Grid Alpha

$202.15/MWh is where SP15 real-time topped out in the trailing 24 hours, on a single peak timestamp of 05:25 UTC, or 22:25 PT Saturday. The window's mean was $44.54/MWh and its low was $8.87/MWh, printed at 08:55 PT Saturday. By 04:00 PT Sunday, the last bar in the series, the hub was printing $54.04/MWh.

The series runs 288 five-minute bars from 04:05 PT Saturday to 04:00 PT Sunday. Its shape is a single-digit morning floor, a spike late Saturday evening, and an overnight Sunday print of $54.04 that sits above the 24h mean and well below the peak. At the 03:55 PT Sunday snapshot, CAISO load was 22,732 MW and net load 20,337 MW, with natural gas the main source. NP15 printed $40.72/MWh at 04:00 PT Sunday, so the north was clearing below the south at the end of the window. Three constraints appear in the congestion feed, GWANETITC, AP1 FC-MK1 500 and SHERIDANYELLOWTA_230, but none carries a reported shadow price, so how much of the SP15 premium over NP15 is congestion cannot be read from the data. The anomaly feed logged at least 1,668 curtailment events over the last seven days, though the count arrives with a five-row sample marker and may be truncated; the sampled rows all sit at 05:00 PT on 2026-09-20 and show wind curtailment as high as 39.73 MW while the solar rows read zero.

What the data does not say matters as much. No source names a driver for the $202.15 bar: outage, constraint, gas, load forecast miss, ramp shortfall, none is confirmed. The series reports a peak value and a timestamp, not whether it held for one bar or several. There is no day-ahead SP15 or NP15 price for Sunday, so the DART implied by the spike is not computable, and there is no SoCal Citygate print to cost the marginal gas unit under $54.04. The angle reads the spike as the evening ramp scarcity window pricing a fat tail; the timestamp places it at 22:25 PT, and whether that slot belongs to the ramp or to something else is the open question, not a settled one.

What I'd watch. If the THSP15GEN-APND five-minute LMP exceeds $54.04/MWh at 22:25 PT on 2026-09-27, the same clock slot as Saturday's peak, the tail is repeating and the one-off reading loses. If that bar sits at or below the $44.54/MWh trailing mean, Saturday's print looks idiosyncratic and the fat-tail case needs a named driver before it gets more weight. Watch SP15 against NP15 at the 04:00 PT bar on 2026-09-28: a gap wider than Sunday's $54.04 versus $40.72, with the three listed constraints still showing no shadow price, is a congestion story hiding behind a missing data field. A low on 2026-09-27 beneath $8.87/MWh at 08:55 PT would say the daytime floor is intact, which is the other leg of the intraday range the angle is pricing.

> One $202 bar with no named driver is a tail to respect, not a regime to assume; the 22:25 PT slot on Sunday decides which it was.

Not investment advice. For informational purposes only.

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

PJM's real-time nodal stack split by $1,078.89/MWh in the 10:50 UTC interval, 06:50 EDT on a Saturday morning, with the PJM-RTO hub at $30.88/MWh. Five buses at the MORRISPA35 KV substation cleared at $759.90/MWh while both GLENDON 34.5 KV transformer banks printed $-318.99/MWh. This is a congestion event, not an energy event, and the decomposition proves it.

The energy component is identical at $30.77/MWh on both legs. Morrispa carries $728.99/MWh of congestion on a $0.14 loss; Glendon carries $-348.92/MWh of congestion and $-0.84 of losses. Every Morrispa bus (WASHBNSP, WASHB2SP, RICHLISP, HOLLNDSP, HIGHSTRT) prints the same $759.90, which is what a group of nodes with a shared shift factor on one side of a single binding element looks like. The negative tail steps down through AMDEASTN115 KV CUSTLD at $-211.13/MWh and the two BELFAST 115 KV banks at $-186.89/MWh, a shift-factor gradient off a common constraint rather than a scatter of unrelated buses. PJM's top-congestion list names two 115 kV elements touching Glendon, GLENDON-NORW and GILBERT-GLEN, plus a Lake George - Tower Road 138 kV constraint under a Lake George - Babcock 345 kV contingency, but all three carry "n/a" shadow prices, so the binding element and its cost per MW of relief cannot be confirmed from the feed.

