Where our numbers come from

Trust in grid data depends entirely on knowing how it was produced. Here is exactly how PowerGridIQ builds the PGIQ Rating and collects, processes, estimates, and labels every figure on the site, with nothing hidden behind a login.

How the PGIQ Rating works (v2.1)

The PGIQ Rating scores each market for siting or scheduling a large new electricity load (data centers, AI clusters, industrial load). It has two layers: a five-pillar view of the fundamentals, and a reliability overlay that adjusts for what a load actually experiences on the ground.

The five pillars (the fundamentals)

Each market is scored 0-100 on five weighted pillars: Access (30%, can a large new load actually connect: moratoria, queues, connection rules), Availability (25%, whether there is enough generation, with room to spare), Cost (25%, the price of power), Momentum (15%, the data-center build already underway, not just announcements), and Carbon (5%, how much carbon the grid emits per unit of power). The weighted blend is the Fundamentals score. Weighting "lenses" (default, cost, carbon, momentum) let you re-weight the pillars for your own priorities.

The reliability overlay

Fundamentals reward cheap, clean, available power, but they do not capture whether the lights stay on. On top of the pillars, each assessed market carries a profile of cited reliability reads across five declared layers: historical bulk-system performance, forward resource adequacy, local delivery performance, contractual service terms, and designed site resilience. They are set out in full under the definitions below, and most markets hold only some of them, which the page states explicitly rather than leaving blank. Each read is None, Watch, Elevated, or Severe, and every input is labelled measured, modelled, assessed or event-based with a source link. Forward adequacy is a forecast and is never labelled measured. The overlay deducts points from the Fundamentals score to produce the reliability-adjusted PGIQ Rating, and both numbers are published. Reliability is framed as a risk to a prospective large new load, using operators' own published figures, never as a verdict on any utility or grid operator.

Tiers, and how Tier 1 is decided

Markets sit in five tiers on the PGIQ Rating: Tier 1 Prime, Tier 2 Strong, Tier 3 Workable, Tier 4 Constrained, Tier 5 Largely closed. Tier 1 is relative, not an absolute cutoff: a market earns Prime only if it is reliability-assessed, carries no Severe flag, and its PGIQ Rating is within 4 points of the highest adjusted score anywhere (or clears an absolute 75). The minimum Elevated reliability dock is larger than that 4-point window, so a market with a real reliability flag cannot sit in Prime beside a clean peer. A Severe flag caps a market at Tier 3 outright. This is why a market with strong fundamentals but a demonstrated reliability problem, for example a grid with a multi-day blackout history, can sit a tier below a cleaner peer.

Regional splits

Where a country's internal divergence is large enough to change a siting decision, we rate the sub-regions rather than a single national average. The country page becomes a non-ranked overview that shows the regional range and links to its regions; the sub-regions are the rated, ranked markets. Current splits: Germany (north and south), Japan (east 50Hz and west 60Hz), Sweden (northern zones and the constrained south), and Australia (New South Wales, Victoria, Queensland).

Coverage, confidence, and dating

The reliability overlay is applied market by market. Assessed markets show their reads and cited sources; a market not yet assessed shows "reliability not yet assessed" and keeps its Fundamentals rating unadjusted. Every market carries a confidence level, and a read built only on modelled inputs is capped at Watch. Event-based reliability flags step down over time absent recurrence, so a grid that fixes itself is not penalized indefinitely. Ratings and reliability reads are dated and versioned, and rating changes are logged as dated actions so the record of calls is auditable over time.

The five definitions everything rests on

Ratified 26 July 2026, version 1.0. These are the meanings behind every number on this site, and they are fixed within a version and never edited in place, rather than irreversible. That distinction is deliberate: a definition that could never change would eventually be wrong and unfixable, while a definition changed in place would silently rewrite every claim and case built on it. So every claim we publish carries the definition version it was produced under, a change creates a new version with an effective date, and the superseded version stays readable alongside a crosswalk stating what moved and what it does to comparability.

1. Cost archetype — 100 MW baseload archetype v1

100 MW is maximum contracted demand, not continuous consumption, at a 95% annual load factor: about 832.2 GWh in a normal 8,760-hour year, with the hourly profile stored explicitly. Service is high-voltage at a disclosed point of delivery and service voltage. N-1 means the nominated firm load remains serviceable following loss of one supply element, a line or transformer, at the supplying substation, restored by automatic or manual transfer within 30 minutes. Recurring buyer cost includes energy, capacity, demand, transmission, distribution, loss, ancillary, reactive-power, tax, levy, carbon and commercial-supply charges. Customer-funded connection, substation and network-upgrade capital is reported separately and never blended into recurring cost: excluded does not mean invisible, because connection capital is frequently the number that decides a project. Energization date, currency, nominal or real basis, FX convention, contract term and escalation are visible inputs set per case, so two cases at different dates can share one archetype.

