V4 · IPP

1. What this is, and why it exists

Follow along. Sign in as ipp_2@webill.net (password supersecret). Path: Generator sites → IPP2 Atlantis Wind → Strategy. You should see the Overview tab, where the chain in this chapter becomes a capacity decision.

Data. full_demo · seed 42 · IPP2 Atlantis Wind · worked figures are illustrative unless a later chapter stamps them otherwise · last checked 2026-07-18.

1.1 The IPP's problem

You own (or operate) a generator — usually solar or wind — that makes electricity whenever the weather allows. Energy with no buyer earns R 0. Your commercial job is to sell the gap: every useful hour that is still unsold, at a realised price that clears your cost of production (LCOE), without promising more than the plant can stand behind.

When you enter the spot market, the same site is the source of §7.2 standing data. Register it from BRP / Market → Market registry → Register generator site (V7 §3.9A): the site supplies meter, technology and emissions; the plant engineer declares ramps and capacities with provenance; only the System Operator certifies reserve. The wheeling site and the Trading Unit stay linked.

That job has two faces that this volume holds together:

Sell-side strategy screens (§4–§6, §5 HAF, §11 Plant, §12 Dashboard) — read the revenue gap (realised vs achieved, unsold %, concentration), check safe contractable capacity, rank buyers against your unsold shape, test TOU counters and firm blocks, and promote only what you can defend.

Allocation, forecast and contracting (§7–§9, §8 Forecast, §13 guard) — each hour, metered injection at your bus bar has to be split among off-takers; next month's generation has to be predicted; and the "still for sale" number has to be discounted for how reliable the plant has actually been. Four questions decide whether that works commercially:

• What is still for sale, and at what realised price? Unsold hours are the prospectus; celebrating achieved price while ignoring them is how plants under-earn LCOE (§4).
• Who gets served first? When generation runs short, someone is curtailed. A rank-1 priority customer expects to be made whole before a pro-rata customer sees a single kWh (§7).
• What will I generate next month, and how much of it is already promised? If your book is subscribed beyond what you'll produce, you must not sign another off-taker (§8).
• How reliable is that answer? A plant that has been volatile, curtailed, or has starved its lower tiers is not as "available" as its nameplate suggests — and a prospective buyer deserves a number that reflects that (§5).

This suite answers all four, and it never guesses: every figure you see is computed from metered data, settled allocations and signed contracts, and is reproducible from a stored snapshot. Trader and off-taker book flow stays in V3 · Strategy; this volume is the IPP seat's home.

1.2 The chain, end to end

Read the screens first when you are learning the revenue gap; then follow the allocation chain. Five programme pieces fit together in a line; each consumes the one before it. The same waterfall algorithm runs twice — once on metered energy (settlement) and once on forecast energy (obligations) — which is why the forecast is credible: it predicts using exactly the rules that will settle.

1. Hierarchy — you place every off-taker into a tier (priority / pro-rata / fixed). This is the rulebook for the split.
2. Settlement — every hour, metered generation is carved across those tiers by the waterfall, and the result is stored per off-taker.
3. Forecast — once a month the engine predicts next month's generation and, running the same waterfall on the forecast, tells each off-taker what to expect and tells you what's left to sell.
4. HAF — a bankable availability score discounts that "left to sell" figure for volatility, curtailment and past delivery.
5. Widget — a prospective buyer sees a single sentence: "You can safely contract another N MWh next month."

1.3 Glossary — every abbreviation, up front

The suite is unapologetically acronym-heavy because each number has a precise definition. Here is the full set; each is explained again in context.

Two concepts worth pinning before you read on:

• Bus-bar energy — every quantity in this suite is measured at the generator before losses; loss factors are applied later, per contract, by settlement's matcher. That keeps the allocation arithmetic clean and comparable.
• Effective-dated versioning — hierarchies and forecasts are never overwritten. A change creates a new version; the old one stays exactly as it was, so history always re-reads the way it settled.

2. Your fleet — the portfolio story

The Fleet tells the story of every generator site your organisation owns, as one picture: energy still looking for a buyer, which buyers the model leans on, and whether putting the plants together looks safer than adding them up one by one.

Think of a shelf of recipe books versus the kitchen. Each plant still cooks from its own book — it forecasts, sells, allocates and settles on its own IPP strategy. The fleet page is the kitchen noticeboard: it does not change a recipe, place a bid, or move energy. It reports.

The IPP’s problem, in plain words

Five plants can each look healthy while the organisation is still carrying unsold energy, leaning on one buyer, or assuming that a Cape wind farm and a Gauteng solar plant cancel each other’s bad hours when they do not. A wall of tiles does not answer those questions in order.

2.1 Read the glance header, then the story

The glance summarized the fleet’s sales opportunity as a percentage of unsold energy, then the four chapters sit under it. It presents the main aggregated statistics of the entire fleet with summarized coverage and calculation notes.

Recompute now and Export CSV are taught in §2.6. The story drawers exploring the fleet live in §2.2–§2.5.

2.2 The sales opportunity

01 / The sales opportunity. You cannot decide what to sell next if you do not know how much energy still has no buyer, and a wall of plant tiles hides that answer.

The big sentence is chosen from the saved fleet snapshot’s unsold % — generation with no buyer, as a share of total generation in that snapshot.

A two-colour bar splits sold (brand green) from still to sell (amber).

The footer says what to do next: check evidence and recompute if there is no percentage; check that production can meet the commitments if everything is sold; otherwise find which sites still have volume, and when buyers can use it.

Unsold energy is not a Rand loss. It is volume that still needs a contract or a market bid. The site Sales desk is where you turn that volume into a plan (§6).

Under chapter 01, the snapshot date reads Fleet model snapshot · DD Mon YYYY, or No saved fleet snapshot yet. Beside it, How sure are we? opens the evidence drawer. Read that date before you compare a sales percentage with a buyer bar: the headline uses the saved snapshot; membership and buyer exposure are evaluated today.

Review the sales plan

Review the sales plan opens the sales drawer.The side-drawer panel lists today’s fleet members, ranked by the energy each still has to sell: the unsold share of its latest forecast × its annual P50. Each row is a plant name, its export megawatts and technology in words (50 MW · Solar PV), and on the right the unsold share. The site has a forecast. Under each row is one link, Open its sales plan → to that plant’s Sales desk.

In the side-drawer a fleet of one still lists that one plant. A site missing today’s snapshot still appears here if it is a fleet member — the missing-evidence card on the page is the warning, not a hidden row. An unsold percentage is a share of the site’s own forecast, so a small plant at 100 % unsold can sit below a large plant at 60 %.

How sure are we?

The sales sentence on chapter 01 can look finished when the picture behind it is old, missing a plant, or still a stand-in for how your plants behave together. Open this drawer before you act on that sentence, and before you trust chapter 03 when it says a bigger fleet is safer.

What it is for. The drawer is the evidence shelf for the noticeboard. It answers three questions in order: which numbers are a saved photograph and which are counted today, how many plants actually contributed, and whether “these plants cover each other” came from their own meters or from a published stand-in.

Five tiles are the compact stats that used to live on the page.

Volume correlation — estimated and n

You need this list because chapter 03’s “a bigger fleet helps” claim is only true when a bad hour at one plant is a good hour at another. If your wind farm and your solar plant sag on the same hours, owning both does not protect you. The model can only say they cover each other if it knows how their generation moves together. That knowledge is volume correlation. Without it, the green “combined fleet” bar in chapter 03 is a story with no evidence.

Everyday picture: two rain gauges on one farm. If they fill on the same hours, a storm hits both roofs. If one fills while the other stays dry, the farm still catches water. Volume correlation is that comparison for generation, one pair of plants at a time. A value near 1 means they move together. A value near 0 means they miss each other. The drawer does not print that number. It prints where the number came from.

Volume correlation is the question (do these two plants move together?). The heading is the answer you can check (whose answer is this — history, a register, or a person?). In Assumption sources. each row is two plant names, each with its organisation, an arrow between them, and a badge:

The register, used only while a pair is still default — no history yet:

Read 0.80 as “these two solar plants in one province are assumed to sag together most of the time.” Read 0.10 as “solar and wind are assumed to miss each other.” A pair the register does not name stays at 0. Those numbers are a published stand-in, not your meters. When shared history reaches 720 hours, a later version can replace the stand-in for reporting. The old version stays, so you can still see what the model used.

A fleet of one has nothing to compare. The list says Fleet of one — no site pairs. Chapter 03 says the same thing in its own words: one site cannot spread risk.

The nightly fleet job, or Recompute now when you cannot wait, may measure pairs on the first of the month (§2.6). Closing this drawer does not measure them.

2.3 Who the fleet depends on

#fleet-buyers-story is labelled 02 / Who the fleet depends on.

The reader’s problem, in a kitchen picture: you can sell jam from three stalls and still have one customer. If that restaurant stops buying, all three stalls feel it. Chapter 02 groups each buyer once, even when they take energy from two plants, so you see dependence, not a duplicated name per site.

The bars (#fleet-buyers-chart) are modelled exposure in Rand — an adverse-price picture. They are not last month’s invoices and not revenue. Think of them as: if prices move against this position, how much of that pain sits with this buyer?

At most three named buyers are drawn; further names collapse into Other buyers (N). A buyer with no positive exposure is counted in a quiet note (N other buyers have no positive exposure reported), not drawn as a bar. If nothing is positive, the empty copy is No positive buyer exposure is reported. This does not establish that buyer risk is absent (#fleet-buyers-empty).

The footer says As sales grow, watch how much depends on each buyer. See buyer exposure (#fleet-open-buyers) opens the buyers drawer.

See buyer exposure

Chapter 04’s second row does not open this panel; it opens Deals › Contracts (§2.5). There is still no Counterparties tab. A buyer that purchases from two plants appears once, with megawatts and adverse Rand exposure summed.

That first sentence is the whole job of this drawer. The number in Exposure is modelled exposure — a picture of if prices move against this open position, how much of that pain sits with this name? It is not last month’s invoice, not revenue you have already earned, and not a cash call you will definitely pay. The nightly fleet job stores the same family as COUNTERPARTY_EXPOSURE_R. Membership and this table are evaluated today; they can disagree with the saved sales snapshot under chapter 01.

Everyday picture: you do not wait for the grocer’s bill to ask “if fruit prices jump, how much of that jump sits with this restaurant?” You take the open order book, pick the unkind price, and write one Rand figure per customer. That is this table.

Inside: heading Counterparty exposure across sites (#fleet-counterparties) and the sentence A buyer appearing at several plants is grouped once here. Column labels stay visible while you scroll (thead is sticky). The columns:

What the Adverse basis cell says

Adverse basis is the book’s unkind price, not the buyer’s invoice. The engine groups each site’s open position by the other organisation, then values megawatts × hours at one tail:

• Short — you still owe that buyer energy. If they stop taking it, you may have to buy replacement. High prices hurt. The book uses the high tail p90.
• Long — you have leftover energy against that name. Low prices hurt: leftover can be cheap to sell. The book uses the low tail p10.

Those grades are price tails (how expensive or cheap the market looks), not generation P50 / P90 (how much the plant makes). They are the book’s bad case. They are not “the buyer’s bad case” on this screen, even though a high p90 price is also a buyer’s expensive hour in the forecast cells.

The cell copy, quoted exactly. The grade is lowercase and bold, centred, with the caption centred under it:

1-in-10 is the plain reading of the tail: p90 is a high price that only about one hour in ten clears above, and p10 a low price that only about one hour in ten clears below. Short means the fleet is short versus that buyer: buying replacement energy can be expensive. Long means leftover energy that may have to be sold cheaply. A SUPPLY pool membership is treated as short; a DEMAND pool membership is treated as long. If one name ends up with both sides, megawatts and Rand still add; the printed grade is the last side the model stored for that row.

Worked numbers: how one Exposure cell is made

Data. Mechanism from the counterparty-exposure engine (hours = 24; default tails p10 = R 800/MWh, p90 = R 1,400/MWh). Toy positions, not today’s IPP 2 invoices. The drawer uses the same arithmetic on whatever open positions the fleet has today.

Same megawatts at p90 cost more than at p10, because replacement is the expensive tail and leftover is the cheap tail. That is the pattern: the caption tells you which unkind price was used; the Exposure column is that price × megawatts × hours.

Done looks like this: you can name who you depend on, you can read whether each row is p90 (short, buying replacement energy) or p10 (long, selling leftover energy), you know whether one organisation dominates, and you have not added those Rand figures to unsold energy or to VaR.

Edges. There is still no Counterparties tab. Fleet risk limits create alerts only (#fleet-limit-breaches) and do not block site bids. A quiet on-page note about zero-exposure buyers is not a hidden table row — open the drawer to see every grouped name the model has. Closing the drawer (X, backdrop, or Escape) returns you to chapter 02; nothing is saved.

2.4 Does a bigger fleet help?

#fleet-risk-story is labelled 03 / Does a bigger fleet help?

The reader’s problem: owning two plants feels safer than owning one. It is safer only when their bad hours miss each other. If the Cape wind farm and the Gauteng solar plant both sag in the same modelled month, a bigger fleet does not help.

Everyday picture: a bakery with two ovens. Oven A can have a bad month. Oven B can have a bad month. If you add “A’s worst month” to “B’s worst month”, you are assuming both ovens have their worst month at the same time. If they bake in different weather, their bad months often miss each other. The bakery’s worst month is then milder than the sum of the two worst months. That milder number is the combined fleet. The sum of the two worst months is Sites added together.

The on-page copy says: Compare the combined fleet’s risk estimate with the sites added together. A smaller fleet bar means the model shows a benefit.

The chart (#fleet-risk-chart) appears when a monthly distribution exists and the correlation matrix was not refused. Two bars:

The caption is One-month risk · 95% model threshold · the distribution date.

If the monthly run is simply missing, the chart is withheld (#fleet-risk-unavailable): A valid monthly fleet distribution and site comparison are needed before showing this chart.

If today’s risk run was refused, the chart is withheld too, and a rose status box sits inside the card instead (#fleet-correlation-refused):

Today’s risk run was refused. Correlation table vN says how closely each pair of sites rises and falls together, and its pairs contradict one another. The model would not build an estimate on it, so there is no fleet risk figure for today.

Fix the manual pairs under Review the assumptions, or remove them so the defaults apply, then Recompute now.

A second link, Review the assumptions (#fleet-open-refused-assumptions), then sits beside Understand the estimate and opens the evidence drawer at the pair list (How sure are we?). Fix or remove the manual pairs there, then press Recompute now (§2.6).

The footer names how many site-pair assumptions still use defaults (N of M site-pair assumptions use defaults. Operating history is needed to judge the benefit.). A fleet of one says A fleet of one has no site-pair assumptions to compare. When every pair has a non-default source: Review the dates and evidence behind the site-pair assumptions before relying on this estimate.

Value at Risk, in plain words

Value at Risk (VaR) is not “the most you can lose” and not cash sitting in a till. It is a modelled shortfall against the average leftover.

Picture the bakery again. The model imagines many months of every oven together: weather, prices, how much each oven baked. Most imagined months cluster around an average leftover. VaR-95 is the gap from that average down to a poor leftover. In 95 of 100 imagined months you miss the average by less than this Rand amount; in the worst 5 you miss it by this amount or more. The platform term is loss relative to expected.

The green Combined fleet bar, and the tile One-month VaR-95, are that gap for a one-month sample. How to read a made-up day versus a made-up month, what to do with the two tiles, and every field in Distribution data are taught under Understand the estimate. The next headings stay with the two bars on the page.

