Urdais Interconnection Queue Analytics — 1.0.0#

Status: approved, version 1.0.0, effective 21 September 2026. Written against the canonical evidence base built by IQ-2 through IQ-4, covering all seven organized U.S. wholesale markets.

This version approves a deliberately small set of metrics for publication and refuses a larger set, each with its reason. The refusals are the load-bearing part: an interconnection queue is unusually easy to describe with a number that is four times too large, and most of what follows exists to prevent that.

This document follows the principles of the Urdais methodology framework. It defines which interconnection queue analytics are defensible, for which markets, from which fields — and, at least as importantly, which are not.

What the evidence actually supports#

Every decision below was made against the seven ingested datasets, not against a desired chart. The single most important measurement is this one:

No quantity kind exists in all seven markets. The best-covered pair reaches four markets, and one of those four cannot be published at all. That single fact decides the shape of this methodology.

1. The active lifecycle#

A request is active when its latest observation is in one of:

requested · study · agreement_pending · agreement_executed · under_construction · suspended

A request is terminal when its latest observation is operational or withdrawn.

Two stages need saying explicitly.

suspended is active. PJM publishes 140 suspended requests and suspension is routinely lifted; a suspended project has not left the queue. It is included in stock and flagged, never silently.

unknown is excluded from every published metric, and is not counted as active. IQ-2 made unknown non-terminal so that a vocabulary gap could not silently shrink a queue; that was correct for storage. For publication the opposite risk applies — counting a request whose state nobody knows would silently inflate a queue — so unknown is reported as an explicit coverage figure beside any stock metric rather than folded into it. MISO's 74 Done-with-no-outcome requests and ERCOT's small-generator sheet are the live cases.

agreement_executed and under_construction are active, not complete. PJM and MISO both publish projects that hold an executed agreement for years, and construction is not operation.

2. Request subtype eligibility#

Hard rule: capacity-rights activity is never new-generation queue capacity. ISO-NE's 662 capacity-rights requests carry 171,358 MW of summer capability against 50,260 MW on requests that actually propose new plant. Including them overstates the New England queue by 4.4×. The database already refuses to put a new-generation quantity on such a request; this methodology adds that no metric may reintroduce them by reading other_mw.

3. Project counts#

The unit is the canonical stable request — one row of pipeline.interconnection_requests — never a raw row, a source row or a queue position.

ISO-NE's queue position is not unique: 92 positions carry more than one row and no column makes them unique. Those 242 rows are retained as evidence and excluded from project counts, reported as excluded evidence rather than as projects. MISO contributes one such collision and NYISO two.

A project with several resources is one project. CAISO has 500 such, PJM 578, MISO 336, ISO-NE 157.

4. MW: no seven-market total#

There is no defensible seven-market active_queue_mw, and none may be published. The reasons are not stylistic:

• PJM publishes four MW columns aggregating to 1,628 GW, 883 GW, 638 GW and 91 GW. Choosing one is a methodology decision, not a reading.
• SPP publishes six spanning 4.2×, and cannot be published at all.
• CAISO's component MW are non-additive: 937 observations have components summing above the project figure.
• ERCOT's single column is a net-change basis for repowering, carries negative values, and is therefore not semantically the same quantity as PJM's MaximumFacilityOutput even though both normalize to maximum_facility_output. A shared quantity kind groups; it does not certify equivalence.
• ISO-NE requires the capacity-rights exclusion before any MW is meaningful.
• NYISO's load MW is a withdrawal from the grid and belongs to a different metric family entirely.

What replaces it: market-specific MW, each naming its field, and project counts as the cross-market comparable.

Market MW decisions#

PJM is deferred deliberately. MWEnergy and MWCapacity are two different service rights and MaximumFacilityOutput is neither; nothing in the source says which one a reader means by "the queue". Choosing the largest because it is largest is exactly the error this methodology exists to prevent. V1 therefore ships no PJM active-MW figure at all, which costs PJM one metric and costs the product nothing it could have stated honestly.

ERCOT is deferred because a net change is not a level. A repowering that reduces output by 53.3 MW is a real published value, and summing it with additions produces a number that is neither the queue nor the change in it. V1 ships no ERCOT active-MW figure.

5. Generation and load never mix#

Two separate metric families, permanently.

Generation and storage queue: request classes generation, storage, mixed. Load interconnection queue: request class load.

A load MW is a withdrawal from the grid. It may never enter a generation total, in any market, now or when other markets begin publishing load queues. NYISO is currently the only market that publishes one.

6. AI and data-centre load#

explicit_ai_data_center_load_requests and explicit_ai_data_center_load_mw, NYISO only, market-specific.

Derived only from the publisher's own end-use code. NYISO assigns DAT-AI to 12 requests and DAT/DAT-CM to 28 more. No inference is permitted from project names, company names, location, size or any model. If a market does not publish an end-use classification, its AI data-centre load is not zero — it is unknown, and must be presented that way.

7. Queue age#

observation_date − requested_on, on active new-capability generation/storage requests with a published request date.

Published as median, p75 and p90. The distribution is skewed — CAISO's p90 is 15.5 years against a 6.5-year median — so a mean would be misleading and is not published.

