From any global event to your local exposure, in six steps or fewer.
Mark Granovetter showed in 1973 that the most valuable information travels through weak ties between clusters, not through strong ties inside them. Chainge's job is to find the path from a global event through five or six weak ties into a specific customer's exposure, every week, with every claim cited, every prediction tracked to a verdict.
Six panels follow. The premise, the input, the engine, the grounding, the matrix, the flywheel.
Weak ties carry the signal. Strong ties carry consensus.
Granovetter's claim, half a century on, still holds. Strong ties inside a cluster carry consensus. Weak ties between clusters carry the surprises. A consultancy with the right kind of team can trace those weak ties for a single client, as an institutional engagement. An AI system can now trace them at scale, for any company that can describe its structural shape.
The qualitative quant; elite foresight, democratized.
strong ties inside a cluster carry consensus. weak ties between clusters carry the surprises.
A profile of your business, run once, then anchored.
A customer answers a structured prompt about their business: what they make, where they sell, what they buy, which regulations apply, where the gas comes from, how they hedge. The prompt runs inside the customer's own AI tool (ChatGPT, Claude, or Gemini), so step one needs no install. Output is a profile JSON describing twelve structural dimensions: product families, production geography, energy and water exposure, commodity hedging, procurement cadence, distribution channels, currency exposure, regulatory regimes, customer concentration.
The profile is the anchor. Every cascade we run for that customer threads through it. No client data leaves the customer.
Twelve dimensions of your structural shape.
Sectors and revenue mix. Production locations. Commodity exposure plus hedges. Customer geography. Regulatory regimes. Labour and capital structure. Procurement cadence. Distribution channels. Currency exposure. Energy and water draw. Customer concentration. Known uncertainties.
The anchor every cascade threads through.
The JSON object is the spine. Every weekly cascade we run for this customer is generated against this profile, with no client data leaving the customer's environment. The diagnostic catches anything you missed; the profile is then locked, with the option to refresh it as your business shifts.
The diagnostic and the weekly briefing arrive through a direct conversation with Niels. You start with a profile of your company.
Behind the briefing, a weekly pipeline.
Every Monday the engine ingests the previous week's news, regulatory filings, prediction-market odds, and macro data. It clusters them into five to eight strategic themes for the week. Then, instead of tracing one cascade per theme, it generates candidate cascades across the customer's full exposure surface: every structural dimension of the profile that a live theme could plausibly reach. Each candidate is a chain of five or six numbered steps, each naming a transmission mechanism and an attenuating probability node by node.
Then the gate. A skeptic pass verdicts every candidate chain: promote, surface, demote, or drop. The system surfaces up to three promote-verdict headlines, plus the full deep field of everything that clears at surface or above, marked as system-generated, and it reports an honest count of what it dropped. Generate wide, then gate hard: the gate is the product, not a formality. A grounding pass then anchors every quantitative claim against the harness layer described in the next panel.
Ingest
News (Politico EU, Euractiv), central bank releases (ECB), regulatory filings (EUR-Lex), global event data (GDELT), macro time series (FRED), prediction markets (Polymarket), think tanks, arXiv.
Cluster
Each item tagged for industry, urgency, novelty, named entities. The week's signals collapse into five to eight strategic threads.
Generate wide
Candidate cascades are generated against the customer's full exposure surface, not a fixed set per theme. Each is a typed object: nodes, transmission mechanisms, probabilities, citations, attenuation rationale. One structured-output call carries the multi-perspective analysis (regulatory, commodity, geopolitical, industrial, technological) inside a single prompt, not across orchestrated parallel calls. Each chain runs five or six steps into the customer's profile.
Gate
A skeptic pass verdicts every candidate: promote, surface, demote, drop. Up to three promote headlines surface, plus the deep field that clears at surface or above, with an honest count of what dropped. A grounding pass anchors each claim against the harness (next panel). No claim ships without a source chip.
A reopening resolves the geopolitics, not the cost.