System conditions do not explain it. Load was 73,995.82 MW and net load 70,931 MW at 11:00 UTC with nuclear the main supply source, a flat Saturday morning by any measure. The outage feed shows at least 144 daily events over the past week, and a 2026-09-20 sample with forced outages between 2,130 MW and 4,662 MW and planned outages of roughly 10,934 to 11,507 MW, but nothing dated to today. The 48-hour news wire has no mention of Glendon, Morrispa, or any PJM congestion event. The driver is unknown, and that is the point: a $1,079 split with no posted shadow price is a local topology problem the market has priced before the operator has labeled it.

With the cause unconfirmed, the question is persistence, and the angle already frames both legs as watch items. If the Morrispa congestion component still exceeds $500/MWh at 09:00 ET, 13:00 UTC, the constraint has survived two hours of the morning ramp and is not a single-interval dispatch artifact; if it has fallen below $100/MWh by then, treat the 10:50 UTC print as transient. On the other leg, if Glendon congestion remains below $-200/MWh at 09:00 ET while the RTO hub stays under $40/MWh, generation behind the constraint is being paid to back down against a flat system price, and the DA/RT basis on any Glendon-side exposure is the number to size. PJM's constraint table is the tiebreaker: a posted shadow price on GLENDON-NORW or GILBERT-GLEN by the 12:00 ET refresh, 16:00 UTC, names the element; a third consecutive "n/a" past that hour means the cause is still unresolved and the spread can reappear in any interval.

> A $1,079 spread over a $30.88 hub is the grid telling you exactly where the constraint sits, before PJM says which element is binding.

Not investment advice. For informational purposes only.

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

An $800.95/MWh nodal spread opened across the CAISO grid in the 10:50 UTC real-time interval, 03:50 PT on a Friday overnight. PODCHWCHL1UNIT-APND printed $266.65/MWh at the top of the stack; CSADIAB7_N001 printed -$534.30/MWh at the bottom. The NP15 hub sat at $39.94/MWh at 04:00 PT, untouched by either extreme.

The decomposition leaves no ambiguity about what moved. Energy at the high node was $26.49/MWh and $25.16/MWh at the low node, a near-identical marginal energy price; the congestion component was $224.41 at CHWCHL and -$561.87 at CSADIAB, and that difference is the spread. Four neighboring high-side nodes (CERTTEED1N001, CERTTEED1N002, CHWCHLA27N001, CHWCGN1B1) cleared at $260.06 to $260.08 with an identical $218.76 congestion component, while four low-side nodes (CONTROLX1N008, N009, N010 and CSADIAB7N003) cleared at -$534.20 sharing the same -$561.87. Identical congestion components across each cluster mean one shadow price is being spread across the same set of shift factors on both ends; the dispatch found a hard limit and priced it through. Two caveats. The high-node stamp is 10:50 UTC and the low-node stamp is 10:20 UTC, so the headline spread may straddle intervals rather than sit inside one. And nothing in the snapshot names the binding element: the live congestion table lists EPENETITC, TEPCBAEXPITC and LADWPNET_ITC, all with shadow price unreported, none obviously local to a Chowchilla-area pocket.

System conditions were unremarkable. Load ran 23,371 MW against 21,462 MW of net load, with natural gas the main source on the margin. That is the point: an $800.95 spread on a flat overnight system is a local event, not a system one. At -$534.30/MWh any resource behind the CSADIAB and CONTROLX nodes that stayed dispatched was paying to inject, which is curtailment by price whether or not a curtailment record follows. The anomaly feed shows at least 1,668 CAISO curtailment events over the last seven days, but the sampled rows from 2026-09-18 12:00 read zero MW for both solar and wind, so the sample cannot tie the count to these nodes.

