Skip to content

v0.2.0

Choose a tag to compare

@samuelduchesne samuelduchesne released this 19 Aug 15:27
· 72 commits to main since this release

The first tagged release of the shoebox: a one-page, client-side EnergyPlus demo laid out as a drafting sheet, continuously deployed at shoebox.idfkit.com. Everything below shipped to main since 0.1.0 — six new channels and controls (adiabatic surfaces, skylights, per-wall glazing studies, custom holidays, layered glazing panes, a picker for the reading hour), a run of accessibility and mobile-layout fixes, and one deliberate breaking change to the permalink format, made before this tag existed to carry it.

⚠️ Breaking changes

1. The run period's beginMonth / endMonth link parameters are gone

Run from and Run to, two sliders describing one unbroken span, are replaced by a twelve-month calendar mask so a January-and-July run — sizing days without the spring between them — is expressible at all. The old pair could not describe that, and rather than carry it forward as a translated months= mask, it was dropped outright:

# Before (v1 link)
#v1&beginMonth=3&endMonth=5

# After (v1 link, same version — the mask needs no new encoding)
#v1&months=001110000000

A link naming either old key is refused whole, with the offending parameter named in the status line, rather than silently reinterpreted. Nothing had been published under the old format before this tag, so there is no link in the wild to migrate and no reason to spend a LINK_VERSION bump on one — this is the one and only release where that reasoning holds.

Added

  • The reading hour drags, and it has a picker. The plate's marker could be placed with a click; it can now be taken hold of and moved along the curve, with every meter on the rail, the balance total and the stamp under the plate re-lettering as it travels. Nothing is simulated — the run is already in hand and the hour is only a way of reading it — so a drag across a Denver year costs a re-read of an array rather than 8,760 hours of engine time. A press that never travels is still a click and still toggles the hour it names; a drag that ends where it began keeps the pin it just placed; and the address bar updates on release, never per frame, like every other gesture here.

    Under the plate there is now a bar carrying the instant, the hold, and two ways of naming another one. Named hours first, because "which hour" is a question the field already has stock answers to: EnergyPlus reports its own component loads at the time of the peak load, heating and cooling apart, so peak heating and peak cooling are the two a modeller arrives with; a results tool's period list offers summer and winter design weeks read off the weather file's statistics, which becomes the hottest and coldest outdoor hour; the zone's own warmest and coolest stand beside them because a free-running desk has no heating or cooling rate at all; and peak solar gain is the hour every glazing and shading control on the desk is arguing about. Each is found in the run that came back and letters the environment it landed in, with the value it reads there — on Denver's design days, peak heating at 24:00 on 21 December is 5.91 kW and peak cooling at 15:00 on 21 July is 2.18 kW.

    An offer the run cannot make states its reason where its stamp would have been and cannot be pressed. Bypass the System strip and both peaks say so rather than quietly handing back the least-cooled hour of the year under a label claiming the opposite — an argmax always returns something, which is exactly why it needs a gate.

    Then a calendar, every option of it walked out of the run's own timestamps, so it cannot name an hour the run does not hold. A date field was rejected when the marker was built, on the argument that it invites February the 30th and hour 25 purely to meet a refusal message; that was an objection to a free field, and there is nothing left here to refuse. It earns its place because the gesture cannot reach everywhere — an annual plate at ten hours to the pixel is physically unable to name 15:00 on 14 February, and a pointer is not the keyboard's instrument at all. It works coarse to fine: an environment opens at its own worst hour, a month or a day at that day's extreme, and only the hour field names an hour.

    The bar sits on the sheet rather than only on the rail for the reason the plate grew its marker in the first place: the rail is inside a console you have to open, and the hour is the single most movable thing about every figure on the page. It also says, on the sheet, when a pin could not be found in the new run and was released — until now the marker simply went from filled to hollow, which is not an explanation.

  • Any of the six surfaces can be made adiabatic, on a key of their own at the foot of the Fabric strip. A shoebox is almost never a free-standing object: it is one bay of a terrace, a middle floor of a stack, a corner unit with two party walls. Until now this desk could only say so about the floor, and every other surface was exposed whether the building it stood for was or not.

