit's label, and the unbiased variant is computed and published alongside (the Tape already does both).
• P-3 · No Frozen Counters. Quantum Algo's premium record is a manually frozen 88W/29L counter with an empty row set behind a headline blended from two incompatible counts; SummitIQ's hero win rate and P&L are hard-coded literals; GainzAlgo's free "reliability test" returns hard-coded 92% accuracy; ApexQuant's demo is a random walk with hard-coded signals. Rule: every published figure is computed from the append-only record at render time. A number that cannot be recomputed from the log does not ship.
• P-4 · No Backtest-as-Record. Quantix's backtest beat its live bots in 133 of 133 pairs while the marketing led with the top slice; TipRanks' own footer concedes its performance is retroactively applied and adjustable; Intellectia's flagship return was accumulated under a retired product design; Zen Ratings runs two irreconcilable vintages of the same headline on live pages simultaneously. Rule: backtests exist only as fixture checks that our math reproduces a published series, are labelled as such, and are never blended with, averaged into, or displayed beside the live record as if comparable.
• P-5 · No Vanishing Losers. Chaikin's model-portfolio page shows open positions only — closed trades drop off the dashboard, making the record unverifiable by design. AltIndex prints 61%, ~70% and 0% win rates on three of its own pages. Rule: the full published set is the denominator, forever; closed and losing units stay in the archive with the same prominence mechanics as winners.
• P-6 · No Score Without a Scorecard. InsiderFinance claims a "proven track record" and "unmatched accuracy" with no published record of any kind; Tradytics' Performance page loads every figure from a sealed API with no counting rules; Leadr's "84% Historical Win Rate" is footnoted to a predecessor product's alerts and is explicitly not the product's own performance. Rule: a component displays its own matured Ledger scorecard, or it displays an explicit "not yet graded — N signals logged, first scorecard on [date]" state. Borrowed records are prohibited.
• P-7 · No Silent Restatement. FlashAlpha's methodology states its history store is not versioned and re-pulls silently overwrite, while its historical-data page claims inputs are never restated; Danelfin's figures moved when its simulations were rerun; Arkham's labels are retroactively revisable with no point-in-time snapshot, so historical joins leak. Rule: restatement is a new row in the record, never an edit; the as-published value is the one that gets graded.
• P-8 · No Direction-Hit Win Rates. LevelFields' headline rate is a one-day direction hit rate over every event in a scenario — no fills, no costs, no exit rule; TradeAlgo's legacy funnel sells a 70% win rate that matches neither its live ledger (48%) nor its printed backtest (69%). Rule: nothing is called a win rate unless the underlying unit carried a complete trade grammar (entry, exit rule, horizon) before the fact.

3. FREE-TIER + FUNNEL PATTERNS

What the free surface does, by archetype.

• Live Terminal: the free surface is a taste of the same data, degraded in exactly one dimension — delay (Bullflow's 15-minute GEX pages with capture time shown; AInvest's delays run to 60 days, which crosses from taste into archaeology), freshness (FlashAlpha's free tier: 15-minute snapshots, 5 requests/day, no card), or breadth (Skylit's free plan restricts ticker coverage). The trial-clock variant (Cheddar 7 days, GEXRadar 24 hours with a card on file) converts worse and teaches the customer nothing about the vendor's honesty.
• Score Board: the free surface is metered depth — Danelfin's 10 reports a month, Simply Wall St's 5 reports and top-4 screener results, Zen Ratings' 10 free stock views. The exception is TickerNerd: the entire current rank dataset is a public file anyone can fetch, and the paywall sits on the portfolio and its ledger instead. Prospero is the limiting case — signals free "always," monetisation deferred to future interpretation features — and its trust record is correspondingly unreconciled marketing.
• Chart Overlay: almost no free surface at all (invite-only is the funnel), except the Saty stack, which is entirely free and open and funnels only to reputation. GainzAlgo's free calculators are the dishonest pole: a funnel surface that fabricates its own product's reliability score.
• Signal Call: the free surface is the room's antechamber — Optionality's free Discord server with the signal channels marked Pro, CoinCodeCap's free Telegram channel as the trial substitute, the 15.87 brief giving the entire method away free.
• Automation Shell: the free surface is the platform itself, metered on capacity — Option Alpha's $0 route via a funded partner-broker account (the broker pays, disclosed on the pricing page), AlgoBars free forever with AI actions metered at compute cost, Capitalise free to users because brokers white-label it, Composer's free tier capped at one automated strategy.
• Reference Library: OptionStrat's free builder with the probability figure rendered as an upsell blank; Finviz's full basic screener, ad-supported, with the wall placed at the exact moment of intent (clicking Custom filters opens the Elite dialog).
• Research Letter: free posts and previews; Kobeissi's Chart of the Week free with an account. The letter archetype's free surface is a sample of voice, not of data.
• Open Framework: everything is free; the funnel is adoption, and the "paywall" is the user's own LLM and data keys.