The user's angle flags curtailment risk with no confirmed driver, so the test is persistence. If CSADIAB7N001 still prints a congestion component at or below -$561.87 in the 12:00 UTC interval (05:00 PT), the limit survived the overnight trough and the daytime solar ramp in that pocket will hit it with more injection, not less. If the CHWCHL cluster's $218.76 congestion component has instead collapsed toward the $39.94 NP15 hub print by 12:00 UTC, treat the 10:50 interval as a dispatch artifact and move on. Watch the day-ahead clears for both node groups against these real-time prints: a DART gap of this size on either side tells you whether the market saw the constraint coming or the real-time engine discovered it alone, and whether CAISO posts a shadow price on any constraint tagged to these nodes.

> A $39.94 hub and an $800.95 nodal spread in the same interval is the definition of a local problem, and until a constraint name appears the only tradable fact is whether the -$561.87 congestion component is still there at 05:00 PT.

Not investment advice. For informational purposes only.

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

At 10:55 UTC the widest real-time spread in MISO was $312.70/MWh, and none of it was energy. AMIL.PEARLG2 printed $210.52 on a congestion component of $168.18; AMMO.HANN_1.AZ printed -$102.18 on congestion of -$143.31. The energy component at both ends was $42.14.

That same $42.14 held at all ten extreme nodes, top five and bottom five, which makes the entire spread congestion plus a rounding of losses. ILLINOIS.HUB cleared $43.34 at 11:00 UTC, within a dollar of the system energy price and blind to everything happening around it. The geography is messier than a clean Illinois-versus-Missouri read: three of the five highest nodes sit in NIPSCO territory in northern Indiana (NIPS.MICHCP12 at $208.40, NIPS.MCHCP.ARR at $203.09, NIPS.BAILY.ARR at $201.97), and the second-lowest node in the footprint is also NIPSCO, MUNSTR.LN at -$77.15 on -$120.21 of congestion. On the Missouri side, AMMO.PENOCTG1, PENOCTG2 and PENOCTG3 all cleared exactly -$47.29 with congestion of -$88.11, the signature of a shared binding element. AMIL.ALSEYCTG6 filled out the high side at $188.85 on $146.58 of congestion.

What the feed will not give you is the element itself. The three constraints carried in the live snapshot are northern flowgates (Lake George, Barton Lake, Arrowhead-Iron Range) and no shadow price is reported for any of them, so the economic depth of the binding is unknown. The named constraints in the trailing seven-day sample, the Chicago_Praxair flowgate into Wilton Center and Crete-St Johns Tap 345 kV, are dated 2026-09-17, not today. That sample counts at least 7,420 real-time binding events and at least 84 generation-outage events in seven days, both flagged as truncated and therefore floors; the largest sampled outage row, 2026-09-18 at 00:00 CT, shows 12,075 MW derated against 13,621 MW forced and 18,332 MW planned, with no labeled aggregation basis. Load was 73,590 MW against 64,010 MW net, gas dominant on the margin. This is 05:55 CT on a Thursday, not a scarcity hour.

The angle reads this as an Illinois/Missouri binding constraint, and the NIPSCO nodes on both tails complicate that: either two elements are binding, or one element reaches a lot further east than the node names suggest. Three checks settle it inside today's session. If PEARLG2's congestion component is still above $100/MWh at 14:00 CT, the element survived the morning ramp and carried into peak. If ILLINOIS.HUB is still inside $5/MWh of the real-time energy component at that same 14:00 CT interval while PEARLG2 holds triple-digit congestion, the Hub is worthless as a hedge against this basis and the exposure has to be taken nodally. If NIPS.MICHCP12 and NIPS.MUNSTR.LN are still printing opposite-signed congestion at 14:00 CT, treat the Indiana split as a separate constraint from the Ameren one and stop pricing them as a single spread. The bearish version: PEARLG2 congestion under $25/MWh by 09:00 CT makes the 05:55 print an off-peak dispatch artifact with no peak-hour tail.

