Four conversations

DanYou've got six minutes and you're spending them on a data pitch?

MayaA small one. I want a data license from a shop called Scopes — they run an independent record of event-market prices against computed fair values. You've seen me use the free side; this is the paid history.

DanWe already pay CME for fed funds. We can see Kalshi's screen ourselves. What am I buying that I don't have?

MayaThe pairing, historically. Every fifteen minutes to six hours, they snapshot Kalshi and Polymarket next to an independent fair value — futures-implied for Fed meetings, nowcast-implied for CPI — plus the bid, ask, and depth at that moment. When the two disagree past a threshold, it's flagged and tracked to resolution. There are hundreds of resolved disagreements on the public scorecard already. I want to test whether those divergences predict anything.

DanPredict what, exactly?

MayaThree candidates. One: resolution — when event markets pull away from the futures curve before a Fed meeting, who ends up right? Their public record so far has the market side winning the majority of flagged disagreements — but that's pooled and sports-heavy; I want the rates-only cut, conditioned on gap size and book depth. Two: drift — does a divergence close in a tradeable way, and toward which side? Three — the one I actually care about: cross-asset lead-lag. If Kalshi's hike ladder moves ninety minutes before the futures curve does, that's an early-warning channel on positioning we already hold. I don't need to trade their markets to profit from watching them.

DanYou can't rebuild this from Kalshi's own API? They sell candles.

MayaCandles, single venue, no benchmark leg, no book states. The fair-value-as-computed-that-morning and the liquidity context were never written down anywhere else. It's not reconstructible at any budget — I checked. That's actually the whole reason it's interesting.

DanFine. Here's my real objection. Every backtest you bring me on vendor history has the same disease — the data's been cleaned after the fact and your model quietly knows the future. Why is this different?

MayaBecause it's the first vendor I've seen that's architecturally paranoid about that. The archive is bitemporal, append-only — corrections never overwrite; they stack on top with timestamps and reasons. There's a query mode — knowledge_as_of — that returns what the record said on March 3rd, not what it was later fixed to say. So the backtest can only see what was seeable. And if the signal works and we go to production, I can hand risk an audit trail proving no revision leaked backward. Try getting that letter out of a scraped dataset.

DanWho runs it? If it's two guys talking their book I don't want their "fair value" anywhere near ours.

MayaThat's the other reason I picked them. Published conflicts policy — the firm and its founder don't trade any market they cover. No picks, no newsletter-guru layer, and their own headline stat currently favors the markets over their own benchmark, which they print at the top of their homepage. People cooking numbers don't publish their losses at 95% significance. Methodology's versioned, rules are public, every figure has a permalink.

DanCost?

MayaTwo-forty-nine a month gets me the API for the research phase — internal-use terms, my card, cancel anytime. If the signal survives, the production license — files, rights, the audit mode — is a real conversation, mid five figures. But I'm asking for the two-forty-nine and three research days. Kill criteria: if the rates-only record has too few resolved flags for power, or the lead-lag Sharpe can't clear costs, I write it up dead and we're out a month of data spend.

DanAnd the downside if it's all noise?

MayaSeven hundred dollars and I've still got the calibration answer — whether these markets are efficient enough to ever bother with — which we currently assert by vibes. Cheapest piece of diligence on this asset class we'll ever buy.

Dan...Do the CPI print first. Then take your three days. If you come back wanting the five-figure license, bring the risk guy into the room.

MayaThat's the plan. The five-figure conversation is with their Reference product anyway — that's the one built for the risk guy.

TomYou want budget for another polling source? We have the wire, we have our own average, and I have forty-one days to the midterms.

PriyaNot a polling source — the disagreement layer. Scopes runs a composite of named public forecast models next to the prediction-market prices, flags the gaps, and keeps a resolution record. I want the institutional election package and the API for our live pages.

TomWe already write "markets say 54%." What's the upgrade?

PriyaRight now "markets say 54" is half a story with no counterweight. The full sentence is: markets say 54, the models say 61, and here's the historical record of who wins that argument. That third clause is what nobody else can give us — they've tracked every flagged disagreement to resolution, both sides published alike. Election night, when the markets swing wildly at 9 PM, I can put a ruler next to the swing instead of just narrating it.

TomSourcing. If I print "per Scopes" and someone asks who they are, what do I say?

PriyaIndependent measurement firm, Hartford. Published methodology, versioned; every number has a permalink and a timestamp; the composite is a median of named models — each constituent listed and linked, so we're never citing a black box. And their conflicts policy is one sentence: they don't trade anything they cover. Standards desk already looked; it's cleaner sourcing than half our polling cites.

TomWhy not build the comparison ourselves? We have the markets screen and we have model numbers.

PriyaTonight's comparison, sure. The history is the product — they've been snapshotting both sides at fixed intervals with the flags pre-registered before outcomes. I can't backfill that, and without the record, "who tends to be right" is a vibe. Also: their archive is append-only — if a number ever gets corrected, the original stays queryable. If we're challenged on a chart three days after the election, I can show exactly what the data said when we published. That's a corrections-desk insurance policy.

TomCost and speed.