2. Connection-process taxonomy v1

Named process rather than queue, because a large-load request does not always enter something a jurisdiction calls a queue: it may be a transmission-service application, a utility service request or a study process. Every timing or capacity figure declares its process type (generation, storage, load, transmission service or utility service request), its population (requested, active, studied, offered, contracted, energized or withdrawn), its exact start and end milestones, requested and energized MW, whether service is interim, conditional or firm, the cohort period, sample size, censoring method, geography, observation date, method and sources. The canonical end-to-end load metric runs from a complete application being accepted to full firm energization of the nominated MW; initial, phased, temporary, bridge and conditional energization are separate, earlier milestones reported separately. "Application to energization" is not a definition, because an application could be an inquiry, an initial submission, a complete application, a paid deposit or an accepted study, and energization could be construction power, a first phase, interruptible service or full firm service. No timing, withdrawal or completion figure may cross process types.

3. Launch carbon scope v1

We report estimated average operational grid emissions intensity in gCO2e per kWh, for a declared geography, period and system boundary, using direct operational emission factors. The generation-versus-consumption basis, imports, delivery losses, temporal granularity, factor set, factor vintage and uncertainty are disclosed or explicitly marked unknown. It is permitted for location-based grid comparison only. It is not a marginal-emissions estimate, a lifecycle assessment, a procurement claim, or by itself a market-based Scope 2 result. Carbon method scope and evidence maturity are separate attributes: average operational intensity can support a project case where the case's declared carbon requirement is specifically average operational intensity, and a case requiring marginal, market-based, lifecycle or hourly-matched carbon returns insufficient evidence for that requirement until the method exists here.

4. Reliability evidence layers v1

Reliability is reported as a profile and never as one combined label, across five layers: historical bulk-system performance, with defined events, geography, period and reporting threshold; forward resource adequacy, an assessed or modelled forecast with horizon, scenario, criterion, vintage and metrics, never described as measured, because it measures nothing that has happened yet; historical utility-territory delivery performance, meaning utility or customer-class SAIDI, SAIFI and CAIDI with the major-event convention stated; contractual service, meaning firm, non-firm or interruptible rights, curtailment conditions, required redundancy, contingency capacity and transfer time; and designed site resilience, meaning utility feeds, UPS, storage, on-site generation, islanding, black start and residual expected unserved energy. Project-ready depth requires the applicable bulk, forward, local-delivery and contractual layers. Local depth additionally requires current local or site evidence and the designed-site layer. A critical unknown produces insufficient evidence or a conditional read, never "reliable".

5. Evidence-tier rules M1–M4 v1

Tiers are assigned independently to each field, geography, effective date and method version, based only on eligible evidence actually stored with provenance. A tier is not a rating strength, a confidence score or a feasibility conclusion. M1 Directional is versioned, cited evidence for discovery and monitoring, not normalized enough for project comparison. M2 Comparable is a normalized value or range with a defined boundary, geography, vintage, method, provenance and uncertainty. M3 Standard-case ready is M2 plus evidence directly applicable to a named archetype and pathway, with assumptions, ranges and dependencies stated; it supports a gated standardized Project Case, not site feasibility. M4 Locally substantiated is M3 plus current primary or contractual evidence at utility-territory, site, bus, substation or service-agreement level, and still requires direct confirmation. The tiers are cumulative, so local evidence alone cannot jump to M4. An aggregate tier is shown only for a named use case, and equals the lowest tier attained across that use case's versioned required-field matrix; the field-level profile is always displayed alongside it. Evidence that is unknown, not assessed, stale, withheld or subject to unresolved dispute can never satisfy a required threshold, and a critical required unknown produces insufficient evidence.

What a utility outage record does and does not establish

The local-delivery layer is deliberately named historical utility-territory delivery performance. It is not site reliability, it is not project reliability, and it is not the reliability available at a given location. A utility's filed SAIDI is an average across every customer in its territory, from a rural single-phase tap to a substation-adjacent industrial plant. Naming it after the site is how a territory average quietly becomes a promise about a point on the map.

