How Kresmion works

1 · Platform overview

Kresmion puts filings, macro, rates, futures, crypto, prediction markets and open-source intelligence on one screen, with consistent units, a source line and a freshness stamp on each number.

The signed-in navigation has eight top-level entries:

• Home: the returning-user dashboard.
• Markets: Equities (overview, signals, filings, markets, screener, supply chain), Macro (overview and daily brief, liquidity, calendar, central banks, money markets, surprise index, correlation, countries, trade), Crypto (overview, whales, exchanges, exchange flow, token due diligence, derivatives, ETF flows, news, DEX), Commodities, Forex and Futures (overview, seasonality, volatility, curve monitor).
• Intelligence: the live feed, signals and signal accuracy, OSINT and news, Congress, insider trades, activist investors, earnings, corporate signals (FDA, bonds, IPOs and offerings, jobs, contracts), market sentiment, prediction markets, narratives and physical intelligence.
• Workspace: Split View, the Terminal canvas, Docs and the Spreadsheet.
• Tools: trading utilities (charts, position calculator, screener, report generator), cross-asset and macro reads (market rotation, correlation regime, sovereign yields, macro event impact, relative strength, relative value, currency strength, yield curve, drawdown), positioning and flows (COT, greeks exposure, dealer positions, short volume, implied probability), insiders and Congress, and supply chains.
• Data: capital flows (COT, short interest, institutional 13F, Treasury TIC, BIS), risk models and data export.
• Community: forum, people, contacts, messages and research notes.
• Portfolio: overview, holdings, activity, analytics, watchlist, evidence boards, wallets, reports, alerts and settings.

Portfolio transactions are stored encrypted, and the portfolio analytics run in your browser: the server receives only the symbols it needs to price.

2 · Keyboard shortcuts

Press ? (Shift + ?) anywhere in the signed-in app to open the shortcut overlay.

Two-key sequences such as g t must be completed within 1.5 seconds. On the Terminal canvas, scrolling zooms toward the cursor, dragging empty canvas pans, dragging a panel's title bar moves it and dragging its corner resizes it.

3 · Update frequencies

Design cadence per surface. Exact intervals vary with load and upstream rate limits; data pages show the actual last-updated time, so a pipeline that falls behind is visible.

Macro desk

Regime, policy rates, surprises, event studies, curves, sovereign credit and capital flows.

4 · Macro regime engine

The regime engine produces one daily risk-on / risk-off score that other surfaces reference (the daily brief, the regime badge, the regime-shift signal type).

It combines fifteen cross-asset and macro inputs in four factors, with fixed weights:

Most inputs are scored on their change over the last five observations (five sessions for a daily series, five weeks for a weekly one), as a z-score against that input's own history of such changes: the latest 400 observations for series read from FRED (net Fed liquidity, rebuilt on weekly balance-sheet dates from daily reverse repo data, ends up with about a year and a half of points), about two years of daily closes for market prices. Three inputs are scored differently because they already measure change: the S&P 500 20-day return and the economic surprise index are z-scored directly (the surprise index against its last 365 days), and PMI on its month-over-month change against every stored monthly change. Every z-score is clipped at plus or minus 3. An inverted input enters with its sign flipped, so a rise counts against the score; the curve is not inverted, so a steepening adds to Growth and a flattening subtracts from it. An input with no reading, or whose latest observation is older than its cadence allows (7 days for a daily series, 21 for a weekly one, 75 for a monthly one), is left out of its factor rather than counted as zero.

Each factor is the average of its available inputs. The raw score is 0.20 × Growth + 0.40 × Liquidity + 0.30 × Risk appetite + 0.10 × Volatility, clipped at plus or minus 3. The published score is smoothed on every run: 70% of the new raw score plus 30% of the previous day's smoothed score. The label is a fixed banding of the smoothed score multiplied by 3: above +1.5 Strong Risk-On, +0.5 to +1.5 Risk-On, -0.5 to +0.5 Neutral, -1.5 to -0.5 Risk-Off, below -1.5 Strong Risk-Off (on the published score the edges sit at about ±0.17 and ±0.5). There is no hysteresis: the label is read from the smoothed score on every run, so a score close to an edge can change label from one day to the next. A regime change is announced on Kresmion's public channels only once the new label has held for five consecutive daily readings. Conviction is the share of the four factors that point the same way (HIGH at 75% or more, otherwise MEDIUM). The engine runs about every six hours and again after major US data releases; each run rewrites that day's row.