Where the walls sit. Five wall placements recur: request volume and history depth (Unusual Whales, FlashAlpha), feature families (Cheddar's Pro-only gamma/dark-pool/Power Alerts), asset class (BlackBoxStocks), capacity (Option Alpha's bot counts, Composer's nodes, TradingKit's credits), and recency (Quant Data's 365-day API history, OptionScout's flow history: today free, 7 days Trader, 30 days Pro). The wall that best matches value delivered is OptionScout's and FlashAlpha's: the present is free; depth, history and speed are paid — because history and speed are genuinely more expensive to serve and genuinely more useful to professionals.

Conversion mechanics — honest vs not.

Honest, and worth copying:

• No-card free tiers with a real product inside (Tapeboard $0 forever, OptionScout $0 with live data on every ticker, Bullflow's no-account MCP tools). The customer can verify the unit before any commitment.
• Method-as-lead-magnet (the 15.87 brief): give the formula, the worksheet and one full setup away; sell the daily labour of doing it, grading it and enforcing its clock. The wall sits on work, not on knowledge.
• Falsify-before-you-pay (Guardian FTMO's $1 backtester edition of the same engine): the free/cheap surface exists so the customer can try to break the claim before buying the unlocked version.
• Arithmetic that reconciles (Quant Data's annual savings, TradeEcho's lifetime-vs-monthly framing, CoinCodeCap's savings and monthly counts — all verified exact in teardown). Pricing honesty is a trust signal customers can check in thirty seconds.
• Refund terms that match the pricing page (TickerNerd's 30-day no-questions refund, stated identically in Terms and FAQ).
• Full-dataset-free, judgement-paid (TickerNerd): publish the data; charge for the curated portfolio, its ledger and the briefs.

Dishonest, and prohibited for us:

• Refund contradictions — Cheddar (FAQ says all sales final, Terms allow refunds), AltIndex (a 14-day guarantee on pricing, "all payments final without exception" in Terms), Zen Ratings (90-day vs 30-day guarantee on its own pages), TradeEcho (Terms vs standalone Refund Policy at the $3,499 tier).
• Stale or bait free surfaces — Bullflow's blog CTAs advertise a trial its FAQ says is not offered; GEXRadar's demo is a captured session (labelled, to its credit) under "real-time" marketing.
• Anchor theatre — TradeAlgo's $20,000 list price against a $4,995 checkout, Tickeron's 50–75% permanent "discounts."
• Gate-the-flagship trials — Cheddar's trial is Standard-only, so the features in the marketing cannot be tried; AlgoPoint's trial excludes its headline component.

4. THREE FREE PUBLIC SERVICE CONCEPTS

All three are built only from what exists: the Morgue Engine's Map (Slice 1, built), Tape (Slice 2, built) and Board (Slice 3, built), the Ledger (append-only log already running; Slice 4 scorecards complete it), the Studies (Slice 5, designed), and the Lector chart-review ranker already publishing alongside the terminal. All three obey the engine's existing laws: written to the record before shown, plain English on the public surface, declared biases on the label, no hindsight metrics — and the prohibitions P-1 to P-8 above.

Concept 1 — THE MORNING MAP

The pre-open positioning brief: where the walls are today, and whether yesterday's walls worked.

• Unit of output — one card per index (SPY, QQQ, IWM), carrying exactly: date and data vintage ("overnight OI, as of [time]"); spot at last close; net GEX under all three conventions (SpotGamma-static, FlashAlpha-variant, tape-signed) with an agreement flag ("3 of 3 agree: negative"); Call Wall strike and its GEX; Put Wall strike and its GEX; zero-gamma flip level and its distance from spot in %; DIX with its one-day-lag label; a one-sentence plain-English regime read ("Dealers are positioned so that moves toward [flip] tend to speed up, moves away tend to slow down"); chain coverage ("86% of OI-bearing contracts priced"); and the grading line: yesterday's card, graded — "Yesterday: price stayed above the flip all session; the Call Wall capped the high within 0.3%."
• Channels — the public web terminal (a dated page per day, permanent URL) plus a free email/RSS edition.
• Cadence — once per trading day, published pre-open (target 08:45 ET), never updated intraday; a correction is a new dated note appended to the same page, never an edit (P-7).
• Curation rule — none, deliberately: all three indices, every trading day, including dull days and days the conventions disagree — disagreement is printed, not resolved. Publication is earned by the data arriving and the card being written to the Ledger, not by the day looking interesting. (This is the anti-cherry-pick: the corpus's letter vendors publish when they have something to say; a map publishes because it is Tuesday.)
• The under-a-minute use — before the open, the reader knows the two strikes that bound the expected day, how far price sits from the flip, and whether this map has been right lately. They set their day's range expectations and know which level break would change the regime. That is the whole interaction.
• Trust mechanic — every card is Ledger-written before it renders, and every card is publicly graded the next morning by the wall-validation harness (Slice 5): did price respect the walls, did the flip behave as a flip. The running grade (e.g. "over the last 60 cards: flip respected 71%, Call Wall touched and held 44%") sits at the top of the page, computed from the record at render time (P-3), losers included (P-5).
• Cost to run — Slice 1 already runs in ~4 minutes for the three chains on data already paid for; FINRA's DIX input is free. The marginal cost is the grading harness (designed in Slice 5) and page/email rendering. No new subscriptions, no per-reader cost beyond static hosting.
• Honest paid tier (future) — intraday Map refreshes, per-name Map cards for the full universe, and an API with history depth — speed and breadth, the FlashAlpha/OptionScout wall. The daily card, its archive and its grades are never paywalled: the public record is the product's proof, and gating it would destroy the trust mechanic that makes the paid tier worth buying.