Specifically, the figure is silent on: performance at a specific substation or feeder; transmission-level service, which is a different product from distribution service; the industrial or large-load customer class on its own; N-1 or other redundancy architecture at the supplying substation; contractual curtailment rights the utility may hold over the load; power quality and momentary interruptions, which the metric excludes by construction; on-site UPS, storage or generation; and forward adequacy, because it is a record of what happened rather than a forecast.

What a single utility's record can and cannot carry. A serving utility is a verified relationship between a site and a territory. We do not have a site, so we never assert one. Where a market's evidence comes from one operator we call it a single-utility observation, and where it spans several we call it market context. A single-utility observation supports a Project Case only when that operator is genuinely the only one serving the market. Where we have excluded a co-serving operator — NOVEC in Northern Virginia, the Jackson, Cobb and Sawnee cooperatives in metro Atlanta — “one utility” describes our compilation rather than the market, and the figure is capped at Comparable (M2). A reader whose site sits in Sawnee’s territory learns nothing from Georgia Power’s filing.

A blend is not a serving utility. Where a market covers several utilities, we publish a customer-weighted figure with its constituents, weights and vintages shown. Customer weighting estimates the experience of an average customer, and a large load is not an average customer: it connects to one utility, not to a weighted population. So a multi-utility blend is contextual evidence — good for comparing markets and watching them move, and it tops out at Comparable (M2) for that reason. Only the serving utility's own filed record can support a Project Case. Where the serving utility is not yet known, the blend is context and nothing more.

Which reporting standard. Utilities file outage figures under IEEE 1366's major-event-day method or under their own; both exclude major event days, and the standard governs how those days were identified. We prefer IEEE 1366 and name the standard used for every figure. Where a utility's IEEE filings carry no values — which is common, and true of Dominion in Virginia for every year from 2013 to 2024 — we use its other filing and say so, because discarding a published figure on a technicality is not caution, it is a gap we created. Where a market's utilities split across the two methods we blend them, label the market mixed-method, and state the customer share on each — and every constituent carries its own standard, so the figure can be taken apart. We considered refusing instead, and rejected it: CAISO already publishes a blend running from 275.6 minutes at one utility to 71.1 at another, all within IEEE 1366, and the gap between the two methods for identifying major event days is far smaller than that. Refusing over the method while publishing the spread would have been inconsistent rather than careful.

The serving utility is published, not just the average. Where a market covers several territories, every market page lists them: each utility’s own filed outage figure, its customer count, its reporting standard and its vintage. This matters because the spread inside a market is usually larger than the spread between markets. Inside NYISO the constituent figures run from 15.0 minutes at Con Edison, an underground urban network, to 167.2 at Central Hudson. Inside ISO New England they run from 49.2 to 478.0. A siting decision in either market is not a bet on the average; it is a choice of territory, and averaging that away would hide the most decision-relevant thing we hold. The market figure remains contextual evidence and remains capped at Comparable (M2). A single territory’s record is that utility’s own filing and can support a Project Case for a site inside it.

United States figures in this layer are utility-reported EIA Form 861 data, normalized through Catalyst Cooperative's PUDL pipeline. Neither the EIA nor Catalyst Cooperative endorses PowerGridIQ, supplies us with data directly, or reviews what we publish; we read a public federal filing that anyone else can read too. Records are keyed on the EIA utility identifier together with state, report year and reporting standard, never on the utility's name, because names drift between filings and two utilities can share one.

How a national overview is adjusted

Some countries are rated by region, because the decision genuinely differs inside them: Germany, Sweden, Japan, Australia and ERCOT each have separate regional ratings, and those are the actionable ones. The national page above them presents the best available region and the tier range across regions, and it is excluded from cross-market rankings.

Its reliability adjustment is the average of its rated regions' adjustments, rounded to the nearest point, and the page names the regional figures it was built from. A national view has to represent all of its regions rather than only the strongest, and an average is decomposable in a way that a single national judgement is not. Until July 2026 a national page instead carried an adjustment drawn from a separate national reliability record which, for three of the five, was never applied to the published score. Adopting the averaging rule was a material methodology change, because the previous method did not specify an overview aggregation rule. Five overviews were evaluated and three scores moved; ERCOT and Australia were unchanged because their existing adjustments already equalled the regional average. See the change log.