The score is computed deterministically. The short interpretation beside it is written by a language model over those numbers, under two rules: every direction word ("rising", "falling", "wider") is a fixed phrase taken from the sign of the input it describes, and a sentence that contradicts the numbers is rejected and replaced by a template that names the largest moves.

Note: the regime describes the backdrop its inputs show on the day. It is not a forecast and not advice.

5 · Central banks and policy rates

Policy rates cover 37 central banks: the 36 jurisdictions in the BIS central bank policy rate series, plus the United Arab Emirates, for which BIS publishes no policy rate. The UAE row is maintained by hand: it carries no source link, is labelled manually maintained, shows the date it was last checked and the bank's own last change, and is marked stale once the next Federal Reserve decision, which the Central Bank of the UAE follows, passes without a new check. Singapore is listed without a rate: its central bank targets the exchange rate and sets no policy interest rate, and SORA, a market rate, is not shown in its place. BIS publishes a decision days after it is taken, so for the major banks and for countries whose BIS series has stopped, Kresmion also reads the bank's own publication and uses the newer observation. A daily check at 22:10 UTC compares the stored rate with the bank's own figure.

The "last change" date is the day the new rate took effect, not the day Kresmion first saw it. A sourced row that has stopped updating is marked stalled rather than shown as current.

6 · Economic surprise index

A surprise is the actual print minus the consensus, in the indicator's own unit, counted only when both exist for the same release. Each surprise is standardised by that indicator's own trailing history of surprises (never by data from after the release), and an indicator with too short a history is left out rather than guessed. Indicators where a higher print means a weaker economy (unemployment rate, initial claims) are sign-flipped, and inflation is kept as its own sub-index rather than netted against growth.

Only the derived index and its components are published. Per-event calendar rows, scheduled times and consensus figures are not redistributed.

7 · Macro event impact

The macro event impact study measures how major instruments moved around scheduled macro events.

• FOMC decisions: every scheduled decision day since 2000 from the Federal Reserve's own calendar, plus unscheduled meetings, with the size of any change taken from the published target rate. Holds are events too. Each decision carries a surprise measure: the change in the 2-year Treasury yield from the session before the decision to the decision day, bucketed into dovish, neutral and hawkish at published thresholds.
• CPI and payrolls: release days from the official release calendars, event-day statistics only (no consensus is used).
• Instruments: SPY, QQQ, IWM, TLT, IEF, HYG, GLD, USO, the dollar index and VIX (in points), with ETF history extended to 2000, plus equity-index, Treasury, metals and energy futures from 2016.
• Horizons: 1, 5, 21 and 63 trading sessions, measured from the last session before the event.

For each event type, instrument and horizon the page shows the event median against the instrument's unconditional baseline over the same horizon, a bootstrap confidence interval on the difference, a binomial test of the hit rate, the median path from 5 sessions before to 30 after, and every individual instance.

Note: these are historical statistics, not forecasts. Each cycle has its own context, and several event types have small samples; n is shown on every figure.

8 · Yield curves

The public yield curve tool shows the current US Treasury curve, the 2-year / 10-year spread with an inversion flag, and the curve one year earlier.

The signed-in sovereign yield tracker follows the 2-year / 10-year spread for the United States (Treasury constant-maturity yields via FRED, history from 1976) and the euro area (ECB all-bonds yield curve, daily, history from 2004). It lists every historical inversion episode with its start, end, length and deepest point, and counts days since inversion and since un-inversion. Historically the re-steepening after an inversion, not the inversion itself, has come closer to US recessions.

9 · Sovereign credit

10-year spreads. For 40 countries Kresmion subtracts one published 10-year government bond yield from another, against the German Bund and against US Treasuries, for the same month. EU countries use the ECB long-term interest rate for convergence purposes; other OECD countries use OECD long-term rates via FRED; the US leg is the Federal Reserve H.15 10-year yield. This is a Kresmion calculation, not a credit rating: no rating agency is involved. Outside the euro area the bonds are in local currency, so a spread also carries currency and inflation expectations.

Ratings. Signed-in country pages show sovereign ratings read from the rated government's own ratings page (debt office, finance ministry or central bank), only where that page's terms permit reuse. A rating action reported in the news is shown only after it has been checked against an official source, and the row names both the report and the source it was confirmed against. Every row names its agency; there is no consensus or average rating.