> A $312.70 spread built on $42.14 energy at every node is not a scarcity signal: it is a transmission element the data never names, and the Hub will not price it for you.

Not investment advice. For informational purposes only.

Yesterday's tape: NYISO LONGIL lmppeak <= 48.75 — verified (observed 35.11); NYISO LONGIL lmppeak >= 324.52 — not triggered (observed 58.18); NYISO LONGIL lmp_peak <= 34.13 — not triggered (observed 58.18).

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

Long Island printed $324.52/MWh at 17:00 EDT Tuesday and then spent the rest of the window doing almost nothing. The 23.1-hour mean through 06:55 EDT Wednesday is $34.13, the low was $15.36 at 09:50 EDT Tuesday, and the latest five-minute bar is $28.31. One interval carries the entire story; every bar around it says Long Island is a cheap zone with nothing to price.

Two caveats before the mechanism. The series is 277 five-minute bars running from 07:05 EDT Tuesday to 06:55 EDT Wednesday, so the "24h mean" is a 23.1-hour mean and the "open" is simply the last print in the feed, not a confirmed opening interval. Neither changes the shape of the day, but a spike-and-flat pattern is easy to overread when the window is short and the sample of evening ramps is exactly one.

The data is more useful for what it withholds. No source identifies what set the 17:00 EDT bar: a unit trip, a cable or interface limit into the island, a load forecast miss, or a gas supply problem are all live candidates and none is confirmed. Duration is also unreported; only the peak bar is in the feed, so whether this was a single interval or a sustained evening event is an open question. The 06:45 EDT system snapshot shows load at 14,904 MW, net load at 14,218 MW, and natural gas as the main source. The constraints NYISO lists at the top of its board are SCRIBA-VOLNEY 345, MEYER 230, and SCH-PJ-NY, all upstate or at the PJM interface, and all with no shadow price attached. Nothing on Long Island is binding this morning. That is fully consistent with a flat tape and says nothing about the evening ramp, which is when the island's import position and local units actually get tested.

The cross-zone read is the tell. N.Y.C. printed $48.75 at 07:00 EDT while LONGIL sat at $28.31, so the island is currently the cheaper of the two despite owning Tuesday's only scarcity print. If LONGIL real-time is still below the $48.75 N.Y.C. morning mark at 16:00 EDT today, the tape is treating Tuesday's bar as noise and the ramp will have to reprice on its own, interval by interval. If any five-minute LONGIL bar between 16:00 and 19:00 EDT clears $324.52, Tuesday becomes a pattern rather than a one-off and the island premium is under-marked. If instead the highest bar in that same 16:00 to 19:00 EDT window stays below the $34.13 mean, the scarcity read is dead and the spike belongs in the trip-and-recover file. Watch the constraint board at 16:00 EDT as well: a Long Island limit appearing with a real shadow price would name the cause the current feed does not.

The flat morning is not evidence against the evening. It is evidence that the morning is fine, which nobody disputed.

> One bar at $324.52 does not make a scarcity regime, but a $28.31 tape that ignores it is an implicit bet that whatever broke at 17:00 EDT Tuesday does not break again tonight.

Not investment advice. For informational purposes only.

Yesterday's tape: ERCOT ENDPARKESS1 negativelmp <= -31.65 — verified (observed -52.58); ERCOT KNAPPRN lmppeak >= 150.08 — verified (observed 814.12).

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

The ERCOT real-time LMP spread reached $557.85/MWh in the 10:50 UTC interval on Tuesday, 05:50 local time in Texas. KNAPPRN cleared at $491.88/MWh while ENDPARKESS1 settled at -$65.97/MWh. That is a pre-dawn print, before solar enters the stack, and nothing in the public feeds names the element that split the two nodes.