    The key is a plan with a section drawn through it. The four walls are the edges of the plan and turn with north, exactly as the glazing key's bars do; the roof and the floor are the two surfaces a plan cannot show — it is a horizontal cut and they are what it cuts through — so they are drawn as the section they would appear in, roof over floor, inside the square. Adiabatic is a doubled line, the way a party wall is drawn on any plan, and open to the weather is a single hairline. The axonometric is the same control: the three faces the viewpoint shows can be clicked to flip them, and every adiabatic surface is hatched like a cut — faintly, through the box, for the three facing away.

    On Denver's two design days, with the ideal unit and the internal gains in the path: the stock free-standing box carries 511.0 m² of exposed envelope at an A/V of 0.481 and asks for 82.5 kWh of heat. Take the east and west walls out as party walls and it is 371.6 m², 0.350 and 39.6 kWh — a little over half the heating for a building nobody moved a millimetre. Take the roof out instead, as a middle floor, and it is 47.8 kWh. Take all four walls out and the heating is 0.0 kWh against 99.0 kWh of cooling: a top-lit core, which is a real building and not one this page could previously describe at all.

    An adiabatic surface has no outside, and the engine refuses an opening cut into one, so the wall that goes out takes its glass, its overhang and its fins with it — the glazing key greys that wall and says The north wall is adiabatic, so there is nothing outside it to open onto, and Skylights blocks itself when the roof goes. The window-to-wall and skylight-to-roof ratios now divide by the surfaces that have an outside, which is what those ratios have always been measured over.

  • The sheet says which build of itself you are reading. The title block's Sheet cell has always carried a revision, and it was the string Rev A, hand-lettered and never once true. It now carries the build: E-01 · Rev 0.2.0 on a tagged release, and E-01 · Rev 0.2.0+cd5881e on the far more common case of a deploy from main with no tag on it, the sha being semver's build metadata — the same declared version, this particular build of it. It links to the release or the commit it names, so a reader who thinks a number looks wrong can now say which drawing the number was on, and the run bundle's manifest carries the same line beside the EnergyPlus version.

    The date under it is the revision's, taken off the commit. It used to be new Date() evaluated in the browser, which lettered "Issued" with the day the page happened to be opened: a drawing dated by whoever picked it up.

  • Every number a slider carries is now a box you can type in. The five seventy-six scales on the desk, the eight walls of its two plan keys and the five dimensions under the drawing — which are five of those same controls, drawn twice — all letter their value exactly as before, and every one of them is now also the way to set it: click the number, type, Enter. A slider cannot say an exact figure — width runs 4 to 40 m across about 200 px, which is 0.18 m to the pixel — so 12.00 m was a hundred presses of an arrow key away.

    Nothing is drawn around the box. No border, no fill, no spinner, no focus ring: the number reads as the lettering it always was, and the only thing that says it can be edited is the I-beam the cursor becomes over it. A sheet made of hairlines cannot afford eighty more rectangles to say the same thing.

    A typed value is brought onto the control's own face — clamped to the stops and snapped to the step, then rounded to the step's own decimals, because 0 + 3 * 0.05 is 0.15000000000000002 and that number would ride the permalink and be written into the IDF exactly as it stands. Anything that is not a number is refused whole and the model's own value comes back, the way a bad link is: no half-reading of 12abc. The unit and the zero word are accepted, since they are what the box says when it is not being typed in — 12 m means 12 m, and Solid closes a wall.

    Two quieter rules keep it honest. Focus shows the value itself rather than its lettering, because a height of 4.572 m reads as 4.57 m on a face ruled to two places and a reader who touched the box and left it alone would have silently trimmed 2 mm off the building — and the box compares what it gave against what it got back, so a touch-and-leave commits nothing whatever. And a redraw never types over the reader: a study tick landing or a station attaching redraws every face on the desk, and the one being edited is left alone until it is let go.