10 · Cross-asset correlation regime

The correlation regime tracker flags pairs whose short-term correlation has moved away from its 12-month norm.

• Universe: 45 instruments across equities (broad and sector ETFs, indices), bonds, commodities, the dollar, volatility and major crypto assets, giving 990 possible pairs. A pair without enough shared history is skipped, so a snapshot holds slightly fewer.
• Math: the current value is a 30-day rolling Pearson correlation of daily returns; the baseline is the trailing 252-day mean of that series. Crypto and equity calendars differ, so returns are computed on each symbol's own dates and joined per pair.
• Break: a pair enters a break when the gap to baseline exceeds 0.40 and the relationship flips sign, doubles or halves; it leaves the break once the gap falls below 0.30.

Each pair's detail shows the 252-day history and up to three earlier breaks of the same pair in the same direction and of similar size.

11 · FRED series and units

FRED series are shown with the unit FRED publishes for them (percent, index level, billions of dollars and so on), held in one table in code and applied everywhere a series is displayed or served. A series' observation date is shown beside its value, because daily and monthly series are often read side by side.

12 · Treasury TIC foreign holders

Source. The US Treasury "Major Foreign Holders of Treasury Securities" monthly report, published about six weeks after the month it describes.

Classification. Each country's month-over-month change is graded on both size and percentage:

• Major accumulation / major selloff: at least $10B and at least 3%, both required.
• Accumulation / selloff: at least $5B or at least 3%, either one.
• No flag: both below the standard thresholds; the row is still shown, muted.

The page lists Treasury's top 20 holders first and the other reporting countries below, on the same thresholds.

Note: single-month moves are noisy. Read two to three months together with the country's currency, policy stance and trade balance.

Equities desk

Company fundamentals, institutional, insider, congressional and activist filings, debt issuance and options-implied distributions.

13 · Company fundamentals and relative value

Fundamentals come from the XBRL data in SEC filings where the filer tags the line. Where it does not, a fallback parser reads the number from the filing text, and that number is served too, for US domestic filers only (never for a foreign private issuer). A parsed number whose scale (thousands or millions) cannot be established is never served, and total debt is shown only when it is tagged in XBRL: when a filer's debt cannot be recognised, the chart is left empty.

Each series is keyed by the period the number describes, not by the filing date. Revenue is shown per quarter: a 10-Q's year-to-date figure is not plotted as a quarter, and the fourth quarter is derived as the full fiscal year minus the same year's nine-month figure, flagged as derived. Balances (debt, inventory, receivables) are shown at the period-end date. A filer without quarterly reports gets fiscal-year points only; a series never mixes the two.

Relative value. A company is compared with peers in the same industry group (the GICS sub-industry where the company is GICS-classified, otherwise the data vendor's industry label; the two label sets are not merged), chosen by nearest market capitalisation on a log scale. Every ratio's numerator and denominator come from the same annual filing, trailing-twelve-month figures are not computed, and a peer that files no annual report with the SEC is listed as dropped with the reason instead of being silently omitted.

14 · 13F institutional tracking

Each tracked filer's reported position is compared with the same filer's prior-quarter position on the same CIK (tickers change with corporate actions; CIKs do not). Changes are classified as new, increased, reduced, exited, institutional surge or institutional exit. When a filer reports the same holding through several subsidiaries, the lines are aggregated per CUSIP first, so the total position is shown. Values are in whole US dollars under the current 13F-HR XML schema. Reported option positions are kept separate from share holdings.

Percentage display. Some funds hold small placeholder positions and later size up, which makes raw percentage changes meaningless.

• Prior below 100 shares: shown as RESUMED instead of a percentage.
• Prior 100 to 1,000 shares: changes above 1,000% show as ">1000%" with a major increase or decrease label.
• Prior 1,000 shares or more: the full percentage is shown, capped beyond ±5,000%.

NEW_POSITION means no prior-quarter line for that CUSIP at all; RESUMED means a placeholder prior below 100 shares.

15 · Insider Form 4 clusters

A cluster is at least 3 distinct insiders trading the same ticker in the same direction within a rolling 30-day window, detected daily at 02:30 UTC from SEC Form 4 filings.