ERCOT is excluded: it publishes no request date at all — 0 of 3,325 requests. This is not a coverage gap to be filled; the GIS report does not contain the field.

ISO-NE's 28 active requests fall below the percentile sample floor in §12; its median is publishable, its p90 is not.

8. Entries: two bases, never mixed#

Source-reported application date — the request date the publisher states. Available for PJM, MISO, CAISO, NYISO, ISO-NE and SPP.

Snapshot first-seen — the first published vintage in which Urdais observed the request. Available for ERCOT alone, which is the only market with a published archive.

These are different measurements and must be labelled as such. Source-reported takes precedence wherever it exists. ERCOT uses first-seen, with two constraints: the series begins 2019-01, and the 366 requests already present in the December 2018 report are left-censored and excluded from entry counts. 2,959 of 3,325 ERCOT requests have a usable first-seen entry.

Source-reported history is not observed history. Today's PJM feed saying a project withdrew in 2019 is PJM's statement, not evidence that Urdais watched it happen.

9. Exits#

Three distinct things, never conflated:

Source-reported withdrawal — the publisher says withdrawn. PJM 4,921, MISO 2,152, ISO-NE 1,170, NYISO 1,749, CAISO 1,764.

Source-reported operation — the publisher says in service. PJM 1,259, ISO-NE 280, SPP 336, CAISO 252, NYISO 153, MISO 92.

Observed disappearance — a request stops appearing in a later vintage. This is not an exit. It is not a withdrawal, not a cancellation and not an operation. ERCOT is the only market where disappearance is observable, and ERCOT publishes neither withdrawals nor operations, so ERCOT contributes no exit metric of any kind.

Withdrawal metrics#

NYISO publishes 1,749 withdrawals and zero withdrawal dates. A count is defensible; a time series is not, and no date may be invented.

10. Operational evidence#

Operational status requires the publisher to say so. Reaffirmed from IQ-2 and IQ-3:

• A proposed or projected COD is never operation. It is evidence a project is late.
• MISO's doneDate is never operation. Of 269 requests carrying one, 201 have a postGIAStatus that is not in service — 104 under construction, 62 not started, 32 withdrawn.
• NYISO status 13 "In Service for Test" is never commercial operation, nor is 15 "Partial In-Service". Only 14 "In Service Commercial" is.
• ERCOT's energization and synchronization approvals are milestones, not operation. ERCOT publishes no actual COD, so no ERCOT completion metric exists.

11. Completion: cohorts only#

A completion rate is never operational ÷ currently active. That ratio moves when withdrawals clear the denominator and says nothing about outcomes.

Definition. Of eligible requests whose application date falls in cohort period C, the share whose latest observation is operational.

Denominator — requests entering in C that are: new-capability subtype; request class generation, storage or mixed; with a published application date; with canonical identity. Excluded and reported: load requests, capacity-rights requests, transmission requests, unknown-stage requests, and unresolved identity collisions.

The maturity rule#

A cohort is mature, and publishable, only when both hold:

1. the observation window is at least that market's own observed p90 time to operation, and
2. the cohort's unresolved share is at most 15%.

Both conditions are necessary, and each binds in a different place:

The rule's necessity is visible in the data it rejects. PJM's 2021 cohort has 1,328 entrants and 0% operational with 79.7% unresolved; published as a completion rate it would read as total failure. CAISO's 2022 cohort has 0% unresolved and would pass condition (2) — on a sample of four.

Mature-cohort results (the published figures):

An immature cohort is never published as final. It may be shown as in progress with its unresolved share stated, or not at all.

Project completion and MW completion are different metrics#

Project completion rate = operational requests ÷ eligible cohort requests. Defensible for PJM and CAISO publicly; SPP internally.

MW completion rate = operational eligible MW ÷ eligible cohort MW. It additionally requires a defensible cohort quantity field, which PJM does not currently have (§4). MW completion is deferred for every market, and must not be derived by substituting a field for convenience.

MISO, ISO-NE and NYISO publish operational status but no operational date and, for MISO and ISO-NE, no reliable entry-to-exit pairing. They support an operational count, not a completion rate.

12. Time to operation#

actual_in_service_on − requested_on, requiring both explicit dates. Published as median and, where the sample floor allows, p75 and p90.

No substitution is permitted: not a proposed COD, not a study completion, not an agreement execution date.

Time to withdrawal — withdrawn_on − requested_on — is a separate metric on the same rules, and is not mixed into time to operation.

13. Technology mix#

Count-based, multi-label. A project is counted once and tagged with every technology its resources name. CAISO's explicit components identify composition; PJM, MISO, ISO-NE and SPP publish compound labels with no recoverable split.

A hybrid is never two projects. Shares are therefore stated as "share of projects including technology X", which sums above 100% and is labelled as doing so, rather than as a pie chart that implies exclusivity.

MW-based technology mix is deferred. It requires attributing project MW to components, which only CAISO publishes and which CAISO's own figures prove non-additive.

14. Comparability tiers#

Every metric carries a tier.