When a cascade starts at a chokepoint like the Strait of Hormuz, the obvious read is a single date: does the strait reopen. Chainge frames it the other way. The originating event sets the scene, but the exposure rides the grounded spine downstream, where crude that has run elevated, risk premia, war-risk insurance, tanker rates and contract repricing all decay on a lag. So even on a full reopening the cost base does not snap back, and a business exposed to it stays exposed for months after the headline clears. The prediction then rests on sustained cost observables, not on guessing one de-escalation date. That is the difference between testing the trigger and testing the destination, and it is why the originating event sits as context, below the grounded spine, not as the floor under it.
Illustrative, a shape and not a forecast. The solid line is the actual cost path, which decays on a lag; the dashed line is the snap-back people assume on a reopening. The shaded gap is the persistent premium your exposure rides, even after the strait clears.
Generation runs wide, the gate runs hard, and the grounding step is where most of the credibility work happens. The harness is the next panel.
Where the credibility comes from. A primary-source harness.
Each cascade claim is grounded against real economic and regulatory data. Triangulation is the discipline: multiple independent sources per claim, with structured disclosure of what each one can and cannot reach. When no source can ground a claim, the cascade surfaces an explicit "unaddressable" block instead of silently omitting it.
A dozen grounding sources sit in the harness today, plus one synthesis overlay and one passive watcher. Each one closes a structural gap the others cannot. The layer is under construction by design: every customer interaction that surfaces an unaddressable claim pulls the next adapter in. The most recent deepening, in the snapshot below as of 2026-06-11, added the bodies of EU law via the official Publications Office route, the Dutch grid regulator's decision register, and the Italian gas-price stack. What follows is that snapshot, not the end state. Five concrete sample citations follow; each comes from a real adapter output, not a promise.
| Source | What it grounds | Shape |
|---|---|---|
| EXIOBASE3 | EU sector-aggregate input-output | single-source direct-query |
| CBS Nederland | NL granular sector and energy specifics | single-source direct-query |
| CEFIC | EU chemicals industry context | single-source direct-query |
| Airline filings | Per-carrier business-model split | multi-firm curated |
| China firms | Firm-level KPI for listed upstream chemicals producers | multi-firm curated |
| E-PRTR | large industrial sites and their pollutant releases | EU regulatory |
| ECHA CHEM | Substance-level regulatory status | substance-level |
| ENTSOG | EU cross-border gas flow at network-point resolution | EU regulatory |
| EUR-Lex bodies | Full text of EU regulations and directives, article-anchored | EU regulatory · article-level |
| ACM register | Dutch grid-regulator draft and final decisions | national regulatory |
| ARERA | Italian PSV gas reference price, monthly | national regulatory |
| EC DG ENER | EU quarterly gas benchmark (TTF), for the derived PSV-TTF spread | EU regulatory |
| MRIO subsegment overlay | Disaggregates EXIOBASE buckets (specialty vs bulk) | synthesis adapter |
| ISPRA SPARQL watcher | Passive monitor for Italian RIR linked-open-data publication | watcher (not active grounding) |
Free public Italian sources nuanced a cluster the earlier framing called unreachable.
The producer's main industrial site is publicly registered under EU hazardous-site rules. It sits below the IED industrial-emissions reporting threshold, with years of sibling-site pollutant releases on the E-PRTR record. The earlier "we cannot reach plant-level data" assumption was wrong.
An SVHC listing on a substance inside a customer's commercial product line.
The ECHA CHEM adapter identifies a candidate substance on the SVHC Candidate List, in scope of a customer product line. The harness pins the substance, the article, the listing date, and the decision window of eighteen months from listing. Traceable to the ECHA register; reachable by any operator who knows where to look, which is exactly the point.
A twenty-point margin swing at a listed upstream producer.
The China-firms adapter surfaces a listed upstream producer's margin trajectory: roughly +13% in FY2021, about -7% in FY2023, near +3% on the latest reading. A twenty-percentage-point industry-cycle swing peak-to-trough, observable from curated public KPI data. The recovery is thin; the latest reading sits well below the FY2021 peak.
The Dutch specialty tail that EXIOBASE alone aggregates away.
