Grid Alpha

PJM Western Hub real-time printed $445.58/MWh in the 18:55 UTC interval on July 21 and is trading at $32.13/MWh as of this morning's feed pull, a collapse of roughly $413. The 24-hour mean over the snapshot window sits at $85.73/MWh, which puts the peak at roughly five times the average and marks it as a single-interval event, not a sustained scarcity episode.

The overnight shape reinforces that read. The series bottomed at $25.14/MWh at 07:30 UTC on July 22, meaning the hub round-tripped the entire spike inside one session: late-afternoon print near $446, sub-$26 by early morning, low $30s now. Nothing in the public record identifies the driver. No unit trip, load event, or binding constraint has been confirmed, which leaves the spike unattributed heading into today's evening peak. The outage tape offers a candidate but not a conviction: PJM's daily outage records logged at least 147 events over the past seven days, and sampled rows from July 16 show forced outages as high as 14,462 MW in a single record. That feed is noisy, possibly truncated, and the sampled records predate the spike, so it reads as an elevated-churn backdrop rather than a smoking gun. What it does establish is that the fleet has been carrying meaningful forced-outage MW into a summer week, which lowers the threshold for any incremental stress on the ramp to clear at scarcity-adjacent prices.

The tradeable question is whether tonight's ramp reprices. If the late-afternoon intervals again clear at multiples of the $86 mean without a named driver, the market is telling you something structural is binding on the evening ramp: an unattributed spike that repeats stops being an anomaly and starts being a feature of the hub's peak-hour supply stack. If tonight clears near the mean, July 21 was a one-off and the $32 tape is the honest price. Two data points would settle this faster than the LMP series alone, and neither is in front of us yet: the day-ahead versus real-time spread at Western Hub for July 21 and 22, which would show whether the day-ahead market saw any of this coming, and today's forced-outage MW into the peak, which would show whether the outage backdrop is building or bleeding off. Absent those, the honest position is agnosticism on the driver and full attention on the 18:00-to-19:00 UTC window tonight. A spike that recurs at the same hour, at the same hub, two days running is a pattern with a mechanism behind it, even if the mechanism has not been named. A spike that does not recur is a footnote, and the $85.73 mean already tells you how the settlement math absorbed it: one violent interval, priced and forgotten by the next morning's low.

> A $413 round trip with no named driver is not a signal yet; a second one at tonight's peak would be.

Not investment advice. For informational purposes only.

Grid Alpha — daily briefs · live dashboards, all nine markets · public call record

The New York City hub printed $927.26/MWh in real time at 22:55Z on July 20, against a 24-hour mean of $71.15: roughly a 13x gap between the peak and the average. The market has already handed it back. The current print sits at $48.36, about 95% below the top.

The same series shows a $32.34 low at 15:35Z, mid-afternoon. The hub traversed essentially its entire 24-hour range between the afternoon trough and the evening peak, then gave it all back into the overnight.

The intraday shape describes a market that was cheap almost everywhere except where it mattered. The mean held near $71.15 even with the spike inside the sample, which says the elevated pricing was concentrated rather than broad; for anyone marking evening exposure off the daily average, the average is dominated by hours that look nothing like the one that mattered. What the snapshot does not carry is attribution: reserve-shortage pricing, a transmission constraint into Zone J, a forced outage, or demand-curve activation all remain live explanations, and none is confirmed. Nor does the feed settle whether the $927.26 print was a single five-minute bar or a sustained interval. Those are different animals: a one-bar excursion is noise that fast-responding assets eat, while sustained scarcity pricing implies reserves were genuinely short into the ramp. The system context leans uncomfortable either way. NYISO told stakeholders on July 20 that it ran through the Independence Day week heat wave on extremely thin reliability margins with worrying resource performance.