Tier A — same semantic metric across all publishable markets. Active request count; entries by count; withdrawal count; technology mix by count; queue age (six markets, ERCOT structurally excluded).

Tier B — comparable across a documented subset. Project completion rate (PJM, CAISO); time to operation (PJM, CAISO); withdrawal flow by date (PJM, MISO, CAISO, ISO-NE).

Tier C — market-specific only, never compared. Every MW figure. AI data-centre load (NYISO). ERCOT entries by first-seen.

A Tier C metric may be displayed beside another market's only with its field named and its basis stated.

15. Minimum samples#

Below the floor the value is not published and not rounded away — the metric reports insufficient_sample with the actual n. ISO-NE's 28 active requests fall below the median floor for queue age; its percentiles are unavailable.

16. Aggregation cadence#

Annual for entries, exits and cohorts. Point-in-time for stock, stamped with the observation date.

Monthly is defensible only for ERCOT, the one market with monthly vintages, and only for stock. Nothing is interpolated: a market with one observed snapshot has one point, not a line. Monthly snapshots are never expanded into daily history.

17. Revisions and vintages#

Analytics use the current canonical revision of a historical source period, and retain provenance to the evidence it replaced.

ERCOT republished six corrections. Where a correction exists for a report period, its content is canonical for that period; the original snapshot and all its raw records are retained and remain queryable. A restated MW or status produces a new observation with a higher ordinal, and the superseded observation keeps its own values and its own frozen evidence.

A metric recomputed after a correction may legitimately change. It must not change silently: every published figure carries the snapshot set it was computed from.

18. Provenance#

Every derived value must be reproducible to canonical requests, their observations, the quantities and resources on those observations, the raw records behind them, and the source snapshots those came from. A metric that cannot name its snapshot set is not publishable.

19. Rights#

Publication eligibility comes from the existing rights machinery — mayPublishSourceValue — and not from any parallel logic here.

SPP is blocked from every public metric, raw and derived. A derived metric does not launder the block. SPP may be computed internally, and any cross-market statement that silently omits SPP is a different claim from one that says so — so a published multi-market figure must name SPP as excluded and why.

20. Proposed public V1#

The smallest credible set, all Tier A or clearly labelled Tier C:

1. Active request count, per market (six markets; SPP excluded and named).
2. Active request count by technology, multi-label.
3. Median queue age, per market, ERCOT excluded with its reason stated.
4. Annual entries by count — application-date basis for five markets, first-seen for ERCOT from 2019, each labelled.
5. Annual withdrawal count for the five markets that publish withdrawals.
6. Mature-cohort project completion rate for PJM and CAISO only, with cohort years and unresolved share shown.
7. Median time to operation for PJM and CAISO only.
8. NYISO explicit AI data-centre load, requests and MW, market-specific.

Everything else waits.

21. Rejected, with reasons#

22. How the 0.1.0-draft gates were resolved#

The draft left six questions open. All six are closed here, and four of them are closed by declining to ship something rather than by finding an answer.

1. PJM's MW field. Not resolved, and therefore not shipped. MWEnergy, MWCapacity and MaximumFacilityOutput remain three different things with no source statement preferring one. PJM publishes no active-MW metric in V1, and this does not block the rest of PJM.
2. ERCOT's net-change quantity. Not resolved, and therefore not shipped. A net change is not a level. ERCOT publishes no active-MW metric in V1.
3. MW completion rate. Deferred for every market, permanently in V1. The two markets that support a project completion rate are exactly the two with no agreed MW field, so the MW version cannot be derived from the project version. It is exposed as a named metric with status methodology_deferred, never as a blank or a zero.
4. ISO-NE's sample size. Resolved by the existing sample floors: 28 active requests is below the median floor of 30, so ISO-NE's queue age reports insufficient_sample with its actual n. This is the designed behaviour, not a gap.
5. Coverage for unknown-stage requests. Resolved: every published metric carries a coverage figure naming how many requests were excluded and why, unknown among them.
6. ERCOT first-seen entries beside application-date entries. Resolved: no. They measure different things and never share a series. Every entry figure carries its basis, and a reader is told which one they are looking at.

23. Known limitations#

• No cross-market MW figure of any kind exists, and none can be constructed from these seven sources. Project counts are the cross-market comparable.
• ERCOT contributes no age, exit or completion metric. It publishes no request date and no actual commercial operation date.
• SPP contributes nothing public. Any multi-market statement names it as excluded.
• ISO-NE's queue position is not unique. 242 rows have no canonical identity and are excluded from counts as stated evidence.
• Completion rates describe mature cohorts only, which means they describe projects that entered years ago. They are not a forecast of what today's entrants will do.

Version history#

1.0.0, 21 September 2026. Approved. Ships project completion, time to operation, active counts, technology mix, queue age, entries, withdrawals, four market-specific MW metrics and the NYISO AI data-centre load. Defers MW completion for every market, and declines to ship a PJM or ERCOT active-MW figure rather than choose a field the source does not prefer.

0.1.0-draft, 21 September 2026. First specification, written against the complete seven-market canonical evidence base. Approved nothing for publication and left six gates open; superseded by 1.0.0, which closes all six. Nothing was ever published under the draft.