Governance

PowerGridIQ is founder-owned and founder-governed. The founder has final authority over its methodologies, ratings, corrections and releases, and may appoint others to advise or to exercise delegated authority. Public consultation is not required. Material changes are versioned and explained, and prior results are preserved.

Governance update, 26 July 2026. PowerGridIQ has retired its mandatory 30-day public request-for-comment process. Methodology decisions are now approved by the founder or a person appointed by the founder. External comments may be considered but are not required before a change. This retirement is stated openly rather than dropped quietly, and the superseded commitment remains in the published record.

What we owe you is not a vote. It is an accurate account of what the method is, what changed, who decided, and what the change affected. You then decide whether to trust it. To comment, dispute a rating, or ask which version applied on a given date, write to hello@powergridiq.com.

Methodology change log

v2.14 — Every remaining US market checked against the right ranking — 1 August 2026. The market lists that v2.13 could not resolve from geography have now been checked one by one, ranked by the electricity large customers actually buy rather than by how many customers a utility serves. Twelve operators were added and twenty-six absences were written down with a reason. Which grid each utility belongs to was read from the balancing-authority code on its own filing, not decided by us: that is what settles whether a utility in Texas is in ERCOT or in one of the three other grids that also reach into the state.

Four outage figures moved. SPP 98.68 → 104.32 minutes, now 20 territories. MISO 121.15 → 119.90, now 25. ERCOT 97.27 → 97.52, now 10. California 177.14 → 177.06, now 4. All four cover every territory mapped to them.

The additions are the kind of operator a customer-count ranking cannot see: cooperatives and public bodies with few customers and very large industrial loads. Two North Dakota cooperatives serving the Bakken oil field, a Minnesota utility serving the Iron Range, an Oklahoma state agency, a Texas cooperative, and the municipal utility in Santa Clara, which serves a substantial data-centre load in the middle of Silicon Valley.

The Houston and Dallas sub-market had no recorded exclusions at all. Its list was right, but nothing said why the four grids adjacent to Texas were left out, or why San Antonio and Austin — both in ERCOT — are not in a market defined as those two load pockets. Eight absences are now written down. A list that is correct and undocumented cannot be told apart from a list nobody checked.

A new refusal rule for cost. Where a market has several utilities selling all-in service and only one of them has a large-customer class we can read, we now decline to publish rather than print that one company’s average under the market’s name. California hit this immediately: adding Santa Clara — correctly — produced a figure resting on one municipal holding about 4% of the market’s load. A caveat under a headline number does not undo the headline. A market that genuinely is a single utility, such as Northern Virginia, still reports, because no reader is being told that one company stands for many.

v2.13 — Four outage figures moved because the crosswalk was wrong — 1 August 2026. Every market’s list of serving utilities had been built from a ranking by customers served. That is the right ranking for an outage figure, which is customer-weighted, and it is the wrong way to decide who belongs in a market: it quietly drops any operator with few customers, however much of the region it actually serves. For markets that are a geography, membership is now resolved from the filings themselves rather than from a list, so an operator cannot be missed for being small. Four figures changed as a result, each now resting on far more of its market than the number it replaces.

Pacific Northwest 144.44 → 152.68 minutes, from 14 territories to 51 of 55, covering 98% of customers. ISO New England 107.58 → 112.25, from 9 territories to 32 of 35. New York 72.28 → 71.99, from 7 to 10 of 11. US Southeast 127.94 → 114.26, from a single utility to six.

The US Southeast change is the largest and rests on a decision we reversed. On 28 July the metro-Atlanta cooperatives were deliberately excluded so the market would read as one utility’s own filed record. That reasoning does not hold: those cooperatives serve metro Atlanta, and a figure that leaves out an operator serving the market describes the market less well, not more. Jackson, Cobb, Sawnee, GreyStone and Walton are now included. The rural cooperatives of south Georgia remain out, because they are in the same state and not in the same market. This market’s evidence role moves from a single-utility observation to market context; it could not carry a project-specific case before and still cannot.

Coverage is now a number, not a gate. The compiler previously refused an entire market if one filing was unusable, a rule that survived only because the old lists held a handful of large utilities that all file cleanly. Against full membership it discarded 46 sound filings out of 50. A market is now published when the measurable part covers at least half its customers, with the share stated on the page, and refused below that. Every market above states its coverage whether or not anything is missing.

v2.12 — When a utility may be named — 1 August 2026. A rule now governs where an operator’s name may appear. We name a utility only where we are republishing that utility’s own filed value, attributed and dated. A name never appears beside a figure we derived. The distinction is about whose mistake a wrong number would be. The per-territory outage table restates each operator’s own annual submission to the US Energy Information Administration; we are a conduit for it, and it stays as it is. A cost band is our construction — we chose which customers to select, the size threshold that selects them, and how to weight them — so a name printed next to one would be us making a claim about a named company on the strength of our own method. The build now fails if an operator’s name appears inside a derived cost section, so a future figure cannot acquire one by accident.

