Congestion is a physical story.
A price separation is the beginning of the analysis. Following it into system lambda, binding constraints, outages, and weather is the rest of it.
Power-market screens are very good at showing that two prices are different. They are much worse at preserving the chain of physical evidence that explains why, which is the part a desk needs before it trades on the move or explains it to anyone else.
In ERCOT the first step is arithmetic rather than intuition. Real-time LMP carries no marginal-loss component, so subtracting the five-minute system lambda from a settlement point's price in the same interval leaves that point's congestion component. That separates a congestion story from a system-wide energy story. It does not yet name the facility responsible.
Follow the price into the physical system.
The reconstruction remains attached to the map and clock.
Conceptual product sequence. The tools and records vary with the market question.
01
Start with the move, not the dashboard.
A desk rarely arrives looking for every available metric. It arrives with a break: a hub moved, a node separated, a day-ahead expectation failed to appear in real time. The first job is to bound that break in time and space at the resolution the market actually publishes — five minutes for ERCOT dispatch prices and bus-level LMPs, fifteen for CAISO's real-time pre-dispatch run, five for NYISO's zonal LBMP, five for PJM's unverified real-time feed.
Resolution and settlement are different questions, and the answer has to survive both. PJM's verified hourly prices arrive roughly eighteen hours behind the operating hour and will not reproduce the five-minute series anyone watched at the time. Deciding up front which record the conclusion has to hold against is cheaper than discovering the mismatch the next morning.
02
The network explains the price surface.
ERCOT publishes the binding constraints themselves. Every dispatch run posts the constraint and contingency pairs that bound, the named facilities, limits and flows where published, violations, and the shadow price, which is the marginal system cost of that constraint in that interval. The record reaches back to 2017, which is enough history to tell a recurring seasonal constraint apart from something the system has not done before.
The honest limit belongs in the same breath. A shadow-price row establishes that a facility bound and what it cost; it does not establish which settlement point it moved, and that attribution requires shift factors. The caution runs in both directions: a wide LMP spread does not name a constraint, and a substation or line record is asset identity and topology, carrying no real-time flow, thermal limit, or loading percentage that could rank an overload.
Other markets expose the same physics through different records. CAISO posts intertie constraint and nomogram branch shadow prices every five minutes; NYISO publishes its real-time and day-ahead limiting constraints; PJM's day-ahead transmission constraints post daily, and its nodal prices arrive already decomposed into system energy, congestion, and loss. The question is the same everywhere. The evidence that answers it is market-specific, which is why the market boundary stays explicit rather than being averaged away.
Geography is where the remaining pieces line up. Transmission by voltage class from 69 kV up through 345 kV and above, substations, plants by fuel, price nodes, weather, and reliability events share one surface, and the discipline is to add only the layers the event requires rather than every feed at once.
A map is not decoration around the analysis. It is where the explanation becomes testable.
03
Weather and outages belong on the same clock.
A cold front, a renewable ramp, a thermal outage, or a transmission event matters only in relation to the interval. ERCOT's hourly resource outage capacity report is bitemporal for exactly this reason: each row carries the projected delivery hour and the vintage that published it, and because the outlook is revised every hour across a seven-day horizon, successive vintages accumulate as distinct rows instead of overwriting one another.
Weather divides the same way. Reanalysis is hourly on a quarter-degree grid but trails real time by roughly four to five days, so it can never describe the current hour; the forecast grid covers the current hour and the days ahead, vintage-stamped by lead time. Reading the two as one weather series is the fastest available route to a confident claim about a day that has not settled.
Outage evidence has a grain problem too. The public generation-outage feed is a five-minute system snapshot of total, planned, and forced capability reductions with dispatchable and renewable splits, covering the current day and the previous six. It is aggregate megawatts submitted by market participants, not per-resource outage identity, so it can establish that supply availability moved without naming the unit that moved it.
Kept on one clock, the difference between a plausible driver and something that merely happened the same afternoon becomes checkable rather than rhetorical.
04
Keep the reconstruction available.
Events are the other half of the story, and they are normalized before they are counted. Every record resolves to one of three forms — an episode, a schedule, or an observation — so a planned future outage is never tallied as an incident that happened and an aggregate system observation is never summed with either. Each link between an event and a physical subject records how it was established: reported by the source, resolved through a deterministic crosswalk, inferred with a confidence, or inherited from scope.
That is what makes a reconstruction worth keeping. The output should not be a screenshot with an arrow drawn on it, but a state: the market, the interval, the selected node and its coordinates, the constraint rows, the calculation, and the reports the numbers came from. It can be shared, revisited after settlement, or handed to the agent for the next question.
Congestion will always take judgment. The interface should spend its effort on the evidence that makes the judgment better.