For tonight, the checkpoint order matters. First, the day-ahead: if the day-ahead market priced the late-evening hours anywhere near scarcity levels, the premium is already partly in the curve and the real-time upside compresses. If it cleared near the $48 tape, a repeat of the July 20 pattern lands entirely in real time, and the DART spread in the hours around yesterday's 22:55Z print becomes the cleanest expression of the setup. Second, whether the resource-performance issues NYISO flagged from early July carried into this event; recurring generator underperformance would make evening scarcity structural rather than episodic, and structural scarcity tends to get priced into the forwards instead of faded. Ancillary prices around the ramp deserve the same attention: if reserves priced up alongside energy, the shortage explanation gains weight over a transient constraint. With no load forecast or outage picture in hand for tonight, the honest posture is conditional: watch the day-ahead clears for the late evening, watch reserve pricing as the ramp builds, and read the duration of any repeat spike in the settled data as the tell on whether this is shortage pricing or a passing excursion.

> A $48.36 tape with a $927.26 memory is not calm; it is a market that prices the ramp only when the ramp arrives.

Not investment advice. For informational purposes only.

Grid Alpha — daily briefs · live dashboards, all nine markets · public call record

A -$857.00/MWh congestion component drove CAISO node AEC3001N001 to a -$802.93/MWh real-time clear in the 10:00 UTC interval, against a system energy component of just $42.07. The snapshot's reported maximum nodal spread reached $875.84/MWh, with three POD_USWND2 wind nodes marking the high side at $72.91/MWh.

The decomposition tells the story cleanly. Energy is $42.07 across the board and congestion-free nodes NARROWS6N010 and BREGGEN7N006 cleared a placid $61.16 with exactly zero congestion, so this is not a system-wide event; it is a localized transmission problem doing all the work. Four GWF Tracy nodes clustered between -$660.59 and -$660.77, every one carrying an identical -$714.00 congestion component, the signature of units sharing the same shift factors onto a single binding element. AEC300's print is deeper at -$857.00, and the feed offers no constraint name, outage notice, or plant trip to explain it; whether it sits on the same element as the Tracy cluster or a different one is simply not disclosed yet. One caveat on the headline number: the high-side wind prints carry a 10:50 UTC timestamp while the negative cluster is stamped 10:00 UTC, so confirm the spread persists within a single interval before treating it as a live arbitrage width. The trailing anomaly feed logs at least 1,668 curtailment events over seven days, but the sampled records show effectively zero curtailed MW, so that count is not yet corroborating evidence of a chronic trapped-generation problem.

The angle has it right: with no confirmed driver, the disclosure is the trade. If CAISO names the binding constraint or posts a transmission outage covering the Tracy corridor, and it carries a multi-day return-to-service, expect day-ahead congestion at these nodes to start pricing what real-time already knows; the DART congestion gap is where the basis risk lives. If instead the next few intervals clear without the -$714 and -$857 components, this was a transient dispatch artifact and nodal basis reverts toward the $61.16 congestion-free reference. What I'd watch: the constraint disclosure itself, persistence of the identical -$714.00 print across the Tracy cluster interval over interval, and whether AEC300's deeper congestion converges toward or diverges from Tracy's as the morning solar ramp builds. CRR positions sourced or sinked at these nodes mark against every interval this congestion survives.

> An -$857 congestion print with no named constraint is a setup in search of a disclosure; the basis trade begins the moment CAISO says which element is binding.

Not investment advice. For informational purposes only.

Grid Alpha — daily briefs · live dashboards, all nine markets · public call record

Ontario's real-time nodal board split by $280.89/MWh in the 04:00 UTC interval, the widest spread on the system. YOUNGDAVIDSN-LT.TTLFDRA printed $285.98 while CARIBOUFALLS-LT.AG123 in the Northwest cleared at $5.09. Strip out the components and this is almost pure congestion: the systemwide energy component sat near $37.07 at both ends.