The shape of the tails says congestion, not system-wide scarcity. BULLCRK12 was second-highest at $299.56/MWh, LAMESASLRG and ALPBESSRN tied at $150.08/MWh, and CROSSTRLRN rounded out the top five at $108.61/MWh; the premium decays quickly away from KNAPPRN rather than lifting the whole system. On the floor, FLUVANNA12 printed the identical -$65.97/MWh as ENDPARKESS1, and a second cluster of AMADEUSALL, ANDMDSLRALL and DERMOTTALL tied at -$31.65/MWh. Identical LMPs at separate settlement points are the signature of resources sitting behind the same binding element and receiving the same shadow-price allocation. A storage node pinned negative is being paid to charge, the classic marker of a pocket that cannot export; the open question is which line. The SCED shadow-price feed for the trailing week lists three 138 kV constraints, 138ALVNAL1 with limits of 394.2 and 386.3, SWIOLIN1 at 255.8 and 265.9, and G138151 at 210.6, but the sample carries no timestamps, shadow-price values or node mappings, so none can be tied to this interval. The same feed logs at least 2,021 unplanned outage events over the week; sampled entries include WZRDESS and CFLATES storage resources dropping their full 150 MW and 100 MW from 05:01 UTC on 2026-09-18, and SWEC wind losing 42 MW of 120 MW. Whether any of those sit in the KNAPP or ENDPARK footprint is unconfirmed, and the trailing shadow-price record runs to at least 20,833 rows, so the driver is in there somewhere, just not surfaced yet.

With no owner for the constraint, the watch list is about persistence rather than direction. If ENDPARKESS1 and FLUVANNA12 still print an identical LMP below -$31.65/MWh in the 10:50 UTC interval on 2026-09-23, the pocket is structural rather than a one-interval dispatch artifact, and the DART at both nodes becomes the spread to price. If KNAPPRN clears above $150.08/MWh in 2026-09-23 HE20, the hour the GridAlpha forward model newly flagged as elevated-price risk with P75 above trailing P95, the overnight premium has carried into the evening ramp and the model's single new flag is doing real work. Watch the KNAPPRN premium over BULLCRK1_2 as the tell for whether this is one node or a wider West Texas pocket; if the gap between them narrows toward zero by the 10:50 UTC interval on 2026-09-23 while both stay above $108.61/MWh, the constraint has widened rather than cleared. The day-ahead prints at both nodes remain the missing datum, and until they surface the basis thesis stays a hypothesis with a timestamp, not a position.

> A $557.85 spread with no named constraint is not a scarcity story; it is a congestion story waiting for its line.

Not investment advice. For informational purposes only.

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

PJM Western Hub touched $341.35/MWh on the 15:15 ET five-minute bar Saturday and opened Sunday at $33.67/MWh. The 23.8-hour window mean sits at $56.98/MWh, which tells you the spike carried almost no duration: one bar far above the mean barely moved the average across 286 prints. Under three hours before the peak, at 12:25 ET, the same hub printed -$0.31/MWh.

That is a scarcity shape, not a fuel shape. A hub that swings from negative to $341.35 and back inside one afternoon is pricing a transient shortfall in dispatchable supply against a mid-afternoon ramp, and the feed gives no causal statement: no reserve-shortage flag, no unit trip, no constraint tag on the peak bar. The three constraints listed at the 07:03 ET snapshot (Lake George-Tower Road 138 kV, the NOTTINGHNOT2301-31_LN element, and Chicago-Praxair3 138 kV) all show shadow prices of n/a, so congestion can be neither confirmed nor ruled out from what is on the screen. The outage side is noisier than usual. The daily outage feed logged at least 147 events over seven days, but the five sampled rows all carry the same 2026-09-14 00:00 ET stamp with forced MW ranging from 2,437 to 4,978 and maintenance MW from 4,597 to 7,071; only the planned figure of 7,501 MW agrees across rows. Whatever the current outage stack is, the feed is not showing it cleanly.