How the two bars are calculated

The fleet run follows the ordinary site strategy and stochastic work:

1. It shares one market-price path across sites (the bakery sees the same price weather).
2. It draws each site’s volume error, then mixes those errors through the current volume correlation matrix. Plants the matrix says move together tend to miss on the same hours.
3. It scores each site’s margin on each path (higher is better).
4. Sites added together: compute VaR-95 on each site’s path margins, then add those VaRs (SUM_OF_SITE_VAR_95_R).
5. Combined fleet: add the sites’ margins path by path, then compute one VaR-95 on that summed list (var_95_r on the one-month distribution).
6. Diversification benefit is max(sites added together − combined fleet, 0). If the combined book looks riskier, the benefit tile shows R 0; chapter 03 still tells you the combined bar is longer.

If the matrix is not a valid correlation — technically, not positive semi-definite, which in plain words means its pairs contradict one another — the platform refuses it rather than quietly changing it. That refusal is the rose box inside chapter 03 (#fleet-correlation-refused): no stochastic snapshot is written from that version, and chapter 03 withholds the chart.

Worked numbers: sites added together versus combined fleet

Data. Mechanism from the fleet-estate seed used for every multi-site fleet book (full_demo · seed 42). Toy path margins, not today’s IPP 2 invoices. The on-page bars use the same arithmetic on whatever one-month snapshot is stored.

The seed walks five made-up months (shown here already multiplied for the one-month horizon):

Average month: plant A R 2,400, plant B R 4,800, combined R 7,200. At 95% with five paths the loss quantile is the worst slip: A R 1,920, B R 3,600, combined R 6,480.

The combined bakery’s bad month is R 960 milder than adding the two ovens’ bad months. That is the pattern to expect when chapter 03 says The model shows lower risk when sites work together. A fleet of one cannot show this split: both bars would be the same plant, and the benefit tile reads R 0 (fleet of one).

Understand the estimate

Understand the estimate (#fleet-open-risk) opens the risk drawer. The title in the green bar is What the risk estimate means. The title-row Recompute now and Export CSV buttons stay in §2.6.

Follow along. Seat: IPP 2 (ipp_2@webill.net, password supersecret) on full_demo · seed 42. Path: IPP Sites → Fleet → chapter 03 Understand the estimate. You should see the title What the risk estimate means, then One-day VaR-95 R 30 and One-month VaR-95 R 720. Open Distribution data for the 1 day and 1 month rows.

Read the headings below in order: a made-up day or month, what VaR-95 means, how the two tiles differ, what to do, then every field in Distribution data.

A made-up day or month

The model cannot wait for the next hundred real days. It writes a pile of made-up days (and, separately, a pile of made-up months) for the whole bakery.

Think of one slip of paper. On that slip the model writes: what the weather did, what prices did, and how much each oven baked. Then it adds up the leftover money the book would have kept that day. That leftover is the margin on that slip — Rand the book keeps after costs, not a page margin. One slip is one path: one made-up day of the fleet, or one made-up month of the fleet.

The pile is not a forecast you will invoice. It is a set of “what if this day (or this month) went like this?” stories, written so a bad leftover can be compared with a typical leftover.

The two piles are baked separately. A made-up month is not “a made-up day written thirty times.” Weather and prices in a month do not behave like thirty copies of one day.

On seed 42 the fleet-estate pack writes five slips per pile (small on purpose, so you can add them by hand). The meaning of 95 is still “95 in 100”: five slips do the same job as a larger pile, with rougher marks.

What VaR-95 means

VaR-95 answers one question: on a bad slip, how far below the average leftover do we fall?

It does not answer “how much cash will I have to pay?” It is not a pile of rands you already hold. It is a gap — how much worse than typical the leftover is, at a chosen mark.

The platform convention is loss relative to expected. In kitchen steps:

1. Take every slip in the pile. Read the leftover on each slip.
2. Find the average leftover (add them up, divide by how many slips).
3. Find the poor mark: the leftover that leaves only the worst 5 in 100 slips below it.
4. VaR-95 is average minus that poor mark.

In 95 of 100 made-up days (or months), the shortfall against the average is smaller than this Rand amount. In the worst 5, the shortfall is this amount or worse. Worse slips remain possible. The first paragraph of the drawer says that: a 95% threshold is not a maximum loss.

Data. full_demo · seed 42 · fleet-estate slips used for every multi-site fleet book. Toy leftovers, not today’s IPP 2 invoices. The drawer uses the same gap on whatever snapshot is stored.

The one-day pile on that seed (already added across the two ovens):

Average leftover: (300 + 270 + 300 + 270 + 360) / 5 = R 300. The poor mark with five slips is the worst leftover, R 270. Then:

VaR-95 = 300 − 270 = R 30

That R 30 is the One-day VaR-95 tile. It means: relative to a typical imagined day, a bad imagined day is R 30 lighter. It does not mean you will be asked for R 30, and it is not unsold energy.

The one-month pile on the same seed is those leftovers × 24 so you can see the arithmetic. That is a demo shortcut, not how a live month is built: a nightly run walks a real month on its own, and you must not turn a live one-day tile into a month by multiplying. Average month R 7,200, poor mark R 6,480, gap R 720. That R 720 is the One-month VaR-95 tile and the green Combined fleet bar taught above.

One-day VaR-95 and one-month VaR-95

The drawer eyebrow is Fleet risk. The heading is Risk estimates and their evidence (#fleet-risk). Under it, one line spells the abbreviation out: VaR-95 (Value at Risk at 95 %) is the loss the model expects to beat on 95 days in 100. Two tiles sit side by side. They are two piles, not one number written two ways.

Under each tile, a note that starts with the date that pile was run:

sum of sites is sites added together for that pile: each oven’s own VaR-95, then those gaps added. benefit is diversification benefit on that pile: max(sum of sites − this tile, 0). On seed 42 the one-day note ends sum of sites R 70 · benefit R 40, and the one-month note ends sum of sites R 1,680 · benefit R 960, each after its run date. Amounts use comma thousands.

When today’s risk run was refused. A rose box (#fleet-risk-refused) sits above the tiles:

These figures are not today’s. Today’s risk run was refused: site-pair correlation table vN contradicts itself, so no new estimate was written. The figures below are from the last accepted run on DD Mon YYYY.

Both VaR tiles then turn amber. The numbers are still the last accepted run’s; the run date at the start of each note says how old they are. Fix the pairs through Review the assumptions and press Recompute now (§2.6).

A bigger VaR-95 means a larger modelled miss versus the average — the poor leftover sits farther below typical. A smaller number means a milder miss. Smaller is not “safe.” It is a shorter gap.

Do not turn the one-day tile into a month by multiplying by 30, or by the square root of 30. Those shortcuts pretend each day is a fresh coin flip. This book is judged monthly, and the month pile is sampled on its own. If the day tile is R 30 and the month tile is R 720, read each as its own sentence.

If a tile has no snapshot, the value is Not available (not R 0). R 0 on a tile that has a snapshot would mean the poor leftover matches the average — a different claim.

What to do with the numbers

These tiles do not place a bid, sign a contract, or move energy. They tell you whether to slow down, whom to ask, and which drawer to open next.

1. Read the dates first. Under the first paragraph: Latest daily distribution: … Latest monthly distribution: … These may differ from the fleet model snapshot date. A zero imbalance cost does not prove operations are free of risk. A missing date prints Open. If a date is Open, that pile was not run — do not compare the tiles with chapter 01’s sales percentage or with each other.
2. Read the one-month tile first. The organisation is judged on a month. That number is the green bar in chapter 03. Use the one-day tile as a same-day twin, not as the number you take to a monthly meeting.
3. Compare the bakery with the ovens. Look at sum of sites under One-month VaR-95. If the tile is smaller than that sum, the model says the ovens’ bad months miss each other (the pattern of R 720 versus R 1,680 above). If the tile is larger, or benefit is R 0 on a fleet of several plants, a bigger fleet is not helping in this run.
4. Open How sure are we? when the story depends on a guess. Press Review the assumptions (#fleet-risk-open-assumptions) — it switches this drawer to How sure are we? so you can read the pair badges without walking back to chapter 01. Open it when any pair still says default — no history yet, when sites are missing today’s snapshot, or when the rose box These figures are not today’s. (#fleet-risk-refused) sits above the tiles. Combined-fleet VaR uses those pair assumptions; a default pair is a register value, not overlapping plant history.
5. Combined versus sites-added-together matters when you are deciding whether a new plant in another weather region actually offsets, or whether two wind farms in the same province just double the same bad week. It does not matter as a scoreboard for a fleet of one: both views are the same plant, and the benefit tile reads R 0 (fleet of one).
6. Take the one-month number to Sales when chapter 01 still shows a large unsold percentage and you need to know whether a new contract would sit on a book whose bad month already misses the average by a lot. Sales still lives on each site’s Sales desk; this drawer does not sell the gap.
7. Take it to the board when the one-month gap is large compared with what the organisation can absorb, when combined looks worse than sites added together, or when most pairs are still defaults. Take the dates and the assumption badges with you, not the VaR tile alone.

Do not add VaR to unsold energy or to buyer exposure. They measure different jobs: volume still looking for a buyer, pain sitting with a named buyer, and a shortfall against an average leftover.

When to worry. A date that says Open; a tile that says Not available; amber tiles under These figures are not today’s.; a combined bar longer than Sites added together; benefit R 0 on more than one plant; many default — no history yet badges. When not to treat it as an alarm. A smaller tile than last month, by itself; a fleet of one (there is nothing to spread); a zero imbalance-cost tile in How sure are we? (that tile is a different job).

Done looks like this: you can say the one-month number in a sentence (“we miss the average imagined month by about R 720 at the 95% mark”), you know whether it is lower than Sites added together, and you have opened the assumptions when any pair still says default — no history yet.

Edges. Recompute now writes fleet snapshots only (§2.6). It does not rewrite site allocation runs. A refused correlation writes no new estimate: the tiles keep the last accepted run’s figures, in amber, under the rose box — or Not available if no run was ever accepted. Closing the drawer (X, backdrop, or Escape) returns you to chapter 03; nothing is saved.

Distribution data

Fold Distribution data (#fleet-fan-data) under the two tiles. The summary label is Distribution data. Inside is a table with uppercase headers. The fold shows only these columns: Horizon, P05, P50, P95. It does not print a mean, a min, a max, a sample count, or extra grades such as P10 or P90 — those can exist on the stored snapshot; they are not on this screen.

Each row is one pile. Labels, quoted exactly: 1 day and 1 month. A pile with no snapshot is omitted (no blank row). If both piles are missing, the table has no body rows and this empty copy prints under it: No fleet distribution has been run yet.

P05 / P50 / P95 here are leftovers on the slips (Rand the book keeps on a poor / middle / strong imagined day or month). They are not generation P50 / P90, not price P10 / P50 / P90, and not VaR-95. VaR-95 is a gap. These three cells are levels.

Amounts use comma thousands (R 6,480). A missing cell on a row that does exist renders as R 0 (the drawer treats an empty leftover as zero). That is not the same empty state as a missing row.

1 day — every field you see

Data. full_demo · seed 42 · IPP 2 fleet book · fleet-estate one-day slips ([100, 120, 80, 90, 110] and [200, 150, 220, 180, 250], then added path by path). Same toy leftovers as the one-day worked example above.

Read the one-day row as a fan of leftovers: poor R 270, middle R 300, strong R 348. The tile R 30 is 300 − 270 on this seed (average minus poor mark). If P05 and P95 sit far apart, the imagined days spread widely; the VaR tile can still be small if the poor mark sits close to the average.

1 month — every field you see

Data. Same seed and ovens; the pack stores the month pile as those day leftovers × 24. That is how this demo was written, not a rule you should apply to a live one-day tile.

Empty states, in one place:

The date line also says the distribution dates may differ from the fleet model snapshot under chapter 01. Read both dates before you compare a sales percentage with a VaR tile.

2.5 The next chapter

#fleet-next-chapter is labelled 04 / The next chapter. The heading is Three practical places to focus. It is not a fourth chart, and it does not reopen the drawers from chapters 01–03. Those chapters explain; each row here leaves the dashboard to act.

Think of the noticeboard again. Chapters 01–03 are the notices. Chapter 04 is three doors out of the kitchen: to the stall that sells, to the order book, and to the delivery log.

Each row has a number, a title, a one-line detail, and one action; the row’s id sits on that action.

Done for chapter 04: you have opened the Sales desk of the site with the most energy to sell, checked which buyers already hold signed and active contracts, and — if the pair list is still mostly defaults — imported metered intervals. The footnote under the four chapters reminds you: unsold energy, buyer exposure and fleet risk are different measures. Do not add them together.

Row 01 — turn available energy into a sales plan

This is the first row on chapter 04. It picks the fleet member with the most energy still to sell — the unsold share of its latest forecast × its annual P50, the same ranking as the sales drawer (Review the sales plan) — and names it.

Done looks like this: you are on the named site’s Sales desk, reading the candidates that fit its unsold hours (§6.4 step 2), and you have not treated the fleet unsold % as cash already lost.

Where this is taught. The ranked site list behind this row: §2.2 Review the sales plan. Selling the gap: §6.

Row 02 — keep buyer dependence visible

This is the second row. It opens the contract register rather than the buyers drawer, because the question before a new contract is who already holds one.

For the modelled Rand each buyer carries, stay on the Fleet page and use chapter 02’s See buyer exposure (§2.3). Do not add those Rand figures to unsold energy or to VaR.

Done looks like this: you can name the buyers who already hold signed or active agreements, and you know whether a new contract would deepen dependence on one of them.

Where this is taught. The contracts register: V7 §6. Columns, adverse basis, and the worked Exposure cells: §2.3 See buyer exposure.

Row 03 — keep the model grounded in operating history

This is the third row. Pair assumptions stay default — no history yet until the plants have enough overlapping metered history. This row is the way to supply it.

To read which pairs are still defaults first, open How sure are we? under chapter 01 (How sure are we?). Imported intervals count towards the next pair estimate, which the fleet job makes on the first of the month (§2.6); they do not change today’s figures.

Done looks like this: the modal has shown Imported N intervals, and you know that a pair moves from default to estimated only once enough overlapping hours exist (720 by default) and the monthly estimate has run.

Where this is taught. The modal, the file schema and every refusal: V2 §7.7. Tiles, Volume correlation, badges, and n: §2.2 How sure are we? and Volume correlation.

2.6 Export CSV and Recompute now

The four chapters and their drawers teach what the numbers mean (§2.2–§2.5). This section is the two buttons on the top right of the Fleet title row: how to refresh the noticeboard, and how to take a copy off the page.

Everyday picture: the Fleet page is a kitchen noticeboard. Recompute now reprints the board from today’s books. Export CSV photocopies three chapters of that board into a spreadsheet. Neither button bakes a plant’s own recipe — that bake is Run Forecast Now on one site’s Overview.

Follow along. Seat: IPP 2 (ipp_2@webill.net, password supersecret) on full_demo · seed 42. Path: IPP Sites → Fleet (/org/ipp-sites). On the right of Your fleet, at a glance you should see the white Export CSV then the dark Recompute now. Click Export CSV: the browser downloads fleet-YYYY-MM-DD.csv. Click Recompute now: it reads Queuing… for a moment, then the flash reads Fleet recompute queued. It runs in the background; refresh in a minute to see the rebuilt model. Reload after the job has finished to see a new snapshot date.