The decomposition is unusually clean. All five of the highest nodes, Young-Davidson, Macassa, Lower Notch and Battle Mountain Gold among them, carry an identical +$239.81 congestion component, the fingerprint of a single binding constraint walling off one load pocket rather than diffuse system stress. The five lowest nodes, Caribou Falls, Whitedog, Minaki, Kenora, all print an identical -$29.30, consistent with trapped generation in the Kenora corridor. Losses are rounding error against these numbers: $9.10 at the high node, -$2.68 at the low one. Here is the puzzle. Every sampled real-time shadow-price event from the past seven days reads zero, out of at least 3,859 logged, which flatly contradicts an active binding constraint producing a $239.81 congestion adder. Either the shadow-price feed is lagging or the sample missed the binding intervals; the LMP decomposition is not ambiguous, so I trust the prices and treat the constraint feed as broken. The planned outage log shows at least 36,388 events over seven days, though that count looks inflated by duplication and tells us little about which facility, if any, is the driver.

The dek already says it: no confirmed driver, so confirmation is the trade. If a Northwest tie or transformer outage surfaces in IESO's outage reports or an operational advisory, the spread has a return-to-service date and persistence you can lean on; if nothing confirms, treat this as a single-interval dispatch artifact and expect mean reversion. Day-ahead versus real-time congestion at these nodes over the next few sessions is the cleanest persistence tell. The structural problem is the vehicle: IESO has no nodal FTR market comparable to the US ISOs, so this basis is largely observable but not directly tradeable, which matters most for anyone with physical assets or contracts settled at Northwest nodes now wearing a -$29.30 congestion component with no obvious hedge.

> When five nodes share one congestion number and the shadow-price feed reads zero, trust the LMPs and go find the outage.

Not investment advice. For informational purposes only.

Yesterday's tape: PJM WESTERN HUB lmp_peak <= 45.23 — not triggered (observed 437.44).

Grid Alpha — daily briefs · live dashboards, all nine markets · public call record

PJM Western Hub real-time LMP printed $376.62/MWh at 20:30 UTC yesterday, the peak of a 24-hour window whose mean sat at just $64.31. This morning the hub is trading $45.23, below its own average, and the entire scarcity premium is gone.

The shape of the series matters as much as the levels. The $376.62 print landed at the very end of the reported window, 4:30 pm Eastern, squarely in the evening ramp, across a series of 327 five-minute bars that otherwise behaved. What drove it remains unconfirmed: the feed carries no constraint tag, no reserve-shortage flag, and no coincident load figure, so the honest read is a ramp-hour excursion of unknown parentage. The outage backdrop offers circumstantial support without proving anything. PJM logged at least 147 daily-outage anomaly events over the past seven days, and the sampled rows from 2026-07-12 show forced outages running as high as 15,702 MW alongside 4,478 MW of maintenance outages. But those samples all share a single timestamp, which smells like revisions rather than five distinct snapshots, and the freshest outage data is six days stale. One more data hygiene note: the reported 24-hour low of $26.52 carries a 2026-07-18T08:00Z timestamp that falls outside the stated series window ending 2026-07-17T20:30Z, so treat the low print, and the feed generally, with a raised eyebrow.

The policy tape isn't helping the calm-market case. FERC Chairman Swett said yesterday that PJM's capacity auction results compound "alarm bells," at the same meeting where the commission set data center reliability standard deadlines. That's not a driver of yesterday's spike, but it's the backdrop against which every PJM scarcity print now gets read: a market where the regulator is publicly worried about adequacy is a market where single-hour excursions get taken seriously rather than dismissed.

The clean test of the user's setup is the day-ahead versus real-time spread at Western Hub through today's evening ramp, and that number isn't in the source material, so it's the first thing to pull. If day-ahead cleared near the $64.31 mean and real time again gaps toward triple digits in the ramp hours, yesterday reads as the market discovering a structural tightness it hadn't priced, and DART sellers wear it twice. If the evening hours clear near the current $45.23, the spike was a one-off and the fade is complete. Secondary watch items: whether PJM posts fresh forced-outage figures that update the stale 2026-07-12 sample, and whether any binding-constraint data emerges to explain the 20:30 UTC print, since a transmission-driven spike and a reserve-driven spike carry very different odds of repeating tonight.

> A $377 print that retraces to $45 inside a day isn't a trend; it's a question, and tonight's ramp gets to answer it.

Not investment advice. For informational purposes only.