  • The layered glazing model has a pane count, and the window says what it came to. Two things were missing from the same strip. The layered model built a double unit and only a double unit — there was no way to ask for a triple — and neither model would tell you the one pair of numbers a window is actually specified by, because the layered one is set in causes (panes, a coating, a cavity) and the simple one hides its equivalent layer.

    Panes is a slider from 2 to 4, and it writes real sheets: n panes of 6 mm clear float with n − 1 air cavities between them, the low-e coating staying on the cavity face of the inboard pane wherever that now falls — surface 3 in a double, 5 in a triple. On the default desk it is worth more than any other control on the strip: U 2.68 W/m²K at two panes, 1.73 at three, 1.29 at four, and 0.93 with a hard coat on the fourth.

    As built is the answer, read off the run rather than computed here. It stands under the controls and above the transmitted-solar meter, and letters the U-factor, SHGC and visible transmittance EnergyPlus itself worked out for the assembly, from the envelope summary the run already writes. Where the opening carries a frame it adds a second line for the whole window by the NFRC method, since that is where the frame's own conductance lands and the glass figures stop being the window's. It obeys the readings' rules like everything else on the sheet: an em dash before the first run, and nothing at all where the channel is out. Under the simple model it reads back the three sliders above it, which is the quiet confirmation that the equivalent layer is what you asked for.

  • Landmarks on the calibration faces. A slider set in W/m²K is a quantity before it is a decision: 1.80 says nothing to a reader who reads low-e double perfectly well, and a face offering only a range and a tick has handed them the number and withheld the vocabulary. Fifty-one faces now carry the published cases they are read against — 149 of them — ruled under the face as dimension lines, with the one the reading is standing in lettered underneath. Dragging the U-factor face reads out single · double, clear · double, low-e · triple, and the same sentence rides in aria-valuetext, so a reader on the arrow keys hears "1.80 W/m²K, double, low-e".

    Some of it was already on the desk as prose — "fresh snow reads near 0.7, asphalt near 0.1", "the stock R13LAYER is 2.29" — where the drawing could not reach it and the slider could not point at it. Those notes are landmarks now, declared once in controls.js and drawn by the console, by the plan key along each wall's own bar, and by the sheet's own dimension sliders.

    Where the cases come from is carried with them: ASHRAE 90.1's fenestration and lighting limits, 62.1's occupant densities and outdoor-air rates, 55's comfort bands, the Handbook of Fundamentals for glazing and metabolic rate, EN 12464-1 for illuminance, the Passive House Institute, the CRRC, the Beaufort scale, and EnergyPlus's own Input Output Reference for the engine defaults. Landmark refuses to be constructed without a source, because a landmark nobody can check is the interface asserting rather than measuring.

    The layered model's pane count is landmarked too, at Double, Triple and Quadruple, so the two glazing models name the same cases: the simple one's U-factor face is read in clear double and low-e double, and the layered one is set in the sheets that produce them. The notes carry what the engine returns for each on the default desk — U 2.67, 1.73 and 1.28 W/m²K — which is also the finding that an uncoated triple lands in the low-e double band, so the coating is the cheaper of the two moves.

    Nothing here reaches the IDF. A throwaway harness built the document at six desk positions before and after and hashed each one: byte-identical, with applyModel still idempotent at every position.

  • The blind's slat angle now says which way it runs. EnergyPlus measures WindowMaterial:Blind.slat_angle from the glazing's outward normal, so 0° and 180° are closed and 90° is fully open — the opposite of what a slider running 0 to 180 suggests, and nothing on the face said so. Both stops and the middle are landmarked, and the strip states the convention.

  • The finding says what the building is before it says what the run made of it. The paragraph under the plate was a reading with no subject: it opened With no heating or cooling anywhere in this model, the envelope alone takes the summer design day's 15.1 °C outdoor swing down to 4.4 °C in the zone over a desk of eighteen strips, none of which it named. It now opens with the shoebox — A single storey of 232.3 m², 15.24 m square and 4.57 m tall, glazed 0.20 south under a 0.60 m overhang — and then reads it.