Concept 2 — THE QUALIFIED BOARD

Today's confluence candidates — only the names that pass the gate, each carrying its trigger, its price map, its base rate and its own track record.

• Unit of output — one row per qualified name, in the engine's existing Board-row grammar: ticker; confluence score with the count of confirming categories ("78 — 7 of 8"); the named trigger and its validation state ("resistance break — validated: close through on 1.8× volume, retest held 10:42"); the Saty ladder (support / trigger / target 1 / target 2 — the target levels are the row's targets, the support is its invalidation); the tape summary (Golden Sweep count, net call premium, mean Heat, each stamped with its source component and evidence grade); the Map context for the name's index; the study line ("this trigger, on this name class: gap filled 61% of 214 historical instances"); and the Ledger line ("this composite's last 90 published rows: 5-day median forward return +0.8%, n=90 — logged as-was"). Beside the Board, the Lector ranker's chart-review ranking for the same names is shown as a separate, clearly-labelled column — two independent judgements, never blended into one score.
• Channels — the public web terminal Board page (dated, permanent), plus a free daily email digest listing only the names that newly qualified that day.
• Cadence — computed on completed sessions and published pre-open for the day ahead; an intraday refresh, when shown, is labelled with its timestamp and does not replace the pre-open record (the pre-open rows are the graded ones).
• Curation rule — the Board's qualification gate, verbatim and unadjusted: Qualify flag and ≥5 of 8 categories confirming and score ≥65 and data-confidence ≥70, over the 60% data-coverage floor — names below the floor publish as "insufficient data," never as a guessed score and never silently dropped. The full ranked list, including every non-qualifier, is published on the same page, so the gate's rejects are as visible as its picks (P-5). On the first live run this gate passed 35 of 1,832 scored names and visibly refused the highest raw score in the universe for failing the gate — that refusal is the curation story, and it is printed.
• The under-a-minute use — read one row: what fired, where it triggers, where it's wrong (the ladder support), what it aims at, how often this setup has worked, and what this exact composite has done since it started publishing. Decide watch / plan / skip. No second page, no interpretation layer, no homework.
• Trust mechanic — every row is written to the Ledger before the page renders; the nightly scorer matures every published row at 1d / 5d / 21d against SIP bars, and the Board's own scorecard (hit rate, mean and median forward return, by-regime split) is the page's header — not a footnote, not a separate marketing page. Until the first scorecard matures, the header says so, with the count of rows logged and the date the first grade lands (P-6).
• Cost to run — Slice 3 already scores the 1,838-name universe on completed sessions; the incremental cost is completing Slice 4's nightly forward-return scorer (already designed) and the digest rendering. Data is already paid for.
• Honest paid tier (future) — intraday Board refresh, the full screener (user-set gates over the same published fields), watchlist alerts when a held or watched name qualifies, and API access. The daily published Board, its full ranked list and its scorecard stay free forever — the record is never the paywall, because the record is why anyone would pay for the speed.

Concept 3 — THE GOLDEN SWEEP WIRE

Every Golden Sweep on the watched universe, pushed as it happens — with the receipt attached, and the outcome appended when it matures.