Grid Alpha — daily briefs · live dashboards, all nine markets · public call record

A $218.73/MWh gap opened across CAISO's real-time footprint at the 10:15 interval this morning, and almost none of it was energy. ELCAJNGT7N011 cleared at $168.73/MWh while DYERSM37N001 printed minus $50.00 in the same snapshot: the widest node-to-node dislocation on the morning tape.

Strip both LMPs to their parts and the split is congestion, not scarcity. Each node carried the identical $29.22/MWh energy component, so the systemwide price was flat between them; the divergence sits in the congestion and loss terms alone. ELCAJNGT booked $125.59/MWh of congestion against a $2.16 loss adder. DYERSM3 ran the mirror sign on price but its own separate math: $-87.37/MWh of congestion and a $-3.61 loss. Two nodes, one common energy base, and a spread that is a transmission-binding signature rather than a heat-driven shortage. The binding element behind the print is not identified on the tape, so what to price here is a constraint's shadow value, not a fuel-stack scarcity.

The week's curtailment tally is loud but mute on this specific print. CAISO's anomaly feed shows at least 1,848 curtailment events over the trailing seven days, but that is a raw count with no fuel-type or locational tie to ELCAJNGT or DYERSM3 and no established causal link to this constraint. Read it as regime backdrop, not as the driver of the split.

The tape gives one 10:15 quote and no duration, so the first question is persistence: does the congestion adder survive the interval or was it a single-print artifact. What I'd watch is whether that $125.59 component holds into the evening ramp, because the residual exposure the angle flags sits in peak-hour length rather than in the systemwide scarcity that has already softened to weak morning numbers. Watch the ELCAJNGT congestion term against the shared $29.22 energy base rather than the headline LMP: the LMP will drift with system energy, but the tradable dislocation is the constraint. If ELCAJNGT's congestion stays bid as the ramp tightens, ELCAJNGT-DYER remains a congestion trade to manage; if it decays back toward that energy floor, the dislocation resolves as noise. There is no day-ahead comparison on the tape to say whether the morning gap was hedgeable, which is itself the reason to treat the evening window as live rather than settled.

Underneath the intraday mechanics, FERC on July 17 ordered CAISO and SPP to report on the seams expected between EDAM and Markets+, a reminder that the region's congestion geometry is being contested at the seam, not just at the node.

> Energy went flat and the tape still split $219: the risk here is the wire, not the weather.

Not investment advice. For informational purposes only.

Yesterday's tape: MISO MICHIGAN.HUB lmp_peak >= 1042 — not triggered (observed 825.1).

Grid Alpha — daily briefs · live dashboards, all nine markets · public call record

Michigan Hub touched $1,042/MWh yesterday against a $238 24-hour mean as MISO demand hit 122 GW, enough to push the RTO into a maximum generation emergency. This is the second heat dome to park over MISO Midwest this month, and unlike June's fast-moving events, this one is dayslong and stationary. The $1,042 print is worth confirming against final settlement data, but the emergency declaration is not in dispute.

The scarcity is a supply story as much as a load story. One reading from MISO's estimated outage feed last week showed 16,405 MW forced out, another 11,425 MW derated, and 6,031 MW on planned outage; the feed contains duplicate rows, so treat the exact totals as directional, but the shape is clear: a large slice of the fleet was already unavailable before this dome arrived. Sustained heat compounds that through thermal derates, and the transmission system is showing the strain in parallel. The real-time constraint feed logged at least 6,386 shadow-price events over the past seven days, with flowgates like Clay-Starkville 161 kV in MISO South binding at consecutive 5-minute intervals. When a constraint binds every interval rather than flickering, the congestion is structural for the duration of the weather, not a dispatch artifact.

Three things determine whether the balance-of-week stays bid. First, forced-outage recovery: if a meaningful share of that 16.4 GW is heat-driven derates rather than hard trips, capacity returns the moment overnight temperatures break, and DART spreads compress fast. If it is unit trips with multi-day return timelines, real-time keeps printing through the day-ahead curve every afternoon. Second, day-ahead peak forecasts against the 122 GW realized number: any forecast at or above it means MISO is planning to operate in emergency territory again, and the DA clear will carry scarcity premium before real-time even opens. Third, import headroom. PJM, the natural emergency supplier to MISO's eastern seam, just cleared its capacity auction at the price cap with a growing reserve shortfall, which is a reminder that the neighbor's surplus is thinner than the interchange history suggests. Two tight pools sharing one heat dome do not bail each other out; they compete for the same marginal megawatt.