This is a limit on attribution, not a claim of safety, and it does not cover the error we are most likely to make. Mis-mapping remains ours. If we attribute a genuine filing to the wrong market, the wrong year or the wrong territory, the number is theirs and the mistake is entirely ours, with their name on it. That happened this week: Dominion’s North Carolina service territory was being averaged into a Northern Virginia figure. The safeguard against that is the territory-to-market mapping and the checks on it, not the provenance rule above.

Correction to v2.10, same date. The entry below describes market pages and the shortlist as ending with an ordered list of the evidence a project would still need. That list was withdrawn from every reader-facing surface on 30 July and the entry should have been amended then. It was a work list written for the people building the compilation, not information a reader came for. The compilation itself continues, ranked by how many markets each gap blocks, and it is kept internally. What a reader sees at the end of a market page is what the market page can support and where it stops.

v2.11 — Evidence-role labels were rendered from a retired vocabulary — 30 July 2026. The v2.5 ontology correction renamed the evidence roles, and five places that render them were not updated. The effect was live until today and ran the wrong way twice over: every market page printed “role not yet declared” while the dataset for the same market declared the role correctly, and on eighteen markets the caveat attached to a blended outage figure was dropped from the second place that restates it. Two gate rules written to catch exactly this tested the retired names and so matched nothing while reporting success. No value, tier, score or rating changed; what changed is that the page now states the role it holds. The gate now asserts the rendered words rather than the internal constant, since a check on the constant cannot see a renderer that ignores it.

v2.10 — Next evidence to obtain — 30 July 2026. Market pages and the shortlist screen previously ended by naming the operator that runs the connection process. That is wayfinding, and a reader capable of siting a large load already has it. Both now end with an ordered list of the evidence a project would need that a market-level read does not hold, each item pairing the gap, the artefact that would close it, and a question that can be sent as written. Two items appear on every market and always will: no source measures the connection wait for a large load as distinct from generation and storage, and none prices customer-funded connection capital, which the recurring cost band excludes by construction. The remaining four disappear from a market as that evidence is compiled. This changes no tier and no rating; it changes what the page hands the reader at the end.

The section formerly headed “who would actually serve you” is now how service and connection are organized. The old heading promised an answer that depends on a specific parcel. The section is deliberately not gated for completeness: where the structure is not compiled, that is a gap in our wayfinding and is stated as one, not as a finding about the market.

v2.9 — Connection pathway — 29 July 2026. Connection is the binding field on 76 of 79 rated markets, and 41 of those hold no connection evidence at all. The response is deliberately not to chase the tier. What a reader needs first is not a percentile wait but who runs the process, what it is called, and where the application starts, which is knowable for every market and does not decay quickly. Market pages now carry that where compiled, and say plainly that its absence is a gap in our compilation rather than a fact about the market.

The pathway record never feeds the connection depth tier. Knowing whom to ask is not evidence of how long they take, and letting wayfinding raise an evidence tier would be the same category error as treating system capacity margin as local access.

v2.8 — Insufficient evidence is a deliverable — 29 July 2026. Where the evidence cannot close a case, market pages previously said so and stopped. That wasted everything already held, at exactly the moment a reader most needs help. Every market page now returns the same eight things, built from evidence actually held rather than from a template: what is established, what is open and which of those unknowns would change the answer, who would actually serve you, credible cost and schedule ranges, what could disqualify the market, five questions to put to the utility or system operator, and the documents that would advance the case.

The questions are generated from the specific gaps found in that market and are phrased against the published 100 MW archetype, so they can be sent as written. Where a market has fewer open gaps than five, the list is completed from the site-level questions that no market-level evidence can answer at any tier: which substation would serve the site, what the service contract promises, and what the milestones are. That is not padding. Those questions remain genuinely open however good the market-level record is, which is the same point the serving-utility correction made in v2.5.

v2.7 — Evidence tiers are field-level disclosures — 29 July 2026. Market pages previously headlined a single aggregate tier for the Standardized Project Case. That let the absence of one label hide everything useful underneath it, and because no market currently reaches project-ready depth, it made a coverage result read as a verdict on the market. Pages now lead with the per-field tiers, which are the actionable statement: Northern Virginia holds cost M3 and connection M3 with reliability at M2, and knowing precisely where the evidence is thin is more useful than a single word.