• Unit of output — one alert card per qualifying print, carrying exactly: timestamp; underlying and contract (strike, expiry, call/put); premium; the Heat Score with its bucket contributions shown (not just the total); the tags earned (Golden Sweep / Top Position / Unusual, each with its rule stated in one line); the side classification and its convention label ("classified at the ask — quote-at-print join, our construction"); spot at print; size vs strike OI; the parallel all-prints context ("this print is also in the all-prints tape; the buys-only variant of this score would read X" — the P-2 label, on the unit); the alert-type record line ("Golden Sweeps on index ETFs, last 90 logged: median 1-day forward move of the underlying +0.4%, n=112"); and two outcome fields that start empty and are filled by the Ledger at +1h and +1d (forward return of the underlying from the print timestamp — fixed horizons, never the high: P-1).
• Channels — a public live feed page on the terminal, RSS, and a free push channel (webhook / chat channel) for subscribers; a once-daily digest edition for everyone else.
• Cadence — real-time intraday as prints qualify, with the feed's coverage stated on the page at all times ("watched: SPY, QQQ, IWM + today's Board names + held names — not the whole market"), and the stated latency measured, not marketed.
• Curation rule — the recovered rule, verbatim, no human in the loop: a print publishes if and only if it is a sweep and size ≥ strike open interest and premium ≥ $1M (Golden Sweep), or its Heat Score is ≥15. Every qualifying print publishes; there is no editorial selection, no "best of" — the rule is the curator, and the rule is printed on the feed.
• The under-a-minute use — an alert lands; the reader sees what traded, how big it is against the strike's entire open interest, how the score was earned, and what this alert type has done lately. They glance at the name's Board row / chart, or they dismiss it. The alert asks for sixty seconds of attention and comes with its own prior.
• Trust mechanic — the alert is Ledger-written before it is pushed (a push that isn't in the record didn't happen), and its outcome fields fill automatically at fixed horizons for every alert, winners and losers alike; the alert type's scorecard is public and is computed from the same rows the subscribers received. No alert can be deleted, and the feed page shows the full day's alerts, not a selection.
• Cost to run — Slice 2 already scores the live tape on the watched underlyings; the incremental cost is the push transport and the outcome-filling job (part of Slice 4). The watched- universe limit is a real cost control: full-OPRA coverage is explicitly out of scope and explicitly stated.
• Honest paid tier (future) — a wider watched universe, user-defined rules over the same published fields (own Heat thresholds, own underlyings), lower-latency transport, and an API. The public wire's archive and scorecards are never gated — a subscriber is paying for breadth and speed, never for the right to check our record.

5. RECOMMENDATION

Launch the staged pair: THE QUALIFIED BOARD first, with THE MORNING MAP as its daily front page. Hold the Golden Sweep Wire until the Board's first scorecard has matured.

Why this pair, decisively:

• The Board is the only concept whose unit is a complete decision — trigger, invalidation, targets, base rate and its own record in one row. The corpus proves that is the scarce thing: 95 vendors, and the trade-complete units fit in one archetype plus two outliers. Publishing measurements (Map alone) or prints (Wire alone) would put us in the two most crowded archetypes, competing on data we already know is a commodity.
• The Map is the context every Board row already cites, it is the cheapest possible daily habit (one card, one minute, pre-open), and its next-morning grading is the fastest trust proof we can run in public — a vendor visibly grading yesterday's own map, every day, is something no vendor in the corpus does.
• The Wire is held back for one concrete reason: a push product multiplies whatever record it carries. Launched before the Ledger's scorecards exist, it would be a high-frequency claim with no matured evidence — exactly the InsiderFinance posture (P-6). Launched after, every alert arrives with a graded prior. Sequence is the strategy.
• Both halves of the pair are already built (Slices 1 and 3); the only missing machinery is the trust spine (Slice 4 scorecards), which the design already specifies. This is a publishing project, not a research project.

First three build steps, in order:

1. Complete the Ledger scorecards for Map and Board rows (Slice 4, scoped to these two layers): nightly forward-return scoring at 1h / 1d / 5d / 21d, the wall-validation grading for yesterday's Map card, and a public scorecard page whose figures are computed from the append-only record at render time. Acceptance: any number on the page traces to a log row written before the timestamp it describes; scorecard math fixture-checked; the "not yet graded" state (P-6) renders correctly for horizons that have not matured.
2. Publish the combined pre-open page on the public terminal: the Morning Map cards on top, the Qualified Board below, the full unqualified ranked list one click beneath it, every row and card Ledger-written before render, in plain English, with coverage and bias labels on the units themselves. Acceptance: the page for a given day is byte-stable once published (a correction appends, never edits), and the gate's refusals — insufficient-data names and high-raw-score non-qualifiers — are visible on the page, not just in the data.
3. Turn on the daily rhythm and measure it for 30 sessions: the free email edition (Map + newly qualified names + yesterday's grades), the scorecard header live on the Board page from the first matured horizon, and a simple public counter of rows published vs rows graded. At 30 sessions, with the first 5d and 21d scorecards public, decide the Wire's launch date from the Tape's measured coverage and latency — and only then design the paid tier's wall (speed and breadth), with the public record explicitly out of scope for paywalling, in writing, on the pricing page itself.