Where the buttons sit

Both live on the Fleet tab only — GET /org/ipp-sites (the same page as /org/fleet). They sit in the title row (#fleet-dashboard-header), in the right-hand action cluster (#fleet-dashboard-actions), after the glance identity and after Explore the fleet when you own sites.

They stay visible on the empty card (No generator sites owned — nothing to aggregate). Explore the fleet hides when you own no sites; these two do not.

Who may click. Anyone signed into the organisation who can open IPP Sites → Fleet. The buttons are not hidden behind a second role, not disabled on an empty fleet, and not reserved for an operator seat.

Recompute now

Recompute now queues the same worker the clock uses at night (WB.Strategy.Workers.RunFleet), scoped to this organisation’s fleet book and today’s date.

It is not Overview Run Forecast Now. That button sits on one plant, shows Running…, and seals that plant’s forecast snapshot. This button reprints the fleet noticeboard. It does not open a spinner, and it does not walk a site’s allocation waterfall.

The flash, quoted exactly:

What the worker does, in order, for this fleet:

1. Reconciles membership — who is an active member today.
2. On the first of the month only, may estimate site-pair volume correlations if that day has no estimate yet. Other days keep the pairs you already have.
3. Stores fleet KPIs (capacity factor, unsold %, imbalance cost, diversification) and grouped buyer exposure.
4. Writes a fleet risk snapshot dated today.
5. Samples the one-day and one-month distributions when the correlation matrix is usable. If the matrix was refused, or the fleet has no members, this last step is skipped; on a refusal, chapter 03’s rose box (#fleet-correlation-refused) explains why.

Done looks like this: you saw Fleet recompute queued. It runs in the background; …; after a reload, chapter 01’s date (#fleet-story-basis) reads Fleet model snapshot · DD Mon YYYY for today; the amber Last fleet run… chip (#fleet-stale) in the page top bar has cleared (it only shows when there is no snapshot, or the latest is more than two days old); and How sure are we? / Understand the estimate show the new tiles. If stochastic sampling was skipped, the KPI snapshot can still be new while the VaR tiles stay Not available.

What it does not do. It does not rewrite a site’s allocation run, site risk snapshot, or site distribution. It does not place a bid, move energy, or change a contract. It does not estimate pairs except on the first of the month. It does not wait on the page until the job finishes — Queuing… lasts only while the job is queued, not while it runs. A second click within about five minutes is treated as the same queued job, not a second run; the flash then says A fleet recompute is already queued or running, so it was not queued twice.

Edges. An empty fleet still queues the job: membership is reconciled and KPIs are stored; the distribution sample is skipped because there are no members. If the job cannot be queued at all, the red flash Could not queue the fleet recompute. … says so; the nightly run still rebuilds the model. Failures while the job runs are written to the operations log, not flashed on this page — if the snapshot date does not move after a reload, ask operations to check the strategy queue (V21 §3). The stale chip is a date label, not a traffic light on the button.

What runs while you sleep

You do not have to press the button every day. The same worker runs for every fleet book at 03:15 (15 3 * * * on the platform clock). That nightly pass does the same five steps: reconcile membership, maybe estimate pairs on the first of the month, store KPIs and buyer exposure, write the fleet snapshot, then sample the one-day and one-month distributions when the correlation is usable.

The button is that pass now, for your organisation only. Use it when the amber chip says the picture is old, when you just recorded a departure, or when you cannot wait for 03:15. Leave it alone when tonight’s run is soon enough.

Where this is taught. The nightly fleet worker is catalogued in V21 §3. The day’s clock preview is V0 §6.

Export CSV

Export CSV downloads today’s fleet evidence as a text file. It is a photocopy of three chapters of the noticeboard — the KPI keys, the active members, and the pair-assumption sources — not the story sentences and not the buyer table.

What happens, in order. The browser requests /org/fleet.csv. The server builds four columns from today’s book (not from the saved sales headline if that snapshot is older). The response is text/csv; charset=utf-8, offered as an attachment. The Fleet page does not flash and does not change.

Done looks like this: your browser’s download list has fleet-YYYY-MM-DD.csv, and opening it shows a header row plus one data row per metric, per active site, and per site pair.

What the file contains

Every field is wrapped in double quotes. A quote inside a value is doubled (" → ""). Rows end with a newline. Encoding is UTF-8. This is the platform’s evidence dump, not a Microsoft Excel template — Excel and Numbers will open it if you pick UTF-8 / comma.

The header, quoted exactly: record, key, value, source.

The metric keys, quoted exactly:

Those KPI rows are computed today for the current month, the same way membership and buyer exposure are evaluated today. They can disagree with a saved sales snapshot under chapter 01. Read #fleet-story-basis before you treat an exported unsold percentage as the headline you just saw.

Worked rows: what a short file looks like

Data. Mechanism from GET /org/fleet.csv (quoted UTF-8). Toy values so you can see the three record kinds — not a claim about today’s IPP 2 totals.

"record","key","value","source"
"metric","FLEET_UNSOLD_PCT","98.8","fleet"
"site","IPP2 Atlantis Wind","50.0","WIND"
"correlation","aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee:ffffffff-0000-1111-2222-333333333333","1","ESTIMATED:720h"

Read it as: one KPI (unsold 98.8%), one active member (50 MW wind), one pair whose assumption is estimated from 720 overlapping hours. The pair key is two site identifiers; the plant names live on the site rows.

Who it is for. A finance or risk person who works in a spreadsheet; an auditor who needs today’s membership and pair sources beside the on-screen tiles; anyone preparing a board pack who must not retype the drawers. It is not a sales invoice, not last month’s settlement, and not a file you send to a buyer.

What it does not do. It does not include the buyer-exposure table (organisation name, MW, adverse basis, Rand). It does not include story headlines, the sold / still-to-sell bar, or Distribution data (P05 / P50 / P95). It does not change the fleet book. It is not the curtailment Export CSV on a plant claim (§10).

Edges. An empty fleet still downloads: you get the metric rows (often zeros) and no site rows. A missing organisation session returns 404 with the body Fleet not found — sign in again and open Fleet from IPP Sites. The file is scoped to the signed-in organisation; you cannot pull another organisation’s fleet by editing the address. Closing the download dialog does not recompute anything.

Words on this screen

The headline uses the saved fleet snapshot. Membership, buyer exposure, and the CSV are evaluated today. Read the snapshot date under chapter 01 (#fleet-story-basis) before you compare a sales percentage with a buyer bar or with an exported unsold key.

Where this is taught. One plant’s profit-and-loss narration is §12 Dashboard. Selling the gap is §6 Sales. The Sites register is V2 §4.1; recording a departure is taught above. Importing metered intervals is V2 §7.7. Run Forecast Now on one plant is §4 Overview. The nightly fleet worker is catalogued in V21 §3.

3. Sites

A site is one plant. The IPP Fleet in §2 is every plant your organisation owns, as one picture. This chapter is the site-level book: how you open one generator and walk its strategy tabs.

Think of the fleet page as the kitchen noticeboard and each site as its own recipe book. The noticeboard tells you what is still to sell across the shelf. You still cook from one book at a time — forecast, sell, allocate and settle on that plant alone.

Open IPP Sites in the Wheeling sidebar. The room has two tabs:

The Sites tab is the register of generator sites (which site needs you, and where to go), taught field by field in V2 §4.1. Press a site’s name for its quick view, then Open site workspace, or follow the row’s Next step. Old /org/fleet still opens the Fleet tab.

The site workspace (/org/ipp-sites/sites/{id}/{tab}) gathers the screens this volume teaches under five tabs. Site details (top right) holds the complete site record and Site lifecycle & fleet membership (§2).

The plant model at the top of Operations lists the plant’s facts with comma thousands and at most three decimals, for example 118,260.0 MWh.

The browser tab names the page the way the sidebar and tabs do: the register reads Sites · IPP Sites · Webill Green, and a workspace reads {Site name} · {Tab} · Webill Green — for example IPP1 Sandton Solar · Performance · Webill Green. A panel hosted inside a tab keeps that title rather than showing its own. The room chip in the top bar (IPP SITES) is a link back to the room’s first page, the Fleet tab.

The top bar carries Manual (#ipp-sites-manual), which opens this volume at the chapter for the view you are on, in a new tab. On the register it opens this chapter and sits right of the N sites chip. On the workspace each tab opens its home chapter — Operations §11, Performance §4, Allocation §7, Commercial §6, Dashboard §12 — and the onboarding form opens V2 §4. The screens that still open on their own under /org/generator-sites/{id}/strategy/… do the same — Allocation · Forecast opens §8, Outcomes §9, Curtailment §10, Plant §11 and the battery contract §20; the sales proposal studio and the proposal review page open V7 §3.

3.1 Getting in — the IPP strategy area

Thread part 4: the IPP programme for IPP2 Atlantis Wind — what the plant has promised, what it can still sell, and why a buyer should trust that answer.

You need to know what your plant has promised, what it can still sell, and why a buyer should trust that answer. Start from the generator: every number in this area belongs to one generator site and its physical PPAs — not a trader portfolio view.

Follow along. Seat: ipp_2@webill.net (password supersecret). Path: Generator sites → IPP2 Atlantis Wind → Strategy. You should see six tabs, left to right: Overview · Sales · Allocation · Curtailment · Plant · Dashboard. Allocation is one tab; it opens on Rules · hierarchy, with a mode switch for Forecast · next month and Outcomes · settled.

Data. full_demo · seed 42 · IPP2 Atlantis Wind · stage S03/S07 · last checked 2026-09-13.

3.1.1 Find the right generator first

1. Open IPP Sites in the org navigation.
2. Select IPP2 Atlantis Wind (not a trader pool, not an off-taker).
3. Choose Strategy in the site navigation — the plant name stays in context as you change tabs.

The site boundary matters: its meter supplies generation, its active physical PPAs supply eligible buyers, and its hierarchy supplies the allocation rule. Mixing plants would break conservation checks later.

If you arrive from a contract register or a counterparty link, follow the generator deep-link rather than guessing from a commercial total.

3.2 Read the tabs in order

The strategy bar is six tabs, left to right. Each tab answers one question. Treat them as a walk, not a menu you skip around: size the gap, try to sell it, then prove who is served, recover a held-back hour if one happened, check what the plant is doing now, and only then read the plant’s story since it switched on.

flowchart LR
  A[Overview] --> B[Sales]
  B --> C[Allocation]
  C --> D[Curtailment]
  D --> E[Plant]
  E --> F[Dashboard]

The old stand-alone HAF (Historical Availability Factor), Hierarchy, Forecast and Allocations tabs have not gone away — their screens now live inside Sales and the Allocation tab, and the old addresses still open the right place.

Think of Allocation as one folder with three sheets. The bar shows the folder once. Inside it, #allocation-mode-bar is the three-way switch: Rules · hierarchy (the written order of who is served), Forecast · next month (what that order leaves to sell), Outcomes · settled (the receipt of a month that already ran). Clicking Allocation always opens Rules (…/strategy/hierarchy). Hover the tab for the same three-line hint the product shows.

Four ways through the area:

1. Sell the gap: Overview (how big, how much it is worth) → Sales (who could absorb it, what the deal earns against holding for SAWEM — the South African Wholesale Electricity Market) → promote a draft into V7.
2. Stand behind promises: Allocation (Rules, then Outcomes, then Forecast) → Dashboard health check → contracting guard (§13).
3. Recover a cap: an ended System Operator cap opens a draft on Curtailment — review, lodge, record payment (§10).
4. See every plant as one picture: IPP Sites → Fleet — not a sixth strategy tab. That four-chapter story is §2.

Note. Bookmarks to the retired HAF tab open Sales; the safe capacity widget is in the Sales pricing tools and its headline is also a link on the Overview gap hero. Bookmarks to Hierarchy, Forecast and Allocations open the matching Allocation mode, with that one tab marked active.

3.3 What you can change

All pages are org-scoped. Another organisation’s generator is not reachable here.

Some seats are read-only: you see the cascade, reports and traces, but Publish, Re-run and acknowledgement controls are hidden and the Rules board shows a Read-only badge. That is not a failed save — it separates commercial review from actions that version a rule or mint an auditable report. Ask a generator administrator to make the controlled change, then return to inspect the new version.

3.4 First five minutes

Where this is taught. Lifecycle spine: V0 §4. Revenue-gap Overview: §4. Allocation rules: §7. Settlement consumption of the carve: V8 §6. Contracts: V7. V3 pointer chapter: V3 §14.

Previously in the thread → V3 §21 · demo walkthrough. Next in the thread → V4 §4 · Overview (then §6 · Sales · §7 · Allocation rules).

4. Overview — revenue-gap-first

Thread part 4, continued: IPP2 Atlantis Wind’s first screen — how much of the plant is still for sale, what that gap is worth, and whether what has sold pays for the whole plant.

Follow along. Seat: ipp_2@webill.net. Path: Generator sites → IPP2 Atlantis Wind → Strategy → Overview. You should see a one-line headline over a sold / unsold bar, a ranked list of next best actions below that card, then the panel When your energy is available — and what those hours earn (#ipp-generation-story): the 24-hour unsold profile, the plant-assumptions fold, and Value of your generation hours. Under that, three small panels: Who buys what you’ve sold, Weather outlook, and Price vs cost, on one line. Explanations (#overview-explanations-open) sits on the assumptions strip — §4.8.

The IPP’s problem, in plain words

You run a plant. Every hour it makes energy; some of it has a buyer under a PPA and the rest is spilled — generated, metered, and paid for by nobody. The trap is that the numbers a plant is usually judged by hide the spill: the price on the energy that did sell can look healthy while the plant as a whole earns far below its cost.

Overview exists to keep the spill in the headline. It answers, in this order: how much of the plant is sold and how much is still for sale; what the unsold part is worth and what it earns today (always R 0); what the one next best action is; which hours the gap sits in; whether the hours you do generate have been historically cheap or dear against the market; who is buying what has sold; whether the next three days of weather will give you energy to sell at all; and whether the whole plant covers its cost. Everything on the page except the generation-hour card is built from the latest forecast run — the strategy engine’s allocation of a typical day of generation across your contracts — so it is a picture of the book as it stands, not a settled statement. The generation-hour card is a historical market-price check on the same plant; it is not this run’s contracted earnings.

What you see (top → bottom)

On the site workspace these panels open the Performance tab (§3); a fold under them, Measured performance & revenue analysis, holds the timeline, revenue sources, losses, availability and forecast evidence.

Concepts and abbreviations on this screen

The revenue-gap trap — achieved can look fine while realised is red

Imagine a bakery that bakes 100 loaves a day but only has customers for 60. The baker can truthfully say “I sell bread at R 65 a loaf” — that is the achieved price on the loaves that sold. But 40 loaves went to waste and earned nothing, so the bakery’s true income per loaf baked is only 0.6 × R 65 = R 39. If each loaf costs R 48 to bake (LCOE), the shop is losing money even while the sales team celebrates R 65.

A power plant works the same way. Sunshine and wind arrive whether or not you have a buyer. Celebrating the price on the sold slice while ignoring the unsold mountain is how plants quietly under-earn their LCOE.

Worked example — why “Realised” is the honest number

Data. Illustrative solar plant sized for teaching the KPI maths (same numbers the Overview cards walk). Demo walkthrough on the seeded cast: §21.

Step 1 — revenue from what sold.

revenue = 26,280 MWh × 650 R/MWh = R 17,082,000

Step 2 — achieved price (what sales celebrates).

achieved = revenue ÷ sold MWh = 17,082,000 ÷ 26,280 = **650 R/MWh** (65 c/kWh)

Unsold hours never enter this fraction — so achieved stays high even when most of the plant is idle commercially.