The aggregate is retained and demoted. It now names the binding field rather than pronouncing on the market, and states plainly that a readiness result belongs to a declared case with its own requirements, siting and utility, which a market-level read does not know. Where a required field was never assessed, the page says so in those terms, because “not answered here” and “answered thinly” are different facts and absence of a finding is not evidence of low risk. No tier, score or rating changed.

v2.6 — Positioning and tier subtitles — 29 July 2026. Two of the five tier subtitles told the reader what to conclude rather than describing the market: Tier 1 read “Best available today” and Tier 2 read “Build with confidence”. They now read Strongest fundamentals and Strong, with known trade-offs. No tier boundary, score or rating changed; this is labelling. The claim that “the same score reads four ways” is also withdrawn: building a load, selling power, planning a grid and monitoring an exposure are different decisions, and the same evidence serves them through different reads rather than through one number.

The wider correction behind both: a PGIQ Rating is compression and navigation. It helps a reader find a shortlist. It does not carry the siting decision, and the evidence underneath it is what goes to the utility.

v2.5 — Serving-utility ontology correction — 28 July 2026. PowerGridIQ had been using “serving utility” to mean “exactly one utility in this market’s evidence”. That is not what the phrase means: a serving utility is a verified relationship between a site and a territory, and we hold no site. The label was doing work it had not earned, and it let two markets reach project-ready depth on the strength of a count rather than a relationship. This is a material methodology change, and it is a correction of our own reasoning rather than of any source.

Evidence roles are now single-utility observation and market context. A single-utility observation carries a Project Case only where no co-serving operator has been excluded from the market. Two markets moved down: Northern Virginia from M3 to M2, losing project-ready status, because NOVEC serves parts of Loudoun and Prince William; and US Southeast from M3 to M2, because Jackson, Cobb and Sawnee cooperatives serve metro Atlanta. Both exclusions were deliberate and both remain, but excluding a co-serving operator makes a figure less representative of the market, not more. US Southeast had been raised to M3 earlier the same day; that movement is withdrawn. No rating scores, pillar scores or rankings changed.

The same correction applies to the EIA-861 cost source, which is separately reclassified as a historical industrial class average, explicitly not eligible for a Project Case, and limited to Comparable (M2).

v2.4 — Serving-utility records published — 28 July 2026. The territories behind every blended local-delivery figure are now published on the market page and in the dataset, as by_serving_utility: 83 utility-territory records, each with its own outage figure, customer count, state, reporting standard and weight. No rule changed and no market moved tier. What changed is that a reader who knows which utility would serve them is no longer restricted to the average, and the evidence they need is a serving utility’s own filed record rather than a population.

Separately, contractual service terms were compiled for the US Southeast, taking that market’s reliability layer from M2 to M3. Since 1 February 2025 a new Georgia Power customer expecting 100 MW or more of peak demand may not take the standard large-power tariffs at all and must negotiate a customised contract carrying up to 15 years of term, minimum billing, collateral before construction and repayment of network investment on early exit. The threshold matches our own 100 MW cost archetype, so a project built to the archetype falls into this class by construction.

v2.3 — United States local-delivery coverage — 28 July 2026. Six previously unmapped United States markets were crosswalked to their reporting utilities, taking the compiled local-delivery layer from five markets to eleven, all 2024: MISO 121.2 minutes, SPP 98.7, ISO New England 107.6, Pacific Northwest 144.4, US Southeast 127.9, Desert Southwest 67.9. This is new coverage, not a methodology change; the aggregation and evidence-role rules are unchanged from v2.2.

Three market footprints were narrowed to match what each market actually claims to be, which is a correction rather than a new rule. US Southeast is Atlanta-anchored, so it is Georgia Power rather than the whole southeast; Duke, Dominion South Carolina, Santee Cooper and the TVA distributors are separate systems. Desert Southwest is Phoenix-anchored, so it is Arizona; Nevada and New Mexico are separate systems. MISO is MISO Midwest, so Entergy's Arkansas, Louisiana and Mississippi operations, which sit behind a contractual transfer limit as MISO South, are excluded. Every excluded utility is listed with its reason in the published crosswalk.