Filters. Open-market purchases and sales only (Form 4 codes P and S). Option exercises, grants, gifts and tax withholding carry no direction and are not counted. Rule 10b5-1 plan trades are flagged where the filer affirmed one, which Form 4 has recorded since the April 2023 amendments; filings without that element carry no flag.

Strength score. Buys and sells are scored separately. Both start from the number of insiders, the dollar value on a log scale, a bonus when a CEO, CFO, President or Chairman takes part, the share of each insider's holdings traded, the value as a share of market capitalisation, and a decay for days since the last trade. Sells are marked down for 10% owners and when every seller sold less than a tenth of their stake. A cluster seen for the first time on a ticker is labelled as such but carries no score weight, because the Form 4 history held is too short to tell a first cluster from an unobserved one.

Each cluster's detail lists up to three earlier clusters of similar strength for the same ticker and direction.

Note: sell clusters include tax sales, scheduled diversification and plan trades. Read the per-insider role and plan flag before drawing conclusions.

16 · Congressional trading

Sources. House members' periodic transaction reports are read from the Clerk of the House, and each trade links to its filing there. Senate filings are not ingested: the congressional data covers the House only.

Amounts. STOCK Act disclosures give brackets (for example "$1,001 - $15,000"), not exact amounts. Kresmion shows the bracket and uses its midpoint only to compare sizes; the open-ended top bracket ("$5,000,001 +") uses its floor, which understates the largest trades.

Anomalies. An outlier is a trade at least 3 times the member's median trade size, for members with at least 12 disclosed trades, within the last 60 days. A cluster is at least 3 members trading the same ticker in the same direction within 30 days; because members have up to 45 days to file, a cluster can become visible weeks after the trades. There is no party segmentation in detection or display.

Note: a disclosure does not imply impropriety. Many accounts are spouse-managed, delegated or in blind trusts. Read the filing before drawing conclusions about a member.

17 · Activist investors

The activist tracker shows SEC Schedule 13D and 13D/A filings, filed when a holder crosses 5% with intent to influence. Schedule 13G (passive) filings are kept for other surfaces and excluded here.

Classification. Each filing gets two checks. First, the "Item 4, Purpose of Transaction" section is extracted with strict boundaries and scanned for activist-intent language (board representation, strategic alternatives, sale of the company, proxy contest and similar); when Item 4 cannot be extracted, the result is "unclear" rather than a guess. Second, an individual filer with director or officer markers is classified as an insider, not an activist.

Confidence. High means at least two distinct activist phrases and no passive-investment language; medium is one phrase; low is mixed language. Only high-confidence filings produce an activist-stake signal.

Filer tiers. A short curated list of known activist funds, keyed on EDGAR CIK:

• Tier 1: Elliott, Icahn, Pershing Square, Trian, Starboard, Third Point
• Tier 2: Saba, Engaged Capital, Cevian, ValueAct, JANA
• Tier 3: Politan, Ancora, and institutional 13D filers such as GAMCO, OrbiMed, General Atlantic and Bulldog Investors

Funds are added only after their CIK is verified. About a fifth of 13D/A amendments reference the prior filing without restating Item 4, and some omit the ownership table; the page shows coverage so readers can see what was observed.

18 · Bond issuance

Bond filings come from SEC EDGAR full-text search on the 424B prospectus family (424B2/3/4/5), parsed in two passes. The first pass filters the search snippet to reject equity and fund takedowns that mention "aggregate principal amount" in boilerplate. For filings that survive without structured terms, the second pass reads the start of the prospectus document and re-runs the extractors; a filing whose body is mostly equity language is tagged "not bond" and hidden from the page while kept for audit.

Principal, coupon and maturity are independent fields: a row may show an amount without a rate. The default view shows rows where all three are disclosed. Tickers are taken from the EDGAR display name where present, otherwise matched by normalised company name.

Three rule-based signals fire: large debt issuance (principal of $1B or more), acquisition financing (use of proceeds parses as an acquisition) and high-rate issuance (coupon at least 200 bp over the current US high-yield option-adjusted spread).

19 · Risk-neutral implied probability

The implied probability tool turns each listed options chain into the probability distribution the market prices for the underlying at each expiry, using the Breeden-Litzenberger identity: the risk-neutral density at strike K is e^(rT) times the second derivative of the call price with respect to K.