Step 3 — realised price (what the plant actually earns per MWh generated).

realised = revenue ÷ *all* generation = 17,082,000 ÷ 43,800 = **390 R/MWh**

Or equivalently: realised = sold fraction × achieved = 0.6 × 650 = **390 R/MWh**.

Step 4 — compare to LCOE.

LCOE is 480 R/MWh. Realised 390 is 90 R/MWh below cost. The plant is underwater at the realised level while the sales team still quotes 65 c.

The page’s price rail (§4.7) shows the same three prices in c/kWh — realised 39 c, LCOE 48 c, achieved 65 c — because 1 c/kWh is 10 R/MWh. Its verdict reads the gap the same way: under-earns by 9 c/kWh.

Step 5 — fill the gap (even at a modest price).

Suppose you sell the remaining 17,520 MWh at a modest 55 c/kWh (= 550 R/MWh) — below your current achieved, but above zero:

extra revenue = 17,520 × 550 = R 9,636,000

new total revenue = 17,082,000 + 9,636,000 = R 26,718,000

new realised = 26,718,000 ÷ 43,800 ≈ **610 R/MWh** — now above the 480 R/MWh LCOE.

The lesson the two cards are glued together to teach: filling unsold volume at a “worse” unit price can still rescue the plant, because realised cares about every MWh generated, not only the pretty ones.

4.1 The gap hero and the next best actions

The top of the page is one sentence and one bar. The sentence is written from the numbers — “Sold 4 %, unsold 96 % — R 5.9 m a month spilling” on a thinly contracted plant; “This plant is fully contracted” on a full book. The bar underneath shows the same split: solid green for sold, hatched amber for still for sale, with a marker at the sold share.

Two columns sit under the bar.

Buyer organisations counts the distinct organisations paying for this plant’s energy in the forecast run, for example 2 · 1 dominant (1 · sole buyer when there is one; 1 dominant when one of several takes more than half the money). One organisation can hold several contracts, so the figure can be smaller than the site’s contract count.

Below the columns, when there are at least two forecast runs, a sparkline shows the unsold share run by run with a one-line verdict — “stable”, “shrinking” or “growing”.

Next best actions

The panel below the hero ranks up to three actions by the rand at stake and links each to the screen where the work is done:

When none fire the panel reads Fully contracted at a healthy margin.

4.2 Unsold profile (24h) and the three small panels

The chart stacks Contracted under Unsold for each hour of a typical day, with a day/night wash so the solar window is obvious. Two period buttons switch the view:

• Typical day — the forecast run’s average day, with tomorrow’s P50 forecast drawn as a dashed line over it. Where the dashed line sits above the stack, the weather says you will have more energy to sell than the run assumed; where it sits below, less.
• Tomorrow (P50) — the weather-driven profile itself, when a forecast has been issued.

The intensity strip under the chart is the same unsold energy as one row of 24 cells, darker where more energy has no buyer. It is the shape the Sales desk scores buyers against.

Under the fold, still inside this same panel, Value of your generation hours (#ipp-capture) asks a different question: when this plant did generate, were those hours cheap or dear against the market? It is history, not this run — taught in §4.4. The three small cards under the panel — buyers, weather, price — then put a name, a weather check, and a cost check on what the run did sell (§4.5, §4.6, §4.7).

How to read it (step by step)

1. Find the tallest amber. That is the hour to sell first.
2. Compare the amber to the dashed P50 line. Tall amber under a low line means the gap is partly a forecast artefact; tall amber under a high line is real, sellable energy.
3. Read the Concentrated in row on the hero. If it names a daytime window, look for daytime load on Sales; if the amber is evening-heavy, a pure daytime factory will not finish the job.
4. Open Solar PV / Wind / … assumptions & forecast basis (#ipp-plant-assumptions) and check whether the chart is using a captured profile or the engine default (below).
5. Read Value of your generation hours (§4.4). A high capture % with a tall amber stack means the hours are valuable but nobody has bought them yet. A low capture % with a full book means you sold into cheap hours.
6. Then the three small panels: who takes the money (§4.5), whether the next three days have energy to sell (§4.6), and whether realised covers LCOE (§4.7).

Plant assumptions & forecast basis

The 24-hour chart is the plant on the road. The fold under it is the spec plate on the door — the engineering facts the forecast uses when it has no better measured series. A solar site that faces west does not spill at the same hours as one that faces north. A wind site with turbines down for maintenance is not the same plant as the nameplate on the directory card. This fold is where Overview shows that difference.

Follow along. Seat: ipp_2@webill.net. Path: Generator sites → IPP2 Atlantis Wind → Strategy → Overview, then the panel When your energy is available — and what those hours earn. You should see a green Wind chip (#ipp-profile-technology) and a closed fold Wind assumptions & forecast basis. Click it. The card #ipp-overview-technology opens inside the fold. On IPP1 Sandton Solar the same fold is titled Solar PV assumptions & forecast basis.

Data. full_demo · seed 42 · IPP2 Atlantis Wind · last checked 2026-09-13.

Where it lives, and what the title is

The fold sits at the bottom of the Unsold profile panel (#ipp-generation-story), under the chart (or under the empty “Run a forecast to see when your generation is contracted or still for sale” line if no snapshot exists yet). The fold is closed until you click it. Capture — Value of your generation hours — sits under the fold, still in the same panel (§4.4). A thermal Dispatch economics card is not in this fold; it is the first card on Overview and is taught in §18.5.

The summary always reads Solar PV assumptions & forecast basis, Wind assumptions & forecast basis, and so on. The name is the same as the chip next to Unsold profile (24h):

flowchart TB
  A[Open Unsold profile] --> B[Read the technology chip]
  B --> C["Click Solar PV / Wind / Hydro assumptions & forecast basis"]
  C --> D{Is a plant profile captured?}
  D -->|Yes · amber pill says profile| E[Read P90/P50 and the fact tiles]
  D -->|No · amber pill says default| F[Engine uses the technology default]
  F --> G[Add profile on IPP Sites]
  E --> H[If a fact is wrong, edit the profile. This fold does not save]

1. Scroll to When your energy is available — and what those hours earn.
2. Read the technology chip. That is which recipe the fold will open.
3. Click Solar PV / Wind / … assumptions & forecast basis.
4. Read the amber pill: profile (you captured facts) or default (the engine is using its built-in shape for that technology).
5. Read P90/P50. That is how much the conservative forecast is allowed to sit below the expected one.
6. Scan the tiles that shape hours (tables below). Treat pad sizes, SCADA strings and nested layout maps as the plant notebook, not as the reason the 09h bar is tall.
7. If the pill says default, follow Add profile to IPP Sites (/org/ipp-sites/sites) and capture the facts in the wizard’s Plant profile step (V2 §4.7, §18).

What the card is for

A plant profile is the optional engineering record on the generator site: how the panels sit, how many turbines stand, how the river is meant to run, what the thermal machine costs to start. The forecast engine already has a default shape per technology (solar peaks in daylight, wind is flatter, hydro run-of-river follows the month, a thermal peaker has a dispatchable day). Capturing a profile does not rewrite a snapshot you have already sealed. It changes the input the next Run Forecast Now (and the nightly run) will use, unless a measured or nominated series already outranks it (§17, §18.2).

Think of it as writing the recipe on the tin. The bakery still has to bake (Run Forecast Now) before Overview’s 24-hour chart changes.

The chrome, word by word

A tile whose value was never supplied shows Not supplied and the missing tag. That is honest emptiness, not a failed load. Nested maps (layout blobs) may print as a technical list — they are stored engineering, not something you edit in this fold.

This fold is read-only. There is no Save here. Changing a fact means editing the site on IPP Sites, then running a new forecast if you need the 24-hour chart to move.

P90 / P50 on this card

P50 / P90 here is a generation grade, not a market-price band. P50 is the middle case: in half of similar months the plant makes at least this much. P90 is the conservative case: nine times in ten it makes at least this much, so P90 sits below P50. The card prints their ratio (P90 ÷ P50). A ratio of 0.86 means the conservative forecast is 86 % of the expected one.

When no profile is captured, the engine uses a technology default:

When a profile is captured and it carries Availability pct, the ratio is the lower of that default and availability ÷ 100. Example: wind default 0.86, availability 97 % → still 0.86. Wind availability 80 % → 0.80. Declared availability can only tighten the conservative case; it cannot loosen it past the technology default.

A live weather model that already supplies its own P90/P50 spread wins over this ratio (§17.3).

Tiles that shape the hours (by technology)

Open the fold on the technology you actually run. The chip tells you which table to use. Capture and validation of these fields is V2 §4.7 and §18; this page only teaches how to read them.

Solar PV

Wind

If turbine count × rated MW is more than 1 % away from the site’s installed capacity, the amber warning appears. That is a data-quality check, not a forecast rewrite. Review the engineering data on IPP Sites.

Hydro (including run-of-river and pumped storage)

Thermal (diesel, gas, biomass, coal)

Hybrid

The conservation line under a hybrid card is one January weekday: source energy minus storage losses and clipping should equal the generation shown. §18.3 owns that model.

What this fold does not do

Worked walk — IPP2 Atlantis Wind

On the seeded wind site the chip reads Wind. Opening the fold shows Wind assumptions & forecast basis. The P90/P50 line uses the wind default 0.86 unless a captured availability is tighter than 86 %. If the site has no wind_profile yet, the dashed empty state appears and the pill says default — the 24-hour stack still draws, from the engine’s wind shape, not from a blank chart.

That is the pattern on your own plant: chip names the recipe, the fold shows whether you wrote it down, and missing tiles tell you which facts are still a guess.

Where this is taught. Filling and validating the profile: V2 §4.7 and V4 §18. How weather turns irradiance or hub height into megawatt-hours: §17. How to read the capture card under this fold: §4.4. The cannibalisation maths: §19.

4.3 LCOE & params

This is the IPP’s cost dial on Overview: two numbers that say what the plant costs to produce and what it would cost to buy energy if you promised firm hours the plant could not make. Everything else on the page — Realised, Unsold, the 24h profile — is about what happened. These two dials are about what “good” means.

Where it lives. Top of Overview, next to Run Forecast Now: an assumptions strip shows the current values, and ⚙ LCOE / params opens the modal titled “LCOE & assumptions” (#ipp-params-modal).

Short steps — open, edit, save

1. On Overview, read the strip: LCOE … c/kWh · replacement … c/kWh (seed defaults: LCOE 48 c/kWh · replacement 210 c/kWh).
2. Click ⚙ LCOE / params (cog). The modal LCOE & assumptions opens.
3. Edit one or both number fields (step 0.1). Leave a field alone if you only need to change the other.
4. Click Save to write the book, or Cancel / the × close control to discard. Success flash: “Assumptions updated.” Failure flash: “Could not update assumptions.”
5. Re-read the strip and the price rail (realised vs LCOE). Press Run Forecast Now only when the sealed snapshot (Realised, Unsold %, profile) itself must rebuild — not merely because you moved the LCOE yardstick.

Save vs Run Forecast Now

Rule of thumb. Save = store the cost dials (strip, vs LCOE, and the “below LCOE” next-best prompt update immediately). Run Forecast Now = rebuild how much energy sold / earned — needed after membership or contract changes, not after an LCOE-only edit.

Each field in detail

The modal exposes exactly two fields (plus Cancel / Save). Seeded defaults come from the IPP book seed (WB.Strategy.Book.SeedIpp):

Why change either dial (or both)

What LCOE / params do NOT change (limits / misconceptions)

Worked micro-example — moving LCOE only

Suppose the sealed snapshot still shows Realised 390 R/MWh (same plant as the §3 worked example) and the strip still shows seed LCOE 48 c/kWh:

Realised stayed 390 the whole time — only the yardstick moved. That is why Overview treats LCOE as a dial, not as generation data.

Follow along. Seat: an IPP operator on the demo cast (e.g. open Strategy from an eligible row on IPP Sites). Path: Overview. You should see the sold / unsold bar, Achieved in the Contracted column, Unsold on the Still for sale side, and realised vs LCOE on the price rail. Open ⚙ LCOE / params — both fields should show the seed defaults (48.0 / 210.0) until someone saved different values. Compare with a trader pool Overview (V3 §3) — traders lead with margin; IPPs lead with the revenue gap. Off-taker ⚙ Assumptions is the buyer twin (escalation % · spill credit) — V3 §13.1.5.

Data. full_demo · seed 42 · stage S09 (settlement matrix) · last re-derived 2026-07-18 against the full_demo manifest. Extreme gap numbers on a thinly contracted demo plant are walked in §21.

Safe contractable headroom (AES / SCE / HAF band) shows on the gap hero’s Safely contractable row and in full on Sales — §5.

The Unsold profile fold Solar PV / Wind / … assumptions & forecast basis is taught in §4.2. How to capture those facts, and thermal Dispatch economics, stay in §18.

4.3 The Capture card

How to read Value of your generation hours on Overview is taught in §4.4. The cannibalisation maths, floor-hour policy and market-indexed vPPA comparators stay in §19.

This heading is kept so older links still land. Do not treat it as a second card.

4.4 Value of your generation hours

The 24-hour chart says when the plant makes energy. This card, still inside When your energy is available — and what those hours earn, asks what those hours were worth on the market. Title: Value of your generation hours. Subtitle on the right: Capture · daily benchmark (D1). DOM id #ipp-capture.

Think of a fruit stall that only opens at noon, when every other stall is also open. Noon fruit is cheap because the street is flooded. A stall that is open at dawn, when almost nobody else has fruit, earns more per box even if it sells the same number of boxes. This card is that comparison for a power plant: did your generation coincide with cheap hours or dear ones?

It is not what your PPAs paid you. It is not the forecast value of unsold energy on the gap hero. The intro on the card says so: Historical market value of your generation hours, not contracted earnings or the forecast value of your unsold energy.

It is also not Bid Desk capture rate (accepted ÷ offered volume). Same English word, different job. Here capture means: the price your generation hours earned, as a share of the average market hour. The glossary name is generation-hour value.

flowchart TB
  A[Hours the plant generated] --> C[Your generation hours R/MWh]
  B[Every market hour that day] --> D[Market average R/MWh]
  C --> E[Capture % = yours ÷ average]
  D --> E
  E --> F[Below 100%: you generated in cheaper hours]
  E --> G[At or above 100%: you generated in dearer hours]

What you see, line by line

How to read the percentage

Capture 100% means your hours were worth the same as the average hour. 84% means they were worth about sixteen percent less — the plant generated when the market was cheap. 110% means the plant generated when the market was dear.

The bakery picture in §3 still holds, but it answers a different question. Realised asks “did the whole plant earn its cost, counting unsold hours as R 0?” Capture asks “when it did generate, were those hours the cheap ones?” You can be fully contracted (realised healthy) and still capture poorly, if the contracts sit in a glut. You can capture well and still spill, if nobody has bought the dear hours yet.

Data. The numbers below are the same mock summer / winter pair §19 uses, so the card and the maths chapter agree. A measured estate can differ.

Summer solar is the glut case: many plants make power together, those hours cheapen, capture falls. Winter is closer to the average hour.

Empty states and pills

The pills can appear together. A published demo price does not turn a profile into metered generation.

Floor hours

Some hours the market price sits on a floor — the lowest the rules allow. The coverage line names how this row treated them:

Changing the convention changes the answer. Old rows keep the convention they were built with.

Where this is taught. How to read this card: this section. The cannibalisation maths, floor-hour policy, market-indexed vPPA comparators and the technology strip: §19. The same card title Capture (not embedded) also appears on Sales and the Dashboard; the numbers are the same daily benchmark.