The largest single exclusion is worth naming. Commonwealth Edison is the biggest utility in MISO's states and is not in MISO: Chicago is a PJM load pocket. Including it would have added 4.2 million customers of a different system and dominated the blend.

v2.2 — Evidence role for utility delivery performance — 27 July 2026. The local-delivery layer was renamed historical utility-territory delivery performance, and every value in it now declares whether it is the serving utility's own filed record or a blend across several utilities. A blend is contextual evidence: it tops out at Comparable (M2) and may not support a Project Case. This is classified as a material methodology change, because it changes what evidence can carry a project-ready tier rather than correcting a value.

Ten markets hold a local-delivery figure. Nine are regulator-published averages across many distribution operators and are now marked contextual; Singapore is marked as a serving utility's record, because SP PowerGrid is the sole distribution licensee there. One overall tier changed: Great Britain moves from M3 Project-ready to M2 Compiled, because its outage figure is Ofgem's average across the fourteen GB licence areas rather than the record of the operator that would serve a given site. No rating scores, pillar scores or rankings changed. Approved by the founder. Prior outputs remain preserved.

The same release changed how the reporting standard is selected. PowerGridIQ previously used only IEEE 1366 filings; it now prefers them and falls back to a utility's other filing where its IEEE filing carries no values, naming the standard either way. This took the layer from one compiled United States figure to five, all 2024: Northern Virginia 131.3 minutes (Dominion alone, the serving utility's own record), CAISO 177.1 across three utilities, ERCOT 97.3 across nine, ERCOT Central 98.3 across three, and NYISO 72.3 across seven. All but Northern Virginia are contextual. ERCOT, ERCOT Central and NYISO blend utilities filing under both methods and are labelled mixed-method with the customer share on each standard shown (ERCOT 59% IEEE, ERCOT Central 44%, NYISO 24%). The spread inside these blends is wide and worth reading rather than averaging away: NYISO runs from 15.0 minutes at Con Edison, an underground urban network, to 167.2 at Central Hudson. That range is the clearest argument for why a blend is contextual evidence and not a read on any particular site.

v2.1 — Reliability overview aggregation — 26 July 2026. PowerGridIQ introduced a rule for aggregating regional reliability adjustments into national overview scores: a national overview now carries the mean of its rated regions' reliability adjustments. This is classified as a material methodology change, because the previous methodology did not specify an overview aggregation rule.

Five national overviews were evaluated and all five are now derived under the new rule. Three scores changed: Germany 57 to 53, Sweden 78 to 76, Japan 51 to 47. Two were unchanged, because their existing adjustments already equalled the regional mean: ERCOT (70) and Australia (51). Ranked markets, regional ratings and coverage tiers were unchanged. Approved by the founder. The prior methodology and outputs remain preserved, and each changed overview carries a dated rating action.

What is free, and what is paid

We keep a clear and deliberately stable line between what we give away and what we charge for, and we are stating it now, before it matters, so it can never look like a change of heart later. Our opinion is free, and stays free. That means every market's rating page, all five pillar scores, the reliability overlay, the outlook, the cited evidence, the methodology, and our own track record. The current read of any single market is free too, including its realized cost band and its connection-friction level on that market's own page. This is the part that should be public, because a rating is only useful if people can see it and cite it.

What we charge for is the compiled, cross-market asset and the dimension of time. A ranked table that puts every market side by side with precise figures, a downloadable export, the history and the quarter-over-quarter change logs, change alerts, and full machine access through the paid tiers of the interface for building software (API): these are the work product a professional is paid to assemble, and they are what the paid tiers and the enterprise feed provide. Put simply, our full read of any single market, its own numbers included, is always free; the compiled cross-market ranking with precise figures, the time series, the change logs, and the alerts are paid. We would rather draw this line openly and keep it steady than blur it and move it.

Three tiers of data confidence

Every value on PowerGridIQ carries one of three badges, and understanding them is the key to reading the site correctly.

We would rather show a clearly labeled estimate than a falsely precise live number. We also publish a public self-assessment of our own data quality: a per-market evidence score, and the measured-versus-modelled split of every data layer, at /data-quality. Each number on a market page carries its method, its vintage, and its source.

How a cost band was built

Every realized-cost band and its build-up carries one of three method labels, shown next to the number and clickable from any market page. They describe how the figure was assembled, not how recent it is.