    What the second sentence carries is decided by difference: a desk has ninety controls and a paragraph has room for three, so the moves are ranked by how far each sits from its own default, and a channel patched into the path outranks any slider — With 0.50 ACH of leakage, gains of 16.0 W/m² over 08:00–18:00 at 12.0 m²/person and an ideal unit holding 20.0–26.0 °C.

    Every compass word is measured rather than named. turn() puts the orientation into the vertices and leaves the wall names alone, so on a building turned 40° the wall the plan calls south faces south-east; the description reads each wall's bearing off its own outward normal and letters it beside the word — glazed 0.45 south-west (220°) and 0.30 north-east (40°). Areas, ratios and overhang reaches are read off the document too, so a channel patched out from under a control describes what the document holds and not what the slider still says: with Fabric and Mass out, the sentence is solid on every face. With every surface adiabatic and the slab swapped for a massless layer.

    Which surfaces have an outside at all is part of it: a wall or roof set adiabatic reads as the north wall adiabatic, since three exposed walls and a party wall is one bay of a longer building and a paragraph that only said what was glazed would letter that fourth wall as solid.

    A setting is described by the object it reached rather than by its own value, which is a sharper rule than it sounds: Available is not a modifier on a unit that has two setpoints, so at "Heat only" — where the model holds a ThermostatSetpoint:SingleHeating and the cooling setpoint reaches nothing — the sentence reads an ideal unit heating to 20.0 °C, and "Occupied" claims occupied hours only when Gains is in the path to give it a band to follow.

    Nothing else is claimed that is not measured — no typology, no assembly names and no verdict — for the reason the rest of the sheet gives an em dash: 12 m² per person is a number, and "an open-plan office" is a building this model was never given.

  • Each wall of a plan key can be swept. The window-to-wall ratio and the overhang projection were the two controls on the desk a study could not be taken of, which is the wrong two: an elevation is where orientation stops being a diagram and becomes a number, and "what does glass on the west cost me?" is the question a shoebox exists to answer. Each of the eight walls now carries its own Study, in the legend under the plan where that wall's number is already lettered, and draws its own curve.

    Four walls are four studies, not one, because a study holds the whole desk still and moves a single parameter — turning all four together would answer a question about a building nobody is designing. So the curves stack under the plan in compass order, each card naming its wall, and each is stale, stopped, cleared and re-swept on its own. On Denver's design days the difference is the point: north from solid to 0.90 moves the summer peak 32.7 → 36.8 °C, and the same glass on the west moves it 32.7 → 51.1 °C.

    An overhang over a solid wall is refused, and says why. Overhangs are cut from the opening they shelter, so a projection set on an unglazed elevation reaches no object in the document — measured, four positions of the west overhang wrote four byte-identical IDFs. That is a sweep of twenty-one identical models at full engine price, so the offer is disabled with the wall's own sentence — The west wall has no opening, so an overhang there hangs on nothing — and the reading and its bar on the plan are greyed alongside it. Until now that control could be turned all day, on a desk that never moved, with nothing anywhere to say so.

  • Skylights, as their own strip at 04, between Glazing and Shading. The roof was the one elevation this desk could not open: it carried an R-value and an absorptance and nothing else, so the whole question of top-lighting — the one an architect asks about any deep plan — could not be put to the engine at all.

    The strip owns a skylight-to-roof ratio, an arrangement (square lights on an n × n grid, or linear rooflights running the width), how many units across, a curb height, and whether the glass is the walls' own assembly or its own unit with its own U-factor, SHGC and visible transmittance. Each light is a FenestrationSurface:Detailed cut into Zn001:Roof001 and wound to match it, so its normal points at the sky; the curb is four Shading:Zone:Detailed faces standing round the opening.