The morning tape is unremarkable. RTO printed $52.58/MWh at 06:55 ET against Western Hub at $33.67/MWh, so the hub opened at a discount to the system rather than a premium, with load at 80,843.79 MW and net load at 78,708 MW at 07:00 ET on gas-led supply. That discount is the tell. If Saturday's spike were a persistent supply problem, Western Hub would not be leading the RTO lower on a Sunday morning; the $341.35 bar reads as a ramp-timing miss, and ramp-timing misses recur on the same schedule.

The angle's underpriced-evening-peak call cannot be checked against a forward: no day-ahead Western Hub price for Sunday's peak hours is in the pack, so treat the thesis as a real-time-versus-mean read. What I'd watch: if any Western Hub five-minute bar clears the $56.98 window mean before 15:15 ET Sunday, the ramp is repricing earlier than Saturday and the evening block is live. If the hub-to-RTO discount that stood at $33.67 against $52.58 at 06:55 ET flips to a premium by 12:25 ET, the constraint story is back in play and the n/a shadow prices deserve a second look. If the hub is still below the RTO print at 15:15 ET, Saturday was a one-bar event and the mean is the right anchor.

> A single bar at $341.35 against a $56.98 mean is not a scarcity regime; it is a scarcity timestamp, and timestamps repeat before regimes do.

Not investment advice. For informational purposes only.

Yesterday's tape: MISO NIPS.MUNSTR.LN negativelmp <= 0 — verified (observed -95.43); MISO CE.CRETE.NU negativelmp <= 0 — verified (observed -34.32).

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

$312.96/MWh separated NIPS.MICHCP12 from NIPS.MUNSTR.LN in MISO's 11:00 UTC real-time interval, 06:00 Central on a Saturday. Michigan City cleared at $234.50/MWh; Munster cleared at $-78.46/MWh. Both nodes sit inside the same NIPSCO footprint, and the energy component at each was an identical $49.24/MWh.

Strip the energy term and the entire spread is congestion. MICHCP12 carried $185.07/MWh of congestion, with NIPS.MCHCP.ARR at $178.89 and NIPS.BAILY.ARR at $179.26 clustered within a few dollars, and NIPS.MAIZEMSV at $152.78 behind them. On the other side, MUNSTR.LN carried $-127.33/MWh of congestion and NIPS.STJOHN.LN $-84.23, and the negative tail runs across the seam into ComEd: CE.CRETE.NU printed $-65.69 of congestion and CE.ZIONEC.MVP $-28.84. Loss terms are noise, $0.19 at MICHCP12 and $-0.37 at Munster. That shape, a lakeshore cluster paying to import while the Lake County nodes are paid to stop injecting, is what a single binding element across northwest Indiana looks like in the price vector.

What the data does not give is the element. The snapshot's three named constraints (MAPLE-08CHR, MTGL-WARR 3, MORO2-RZZY) are not NIPSCO facilities, and all three show shadow prices as n/a. The angle's "constraint behind Munster" is therefore inference from the congestion signs, not a flowgate ID, and nothing in the pack says whether this is one five-minute print or the tenth in a row. System load was 70,273 MW at 10:55 UTC against a net load of 65,270 MW with coal as the main source; neither figure suggests footprint-wide stress at this hour. ILLINOIS.HUB printed $31.5/MWh, below the $49.24 nodal energy component, and the pack does not resolve whether that gap is hub-level negative congestion or an interval mismatch between feeds. The most recent outage estimate, from 2026-09-13, shows forced outages of 5,223 MW, planned outages of 9,937 MW and derates between 6,081 and 6,159 MW, with no Michigan City or Bailly unit named. Real-time binding-constraint events over the last seven days number at least 11,202, a lower bound since only a five-row sample was returned.