• Universe: 113 US underlyings chosen for options liquidity (index, sector and macro ETFs and liquid single names); the tool's ticker selector is the exact list. One snapshot per day at 21:00 UTC, all listed expiries. Expiries under 7 days are not fitted.
• Quotes: out-of-the-money options only (puts below the forward, calls above), with a bid and an ask and a spread under half the mid.
• Forward: taken from put-call parity on the strikes nearest spot, so dividends and borrow costs are priced in.
• Fit: each mid is converted to a Black-76 implied volatility, and a smoothing spline is fitted to implied volatility against log-moneyness, weighted by each quote's bid-ask spread so the curve stays within about one spread of every quote. Call prices are rebuilt from the smoothed curve and differentiated. The rate r is the 3-month Treasury bill yield.
• Checks: if negative density (an arbitrage in the fit) exceeds 3% of the total, or the total probability falls outside 1 ± 0.10, the expiry is shown as uncomputable with the reason. Nothing is rescaled to force a fit.

Note: risk-neutral probabilities include the premium investors pay for protection and tend to overstate tail risk relative to realised frequencies. They are prices, not forecasts. Thin chains are often uncomputable by design.

Futures and commodities desk

Settlement history, seasonality, volatility, positioning, inventories and the forward curve.

20 · Futures and commodities analytics

Futures analytics are computed from stored daily settlement history (front-month continuous contracts and individual delivery months). Every figure describes what prices and positioning have done.

• Seasonality: month-end to month-end returns, averaged per calendar month across the available years, with the share of positive months. A cell needs at least three years of data.
• Realised volatility: RV20 and RV60 are the standard deviation of daily log returns over 20 and 60 sessions, annualised by the square root of 252. The RV20 percentile ranks today within the trailing year; at or below the 10th percentile the market is labelled compressed. ATR14 is shown as a share of the last close.
• Speculative positioning: weekly CFTC Commitments of Traders (legacy, combined). The z-score compares the speculators' net share of open interest with its trailing three-year mean; the COT Index places the net position on a 0 to 100 scale over the same three years.
• Inventories: weekly EIA crude (excluding the SPR), gasoline, distillate and lower-48 natural gas storage, each against the average of the same week over the prior five years.
• Forward curve: a daily snapshot at 22:30 UTC records settlements for 16 roots: WTI and Brent crude, natural gas, heating oil and gasoline; gold, silver, copper and platinum; corn, soybeans and wheat; coffee, sugar, cocoa and cotton. Roots listed every month are walked 12 contracts out, the others 8. The public curve tool shows WTI, natural gas, gold and corn; signed-in users see all 16 in the curve monitor, and each delivery month has its own contract page. History for the twelve roots added most recently starts in late September 2026.

A curve is in contango when the second month settles more than 0.25% above the front month, in backwardation when more than 0.25% below, and flat in between. The annualised roll yield is (front / next) raised to (12 / months between them), minus one.

Three deterministic signal types use this data: a positioning flip (net non-commercial position crossing zero), a volatility compression flag (RV20 at or below the 5th percentile of its trailing year) and a curve flip (the contango / flat / backwardation label changing between snapshots, on WTI, natural gas, gold and corn).

Crypto desk

Spot bitcoin and Ethereum ETF flows and the fund registry.

21 · Crypto ETF flows

The fund list is a dated registry: each fund is identified by its asset and ticker together (some trusts share a ticker with the coin they hold), and a fund is added or retired with an effective date, never deleted.

A day's aggregate is called a total only when every fund active in the registry that day has a known flow. Otherwise it is a reported subtotal and says how many of the active funds it covers, their share of assets and which tickers are missing. A subtotal is always labelled as a subtotal.

Flows are never estimated: a fund-day without a reported flow stays empty. The API withholds any daily flow that depends on a source Kresmion is not licensed to redistribute and says so in the response. Fund facts come from issuers' own pages (for example iShares), SEC filings and a market-data vendor.

Intelligence desk

Open-source intelligence, compound signals and prediction markets.

22 · OSINT pipelines

Open-source intelligence comes from three lanes on independent schedules:

• Official and publisher feeds: first-party RSS and Atom feeds read on their live terms, including the SEC, CFTC, OCC, US Department of Energy, BLS (employment, CPI, PPI, job openings), UK OFSI sanctions, the BIS, the Reserve Bank of Australia and the European Commission, plus a small set of publishers' own feeds. Rows from official feeds carry the publisher's name and attribution.
• Telegram: a vetted list of public channels.
• Web news index: GDELT's multilingual news index.