    The curb is the part that earns its keep, and it is written as real geometry rather than as the one field that would say the same thing. A WindowProperty:FrameAndDivider has an outside_reveal_depth that shades an opening exactly as a curb does, in one number — but this sheet draws by reading coordinates back off the document, so a curb expressed as a number would shade the run and be invisible on the drawing, and a curb you cannot see is the one you forget you set. Measured on the Denver design days at a 10 % roof ratio, four lights: transmitted solar falls from 57.0 to 33.5 kWh over the pair as the curb goes from flush to 1.2 m — 41 % — and the summer zone peak with it, 46.5 to 39.8 °C, monotone at every step between.

    What the rooflights themselves are worth is the finding the strip exists to produce. At the default 6 % ratio on the stock Denver box, transmitted solar goes from 5.1 to 33.6 kWh over the two design days — 6.6× — against a wall ratio of 5 % that contributes the smaller half of it, because a horizontal opening faces the one part of the sky that is never behind a neighbour and never off to one side. Swept from solid to 30 % the summer high climbs to 63.3 °C while the winter low moves 0.4 K, which is the asymmetry a top-lit plan lives or dies by and which no wall control on this desk can show.

    Two joins with the rest of the desk, both stated in the interface rather than left to be discovered. Daylight now takes a rooflight as an opening, so a building with no window in any wall can still be top-lit and dimmed — and its precondition now asks whether the opening is actually in the document rather than only whether a slider is off zero, which it did not before. Blinds reach the rooflights when their glass is the walls' assembly and cannot when it is their own: simple glazing is one equivalent layer with no cavity to hang a slat in, and naming such a surface on a WindowShadingControl is a severe error rather than a blind that quietly does nothing. So the shading control is written for the surfaces it can serve, and the glass selector says which those are. Its precondition asks the same question Daylight's now does, of both kinds of opening: that the channel owning the glass is actually in the path, not merely that a slider is off zero — a blind with nothing to hang in used to read as engaged while writing no shading control at all.

  • Holidays you define. The Run strip's Holidays switch reads From file, Listed or None, and under it is the list itself: dates you type, days you remove, and five published calendars — United States, Canada, England and Wales, France, Germany, the same five regions the tariffs cover — that stamp themselves into it. Each entry becomes a RunPeriodControl:SpecialDays, so a fixed date (12/25), an nth weekday (4 Thu in Nov), a last weekday (Last Mon in May) and a multi-day shutdown (12/24*9) are all sayable. The list is drawn against a twelve-month rule, and the two kinds of date are marked differently on it: a fixed date gets a tick where it falls, an nth-weekday rule a hollow mark at its month's centre, because that is genuinely all the run period knows about it. What you type is what a scheme link carries — the field and the address bar speak one grammar — and a malformed entry is refused in words, in place, rather than clamped or dropped.

    The companion is on Gains, where the occupancy lives: Holidays as As weekend, Closed or Open. That control is the point of the exercise. Until it existed, every Schedule:Compact on the desk ran For: Weekdays … then For: AllOtherDays, and AllOtherDays swallows a holiday alongside Sunday — so observing a holiday and ignoring one produced the same building, and the switch that offered the choice was decorative. It now writes a For: Holidays row. Measured on a Denver year with the weekend open and the eleven federal holidays plus a nine-day Christmas shutdown listed: 488 against 499 MJ/m², a 2.2 % difference that was previously unreachable. At As weekend no row is written and the IDF is byte for byte what it was.

    Attach a year and each entry is lettered with the day it actually falls on — 3 Mon in Jan · Mon 16 Jan — and ticked there on the rule. Anything outside the run is struck through and counted, because EnergyPlus drops such a day without a word — and there is no reading of it anywhere else: the error file is silent and the input echo lists every special day under every run period whether it lands or not.

    The reading is in days, as a set, because that is the unit that reaches the engine. One row can be a nine-day shutdown, and a shutdown beginning 24 December is simulated up to the year end and then dropped, so a row can be partly in — it says Sun 24 Dec · 8 of 9 on its own line rather than being struck through, which would be as wrong as saying nothing. Overlapping rows are unioned: the same shutdown swallows Christmas and, wrapping past the year end, New Year, so eleven federal holidays plus that shutdown is eighteen days and not twenty. Before a weather file supplies a calendar none of that is knowable, so the reading names what it can count — "12 holidays" — and becomes days once there is a year to count them in.