On any market page, the component split within a build-up is our estimate; the total is the triangulated midpoint. The full measured-versus-modelled split of every data layer sits in the open data-quality audit.

When an operator is named. We name a utility only where we are republishing that utility’s own filed value, attributed and dated — the per-territory outage figures are its own annual submission, restated. A name never appears beside a cost figure, because every cost figure here is our construction: we chose which customers to select, the size threshold that selects them, and how to weight them. Cost is therefore reported for the market, not for a company. This limits what we attribute to whom; it does not make a mapping error less ours.

How we calculate derived metrics

Several of our most useful figures aren't published directly by anyone and have to be computed. Carbon intensity is estimated by multiplying each fuel's share of the current generation mix by a standard emissions factor (grams of CO₂ per kilowatt-hour) and summing the result. System-level room is the gap between an estimate of usable capacity and estimated peak demand across the market as a whole. It is a screening estimate of system adequacy, not local deliverability: it says nothing about whether the substation you would connect to has capacity available, and we do not assess local capacity. A market can show system-level room while a specific site cannot be served. Treat local capacity as not assessed until the utility confirms it. Grid stress combines utilization, reserve margin, and price signals into a single comparative score. Demand pressure is built from cooling and heating degree days, which we compute in real time from Open-Meteo temperatures using the standard 18°C baseline that grid operators use worldwide.

How developments are recorded

A development is a dated, cited change in the world that bears on siting a large load: an approval, a contract, a plant, a rule. The unit of record is the event, not the market. One event can affect several markets, and when it does we publish one event carrying a market impact for each — never one event per market. A shared event keeps a single event_id everywhere it appears, and both affected market pages publish together or not at all. Publishing one province and not the other would state a partial truth that reads as a misleading one.

Capacity is three different numbers and we keep them apart. The project nameplate is the physical plant. The physical location is the single market that hosts it. A market allocation is what one market is entitled to out of that plant. A 500 MW facility in New Brunswick, with 400 MW tolled to the provincial utility and up to 100 MW contracted to Nova Scotia, is a 500 MW project — not a 400 MW project, and not two projects. Adding the allocations across market pages would count the same turbines twice, so our API flags every shared-event row and our build fails if the allocations ever exceed the nameplate.

Where a figure is a ceiling rather than a quantity we say so and keep saying so. “Up to 100 MW” and “100 MW” are different claims, and only the first is supported when a contract states a maximum. The qualifier travels with the number into the API, the exports and the page text.

Planned and contracted capacity is not supply. Nothing that has not been built appears in current capacity, current availability or delivered energy, whatever its paperwork says. A development can be fully verified and still change no rating, and that is the normal case: an approval removes a blocker, it does not generate electricity.

Verification is a property of each claim, not of the event. Every material figure carries its own source, source class and verification state, and the event's state is derived conservatively from the weakest of them. We distinguish verified for context — supported well enough to describe to you — from verified for analytical use, which is the higher bar a figure must clear before it may feed a number, a pillar or a rating. An event can sit in the first state indefinitely. Being able to tell you something is not the same as being entitled to calculate with it.

Some older records predate parts of this schema. Where a development was written before we recorded a project stage or a delivery assessment, the canonical API marks it not_assessed_pre_schema rather than leaving a bare null. That distinction matters: a blank there means we did not examine the question, not that the event had no stage and no effect. We have not gone back and inferred answers for those records, because inventing 62 judgements during a schema change would be new analysis wearing the costume of a migration.

Our primary sources

We draw on Open-Meteo for weather; the EIA API v2 for U.S. demand across ERCOT, CAISO, PJM, MISO, NYISO, ISO-NE, SPP and PACW; IESO for Ontario; ENTSO-E for fourteen European countries; Hydro-Québec and AESO for other Canadian provinces; AEMO for the Australian NEM; Transpower for New Zealand; and the Eskom portal for South Africa. Capacity baselines for regions without live feeds come from grid-operator annual reports, cited per region.

Known gaps

Honesty about limits is part of the methodology. ENTSO-E does not publish Swiss generation, and reports Greek thermal generation with enough lag that the real-time mix is unreliable; both appear as structural estimates. Japan's OCCTO data is restricted to licensed market participants, so no public real-time feed exists. Plant-level output for regions outside the major North American ISOs is modeled from fuel-mix percentages and must never be cited as a real plant record. We flag all of this directly in the interface rather than burying it.

If a number on PowerGridIQ ever seems surprising, check its badge first: it will tell you whether you are looking at a measurement or a model.