The social-media lane read public X posts through a third-party mirror until that mirror stopped on 2026-08-24. The lane has delivered nothing since, and no replacement source has been added.

Events are deduplicated, language-detected and translated on ingestion, and the original text is always kept behind a "Show original" toggle. OSINT titles are the source's own words, shown under an "External report" badge. A clustering layer groups related events that share entities and arrive close in time; clusters above a relevance threshold get a one-sentence summary from a language model, which may only summarise what the cluster contains. Source credibility is scored 0 to 100 on track record, attribution and editorial standards, and sources below a floor are excluded. The source list itself is editorial work and is not published.

23 · Compound signals

A compound signal fires when independent source families (for example OSINT, positioning and price action) point at the same region or asset within a short rolling window. Several events from the same family count once. Conviction (high, medium, forming) reflects the number of source families and the strength of each contribution; the exact weights and noise floors are not published.

24 · Prediction markets

Kresmion observes prediction markets read-only: the probability a venue's traders assign to an event, how it moves and the order flow behind it. There is no trading integration.

• Venues: Polymarket is the primary venue. Kalshi prices appear only for signed-in users, where a Kalshi market is matched to the same question: in the market view and the cross-venue consensus and divergence views. The public /odds pages, including the midterms board, show Polymarket only. Kalshi data is not served through the API or MCP, which carry Polymarket only.
• Coverage: new Polymarket markets are discovered every hour; prices of markets already held are sampled every 5 minutes for the active tiers and at least every 6 hours for the long tail; a full sweep once a day picks up resolved markets and delistings, and resolutions are backfilled every 30 minutes.
• Cross-venue consensus: matched questions are merged into one consensus row; sibling markets are never blended, sports markets are excluded, and each row keeps a 24-hour baseline.
• Toolkit: market movers (largest 24-hour and 7-day repricings above a volume floor), a resolution calendar, crowd calibration (Brier score and reliability buckets on resolved markets, read at half of each market's observed life, capped at seven days and never under 15 minutes, so a short-lived market is read hours or minutes before it resolves), and live order-book depth with estimated slippage for standard sizes.
• Wallet and algorithmic flow: for the most active markets, public trades and holder positions are read every 30 minutes and aggregated by wallet. Detectors flag execution patterns visible in the public tape (TWAP-like clips, absorption, bid ladders). The flags describe order-flow structure and carry no directional claim.
• Midterms board: the 2026 US midterms page shows Senate and House control markets with 24-hour and 7-day moves and depth, from Polymarket only.
• Resolutions: public odds pages report an outcome only when the venue has resolved the market; an outcome inferred from a final price is not shown as a result.

A move of 15 percentage points or more within 24 hours creates an event in the intelligence feed. Published briefs quote venue probabilities with a fetch timestamp.

Note: prediction-market probabilities are crowd prices under each venue's incentives and liquidity. They are not forecasts by Kresmion and not financial advice.

Signals desk

How signals are counted, retracted and scored.

25 · Signal hygiene and track record

A signal is a dated, sourced event record: a filing, a threshold crossing, a positioning change. Signals are descriptive; none is a recommendation.

• Withdrawn signals are removed from every public feed, count and score. A signal written more than once by a re-run is one claim, scored once on its first copy.
• No confidence score: Kresmion does not publish a model confidence figure for signals, because none is measured.
• Track record: for each signal with a resolvable symbol, the forward return is recorded at +1, +7 and +30 days. Entry is the first daily close after the signal day, to avoid look-ahead. Losing signals are stored and shown exactly like winning ones. Per-type aggregates are served through the API and MCP.
• Signal accuracy page: rolling hit rates per signal type and direction at 1, 5 and 30 days (close at firing against the close N days later). Groups with too few firings are suppressed, and a cell reads "ins. hist" until enough firings have aged into that horizon.

Note: past hit rates say nothing about future returns, and a hit rate ignores size: many small wins and a few large losses can look better than they are.

Portfolio and workspace

Personal holdings, reports and messaging.