    All of it is checked against the engine rather than against itself: three runs, read back off their own Site Day Type Index series, agree with the desk's count exactly — 4, 9 and 10 days.

    One thing the arrangement cannot do, said here because the interface says it too. Easter is not expressible — an IDF date field carries no year, so Good Friday, Easter Monday, Ascension and Whit Monday cannot be written, and neither can Victoria Day, which is the Monday preceding 25 May and so is neither the third Monday nor the last. Each calendar therefore declares its whole national list and states, on the offer and before you press it, which days it is short and why: DE 5/9, CA 8/10.

    Attaching a station now also reads the EPW's own HOLIDAYS/DAYLIGHT SAVINGS record — not to fill the list, but to say what is in it. Every TMYx file names no holidays and no daylight saving period at all, measured across Denver, Berlin and the five files shipped with EnergyPlus, so From file has always been reading an empty list and reporting nothing about it. The strip now states it. A file that does name days offers them as one more stamp.

  • A register of standards, split the way the code splits them. The specifications live on the console head — five compact accordions beside the desk's own subheading, each folded to a name and a live conformance chip. Standards are laid over the drawing rather than in place of it: applying Passivhaus Classic writes the ten controls it has an opinion about — insulation derived from its U-values with the surface films taken off, a window at U 0.80 with the g ≥ 0.5 its own inequality demands, an airtightness converted from n50 0.6 by the divide-by-twenty rule, mechanical ventilation at 75 % recovery — and leaves the massing, the site, the context, the engine settings and the tariff exactly where you put them, so what would it take to build this to Passivhaus is a question you can ask of the building already on the sheet. Every number carries the arithmetic that produced it, the way the bill's rates do. Whether the desk still meets a specification is measured off the controls each time they move rather than remembered: nudge a wall and the chip falls adrift by itself, naming the clause that drifted. Alongside Passivhaus sit EnerPHit, LETI's commercial office targets — which set no control at all, and are in the list precisely to show that a specification and a target are different things — and two of this sheet's own partis, a shaded free-running heavyweight and an all-glass office plate, labelled as the sheet's own rather than borrowing anybody's authority.

  • Peak loads, read beside the energy. A demand is what a building costs to run; a load is what has to be there on the worst hour, and it is the number the plant, the risers and the distribution are actually sized from — so the scoreboard now carries peak heating and peak cooling in W/m², and the shelf keeps the heating peak beside each scheme's energy. Passivhaus's heating load ≤ 10 W/m², previously listed among the things this sheet could not judge, is now one of its lines, alongside the demand it is an alternative route to rather than a second hurdle. Unlike every other criterion here a load does not need a weather file: sizing days are precisely the conditions plant is designed against, so Target.needs distinguishes 'year' from 'run' and the board answers something on a desk that has never been near an EPW. It costs no new output request — the hourly system transfer rate the balance rail already draws is the whole of it. Worth seeing: the stock desk built to Passivhaus in Denver clears the heating demand at 8.6 kWh/m²·yr and misses the load at 13.9 W/m². Peaks get treated as an afterthought; this is what that costs.

  • The targets became a scoreboard on the sheet, under the results they are read from — every standard's criteria against the one run, all at once, since nothing is remembered and there is no "applied standard" to filter by: one solved year, every published line it would clear or miss. Criteria read as asks for · reads · margin, with one redline mark on a miss and none at all on a pass; a criterion whose limit PHI sets per building prints its reading with no verdict, and one the run cannot answer says what to do about it instead of standing as an em dash. Each standard's accordion also lists what this sheet cannot judge — the blower door, the primary energy, the thermal bridges, the peak heating load — because a panel showing only the half it can answer would read as a certification, and a one-zone shoebox with ideal loads is not one.