4.5 Who buys what you’ve sold

Three small cards sit under the generation-story panel, in this order: buyers, weather, price. The first is Who buys what you’ve sold (#ipp-buyers). Intro (#ipp-buyers-intro): Revenue share by buyer over the forecast run — the basis of the revenue HHI.

If you sold 100 loaves and 80 of them went to one café, you do not have three customers in any useful sense — you have one customer and two hobbies. This panel is that picture for the plant: who takes the money from what the latest run already sold, and how concentrated that money is.

It only counts priced ALLOCATION lines — energy that found a buyer at a sale price. Spill (unsold) is not a buyer. A line with no sale price does not appear.

flowchart TB
  A[Priced ALLOCATION lines in the latest run] --> B[Group by destination membership]
  B --> C[Name: PPA other side, else pool, else Unnamed buyer]
  C --> D[Sum rand, largest first]
  D --> E[Bar per name + share % + R]
  A --> F[Snapshot Revenue HHI]
  F --> G[Meter 0 to 1.0]

Each row

When there are priced sales, each row is:

Rows are largest first. Two memberships that resolve to the same organisation name are added together on the bars.

Empty state: No priced sales in the latest run yet. The plant may still be generating — those hours are unsold, so they have no buyer to name.

The HHI meter

Under the bars: HHI as a two-decimal number, a green → amber → rose bar from 0 to 1.0, a tick at the current value, and the legend 1.0 = one buyer.

HHI (Herfindahl–Hirschman Index) here is Revenue HHI: the sum of each buyer’s squared share of revenue. It lives on the latest risk snapshot. Overview does not recompute it from the bars; the bars and the meter both start from the same run.

The meter is the same number as the old Revenue HHI KPI. Next best actions fire Diversify: buyer takes … of revenue when HHI is greater than 0.5, and send you to Allocation · Rules (§7).

Careful. The bars name organisations (memberships with the same display name are merged). The meter counts destination memberships in the snapshot. If two memberships of the same buyer keep different names — one a pool, one an organisation — the bars can look more concentrated than the meter, or the other way round. Believe the names for “who”, and the meter for “how concentrated the snapshot thinks you are”.

Worked walk — two buyers

Suppose the latest run sold R 1,000,000 of priced energy: OffTaker C Midrand DC takes R 700,000 (70%), a second buyer takes R 300,000 (30%).

1. The first row is OffTaker C, 70% · R 700 k, green bar.
2. The second row is the other name, 30% · R 300 k, slate bar.
3. HHI = 0.70² + 0.30² = 0.58. The tick sits just past halfway, into the amber.
4. Because 0.58 > 0.5, next best actions include diversify.

That is the pattern on your own plant: largest name, then the rest, then whether the meter has crossed one-half.

Where this is taught. Reading this panel: this section. The lifetime sources ledger on the Dashboard (every physical PPA since commissioning, plus RECs and SAWEM with their zeros): §12.4. How allocation decides which membership is filled: §7. Fleet-wide counterparty exposure: §2.

4.6 Weather outlook

The middle small card is Weather outlook (#ipp-weather-outlook). Intro (#ipp-weather-outlook-intro): Will there be energy to sell? P50 generation for the next three days against what is already committed.

The 24-hour unsold chart is a typical day. This card is the next three calendar days of weather-backed generation. A windless Tuesday does not care that your typical Tuesday is windy. If tomorrow’s P50 sits under what you have already sold into those hours, you have nothing extra to offer — and you may be short against the book.

P50 here is a generation grade: the expected case, output met or exceeded about half the time. It is not a market-price P50 from the Bid Desk.

flowchart TB
  A[Weather feed for this site] --> B[P50 MODEL series]
  B --> C[Tomorrow tile]
  B --> D[Next weekday tile]
  B --> E[Day after that]
  F[ALLOCATION kWh in the latest run] --> G[Committed on each tile]
  C --> H[Compare P50 to committed]
  D --> H
  E --> H
  H --> I[Amber if this day is under 80% of the best of the three]

Grade, freshness, skill

Above the tiles, when a series exists:

The skill half of that line:

nMAE (normalised mean absolute error) is “how far off, as a share of typical output”. Smaller is better. P90 coverage is the share of hours where actual generation was at or above the conservative P90. If the P90 is well built, that share sits near 90%. N hours is how many hours went into those two scores.

After the middle dot, the P90 origin, in words:

Overview and the Dashboard print the same English labels.

The three tiles

The card always looks three days ahead from now (not including today). Empty days — no weather-backed model points for that calendar date — are dropped, so you may see fewer than three tiles. A day the provider has only partly issued keeps its tile but no total: the forecast horizon (72 hours by default) often ends a few night hours into the last day, and that day shows — with Forecast not issued yet · 2 of 24 hours in rather than a 0 kWh that would read like a dark day.

A day that is merely lower than committed is not automatically amber. Amber is relative to the other two outlook days: a still week with one calmer day flags that calmer day. Compare the big P50 to the committed line yourself: if committed is larger than P50, the book is selling more than this weather now says the plant will make.

Careful. The 24-hour chart’s contracted stack is a typical day from the run. The outlook tiles are calendar weather. If the run’s hours do not fall on tomorrow’s clock, committed on a tile can be small or a dash of kWh even while the typical-day chart is fully stacked. Believe the chart for the shape of the book; believe the tiles for whether the next three days have wind or sun.

Empty state (#ipp-weather-no-forecast): No weather forecast for this site yet — run a forecast to fetch one. Press Run Forecast Now on Overview, or wait for the nightly run. The card will not draw three zeroes and call them a forecast.

Footer: Weather data by Open-Meteo.com (CC BY 4.0) — the attribution the live default provider requires, shown whether tiles are present or not. The demo estate uses a mock that never opens a socket; the sentence still names the licensed source the live adapter uses.

Worked walk — one quiet day

Suppose the three P50s are 12.0 MWh, 5.0 MWh, 11.0 MWh, and each day has 4.0 MWh already committed.

1. Best of the three is 12.0. Eighty percent of that is 9.6.
2. Tomorrow (12.0) and the third day (11.0) stay ordinary.
3. The middle day (5.0) is under 9.6, so it goes amber with Lower output · 4.0 MWh committed.
4. 5.0 is still above 4.0 committed, so there is a little extra to sell even on the quiet day — just much less than on the other two.

If that quiet day’s P50 had been 3.0 MWh against 4.0 committed, you would be short against the book on that calendar day, whatever the typical-day chart says.

Where this is taught. How irradiance or hub-height wind becomes megawatt-hours, P50/P90 grades, issued-at versus fetched-at, and live versus mock providers: §17. Weather as a platform object: V0 §11. The same panel assembler also feeds the Dashboard weather card.

4.7 Price vs cost, on one line

The right-hand small card is Price vs cost, on one line (#ipp-price-rail). Intro (#ipp-price-rail-intro): Realised counts unsold hours at 0 c — that is why it sits so far below what buyers actually pay. All prices in c/kWh.

Four prices, one ruler, all in c/kWh — the same unit as the assumptions strip and the LCOE modal. Each mark prints its value in bold — for example 3.5 c, 48 c, 85 c or 210 c — with its name (realised, LCOE, achieved, replacement) next to it. The bakery from the top of the chapter is this picture: the price on the loaves that sold, the cost to bake, the true income per loaf baked, and what it would cost to buy bread in if you had promised a café a delivery you could not bake.

Marks with a zero value are omitted, except realised. They are sorted left-to-right by size. Labels alternate above and below the rail so they do not sit on top of each other, and the labels nearest either end read inwards so they stay inside the card. The rail’s right-hand end is the largest mark — so replacement (seed 210 c) often sits far right and bunches realised, LCOE and achieved on the left. That is scale, not a statement that replacement is a sale price.

flowchart LR
  A[Realised] --> D[Verdict]
  B[LCOE] --> D
  C[Achieved] --> D
  E[Replacement] --> F[Scale only on this rail]
  D --> G[Plant covers its cost?]
  D --> H[Sold energy above LCOE?]
  D --> I[Unsold the main leak?]

The verdict sentence

Under the rail, one or two sentences (price_verdict), also in c/kWh:

First sentence — the sold slice.

Second sentence — the whole plant. The wording depends on realised versus LCOE and how large the gap is:

X is LCOE minus realised in c/kWh, to one decimal (a whole number drops the decimal). P is the unsold share. A thinly sold demo plant reads, for example, Energy that sells earns 1.8× LCOE. The plant under-earns by 44.5 c/kWh mainly because 96% of it earns nothing. That second sentence is the revenue-gap trap from the top of the chapter, in one line.

Worked walk — the same 43,800 MWh plant

Using the chapter’s worked plant (60% sold at 65 c, LCOE 48 c, replacement 210 c):

The rail runs to 210 c, so 39 / 48 / 65 sit in the left third. Verdict:

1. 65 ÷ 48 ≈ 1.4, so Energy that sells earns 1.4× LCOE.
2. Unsold is 40% (> 20%), and 48 − 39 = 9, so The plant under-earns by 9 c/kWh mainly because 40% of it earns nothing.

That is exactly Steps 3–4 at the top of this chapter, drawn.

Saving a new LCOE in ⚙ LCOE / params moves the LCOE mark and rewrites the verdict without a new forecast run. Realised and achieved stay put until Run Forecast Now. Replacement on this rail is the same dial; Overview’s realised-versus-LCOE comparison does not use it — Firm viability does (§4.3, §6.8).

What this rail is not

The five snapshot figures that used to sit as cards under a fold — Realised, Achieved, Unsold, Revenue HHI, vs LCOE — are the same numbers already on this page: unsold on the gap hero, achieved in the Contracted column, HHI on the buyers meter, realised / LCOE / achieved on this rail.

4.8 Explanations sidebar

Every abbreviation on this Overview also lives in a right-hand drawer so you do not have to leave the page to look a letter-cluster up.

Explanations (#overview-explanations-open) sits on the assumptions strip, to the left of ⚙ LCOE / params and Run Forecast Now. It does not start a forecast and it does not open the assumptions modal.

Think of it as the card catalogue for this one screen: press the button, find the word, close the drawer, keep working.

flowchart LR
  A[Press Explanations] --> B[Right-hand drawer]
  B --> C[Scroll A–Z]
  C --> D[Close X / backdrop / Escape]
  D --> E[Overview as you left it]

The drawer is the same chrome as Allocation Explanations and the site-card drawer on Generator Sites: it opens on the right, branded green header, title Explanations, X top-right. Eyebrow: Generator Overview. Close: header X (#overview-explanations-drawer-close), the dimmed backdrop, or Escape.

The body is one alphabetic list (leading “The” ignored, case-insensitive). First term Achieved, last Worth at LCOE. Each abbreviation is spelled out on first use in its own row — AES (Available Energy for Sales), HAF (Historical Availability Factor), LCOE (Levelised Cost of Energy), SCE (Safe Contractable Energy), SAST (South African Standard Time). The OffTaker Overview uses the same button and drawer with a different list (V5 §4.11).

The list also carries the words this tab borrows from its neighbours: ARF (Allocation Reliability Factor) — the share of a buyer’s demand actually met, shown on each Allocation member card; Buyer organisations — distinct organisations paying in the forecast run, which can be fewer than the contracts; and VaR 95 (Value at Risk) — the fleet dashboard’s risk estimate. The tier word is Pro-rata, spelt as the Allocation tab spells it (never PRO_RATA): Safely contractable quotes next month as a Pro-rata buyer. R/MWh says 10 R/MWh is 1 c/kWh, and that the capture benchmark reads in R/MWh while the price rail, the Reprice or re-contract next best action and the tab’s other prices are c/kWh. Realised says unsold hours count as 0 c.

Spot-check ids: #overview-explanations-term-aes-available-energy-for-sales, #overview-explanations-term-lcoe-levelised-cost-of-energy, #overview-explanations-term-sce-safe-contractable-energy, #overview-explanations-term-arf-allocation-reliability-factor, #overview-explanations-term-buyer-organisations, #overview-explanations-term-var-95-value-at-risk.

The concepts table above is still the teaching home for why realised differs from achieved. The drawer is the lookup.

5. HAF — the capacity check

Thread part 4, continued: before IPP2 Atlantis Wind promises a buyer another block, how much can it safely still sell?

The seller’s problem, in plain words

A buyer asks: “Can you give me 4 MW firm next month?” Your paper answer is the forecast’s Available Energy for Sales — what is left after the book you already hold. But paper availability is not bankable availability. A plant that made 70 % of its forecast last quarter, or whose lower tiers were starved in three of the last six months, should promise less than its AES says.

HAF — the Historical Availability Factor — is the discount that turns paper availability into a bankable number. It multiplies four factors, each in [0, 1]: how much of the book the plant physically covered (GAF), how stable its generation has been (SF), how reliably each tier was actually served (ARF), and how exposed the buyer’s chosen tier is (CSF). The result, applied to AES and capped by capacity headroom, is SCE — Safe Contractable Energy: the one sentence a prospective buyer gets, “you can safely contract another N MWh next month”.

How to view the HAF

HAF (Historical Availability Factor) is not a tab of its own. The strategy bar is Overview · Sales · Allocation · Curtailment · Plant · Dashboard (§3.2). There is no HAF button to click.

Think of Overview as the shop window: one number that says how many loaves you can still honestly promise. Sales is the back room: you pick which promise level the buyer would sign under, and you see why that number is that size. The Dashboard fold is the notebook that explains a weak month.

Follow along. Seat: ipp_2@webill.net (password supersecret). Path: Generator sites → IPP2 Atlantis Wind → Strategy → Overview. In the gap hero’s Still for sale column, find Safely contractable (#ipp-safely-contractable). Click that number. You land on Sales. Then click Firm viability · safe capacity (#subtab-firm-btn) in the Pricing tools bar (#sales-pricing-tools). Scroll to Safe contractable capacity (#sales-capacity). You should see the capacity widget — or the grey No data · No history chip on a brand-new site.

flowchart TB
  A[Overview · Safely contractable] --> C[Sales]
  B[Sales tab] --> D[Pricing tools]
  D --> E[Firm viability · safe capacity]
  C --> E
  F["Old bookmark …/strategy/haf"] --> C
  E --> G[Capacity widget: SCE · band · factors · trend]
  G --> H[Dashboard health fold if you need why]

From Overview — the headline

1. Sit as an IPP operator. On the demo that is ipp_2@webill.net.
2. Open IPP Sites (or Generator sites) → IPP2 Atlantis Wind → Strategy. Overview is the landing tab. URL: /org/generator-sites/<site-id>/strategy.
3. In the gap hero, Still for sale column, read Safely contractable. That is next month’s SCE (Safe Contractable Energy) for a Pro-rata buyer: the band chip, then a line such as 5,752 MWh next month as a Pro-rata buyer.
4. Click the number (#ipp-safely-contractable). It opens Sales (/org/generator-sites/<site-id>/strategy/sales) so you can choose the tier.

From Sales — the full widget

1. Same seat and plant. Click the Sales tab, or arrive from the Overview link above.
2. Find Pricing tools (#sales-pricing-tools) above the deal builder. Two buttons: TOU pricing and Firm viability · safe capacity.
3. Click Firm viability · safe capacity (shield icon, #subtab-firm-btn). It opens as a drawer from the right, over the desk; Escape, the X or a click on the dimmed page closes it.
4. The firm-block card comes first in the drawer (#sales-firm-tool — that card answers “should I sign this evening block?” and is taught in §6.8). Scroll down the drawer to Safe contractable capacity (#sales-capacity). Intro id #sales-capacity-intro. The widget itself is data-role="capacity-widget".
5. Guarantee lines, if any, sit in #haf-availability above the widget. With no enabled guarantee the box says No enabled guarantee.