26 · Portfolio and on-chain wallets

Portfolio holdings can use FIFO or average-cost lots, computed in the browser from your encrypted transaction ledger. On-chain wallets use FIFO lot matching: an incoming transfer opens a lot at its USD value on arrival, an outgoing one closes the oldest lot first, a swap is a disposal of one token and an acquisition of the other at the swap-time value, and transfers between your own wallets are ignored. A token with no recent price is shown at cost with a "no mark" flag; a price is never guessed.

Note: this is a calculation over your records and public blockchain data, not tax advice.

27 · Reports

The report generator renders PDF reports from the same tables the dashboard reads. The figures are computed deterministically; an optional AI narrative, where offered, is kept separate from the numbers. Generation is capped at five reports per user per day.

• Portfolio risk: volatility, correlation matrix, drawdown profile and historical Value at Risk at 95% and 99%, on your wallet holdings, your watchlist (equal-weighted) or a custom list with optional weights.
• Asset deep dive: price action, signals, institutional positioning, insider trades, OSINT and related prediction markets for one ticker.
• Macro regime briefing: the regime score, factor breakdown, COT extremes and Fed-rate odds from prediction markets.
• COT positioning: every tracked CFTC futures market sorted by three-year z-score.
• Capital flows digest: the latest 13F quarter's largest increases, reductions, exits and new positions by dollar change.
• Watchlist morning brief: one summary per watchlist asset with price moves, recent alerts, related prediction markets and the regime.
• Prediction market microstructure: cross-venue divergence, repricings, algorithmic-flow flags, calibration and wallet flows.

Market-wide reports are generated once per UTC day and shared; personal reports are generated per user and never cached. Reports can be emailed on a daily, weekly or monthly schedule (up to five active schedules per user).

28 · Messaging and shared objects

Kresmion messaging is a terminal-native chat. A ticker, signal, chart or board dropped into a conversation renders as a live object with the same source attribution it has elsewhere: a shared signal shows its originating filing or feed, a shared price shows its as-of time and source. Kresmion adds no interpretation to a shared object. For how personal data and messages are handled, see the privacy policy.

29 · Reading public research data

Institutional holdings are labelled by reporting quarter and security identifier, with reported option positions separate from share holdings. Tool snapshots state their observation dates; a cached value is not an intraday quote. Read about the research publisher and the editorial approach.

30 · Data sources

• SEC EDGAR: 13F-HR, Form 4, Schedules 13D/13G, 10-K, 10-Q, 8-K, 424B prospectuses, XBRL financial data and the company ticker map.
• FINRA: short interest and daily short sale volume. Source: FINRA. Short interest data owned by FINRA.
• Clerk of the US House of Representatives: members' periodic transaction reports.
• FRED (Federal Reserve Bank of St. Louis): macro series, Treasury yields (Federal Reserve H.15), OECD long-term rates and the Fed balance sheet. This product uses the FRED® API but is not endorsed or certified by the Federal Reserve Bank of St. Louis.
• Federal Reserve and New York Fed: the FOMC meeting calendar and money-market reference rates.
• US Treasury: TIC foreign holdings and the Treasury cash balance.
• ECB: the euro-area yield curve and long-term interest rates for convergence purposes. Source: ECB statistics.
• OECD: long-term interest rates for non-EU countries, via FRED.
• BIS: central bank policy rates and international banking statistics.
• Central banks: policy-rate publications of individual banks, used where they are newer than BIS.
• Government debt offices and finance ministries: sovereign ratings from their own pages, where the terms permit reuse.
• CFTC: Commitments of Traders.
• EIA: weekly petroleum and natural gas inventories.
• Official feed publishers: SEC, CFTC, OCC, US Department of Energy, BLS, UK OFSI, BIS, Reserve Bank of Australia and European Commission releases.
• GDELT and public Telegram channels: open-source intelligence.
• CoinGecko: crypto prices and market caps.
• Derivatives exchanges' public APIs (Binance, OKX, Bybit, Hyperliquid): funding rates, open interest, long/short ratios and liquidations.
• iShares: shares outstanding for its bitcoin fund.
• Polymarket: market catalog and prices, order books, public trades and holder positions.
• Kalshi: market prices for matched questions, shown to signed-in users only.
• NASA FIRMS and Open-Meteo: satellite fire detections and weather for physical intelligence.
• A market-data vendor: options chains, futures settlements, the cross-asset daily closes and ETF prices.

Every outbound request carries a contact-tagged User-Agent so source operators can reach Kresmion. Documented rate limits are respected, requests back off on errors and stop on opt-out.