The asymmetry favors staying attentive rather than fading the move. A single $1,000-plus print in isolation is a mean-reversion setup; a $1,000-plus print during a declared maximum generation emergency, with the dome forecast to sit, is the market telling you the tail is live until the weather says otherwise. Watch the emergency declarations themselves: a step down from max gen to a conservative operations posture is the earliest tradeable signal that the scarcity premium should come out of balance-of-week.

> When the second heat dome of July finds 16 GW of the fleet already forced out, $1,042 is not an anomaly; it is the price of scarcity the market was built to reveal.

Not investment advice. For informational purposes only.

Grid Alpha — daily briefs · live dashboards, all nine markets · public call record

PJM logged 16,648 MW of forced outages in a single snapshot on July 9, with another 4,620 MW parked in maintenance. Strip that much iron off a summer stack and the evening ramp gets thin quickly: it is the physical hole sitting behind the peak print the dek flags. The realized RT figure itself is outside my verified feed, so read the price as the symptom and the outage stack as the cause.

The scarcity here is structural, not a one-session accident. PJM's 2028/29 Base Residual Auction cleared 6,831 MW short of the reliability requirement and pinned the $325/MW-day collar cap, the market's blunt statement that capacity is short and already priced to the ceiling three delivery years out. Demand is chasing that shortfall rather than easing it: the Database of Emerging Large Load Tariffs now counts 104 approved or proposed tariffs across 37 states, a load signal that does not soften into the shoulder months. The anomaly feed flagged at least 144 outage events over the last seven days, and that count reads like a truncated sample rather than the full tally. It was forced outages, not planned ones, that drove the July 9 reading: forced is the category that clusters when heat leans on aging thermal units, and it is the category that does not give you a schedule.

Watch the forced-outage line into the afternoon. If it holds near last week's peak while load climbs, the reserve margin compresses into the 4-to-8 PM block and the DART spread widens on any incremental trip. The honest caveat is that the 16,648 MW snapshot is five days stale, and the feed carries no current-session load forecast or reserve-margin figure; treat the repeat thesis as conditional on outages staying elevated, not as confirmed for today. If units return to service and load eases off the peak, the evening tail flattens and the spike does not recur. The asymmetry is what makes it worth watching: the downside to being wrong on a quiet evening is small, the upside to being right on a tight one is another four-figure print.

> A market that clears its capacity auction at the cap with a 6.8 GW hole has no cushion left for the evening ramp.

Not investment advice. For informational purposes only.

Grid Alpha — daily briefs · live dashboards, all nine markets · public call record

CAISO's real-time market showed a $484.72/MWh nodal spread this morning, with TORTILLA1N007 clearing at $346.62/MWh while HPLNDJT6N001 printed negative $138.10. One caveat before anyone anchors on the headline number: the high-side nodes are stamped 09:45 UTC and the low-side nodes 10:15 UTC. The spread mixes two intervals, so the true simultaneous number is unverified even if the directional story is intact.

The decomposition is where the signal lives. TORTILLA's $346.62 is almost entirely congestion: $298.90 of it, against an energy component of just $31.56 and losses of $2.79. All five top nodes, three TORTILLA points plus two LNGBATGN points, printed identical LMPs and identical component splits to the penny. That is the signature of a single binding constraint with matched shift factors, not five independent scarcity events. The mirror image sits 30 minutes later: HPLNDJT went to negative $138.10 on a negative $182.80 congestion component despite positive $34.77 energy, and four more nodes clustered behind it in a graduated sequence, GRANITE at negative $111.35, two CLERLKE points at negative $101.80, and HARTLEY at negative $96.08, all on the same $34.77 energy component. That pattern reads as a group of nodes trapped in an export-constrained pocket where the marginal MW carries negative system value. The 7-day curtailment log shows at least 1,668 CAISO events, but the sampled entries are near zero; the largest shown is 3.208 MW of wind. The count may be truncated by pagination, so the curtailment data neither confirms nor rules out bulk oversupply behind the pocket.