The MRIO subsegment overlay combines four underlying sources (CEFIC, EuPIA-style references, UN Comtrade, pooled listed-company segment reporting) to disaggregate EXIOBASE3's Chemicals-nec bucket. EU27 specialty share within NACE 20: 34% (range 32-37%, medium confidence). NL specialty share: 38% (range 35-42%). The four-percentage-point gap reflects the Dutch specialty concentration that the national aggregate alone smooths over. Confidence bands disclosed, underlying sources cited, methodology locked.
A cross-domain weak tie from a geopolitical event to industrial feedstock supply, in observable network-point data.
ENTSOG transparency surfaces the post-2022 European gas supply transition in concrete flow data: northern pipeline entry points up year on year, the Russian-gas-via-Austria route sharply down, coastal LNG terminals up. Industrial gas supply rerouted from Russian transit to Northern European pipeline plus LNG. Observed in public European transmission system data.
More sources raise the floor. Only the operator's own data raises the ceiling.
When the bodies of EU law came into the harness, the effect was measurable and bounded in the same breath. Four cascade nodes across two operator profiles moved from cited-by-identifier to grounded-in-text, so the mid-chain reasoning got cleaner and better evidenced. The confidence ceiling on those chains did not move. It is set by claims that only an operator's own data can settle: a grid operator's internal congestion mechanics, a manufacturer's actual cost base. Shared public sources buy breadth and cleanliness; only operator-specific grounding lifts the cap. That honest split is the whole reason the feedback loop in the next panel exists, and it is more telling than a longer source list. The Italian price stack records the same discipline: where redistribution terms gate a feed (the GME spot exchange), the harness marks an honest gap rather than reaching around it, and grounds Italy's cost base from the public ARERA and Commission series instead.
Public data grounds the chain. Your own data grounds the last step.
A single discovery run on a representative archetype, a model investor in Dutch AI data centres, drew this line itself. An EU power-market rule travels across the grid and reporting domains to the cost of keeping a site powered and compliant. Read it top to bottom.
The four upstream steps are grounded on public primary sources, EU law and European grid data, and that floor is what the harness reaches for any profile. The final step, the one that lands on the site's own economics, the engine holds back at lower confidence on purpose, because it cannot evidence it from public data. That one held step is exactly what an operator's own data grounds. The engine drew the line, not the marketing.
Predictions become a track record. Every horizon lands in a public ledger.
Every cascade prediction ships with a locked horizon, capped at 90 days. When the horizon arrives, the prediction is scored against primary-source data: right, partial, wrong, or unclear. The verdict is mechanical where possible (a FRED series threshold, a Polymarket resolution, a EUR-Lex filing), observed against operator judgment where not. The original text is never edited. The ledger reads as a record, not a story.
Two states travel together on every row. The prediction has its own lifecycle (pending, resolved, superseded). The underlying situation has its own (active, resolved). A prediction window can close while the situation continues; the matrix shows both, separately.
The first prediction window closed 2026-06-01: Polymarket Hormuz May 31, unclear. The strait situation remains active. Several prediction horizons run on the same situation, and more clusters are live in the matrix; the current tally sits on the track record.
| Source | Prediction | Status |
|---|---|---|
| Polymarket | Hormuz tension lifted before 2026-05-31 |
horizon closed: unclear
situation active
|
| Polymarket | Hormuz tension lifted before 2026-06-30 |
horizon pending
situation active
source disputed
|
| Polymarket |
▸
3 related horizons
June 7 · June 15 · July 31
June 7horizon pending
June 15horizon pending
July 31horizon pending
|
situation active
|
| Chainge chain (Anne lens) | Margin compression at Brabant specialty-chemicals SMEs Q3-Q4 2026 |
horizon pending
situation active
|
| Operator note (O'Leary, Ryanair) | Weaker EU budget carriers face refinancing pressure in Q3 earnings |
horizon pending
situation active
|
| Chainge chain (DSO operator lens) | Dutch regional DSO regulated revenue under ACM congestion framework |
horizon pending
situation active
|
two states on every row. the prediction window has a lifecycle; the situation has its own. one cluster shown; matrix updates weekly. the prediction markets we ingest already picked their horizons; we surface the distribution rather than asserting our own.
Feedback compounds the cascade. Track record compounds in parallel.