Old bookmark

/org/generator-sites/<site-id>/strategy/haf still exists so old links keep working. It flashes #ipp-haf-redirect (Redirecting to Sales…) and opens Sales. It does not reopen a stand-alone HAF tab.

What you see when the widget is open

You are in the right place when you see these pieces. This chapter already owns what each one means — this list is only so you can match the screen.

To ask why the score moved, open Dashboard → the health chapter → Availability detail (§5.8, §12.7). Compact chips of the same widget also appear on generator search cards.

When you meet HAF now

What you see in the capacity widget (top → bottom)

The rest of this chapter defines every term (§5.1), teaches the gauge and trend (§5.2, §5.3), walks the scenarios (§5.4), details the four factors (§5.5), works one plant to three tier answers (§5.6) and lays out the per-tier SCE menu (§5.7).

5.1 Every term and concept on this screen

Read these as glossary cards for the labels you will actually see. Full factor maths live in §5.5; the worked one-plant example is §5.6.

5.2 How to read the gauge

What the dial is. An ECharts gauge (gauge_chart) for the selected tier’s HAF, scaled 0–100 with unit %. There is no target tick on this screen — the fill is a simple progress arc to the needle (sky blue on a grey track). The band badge under the dial carries the Good / Moderate / … colour meaning; do not expect the dial rim itself to be striped into band colours.

Centered stack (top → bottom). Nothing sits beside the dial — the full widget stacks in one centred column:

1. Bold title: Historical availability factor (HAF)
2. SCE message: eyebrow You can safely contract another · huge N MWh (comma thousands) · next month as a {Tier} off-taker, the tier in words — next month as a Pro-rata off-taker
3. Gauge dial (centre %)
4. Band badge under the dial
5. Tier chips → evidence (and the obligations note, when it applies) → How this is worked out, folded → trend

Step-by-step

1. Read the SCE message above the dial. That MWh is what you can put in front of a buyer for this tier — already AES × HAF (and capacity-capped).
2. Read the centre %. That is HAF_tier × 100, rounded to one decimal (e.g. 56.3% for HAF 0.563).
3. Read the band badge under the dial. Map % → band with the table in §5.1. High / Good = grow carefully; Moderate = shrink offers / price risk; Low = stop growing; No data = no scored dial (see scenario below).
4. Click another tier chip. Gauge %, SCE message, band badge, CSF in the folded decomposition, and the dark chip all update. PRIORITY should show a higher % and SCE than PRO_RATA on the same plant.
5. Cross-check evidence. If AES is 0 or tiny, a pretty gauge still means “almost nothing left to sell.” If Remaining capacity is — or very low, firm offers are capacity-bound.

Good vs bad (cheat sheet)

Progress fill (HAF %) Band chip (meaning)

Good · Reliable — needle high, lime chip

Represents: PRIORITY-class score on the worked plant (~0.704). React: SCE is the menu; Sales may offer near that MWh for this tier.

Moderate · Caution — mid dial, amber chip

Represents: default PRO_RATA view on the §5.6 worked example (HAF 0.563 → SCE 845 MWh). React: sell, but size to SCE and expect the guard to warn above it.

Low · Not bankable — needle left, red chip

Represents: HAF under 0.50 — book too big, delivery weak, or both. React: pause new volume; open Dashboard / §5.5 to see which of GAF, SF, ARF is dragging.

5.3 How to read the trend graph

What the chart is. When the plant has monthly HAF history, a multi-line chart appears under the decomposition (multi_line_chart, id haf-trend-spark). If there is no history, the chart is simply not rendered (no empty-plot placeholder on this tab).

What each axis is

So: tier buttons change the gauge and SCE; the trend line is the plant’s PRO_RATA history — useful as a reliability spine, not as a per-tier rewrite.

Step-by-step

1. Find the latest month on the right (e.g. Dec. ’25). Compare its height to the gauge’s PRO_RATA reading — they should rhyme.
2. Scan left for dips. A sharp drop often marks a curtailed or oversubscribed month entering the trailing window (§5.8 · §12.7).
3. Check the Y scale. Values are % (e.g. 56, not 0.56). Three split lines are typical on the LiveView chart.
4. Do not confuse with Overview’s 24h Unsold profile. That chart is hourly kWh sold-vs-unsold (§4.2); this one is monthly bankability %.
5. If the chart is missing and the chip is not No data, history may still be thin on PRO_RATA points — use the gauge + evidence rows, then Forecast / backfill (§14).

HAF series (%)

How the axes read

Represents: X months as Mon. ’YY; Y percent; one HAF line. React: read slope and the latest point before you trust a single gauge snapshot.

5.4 Scenarios — what you might see, and how to react

Illustrative pictures only — numbers rhyme with the §5.6 worked plant where noted (AES 1,500 MWh · PRO_RATA HAF 0.563 · SCE 845 MWh).

HAF % Healthy / bankable Caution / weak

1 · Healthy Pro-rata leftover (Moderate)

Represents: default tab — SCE message above, gauge mid, amber badge under the dial, trend gently soft. React: take Overview’s unsold shape to Sales, but size new PPAs to ~845 MWh PRO_RATA (or click PRIORITY / FIXED for those menus).

2 · Same plant, PRIORITY selected

Represents: tier click only — CSF rises to 1.00; message, gauge and badge jump; decomposition’s last factor changes; trend line stays the PRO_RATA history. React: only quote the PRIORITY SCE if the contract will actually be PRIORITY.

3 · Trend dipping (curtailment entering the window)

Represents: monthly bankability sliding — often SF (volatility / curtailment) or GAF (book vs generation). React: do not size Sales off last summer’s SCE; open Dashboard factor bars and Forecast for the weak months (§5.5 · §13 alerts).

4 · AES gone — oversubscribed book

Represents: evidence row AES at 0 (or SB negative on Forecast). Gauge may still show a mid band from history, but headline SCE is ~0. React: stop new offers; fix the book / generation story on Forecast — HAF cannot invent leftover energy.

5 · No data — young plant

Represents: < 3 months of history — widget no_data state; no gauge, no scored SCE, no trend. Fallback ceiling uses AES (or FG) with an explicit caveat. React: contract carefully on paper ceilings; the guard stays quiet on HAF oversubscription until history exists (§13).

6 · Low band — do not grow

Represents: product of weak factors under 0.50. Decomposition shows where it hurts. React: freeze Sales growth; repair coverage / delivery / volatility before quoting new MWh — see §5.5.

How to use it (short path)

1. Open the widget as above: Safely contractable on Overview, or Sales → Pricing tools → Firm viability · safe capacity, after you already know there is a revenue gap, or when a buyer asks “how much can you really stand behind?”.
2. Select the tier you expect to offer (PRIORITY buyers see a higher HAF / SCE than PRO_RATA on the same plant — §5.6).
3. Treat the SCE headline as the menu size for new contracts; the contracting guard (§13) will compare proposal volume against this same evidence.
4. When the band is Low, AES is zero (oversubscribed), or the trend is falling hard, pause new offers and open Forecast / Dashboard for the underlying month.
5. For factor maths, bands and the annotated widget walkthrough, continue below (§5.5–§5.9).

Where this is taught. How to open the widget: above. Tab order: §3.2. Factor maths: §5.5. Widget annotation: §5.9. Guard: §13. Revenue-gap KPIs remain on §4 Overview.

5.5 The four factors, in detail

December says 1,500 MWh is for sale. Should a buyer believe it? HAF (Historical Availability Factor) is a score from 0 to 1 that discounts your paper availability for the things a buyer actually cares about — and it is contract-aware: a priority buyer sees a higher HAF than a pro-rata buyer on the very same plant, because they would experience a very different plant.

HAF is the product of four discounts, each clamped to [0, 1]. The clamp matters: each factor is a discount, never a bonus — a plant that outperforms is simply “fully available”, not more. And the product means a weakness in any one dimension compresses the whole score: a stable, well-allocating plant that is simply oversubscribed is not safe to contract, and neither is an under-subscribed but wildly volatile one.

Data. HAF worked values below are illustrative and use the §8 forecast example · full_demo seed 42 supplies the screen context · last checked 2026-07-18.

5.6 HAF worked — one plant, three answers

Continuing the same solar IPP into August:

• GAF — trailing generation (13,200 + 12,800 + 12,100)/3 = 12,700 MWh against subscribed demand 16,000: min(1, 12,700/16,000) = 0.794. The book is bigger than the plant — that alone knocks 21% off availability.
• SF — the 12-month series has σ = 454.6 around μ = 12,700 → CV = 0.036, so SF = 1 − 0.036 = 0.964. A steady plant.
• ARF — off-takers received 92% of expected demand over the trailing year: 0.920.
• CSF — depends on where the prospective buyer would sit.

Why does the score move? A curtailed month raises volatility in the detrended series and pulls SF down. An oversubscribed season starves the fixed tier and pulls its ARF down while GAF drops too. Every movement is traceable to settled evidence — never to a manual adjustment.

5.7 Safe Contractable Energy — the per-tier menu

SCE is the number a prospective buyer should trust — December’s 1,500 MWh of AES discounted by the HAF of the tier they’d be buying into, and capped by capacity headroom (energy without firm capacity is not sellable as firm):

SCE = min( AES × HAF_tier, remaining capacity headroom )

Read it as a menu: the same December surplus supports a 1,056 MWh priority deal, a 951 MWh fixed block, or an 845 MWh pro-rata share — the difference is purely how much protection each seat in the waterfall gives the buyer. The contracting guard (§13) compares proposal volume to this same SCE.

5.8 The availability detail on the Dashboard

The Dashboard’s Availability (HAF) block is folded under Chapter 5 (health check): gauge, tier fulfilment matrix, shortfall / spill log. Open the fold when the Availability · HAF vital sign is amber or red and you want to know which factor moved. Field guide: §12.7.

Each panel is taught in §12.7. The Availability vital sign above the fold summarises the same data in one number and a sparkline, and the availability slide row in the loss ledger prices any decline from the best month (§12.5).

5.9 The capacity widget — annotated

The widget is the seller’s answer to “how much can I promise?”, rendered identically wherever it appears. On Sales open Firm viability · safe capacity in the pricing tools bar; in the drawer, the widget sits under the firm-block card as Safe contractable capacity.

1. Tier selector. Click the tier the buyer would sign under. Only CSF changes between tiers; GAF, SF and ARF are the plant’s.
2. Headline. SCE for that tier, min(AES × HAF_tier, capacity headroom). On a PRIORITY promise it is larger; on PRO_RATA smaller.
3. Band chip. High · Bankable, Good · Reliable, Moderate · Caution, Low · Not bankable or No data · No history — the sales meaning of each is in §15.2.
4. How this is worked out. Folded under the evidence: the four factors, so a buyer’s adviser can check the arithmetic, and a legend of every abbreviation on the widget.
5. Badges. FINAL or PROVISIONAL with completeness %, and the capacity headroom in MW that caps the headline.
6. Trend. Twelve months of HAF; a downward run is the cue to open the Dashboard fold before quoting.

Careful. A firm block the buyer asks for must fit inside the SCE for their tier — the firm-viability card beside the widget checks price, the widget checks volume. The contracting guard (§13) repeats the volume check when the proposal is built.

Where this is taught. The Sales pricing tools: §6.7 and §6.8. Forecast AES before the discount: §8.1.6. The Dashboard fold: §12.7.

Guarantee versus measured availability

The Sales capacity section shows each active enabled contract’s guaranteed percentage, trailing measured percentage and periods below the guarantee. With no enabled guarantee, it says so; missing measurements are shown as no evidence. The old HAF route redirects here. GAF, stability, allocation reliability, tier factors and safe contractable energy retain their existing arithmetic.

6. Sales — the Sales Desk

Thread part 4, continued: IPP2 Atlantis Wind has energy nobody is buying. Who should buy it, on what terms, and is any of that better than selling it on the day-ahead market?

Follow along. Seat: ipp_2@webill.net. Path: Generator sites → IPP2 Atlantis Wind → Strategy → Sales. You should see a five-step rail across the top, the gap band with Option 0 · hold and sell on SAWEM, a candidate finder on the left (above the builder on a window narrower than 1280 px), and an empty deal builder that asks you to pick a candidate.

Data. full_demo · seed 42 · IPP2 Atlantis Wind · stage S09 · last re-derived 2026-09-03. The DAM price forecast the desk values the gap against is the trailing-quantile SMP forecast (WB.Brp.PriceForecast). Examples that name IPP1 Sandton Solar — forecast run v101, SIM Menlyn Park Load, SIM Gateway Umhlanga — were read on seat ipp_1@webill.net on 2026-09-28 and move with that seat’s data.

6.1 What this screen is for

The desk answers one question in five steps: given the energy I cannot place today, which deal — or no deal — leaves me best off after risk?

1. The gap and its market value. What you are selling and what the day-ahead market would pay for those same hours. The market is Option 0: the baseline every contract must beat.
2. Who could absorb it. Off-takers and traders ranked by fit to the unsold shape, with how much of the gap each would actually take and which contract routes are open to them today.
3. Build the deal. Capacity, strike, term and shape, seeded by the advisor, on the route you choose — physical PPA, Eskom VPPA or via a trader.
4. Profit and risk vs holding. One simulation, judged against Option 0: advantage and risk as two scores, five risk gauges, and the price at which the market would have been the better choice.
5. Compare and decide. Every simulation and the hold option on one table and one chart, ranked by balanced, advantage, lowest risk or certainty. Then Build Proposal — nothing here signs anything.

The old TOU pricing and Firm viability tabs are now pricing tools. Each opens as a drawer from the right, over the desk, so the terms it feeds stay where they are; they are taught in §6.7 and §6.8. The capacity widget sits inside the firm tool (§5).

On the site workspace the desk is the Commercial tab’s third panel, headed Sales desk (Sell available energy: find the next buyer, test the deal and compare it with holding for the market.), under open proposals and Contract commitments (#ipp-contract-commitments). There the step rail is left out and the gap band (step 1) is folded under Available energy, market pricing & risk context (#sales-market-context); once you open it, it stays open while you work. Contract commitments is the headroom check before you sell: four tiles — Physical PPAs, Virtual PPAs, Contracted energy and Remaining saleable energy — and View site contracts into Deals. Contracted energy takes its annual figure from the best evidence it has, in this order:

Whenever the figure is estimated, Remaining saleable energy (annual P50 less commitments and reservations) adds · estimated to its note: Annual P50 less commitments & reservations · estimated. Reservations awaiting signature are estimated the same way. On the seeded IPP1 Sandton Solar the tile reads ≈ 4,866.9 MWh/yr from the latest forecast run — the 4.9 GWh Performance shows — and Remaining saleable energy reads 113,393.1 MWh/yr.

6.2 What the desk looks like (top → bottom)

6.3 Ideas you need (in plain language)

6.4 The process, step by step

Step 1 — read the gap and its market value

The headline says how much energy a year is unsold and that every deal is judged against what SAWEM would pay for it. One sentence then puts that market in c/kWh: At the day-ahead forecast your unsold hours earn 92.1 c/kWh at P50; the whole day averages 93.4 c/kWh. Under it, the gap chart draws the unsold kWh by hour (amber) with the DAM price band for the same hours (blue): when the amber peak sits where the blue is lowest the chart says so — gap peaks when the market is cheapest. The capture card beside it, Your technology’s realised capture · last 30 days, is history, not the forecast (Captured forecast and realised capture).