PriyaThe package is hundreds, not thousands — cheaper than one freelance night. API's built for live pages; delayed tier's free, licensed tier is real-time. And one more thing: every outlet is going to write the "did the prediction markets beat the pollsters" piece on November 4th. Whoever licensed the scorekeeper writes it with receipts. I'd rather that be us.

TomGet me the standards memo by Friday. And Priya — if their record says the markets beat the models, we print that too.

PriyaThat's the whole point of them. They already do.

MarcusQ4 roadmap says "divergence column." Explain why we're licensing it instead of building it — we compare odds for a living.

LenaWe compare books to books. The feature users are asking for is prediction-markets versus the sharp line — Kalshi's football prices against the de-vigged reference — because half our users are now seeing event markets and asking "are these prices any good?" We could compute today's gap in a sprint. What we can't build is the credibility column: Scopes has flagged and resolved hundreds of these disagreements, publishes the running verdict, and the whole record is statistically significant. "This gap is 6 points, and here's how gaps like it have historically resolved" — that second clause is licensed or it doesn't exist.

MarcusKalshi has an API. Why not go direct?

LenaKalshi gives us Kalshi — one venue, no benchmark, and it's their own exam being self-graded. Scopes is cross-venue with an independent fair value, and the independence is the feature: our users trust us because we're not a book; a divergence column is only worth shipping if its reference layer has the same property. Also — bluntly — their headline record hasn't flattered their own benchmark, and they publish it anyway. That's the trust profile I want under our brand.

MarcusData rights. Last vendor integration nearly burned us.

LenaAlready scoped — and honestly it's why I shortlisted them. Their tiers are clean: API plans are internal-research only; embedding is explicitly a redistribution license — their Reference product — with the terms machine-readable in the API responses. No ambiguity to lawyer through; they built the wall before we hit it. We'd be a named redistribution licensee, attribution required, which suits us — the attribution is the credibility.

MarcusNumbers.

LenaLicense cost is real but fixed; engagement modeling says the column lifts session time on event-market pages — that's the fastest-growing segment we don't yet own a feature for. And there's a marketing beat: "powered by the independent scorekeeper" is a launch story. First-mover matters — whoever embeds the referee first gets to look like the platform that brought standards to the category.

MarcusTerm sheet by month-end. And ask them if the license covers the historical chart, not just the live number — the history's the part users will screenshot.

LenaThe history's the part they sell. That's rather the whole company.

DevBefore we roll the put ladder — I want to add a data feed to the program. Small money. Scopes.

RuthThe scorekeeper site? I've seen their Fed pages. We hedge a variable annuity book, Dev. What do event markets have to do with my guarantees?

DevThe guarantees, nothing. The price of protection, maybe a lot. Look at the quarter we're having: book's sliding toward the GMWB attach points, the committee wants the hedge ratio up, and every roll we do, vol's bid another point. My job this month is buying fear without overpaying for it. Scopes runs an S&P desk — Kalshi's index-threshold ladders against a fair value computed from the options market itself. When those two disagree, it's measuring something I can't see anywhere else: the gap between what the event-market crowd fears and what the options market is charging for it.

RuthTwo prices for the same fear. And when they disagree, which one's lying?

DevThat's exactly the research question — and they've got the only dataset that can answer it. Every divergence gets flagged, timestamped with the book depth, and tracked to resolution. What I want to test on the archive: when event markets price the downside thresholds rich to options-implied, does vol reprice toward the crowd in the following sessions, or does the crowd come back to the surface? If there's lead-lag either way, that's execution timing — which puts to roll this week versus next. On a book our size, shading one quarterly roll by half a vol point pays for their data for a decade.

RuthOr the gap is noise from retail money chasing headlines.

DevCould be — that's a kill criterion, not a reason not to look. Their resolution record is the prior: hundreds of flagged disagreements scored publicly, either side's answer shown alike. If the downside-threshold flags show no predictive structure, I write it up dead in three research days on a $249 key.

RuthHere's my scar tissue, so you understand the bar. October '08, this desk flew blind for a week — we knew what we'd pay for protection, we had no independent read on what fear was actually worth, and we overpaid catastrophically on some rolls and under-hedged on others. What I never had was a record — something outside our own screens saying "here is what was priced, when, and who turned out right." So the question I actually care about: when the actuaries and the auditors go through this program next quarter, what does this feed give them?

DevThat's the part I knew you'd like. The archive is bitemporal — append-only, corrections stack with timestamps, and there's a query mode that reproduces exactly what the record showed at any past moment, unretouched. If we ever fold this into execution policy, the hedge-effectiveness memo can cite an independent, point-in-time reference for why we timed entries — not our own trader's recollection. Insurers buy evaluated pricing for the same reason; this is that instinct, applied to probabilities. And their independence survives diligence: published conflicts policy, they trade nothing they cover, methodology versioned to a rules hash.

RuthCost path?

Dev$249 research key now, internal-use terms. If it enters the program, the production conversation is their Reference product — files, audit mode, a license the compliance office can hold in its hand. Mid five figures, against a hedge program that spends that on a slow Tuesday.

RuthRun it. And Dev — test the ugly weeks first. Any feed can look smart in a drift market. I want to know what their record showed on the days the world was ending.

DevThat's the one thing their archive can't fake. The bad days are in there at fifteen-minute intervals, book depth and all.

RuthThen it's already more honest than most of what we paid for in '08.