  • Chasing a standard. Any standard on the scoreboard can be armed with the same square marker the run ledger and the console's patch buttons use, which reduces it to its single worst line and letters that up beside the drawing — Peak heating load reads 14.8 against 10, over by 4.8 W/m² — with a ghost of where the margin stood when you took hold. The scoreboard reads a run; this reads a gesture, so the answer to "is what I am dragging right now helping" is under the hand rather than a screen away. The worst line is chosen by ratio rather than raw difference, since being 3 over means one thing against a limit of 15 and another against 55, and the line always says how many of the standard's criteria it is speaking for — a verdict from the two a design day can answer must not read as a verdict on a standard that states four. Chasing is the bill's pin in another column: chosen, visible, unchoosable, and making no claim about the building, so conformance stays the measured thing it was.

  • Kept schemes. Save this scheme joins the run bundle and the scheme link in the run log, and keeps the whole desk — stored as the very fragment the link button copies, so keeping and sharing are one format and a scheme kept today still opens on a page that has since grown a channel. Each kept scheme carries what it was reading when it was kept, and the register differences the sheet against it — but only where the two are like for like, the same kind of run in the same currency over the same end uses, which is the bill's own refusal restated on data that survives the browser's storage. Restoring a scheme that names the attached station is instant; one that names a different station is a different climate, so it goes through the link and the page reloads into the decode that already knows how to refuse a weather file it cannot fetch. Schemes are named after they are kept, not before, because being asked to think of a name mid-thought is what stops anybody keeping anything. The shelf holds twenty-four and, when full, says so rather than quietly dropping the oldest; a shelf that cannot be read is refused whole with the reason in its place, never drawn as an empty one.

  • Parameter studies. Every scale on the console now carries a quiet Study action that sweeps that one control across its full range — about twenty solves of whatever run the sheet would make, a second or two of engine on the design days, twenty seconds or so counted out per run on an attached year — and draws the response as a small curve under the control, with a redline tick standing where the control stands now. What the curve reads depends on the desk: free-running, the highest hour of zone temperature in the warm pen and the lowest in the cold one (the summer and winter design days' extremes, or the year's, sizing days excluded); with ideal loads in the path and a year attached, the demand intensities instead — TEDI in the warm pen, CEDI in the cold, and the building EUI in graphite, in kWh/m²·a of the bill's building section, delivered demand before any plant. The desk itself never moves during a sweep: the model ends byte-identical to where it started, the address bar and every readout hold still, and any real gesture cancels the sweep instantly. A study names the desk it was swept against and dims once any other control moves; dragging the swept control just walks the tick along a curve that is still true. Studies are cleared by a station change (they were swept under the old climate) and absent on priced channels, whose controls never reach the engine. A failed sample is drawn as a gap in the curve, never an invented point.

  • Shareable scheme links. The address bar now carries the whole desk — every control off its default, the patch state, and the attached TMYx station with its exact year window — as a readable URL fragment (#v1&width=20&wwrS=0.35&stn=725650), rewritten on every gesture release. A Copy scheme link action joins the run ledger beside the bundle download, and the bundle's manifest now cites the link, so a run can be reproduced locally and re-solved live from the same download. Opening a link rebuilds and re-solves the scheme; a link that cannot be honoured — an unknown key, an out-of-range value, a station whose archive cannot be fetched — is refused whole with the reason in the status line, never half-loaded. The encoding is versioned so links keep working as controls are added.

  • Downloadable run bundle. A Download run bundle button on the run ledger packages the exact IDF and EPW handed to the engine together with the tabular report and console log it wrote and a plain-text manifest, so any run can be reproduced in a local EnergyPlus rather than taken on faith. A design-day run carries no weather file, and the manifest says so instead of shipping a fabricated one.

Fixed

  • Bypassing Fabric fatalled any desk with an opening on it. Patching that channel out sends every surface adiabatic, and EnergyPlus refuses a FenestrationSurface:Detailed or a Shading:Zone:Detailed whose base surface is one:

    ** Severe ** FenestrationSurface:Detailed="ZN001:WALL001:WIN001", invalid Building Surface Name="ZN001:WALL001". ** Fatal  ** GetSurfaceData: Errors discovered, program terminates.
    