The price-rail card answers what a kWh of the gap is worth. Its markers alternate above and below the line, so two close prices never print over each other, and a legend under the rail lists every price with its colour and meaning (#sales-price-legend). Then:

The price forecast is refreshed daily from the day-ahead market’s clearing prices. A refresh over days with no clearing prices writes a run whose every price is 0. The desk skips such runs for the newest run that prices an hour, and says so under the card: The 4 newest forecast runs carried no prices, so this older one is used. (The newest forecast run carried no prices, so this one is used. for one.)

Option 0 · hold and sell on SAWEM is the third card: the P50 value a year, then P10 and P90, then the note that decides its risk — no counterparty, no term, no certainty for financing when you are registered, needs BRP registration before you can bid when you are not. When no forecast run prices the month the card says so, the chart card’s sentence turns amber — No day-ahead forecast prices these hours yet, so there is no market price to compare a deal with. — and deals are shown against spilling the energy for nothing.

Step 2 — who could absorb the gap

One list mixes off-takers and traders. Simulated candidates sort first by their simulation rank; everyone else sorts by shape fit. On a window 1280 px or wider the list runs down the left of the builder; narrower, it sits above the builder, two cards a row. Each card shows:

• the name, a role line (off-taker · Retail · Municipality · City of Tshwane · Cross-municipal (via Eskom transmission) · 6,570,000 kWh/mo), and a shape fit score with a sparkline of the buyer’s load over your unsold hours;
• a could take bar — could take 61.9 GWh/yr · 54 % of the gap — and, when the buyer states an annual volume, what it asks for: asks for 5,256 MWh/yr. The two differ: could take is what your unsold hours could give this buyer, asks for is what the buyer wants;
• route chips — PPA, Eskom VPPA ready, or VPPA · IPP network operator must be Eskom when the route is closed, plus metered history or no history · contract profile;
• Simulated ×n, the #rank badge and the run’s Terms badge once you have simulated it, and the best run’s line (Best: run 1 · 50.0 MW · 60.0 c/kWh · 10y · +R 3.6 m margin a month · unsold −71.5 pp);
• a folded Details & buyer preferences block: organisation, address, connection, business, monthly consumption, topology, and the buyer’s declared tariff, spill, curtailment, REC, term, volume and shortfall preferences. Once opened it stays open while the page updates.

Search matches the words on the card — name, organisation, municipality, business category and address — so manufacturing finds every manufacturer. Or filter with the chips: Suggested (everyone, ranked), Daytime shape, Absorbs ≥ 25 %, VPPA-ready, Traders, Simulated. The count chip (#sales-candidate-count) reads 16 candidates, or 3 of 16 while a search or filter hides some. When nothing matches, the empty list says No candidate matches this search or filter. and offers Show all candidates, which clears both.

Buyers of the same category share a load shape, so they share a shape fit (several cards can all read 90.3). Among equal fits the list ranks by how much of the gap each would absorb, largest first. The footnote under the list says so, and says what the two numbers mean: Shape fit compares a buyer’s hourly demand with your unsold hours (100 means the same shape). “Could take” is the energy the allocation would actually send them.

Each buyer’s name is a real button (#offtaker-candidate-{id}-select, or #trader-candidate-{id}-select on a trader card): Tab reaches it, and Enter or Space selects that buyer. Clicking anywhere on the card still selects it too. Selecting scrolls the builder into view, so a card far down the list never leaves its terms off screen.

Step 3 — build the deal

Click a card. The builder opens as Build the deal · name with the candidate’s fit and a Clear control. Then:

1. Contract route. Physical PPA · Eskom VPPA · Via trader. A route the candidate cannot take is disabled, and hovering it says Closed today. See what it still needs below. The suggested route is preselected — VPPA when it is open and the buyer wants an indexed or TOU-linked price, otherwise PPA; traders go via trader. The route advice paragraph under the terms explains why. When a physical PPA is chosen and the Eskom VPPA route is closed for one reason, the advice names it (The virtual route is closed today: …). When it is closed for more than one, the advice says The Eskom VPPA route is closed today; the checklist below is what it still needs. The checklist is folded, because it is about a route you did not choose: Eskom VPPA is closed today: 10 things it still needs (#sales-route-blockers). Open it to see each blocker on its own line; it stays open while you change terms.
2. Four term cards. Each has a number, a slider and a hint from the advisor, and each checks what you type against the site:
   • Capacity (MW) — hint: knee at X MW · ceiling Y MW saleable. The knee comes from the last capacity sweep; the ceiling is the plant’s saleable MW. The seed covers the buyer’s stated annual volume at the plant’s capacity factor, never more than the plant can sell. On IPP1 Sandton Solar (27 % capacity factor), SIM Menlyn Park Load asks for 5,256 MWh a year, so it seeds 2.22 MW (5,256 ÷ (0.27 × 8,760)), not the whole plant. Above the ceiling the card says More than the 50 MW this plant can sell. Lower it to simulate. in red, Simulate Sale is disabled beside Lower the capacity to simulate., and the server refuses the run as well. Below the ceiling but more than a quarter above the buyer’s ask, an amber line says how far over; at 20 MW: About 47,304 MWh a year, 9.0× the 5,256 MWh the buyer asks for.
   • Strike / price (c/kWh) — hint, for example floor 50 c · SAWEM P50 92.1 c · replacement 210 c. The SAWEM P50 part is left out when it would be 0 c (no forecast hour overlaps your unsold hours). The seed is the price-card ask, never raw LCOE. A strike under the floor reads Below your 50 c floor (LCOE plus margin): every kWh sold loses money.; one over the replacement cost reads Above the 210 c replacement cost: the buyer can buy cheaper elsewhere. Both are amber warnings: you can still simulate.
   • Term (years) — hint: the buyer’s asked term and whether they want a fixed or indexed price.
   • Shape — As-generated or TOU-shaped. Either way the simulation pays the strike on every kWh as generated; TOU-shaped records that the strike stands for a multiplier set’s average price over your generation, which Use as strike in TOU pricing sets (§6.7). The hint reads the best TOU set at the strike you are building (A time-of-use tariff at this strike would average N c), or says No time-of-use set beats a flat price in this plant’s hours. It ends with the firm evening block in words: covered by generation, only with a battery, or not viable, the plant can’t cover it.
3. Demand pool. Which demand bucket the simulated sale is booked into; its allocation rules decide which hours reach this buyer. The route preselects the matching pool.
4. Simulate Sale. A busy overlay holds the page while the engine runs the deal and, for the risk read, a second pass at P90 supply. After the first run the emphasis moves: Build Proposal becomes the green button and Simulate again, which creates the next run, the plain one beside it.

Step 4 — profit and risk vs holding

Each Simulated Impact card carries the run (Run 1), the Terms badge and the route badge, then:

• Four deltas with before → after: margin for the month window (and the annualised figure) in short Rand, +R 3.1 m; realised price plant-wide in c/kWh; unsold share of site generation; revenue HHI (buyer concentration).
• The verdict sentence — Beats holding for the market by R 9.8 m a year at P50 — and by R 22.6 m in a poor price year — with advantage N / 100 and risk N / 100 chips. When the two answers point different ways the sentence says so: Loses R 17.5 m a year against holding at P50 — but beats it by R 37.4 m in a poor price year (a poor price year, P10, is when holding earns least).
• Five risk gauges — Margin@Risk, Concentration, Shape / firmness, Price lock, Counterparty — each with its value and a one-line note.
• The forecast it was judged against — Compared with the day-ahead forecast built 19 September 2026, 9 days ago; newer forecast runs carried no prices. The age appears when the run is more than a week old. With no forecast at all the line turns amber: No day-ahead forecast prices these hours, so this compares the deal with spilling the energy for nothing, not with the market.
• The vs-market line — the advantage at P50, at market P10 and at market P90, and you lose to the market only if SAWEM averages above N c in your hours.
• Build Proposal, which runs the price and capacity sweeps and opens a private draft in the Deal Room (V7). Nothing reaches the other side until you send it.

Step 5 — compare and decide

The comparison appears after the first simulation, and it survives leaving the page: runs are saved, and the desk brings back every run for the current window that has results, including runs made in an earlier visit or by a colleague. The Simulated filter lists their buyers.

Compare and decide comes first, full width. The table ranks each deal with the hold option as row 0: deal and terms (SIM Gateway Umhlanga · run 1, 50.0 MW · 60.0 c/kWh · 10y, its Terms badge), route, vs hold /yr, P10 floor, advantage and risk (with bars on a window 1280 px or wider), and the run’s action. Switch the sort with Rank by: balanced (advantage less half the risk), advantage, lowest risk, certainty (P10). The Same terms chips filter the table — and the candidate list — to one Terms badge or to the selected candidate’s latest terms, so you compare buyers on identical terms.

The last column says what you can do with each run:

Discarding a draft in the studio sends its run back to the comparison, ready to build again.

The Risk against advantage chart follows, with its explanation beside it (on a window 1536 px or wider it sits to the left of the table instead). It plots every deal by risk (right is riskier) and advantage over holding (up is more money); the shaded corner is where you want to be, and the dashed line is the hold option at advantage 0. Dots carry the table’s rank numbers, so the table is the legend: Each dot’s number is its rank in Compare and decide; 0 is holding for the market.

What the system recommends, and why

The desk recommends in four places, and each recommendation shows its working:

The desk never recommends signing. The comparison ranks; Build Proposal starts the negotiation flow.

6.5 Same-terms comparison and the Terms badges

Two deals on different capacity, price or term are different deals, not different buyers. The desk fingerprints every simulation’s terms and labels identical sets Terms A, Terms B, and so on (the Params badges of earlier releases). The badge appears on the candidate card, on the Simulated Impact card and on the comparison row; the Same terms chips under the comparison heading filter to one badge, to same as selected, or back to all deals. Filtering also narrows the candidate list to the buyers who carry that badge, so you can walk a shortlist on one set of terms.

6.6 Worked example — sell midday unsold into a high-fit factory

Data. Illustrative, on the shape of a 50 MW solar plant with 96 % unsold, LCOE 48 c, SAWEM P50 72.7 c in the unsold hours.

Simulate. The engine allocates the deal’s energy and prices it: margin uplift in the month window +R 2.7 m (≈ R 32.8 m a year); the hold value of the same energy at P50 is R 23.0 m a year, so the advantage is +R 9.8 m a year. At market P10 the advantage widens to +R 22.6 m; at market P90 it narrows to −R 4.1 m — you lose to the market only if SAWEM averages above 78 c in your hours. Risk components: Margin@Risk 14 % (P90 volumes), concentration 0.62 after the deal, shape low (as-generated), price lock 7 years fixed, counterparty ready — blended risk 38. The comparison places it first on balanced; a second run at 81 c on a physical PPA ranks higher on advantage and lower on lowest risk because its concentration is worse.

Finding the price where the margin reaches zero

The vs-market line prints it: you lose to the market only if SAWEM averages above N c in your hours. It is the strike at which the deal’s margin uplift equals the hold value of the same energy — d_margin ÷ (absorbed MWh × 10 × (1 − imbalance)). Move the Strike / price slider and simulate again to see the verdict flip from beats holding to loses against holding; the floor on the price hint is the book’s LCOE + margin, below which no deal should go regardless of the market.

6.7 TOU pricing — is that multiplier set worth it?

Route: …/generator-sites/<id>/strategy/sales · Tab: Sales → TOU pricing pricing tool, a drawer from the right (#sales-tou-tool)

On-screen chrome: base-price input + ranked multiplier table (tou_multiplier_table) with a "Best set: …" narration line, and a Use as strike button on each row while a candidate is selected.

The seller's problem, in plain words

You sell electricity. A buyer often rejects a simple flat price ("65 c/kWh, every hour") and counters with a time-shaped offer:

"We'll pay 1.3× your base in peak hours, 1.0× in standard, and 0.7× in off-peak."

That can sound generous — peak is "expensive", so a 1.3× peak uplift looks like a win. For you, it is only a win if your plant actually generates in the hours that carry the uplift, and does not dump most of its energy into the discounted hours.

A solar plant makes almost nothing at 19:00 (evening peak) and a great deal around midday (usually standard on the Megaflex weekday clock). Accepting a big peak multiplier while giving away off-peak (and sometimes standard) can quietly cost hundreds of thousands of Rand a year — even when the brochure looks attractive.

The TOU pricing pricing tool on Sales exists for one job: take your plant's generation shape, apply each candidate multiplier set hour by hour under the same SAST TOU calendar the engine uses everywhere else, and rank the sets by effective price — best first — so you can see which counter is actually worth it.

This tab does not sign a deal. It is a pricing calculator beside the deal builder (§6). Use it when a buyer (or you) floats a TOU-linked counter; Use as strike carries the set you pick into the deal, and the negotiation carries the shape (V7).

When to open TOU pricing vs the deal builder vs Firm

6.7.1 Every concept on this screen

Read these as glossary cards for the labels you will actually see. The full Megaflex hour map lives in V2 §8.2; this section only needs enough clock to price a counter.

6.7.2 The Megaflex day — when is peak?

You do not need to memorise every cell. You do need the intuition that evening peak is often when solar is asleep, and midday is usually standard, not peak.

6.7.3 How effective price is calculated

Every hour of a representative year gets:

1. a generation amount (kWh) from the plant's P50 shape, and
2. a TOU tag (peak / standard / off-peak) from the engine calendar.

For one multiplier set:

effective price = Σ (generation(h) × base × multiplier(TOU of h)) ÷ Σ generation(h)

In words: weight each hour's base × multiplier by how much energy that hour produces, then divide by total energy. Flat multipliers (1 / 1 / 1) always collapse to exactly the base.

Display note. The advisor stores effective price to 2 decimals (e.g. 63.89). The table's ckwh/1 formatter may show one decimal on screen (63.9). Δ R/yr shows whole Rand with comma thousands (e.g. −R 486,369). This chapter quotes the 2-dp advisor figures so the arithmetic matches the tests.

6.7.4 How to use the TOU pricing tool

Follow along. Seat: IPP demo user on a solar generator strategy book (full_demo · seed 42). Path: Generator sites → (site) → Strategy → Sales → TOU pricing. You should see a base input and a three-row table with a "Best set:" line.

Step-by-step

1. Open Sales, pick a candidate, then click TOU pricing in the Pricing tools bar above the deal builder. The tool opens as a drawer from the right, over the desk; Escape, the X or a click on the dimmed page closes it, and focus returns to the button.

2. Read the base price. It opens at the strike you are building. Confirm it is the sticker you are negotiating from, or type the one you want to stress.

3. Optionally type a new base (step 0.1). The table recomputes as you change it — no separate Submit.

4. Read the Best set line first. That is the winner at this base.

5. Scan the table top → bottom (already sorted best-first):

   Column	What to look at
   Peak / standard / off-peak	The row’s three multipliers, in that order
   Effective (c/kWh)	Shape-weighted c/kWh — higher is better for you
   Δ vs flat, R/yr	Yearly Rand vs flat — negative means "worse than flat"

6. Interpret the emerald row. If flat (1 / 1 / 1) is on top, the "attractive" counters lose money on your shape — push back or restructure (raise standard, soften the off-peak cut, or stay flat).

7. Carry the set into the deal. Use as strike on a row puts that set’s effective price in Strike / price, sets Shape to TOU-shaped, closes the drawer and confirms: Strike set to 58.3 c/kWh, the 1.3 / 1.0 / 0.7 set’s effective price. Then simulate. In the negotiation, price terms that match where your energy lives; do not gift multipliers for hours you do not generate into.