What is missing is the driver. No constraint name, no outage filing, no de-energization report appears in the source material, and the angle here is the right discipline: this is a flag to investigate, not yet a basis trade to size. What I would watch, in order: whether CAISO's binding constraint data names the facility behind the $298.90 positive component; whether the negative cluster is geographically contiguous, which would confirm a single pocket rather than coincidence; and whether day-ahead LMPs across these node pairs show the same split. If DA did not price it, DART on these nodes was violent and the congestion is either a forced outage or a real-time modeling artifact. If the pattern persists into tomorrow's day-ahead run, CRR and FTR paths spanning the split reprice, and the spread stops being an intraday curiosity and becomes a structural position.

The identical component stacks on both sides of the split are the tell. When five nodes clear at exactly $346.62 with exactly $298.90 of congestion, and five more sink on the same $34.77 energy print, the market is pricing one piece of transmission, and everything else is noise around it. Until that piece has a name and a duration, the honest trade is the homework: pull the constraint list, check the outage cards, and verify whether the two timestamps ever overlapped at full width.

> A $485 spread built almost entirely of congestion is a transmission story, not an energy story; until the constraint has a name, it is a signal to investigate, not a position to size.

Not investment advice. For informational purposes only.

Grid Alpha — daily briefs · live dashboards, all nine markets · public call record

SPP's real-time board printed a $729.84/MWh maximum nodal spread at 10:50 UTC, all of it inside the WAUE footprint. WAUE.BEPM.MADISON cleared at $621.13/MWh while WAUE.BEPM.GARVINWH settled at negative $108.71, against a system energy component of just $17.01/MWh. On a mild-load Sunday morning shoulder, that is not a scarcity print; it is a wires print.

The decomposition makes it unambiguous. Congestion contributes $604.34 of MADISON's price and negative $124.57 of GARVINWH's; losses are noise at under $1.30 everywhere on the list. The premium side is remarkably isolated: the next-highest nodes, WAUE.BEPM.SMS1 and SMS2, sit at $64.82/MWh with only $48.03 of congestion, a nearly tenfold drop from MADISON. That shape says one binding element is pinning a single delivery point, not a broad regional constraint. The discount side is wider: WAUE.MRES.WTRN prints negative $61.37 on negative $77.25 congestion, and at least five WAUE nodes carry negative congestion components. That pattern reads as trapped generation on the export side of the constraint, likely wind given the node names, though the feed carries no generation data to confirm it. The source names no flowgate. The ISO-wide outage feed shows sampled coal outage readings between 1,293 and 3,312 MW on July 5, but it is fuel-level only, a week stale, and cannot be tied to WAUE or to any transmission element. The 1,943 VER curtailment events logged over the past seven days look like routine interval records; the sampled solar rows are all zeros. So the mechanism is visible in the price decomposition, but the physical cause is not yet in any public feed here.

Persistence is the first question worth answering. A single 5-minute print at this magnitude is a data point; three consecutive hours is a structural setup, and the day-ahead prices for MADISON and GARVINWH tell you which one this is. If day-ahead did not carry this congestion, MADISON is running an enormous real-time premium to its DA clear, and the DART behavior over the next few days shows whether the market learns the constraint or keeps missing it. If the negative pocket at GARVINWH and WTRN deepens into higher-wind hours tonight, that supports the trapped-generation read and points at whatever path connects those nodes to MADISON as the element to watch. TCR positions spanning that cut are the natural expression once SPP publishes a constraint name; until then the setup is real but blind, which is exactly the risk the angle flags.

> A $730 spread on $17 energy is pure transmission: until SPP names the constraint, the information is in the persistence, not the print.

Not investment advice. For informational purposes only.

Grid Alpha — daily briefs · live dashboards, all nine markets · public call record