A curated cascade is shared with the customer roughly monthly. Their response refines the mechanism, validates the headline, or flags an angle the engine missed. The refinement feeds back into the next cascade iteration for that customer. Over time, two things compound at once. The per-customer cascade gets sharper, because the operator is teaching the engine where its model is too coarse. The methodology track record gets longer, because every prediction the engine ever shipped is in the public ledger, scored against horizon-locked verdicts.
That double compounding is the moat.
Profile
Customer input, twelve dimensions, locked.
Cascade
Engine output, five or six steps, grounded.
Curated share
Monthly. One chain in full, the rest as a digest.
Operator feedback
Refines the mechanism. Validates the headline. Flags missed angles.
Operator feedback feeds back into the next cascade iteration for that customer. The original chain text stays in the ledger, unedited, with the verdict horizon intact. Track record compounds in parallel with the per-customer refinement.
A regional DSO refining the transmission mechanism in Cascade A.
A Northwest European regional DSO, with a concentrated distributed-generation customer base, refined the mechanism of Cascade A on 2026-06-02. The cascade had anchored on classical spark-spread economics; the operator pointed at total CHP flexibility value (imbalance, emergency power, flexibility, plus spark spread) as the load-bearing trigger. Same target exposure, sharpened mechanism. The next cascade iteration carries the refined trigger. The original chain text stays in the ledger, unedited.
Six questions readers ask first.
Six things readers ask before going further. Same answers I'd give in conversation.
-
How is this different from McKinsey or any consultancy? Q01
The qualitative quant; elite foresight, democratized. Consultancies sell cross-domain synthesis built on largely public information as an institutional engagement. Chainge runs the same synthesis with every claim cited and every prediction tracked to a verdict. The methodology is auditable, the citations are auditable, the verdict is timestamped. McKinsey is a black box you trust because of brand. Chainge is a glass box you trust because of receipts.
-
How is this different from ChatGPT, news, and my own analysis? Q02
Different comparison set. ChatGPT reads headlines about headlines, with no horizon and no track record. Chainge reads primary sources: EUR-Lex for regulations, FRED for macro time-series, GDELT for geopolitical events, Polymarket for crowd-priced probabilities, arXiv for emerging research, and a primary-source harness grounds every cascade claim. Every chain carries a source chip. Every prediction carries a verdict horizon. The right way to think about Chainge is as an institutional research desk that runs every week against your profile, not as a smarter chatbot.
-
Where's the track record? When do I see verdicts? Q03
First verdict landed 2026-06-01 (Polymarket Hormuz May 31, unclear). Every prediction carries a verdict horizon, capped at 90 days. Short-horizon predictions resolve faster; long-horizon predictions resolve at the cap. The matrix runs several active clusters, tracked live on the record. No retroactive editing. No narration after the fact. Come back as horizons land; the ledger speaks for itself.
-
What happens when a chain is wrong? Q04
Every chain ships with resolution markers: regulatory deadlines, commodity price thresholds, prediction-market levels. When a marker resolves, a deterministic function returns right, partial, wrong, or unclear. Not judged by an AI. Wrong chains are published wrong. The next week's briefing leads with the verdicts that landed. Two consecutive wrong calls on the same domain trigger a methodology review, not a quiet edit. The failures stay visible. That's how the methodology compounds.
-
What if a cascade is correct but I can't act on it in my market? Q05
Chainge surfaces cross-domain links; it doesn't promise every link is actionable in every operator's market. Some chains resolve into exposures you can see but can't trade against. The briefing flags these conclusions explicitly as structurally unactionable. The value is still real: a known exposure you can't hedge is a known exposure you can communicate to your board, your customers, or your insurers.
-
How do I get access? Q06
Access is a direct conversation with Niels. You start with a profile of your company, and the weekly cascade runs against your specific exposure from there.
See it work, end to end.
Anne runs Brabant Advanced Materials, a specialty chemicals SME in Eindhoven. We took her through profile ingest, the diagnostic, and the five-hop cascade that came out, including the regulation the diagnostic caught that she hadn't listed and the German lobbying fight that decides her outcome.