8. If the conversation turns to a firm evening block (not multipliers), switch to Firm viability (§6.8) — different question.

6.7.5 Worked examples

Data. Advisor maths · representative year 2026 · solar fixture P50 43,800 MWh/yr · 20 MW export (same shape family as IPP-STRATEGY §3.1 / PricingAdvisor tests). Shares below are engine-computed under the live SAST calendar (≈ 13% peak · 64% standard · 23% off-peak for this solar shape).

Example A — base 65 c/kWh, demo solar plant (the screen's headline)

Under the engine's SAST TOU calendar this plant's generation lands ≈ 13% in peak, 64% in standard, 23% in off-peak hours. The three candidate sets:

What it means for the seller. Saying yes to 1.3 / 1.0 / 0.7 because "peak is 30% higher" would cost about R 812k/yr versus staying flat at 65.

How to react. Counter with flat, or insist on a higher standard factor and a shallower off-peak cut. Do not treat the peak 1.3 as compensation for the 0.7.

Example B — a tiny day, so the formula is visible

Ignore seasons for a moment. Suppose one day the plant makes exactly 100 kWh in three hours, base 65 c/kWh, set 1.3 / 1.0 / 0.7:

Effective = 6,305 ÷ 100 = 63.05 c/kWh — already below the 65 flat sticker, even though peak paid 1.3×. The 70 kWh at 1.0× and 20 kWh at 0.7× dominate.

What it means. Volume beats brochure multipliers.

How to react. Ask: "What share of my year sits in each bucket?" — which is exactly what the tool answers for the full year.

Example C — same solar plant, raise only the standard factor

The tab's fixed candidates do not include a custom set, but the maths still teaches the negotiation. Using Example A's ~13 / 64 / 23 split as rough weights on base 65:

You do not need the imagined row on screen — the lesson is: move money into the bucket where your MWh live (standard for solar), and stop paying for peak theatre.

How to react in the room. "I'll take a standard uplift; I won't finance your peak story with my off-peak and midday."

Example D — same three sets on a wind-shaped year

Same base 65, same three candidates, but a wind P50 shape at the same annual energy. Engine shares land roughly 15% peak · 32% standard · 53% off-peak — overnight energy is huge.

What it means. The 0.7 off-peak discount now hits more than half of generation. Both "peak premium" stories are catastrophic for this shape — worse than for solar.

How to react. Flat (or an off-peak uplift) matters more than peak theatre. Open this site's Sales → TOU pricing rather than copying another plant's spreadsheet. The tab always reweights to your technology shape.

Example E — change the base, ranking stays shape-driven

Type base 80 instead of 65 on the same solar plant. Every effective price scales with the base; flat remains exactly 80.00 and still ranks first against these two discounted sets — the gaps widen in Rand terms because Δ R/yr = (effective − base) × annual kWh ÷ 100.

What it means. Raising the sticker does not rescue a bad multiplier shape; it magnifies the yearly damage of a bad shape.

How to react. Fix the multipliers (or stay flat); do not hope a higher base hides a 0.7 off-peak.

6.7.6 What good looks like — seller cheat sheet

Rule of thumb. Multipliers are only as valuable as the kWh that touch them. The TOU tab is the scale that weighs those kWh.

6.7.7 Bridges and caveats

• Back to the desk. Once terms look right, Use as strike on the winning row, then Simulate the buyer (§6). The simulation pays that effective price on every kWh as generated: the set’s average over your generation, not the set hour by hour. The TOU-linked shape itself belongs in the real proposal.
• Firm blocks are separate. Evening guaranteed MW is §6.8, not this toolle.
• Candidate sets are fixed on this screen. SalesLive compares the three solar-weighing sets listed in §5.7.1. Custom multipliers belong in negotiation / contract terms (TOU_LINKED in V7); use this tool to learn which direction helps your shape.
• Calendar truth. Effective prices use the engine's SAST / Megaflex classifier — the same family as metering. Older brochure numbers that assumed "solar is ~98% standard" will not match; trust the tool (recorded divergence vs an early spec illustration — the product shows live advisor output).
• Overview context. Why filling unsold volume matters even below a flattering achieved price: §4.

Because the maths is your real shape against the real calendar, run your own numbers — not the brochure's, and not another plant's.

6.8 Firm viability — should I sign a firm block?

Route: …/generator-sites/<id>/strategy/sales · Tab: Sales → Firm viability pricing tool, a drawer from the right (#sales-firm-tool)

On-screen chrome: intro copy (#sales-firm-intro) · Firm price (c/kWh) input · firm-viability card (verdict badge · Expected margin in c/kWh · Chance the plant falls short as a percentage · optional Margin with the battery) · battery footnote ("Battery marginal N c/kWh applied." / "No active battery on this site.")

The seller's problem, in plain words

A buyer sometimes wants more than a flat or TOU-shaped price. They want firm power in named hours — for example, "guarantee us power from 18:00–20:00, every weekday we care about." They will often pay a premium for that promise, because their factory / shift needs those hours whether or not the sun is shining.

That premium is only real money if you can physically stand behind it. If your solar plant is asleep at 19:00 and you still signed "firm evening", you must buy replacement energy from the market (or discharge a battery) to honour the contract. Buy-in at 210 c/kWh against a firm sale at 140 c/kWh is not a premium — it is a 70c loss on every short kWh.

The Firm viability pricing tool exists for one job: take a proposed firm evening-block price, compare it to your plant's generation shape in those hours, fold in the book's replacement cost, and tell you — before you sign — whether the deal is:

This tab does not sign a deal. It is a physical + economic check beside the deal builder (§6) and TOU pricing (§6.7). Use it when the conversation turns from "what multiplier?" to "can you guarantee these evening hours?"

When to open Firm viability vs TOU vs the deal builder

6.8.1 Every concept on this screen

Read these as glossary cards for the labels you will actually see. The engine check behind the card is PricingAdvisor.firm_block_check/5 (same maths the demo walkthrough quotes).

expected margin = firm price − replacement cost × P(short)

6.8.2 The expected-margin formula

Buyers pay a premium for firm delivery in named hours. The premium is only real if your plant can physically stand behind it — and this card checks, before you sign:

expected margin = price − replacement cost × P(short)

Why multiply by P(short)? If you are short half the time (P(short) = 0.5), you do not pay full replacement on every kWh — you pay it on the expected short share. If you are short always (1.0), the formula collapses to price − replacement — the harsh evening-solar case.

Display note. The advisor rounds expected margin and BESS margin to 2 decimals; P(short) to 3. The card shows the margins with their sign in c/kWh (ckwh/1 may drop to one decimal) and P(short) as a whole percentage (1.000 reads 100 %). This chapter quotes the advisor figures so arithmetic matches the tests.

6.8.3 How to use the Firm viability tool

Follow along. Seat: IPP demo user on a solar generator strategy book (full_demo · seed 42). Path: Generator sites → (site) → Strategy → Sales → Firm viability. You should see the intro line about accept / needs storage / decline, a firm-price input at 140, and a card — typically ❌ Decline with Chance the plant falls short 100 % and Expected margin −70 c/kWh when there is no battery.

Step-by-step

1. Open Sales, then click Firm viability · safe capacity in the Pricing tools bar (shield icon). The tool opens as a drawer from the right; Escape, the X or a click on the dimmed page closes it.

2. Read the intro (#sales-firm-intro): it states the three outcomes in one sentence — accept if you can cover it, needs storage if a battery would, decline if neither pays.

3. Confirm replacement cost (if the deal is sensitive). Glance at Overview assumptions / open ⚙ LCOE / params (§4.3) — default 210.0. This tab does not edit it; wrong replacement → wrong verdict.

4. Read or type Firm price (step 0.1). Default 140. The card recomputes as you change it — no separate Submit.

5. Read the verdict badge first, then the numbers:

   Line	What to look at
   Expected margin	As-generated cents after shortfall tax
   Chance the plant falls short	How often the plant cannot self-cover the block
   Margin with the battery (if shown)	Margin if the shortfall is served from storage

6. Read the footnote. "No active battery…" means Decline cannot flip to Needs storage unless you set bess_marginal_ckwh (or add a battery where the domain allows). "Battery marginal N…" means that cost is already in the Margin with the battery line.

7. Interpret, then act:

   Verdict	Typical next move
   Accept	Firm term is physically standable — carry price into the deal builder / negotiation
   Needs storage	Do not sign the evening block alone; couple it with a storage decision (or walk)
   Decline	Refuse the firm evening promise, or restructure (different hours, as-generated, or much higher price)

8. If the conversation is only about multipliers (peak / standard / off-peak), switch back to TOU pricing (§6.7) — different question.

6.8.4 Worked scenarios

Data. Advisor maths · evening block hours [18, 19, 20] · solar fixture P50 43,800 MWh/yr · 20 MW export (same family as IPP-STRATEGY §3.3 / PricingAdvisor tests). Replacement 210 c/kWh unless noted. Battery marginal 88.9 c/kWh when storage is in play (M4 stored-marginal illustration).

Scenario A — solar plant, firm evening block, no battery (Decline)

expected margin = 140 − 210 × 1.0 = −70 c/kWh

What we ascertained. The premium is fake once shortfall is priced: you sell at 140 and buy at 210 whenever the sun is down — which is the whole block.

How to react. Do not sign firm evening on plant alone. Push as-generated / daytime firm / TOU instead, or bring storage into the conversation (Scenario B).

Scenario B — same evening block, with a battery (Needs storage)

BESS margin = 140 − 88.9 = +51.1 c/kWh

What we ascertained. Market buy-in at 210 destroys the deal; battery discharge at ~89 clears a healthy margin — but only if that storage is real (asset or modelled param).

How to react. Sign the firm evening term together with the storage plan, or neither. Do not treat "Needs storage" as a soft Accept.

The battery marginal comes from the site's active battery asset (its degradation cost) when one exists; a solar site evaluating adding storage can set the candidate marginal as a book param (bess_marginal_ckwh) — so "would a battery make this deal work?" is answerable before any battery is bought.

Scenario C — midday block the plant can self-generate (Accept)

What we ascertained. Firm is not automatically dangerous — firm evening on solar is. When the promise sits on the generation mountain, P(short) collapses and the full firm price is expected margin.

How to react. Firm daytime / shoulder blocks that match your shape can be Accept at the same sticker that Declines at 18:00. Use the tool's evening check for evening offers; do not assume the same verdict for every named window.

Screen note. SalesLive's Firm tab is wired to the evening block [18, 19, 20]. The midday Accept case is the advisor's contrast example (same function, different hours) — it teaches why hour choice matters even when the on-screen block is evening-only.

Scenario D — raise the firm price, evening still Declines

Same solar evening, replacement 210, still no battery. Type firm price 200 instead of 140:

expected margin = 200 − 210 × 1.0 = −10 c/kWh  →  still Decline

What we ascertained. A modest premium over LCOE / daytime offers does not rescue a P(short) = 1 evening block while replacement is 210. You need either storage, a much higher firm sticker, or a different block.

How to react. Do not "nudge the price" and hope; watch whether price actually exceeds replacement × P(short).

Scenario E — soften replacement cost (LCOE / params dial)

Keep firm price 140, evening solar, no battery. On Overview set replacement from 210 → 120 (stress: "what if buy-in is cheaper?"):

expected margin = 140 − 120 × 1.0 = +20 c/kWh  →  Accept

What we ascertained. The verdict is only as honest as your replacement assumption.

How to react. Change replacement only when the board pack / market view really changed (§4.3). Do not "fix" Firm viability by quietly lowering replacement.

Scenario F — Needs storage flips back to Decline (expensive battery)

Evening solar, price 140, replacement 210, but battery marginal 150 c/kWh:

What we ascertained. "Has a battery" is not enough — the marginal must sit below the firm price. An expensive discharge path still declines.

How to react. Fix storage economics or refuse the firm evening term; Needs storage is not automatic.

6.8.5 What good looks like — seller cheat sheet

Rule of thumb. A firm premium is real only when price > replacement × P(short) (as-generated), or when storage clears price > battery marginal. Everything else is theatre.

6.8.6 Bridges and caveats

• Back to the desk. Once the verdict is Accept (or Needs storage with a real storage plan), return to §6 and Simulate the buyer under terms that match — this tool never promotes a proposal by itself.
• TOU is separate. Multiplier counters live in §6.7. Firm viability answers "can I guarantee these hours?", not "which peak factor is best?".
• Replacement lives on Overview. Edit replacement_ckwh under ⚙ LCOE / params (§4.3); Firm viability only consumes it.
• Evening block is fixed on this screen. SalesLive checks hours 18 / 19 / 20. Midday Accept (Scenario C) is the advisor contrast case — same maths, different hours — not a second dropdown on the tab.
• Threshold is half of peak. An hour is "short" when generation < 50% of the plant's peak hourly sample in the representative year — a deliberate conservative self-firm bar, not a weather forecast.
• Demo seat. full_demo solar often shows Decline at 140 with no battery; Needs storage appears when a marginal is available — V3 §21.

Because the check is your shape against your replacement and (optional) battery marginal, run your own numbers — not the buyer's brochure firm premium.

Captured forecast and realised capture

The hold comparison uses the forecast price weighted by your unsold hours; the whole-day average in the same sentence (…; the whole day averages 93.4 c/kWh.) gives every hour equal weight. A third figure, the capture card Your technology’s realised capture · last 30 days, describes observed market prices in R/MWh. It is historical evidence, not a promise about the hold option, and the verdict never uses it: the verdict names the forecast run it was judged against.

7. Allocation · Rules — the cascade

Thread part 4, continued: when IPP2 Atlantis Wind makes less than its buyers want, who is served first?

Follow along. Seat: ipp_2@webill.net. Path: Generator sites → IPP2 Atlantis Wind → Strategy → Allocation, mode Rules · hierarchy. You should see the mode switch (Rules · Forecast · Outcomes) with a status note under each, an Explanations button on the right of that bar, the versions rail, the cascade board with Priority, Pro-rata and Fixed buckets flowing into Surplus, and a right-hand rail with Changes vs …, If this draft had been in force in … and Where the cascade runs dry. Explanations (#allocation-explanations-open) opens a right-hand drawer of every abbreviation and term used on Allocation — one alphabetic list from AES through Surplus, including amber reached Surplus — did not run dry.

Data. full_demo · seed 42 · IPP2 Atlantis Wind · stage S07 · last checked 2026-09-03.

The seller’s problem, in plain words

Three buyers hold PPAs on one plant. In a good hour there is energy for all of them and some to spare; in a poor hour there is not. Somebody has to decide, in advance and in writing, who is short. That decision is the allocation hierarchy: every buyer sits in a tier — PRIORITY (ranked, filled in order), PRO_RATA (an equal tier that shares the remainder proportionally), FIXED (a capped entitlement, filled last) — and metered generation cascades down the tiers each hour. Whatever is left is Surplus: unallocated generation, which is exactly the gap the Sales desk sells.

The same cascade runs twice. Settlement runs it on metered energy, hour by hour (§9); the forecast runs it on next month’s expected generation (§8). Rules is where you write it, and it is versioned: publishing writes a new effective-dated version and never recomputes a month that has already settled.

What you want to achieve on this screen

• Put every active PPA’s buyer in the right tier, in the right rank, with the right Fixed entitlement — and see what that rule did last month on the same card.
• Check a draft against reality before publishing: the last settled month re-carved through the draft, member by member.
• Publish with an effective date, and read back what changed against any earlier version.

When to open Rules vs the other modes