The trade in the angle lives or dies on persistence, so the checks are simple. If MICHCP12's congestion component alone still exceeds the $49.24 energy component at 09:00 Central, the constraint has survived the morning ramp and is not a dawn dispatch artifact; if MUNSTR.LN's LMP is still below zero at the same clock time, supply on the Lake County side is still being backed down rather than cleared. Watch MAIZEMSV against DUNNBR1SF as the gradient: DUNNBR1SF carried only $68.77 of congestion versus $152.78 at MAIZEMSV, and if that gap closes by 12:00 Central the binding element is relaxing from the outside in. The Munster-to-Crete link matters for anyone holding cross-seam TCR or FTR paths; if CE.CRETE.NU is still negative at 09:00 Central, the constraint is pricing into ComEd, not just NIPSCO. The Independent Market Monitor said this week that MISO's repricing of several summer emergency hours eroded faith in the markets, which raises the bar on how any sustained intra-NIPSCO dislocation gets explained after the fact.

> A $312.96 spread on identical $49.24 energy is a congestion signature without a named constraint; until the flowgate prints, treat it as an unlabeled setup, not a thesis.

Not investment advice. For informational purposes only.

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

Clifty Creek 1 cleared at -$391.60/MWh in PJM's 10:40 UTC real-time interval, 06:40 ET on a Friday morning, while Lawrenceburg CC 4-6 cleared at $468.07/MWh. That is a nodal gap of $859.67/MWh, the widest in the system for the interval. The user's $860 headline figure rounds the sourced $859.67; the feed itself does not print $860.

The decomposition does the work. The energy component is $197.73/MWh at both extremes, so none of the gap is system energy. Lawrenceburg carries a congestion component of +$279.45/MWh and a loss component of -$9.11/MWh; Clifty Creek 1 carries -$575.38/MWh of congestion and -$13.95/MWh of losses. The signature is uniform across each cluster: Lawrenceburg CC 1-3, CC 4-6 and the LAWRENC218 kV nodes all show the same +$279.45 congestion and $468.07 LMP, while Clifty Creek 1, 4, 5 and the CLIFTYCR15.5 kV nodes all show -$575.38 congestion, with units 4 and 5 at -$391.58/MWh on a loss component of -$13.93. A generator on the Clifty Creek side is being paid to not inject; a generator on the Lawrenceburg side is being paid a premium to run. That is the shape of a constraint separating the two, but the feed does not say which one. PJM's top-congestion list at 11:00 UTC names Sugar Creek Dresser 345 l/o Nucor Cayuga 345, DOUBST2500500-2_XF and Chicago-Praxair3 138 kV l/o Wilton Center-Dumont 765 kV, all with shadow prices reported as n/a; none of the three is an Ohio Valley element. Whether the Clifty Creek units are online and curtailed or simply offline is not in the data either.

The system backdrop is unremarkable by comparison. The PJM-RTO hub real-time LMP was $199.04/MWh at the same 10:40 UTC stamp; the hub aggregate does not carry the nodal extremes. Load was 98,396 MW at 11:00 UTC with net load of 96,981 MW and gas the main fuel source. The outage feed offers no clean trigger: it logged 147 anomaly events over the past seven days, but the five sampled rows for 2026-09-12 04:00 UTC share one timestamp and disagree, with forced outages ranging from 3,354 MW to 11,023 MW, maintenance from 4,047 MW to 10,050 MW and planned from 6,680 MW to 16,056 MW. That feed cannot confirm or rule out a transmission or generation outage behind the spread, and only the single 10:40 UTC interval is sourced, so persistence into day-ahead is unverified.

What I'd watch is whether the sign flip survives the morning ramp and whether PJM names the element. If the Clifty Creek 1 congestion component is still below -$300/MWh at 09:00 ET (13:00 UTC) today, the constraint is structural for the session rather than a single-interval dispatch artifact, and the Lawrenceburg premium should hold above $250/MWh of congestion in the same interval. If a shadow price above $100/MWh appears on any 345 kV or 765 kV constraint in the Ohio Valley footprint in PJM's 10:00 ET constraint posting, that identifies the path the angle points at. If instead Clifty Creek's congestion component is back inside -$50/MWh by 09:00 ET, the spread was a transient and the trapped-generation read does not carry into the afternoon. Watch the Lawrenceburg CC 4-6 LMP against the PJM-RTO hub print at 12:00 ET: a Lawrenceburg premium above $150/MWh over the hub at midday would confirm the congestion has outlived the shoulder.

> Congestion, not energy, is the whole $859.67: the grid has drawn a line between two Ohio Valley plants and is paying one to run and the other to stop.

Not investment advice. For informational purposes only.

Yesterday's tape: ERCOT RUSSEKSTRN lmppeak >= 236.47 — verified (observed 366.4); ERCOT CATARINAB1 negativelmp <= -46.25 — verified (observed -313.82).

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

ERCOT's maximum real-time nodal spread reached $556.90/MWh in the 16:40 UTC interval on 2026-09-17, which is 11:40 CDT on a Thursday. RUSSEKSTRN set the top of the stack at $325.48/MWh and CATARINAB1 set the floor at -$231.42/MWh. That is one five-minute snapshot; no source in the fact pack says whether the gap held in the intervals that followed.

The tails were wide on both sides. SEVNFESRN was second-highest at $260.03/MWh, and LARDVFTNG5, LARDVFTNG4 and DCL each printed an identical $236.47/MWh. On the low side, MCLNSLRRN cleared at -$150.50, CARRIZOESR1 at -$129.03, and BRPPBL1RN and BRPPBL2RN both at -$46.25. What the pack does not contain is the binding constraint. The trailing seven days show at least 19,015 SCED shadow price events, and the sampled constraint names (NTOH at a 5,406 limit, WAPWLY72A at 1,790.7, THWZEN98A at 1,279.6, 138ALVNAL1 at 391.8, GABRIEAT1 at 46.8) arrive with no shadow price values, no timestamps, and no stated link to the nodes above. The outage log is similarly unanchored: at least 2,209 unplanned outage events over seven days, with a sample dated 2026-09-13 at 00:01 CDT showing MCSES (568 MW gas) at zero availability under Maintenance Level 1, FORMOSA (83 MW) and FLCNS (70 MW) at zero under outage extensions, and derates of 80 MW at MLSES and 32 MW at JADESLR. Nothing in the pack says which of those sit electrically near Russekst or Catarina, or which remain out today. Both event counts carry a sample marker and may be truncated by pagination, so read them as floors.

The forward view is system-level only. GridAlpha's pre-DAM lambda forecast for 2026-09-18 runs $21-$58/MWh with a $30 mean and an evening peak up to $58; the model's trailing 30-day MAE is $8.6/MWh against $11.8 for the naive baseline across 684 scored hours. Elevated-price risk is flagged for HE20 on 2026-09-18, HE20 on 2026-09-20 and HE20-21 on 2026-09-23. None of that is a nodal or South Hub forecast, so it says nothing directly about whether the Russekst-Catarina basis reopens.

On the angle's basis-risk framing, three checks are worth running. Persistence first: if RUSSEKSTRN prints above $236.47/MWh, the level of the LARDVFTN and DCL cluster, in the 16:40 UTC interval on 2026-09-18, the same clock slot as today's snapshot, then this is a repeating pattern rather than a one-interval spike; the matching test on the low side is whether CATARINAB1 sits below the -$46.25 BRPPBL floor in that same interval. DART second: the 2026-09-17 day-ahead LMPs at RUSSEKSTRN and CATARINAB1, once pulled, measured against $325.48 and -$231.42 tell you how much of the spread was priced in advance and how much was pure real-time basis. Shadow prices third: if a named SCED constraint shows a nonzero shadow price timestamped to 16:40 UTC, that becomes the driver the pack currently lacks; until then the cause stays unconfirmed. Against the 2026-09-18 HE20 risk flag, the $58 system-lambda ceiling is the reference line, and any excess at a node over the system number in that hour is, by construction, the congestion and loss component rather than the energy component.

> A $556.90 spread with no named constraint is a basis-risk measurement, not a story; until a shadow price and a timestamp attach to it, manage the exposure and withhold the narrative.

Not investment advice. For informational purposes only.

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