    One wall window was enough — the desk ships with one — so the flask the Fabric strip advertises was only reachable by patching Glazing, Shading and Skylights out by hand first, and reaching for it any other way stopped the run. It was recorded as unfixable through Channel.requires, on the grounds that a precondition can only ask about channels already decided and Fabric is declared at 07, below all three that would need to ask. That was the wrong reading of the machinery: being bypassed is an input to that loop rather than something the loop decides, so it can be asked of any channel in any order. requires.test now takes an off(id) beside its on(id), and Glazing and Skylights refuse themselves with a sentence each instead. The appliers ask the document as well — the boundary applyFabric has already written, rather than a parameter — so an opening is never written where it cannot stand, however it came to be there.

  • The low-e coating note named the wrong coating. The Glazing strip said "0.04 is a hard coat"; 0.04 is a soft coat's emissivity. A pyrolytic hard coat is fired on at the float line and sits near 0.15 to 0.20, four times that, and the two are chosen for different reasons — one survives handling, the other performs. Both are landmarks now, with uncoated float at 0.84.

  • Five landmarks could be seen and never reached, which is how the rule that catches it came to be written. input[type=range] only ever returns min + n·step, so a case declared at a published figure that falls between two positions draws on the face, names itself in its tooltip, and can never once be the reading — the reader is shown a place they cannot stand. The BLAST infiltration constant (0.606 against a 0.01 step), the DOE-2 wind term (0.224 against 0.005), BLAST's stack term (0.03636 against 0.001) and two of ASHRAE's lighting allowances (imperial figures landing at 6.89 and 10.76 W/m² against 0.1) were all of them. They are declared now as the narrow band the step grid actually makes, with the published figure in the note, and readLandmarks throws at module load for any that is not reachable.

  • A fin was read as an overhang. A wall carries up to three shades and the overhang is written first, so the quantities panel took the first shade on the south wall and reported its reach — which on any elevation carrying fins and no overhang is a fin's depth, lettered as Overhang, south 0.40 m · PF 0.29 over a window with nothing above its head. The overhang sits at one height and a fin runs sill to head, so the two are told apart by their own geometry now rather than by which was written first.

  • "Heat only" and "Cool only" fatalled the engine. The System strip's Available selector wrote the thermostat's control type number — 1 for heating only, 2 for cooling only — but left the control itself named as the ThermostatSetpoint:DualSetpoint the Always setting uses. EnergyPlus reads that number as a thermostat type and then looks for a control of that type in the zone's own list, so a 1 over a dual setpoint is not a dual setpoint with its cooling half suppressed; it is a control of a type the zone does not have, and the run stops in get-input before any environment starts:

    ** Severe  ** Control Type Schedule=CONTROL TYPE **   ~~~   ** ..specifies 1 (ThermostatSetpoint:SingleHeating) as the control type. Not valid for this zone. **  Fatal  ** Errors getting Zone Control input data.
    

    So two of the four settings of that control could not be solved at all, on any weather, on any desk — the sheet reported only that the engine had crashed. Each setting now writes the setpoint object its number names, and the meters say what the labels do: on a Boston 725090 year Heat only heats and never cools, Cool only the reverse, and Always does both.

  • The results schedule and the bill collided on a phone. Both are tables of right-aligned figures, and a bill of a year's run under a pinned scheme wants seven columns: an end use, three bases and a change against each. At 390 px that is about twice the sheet's width, so the figures ran into one another — 18,456 −3,193$1,090 −$1893,727 −645 was one row of it — and the results schedule pushed its unit column clean off the sheet, which is the one failure a page whose claim is that every figure means something cannot have.

    Below 620 px a row now folds the way the desk's strips fold below --index: the quantity keeps its own line, and every figure stands on a line of its own beneath it, carrying the head it was under. Nothing is dropped and nothing scrolls out of sight. The head each cell carries is set where the cell is built, so the words over a column and the words beside a figure are one string and cannot drift, and keepTableSemantics states the table roles outright because display: grid on a row drops them in every engine — a reader on a screen reader would otherwise lose the structure at exactly the width where the figures need it most. Above the breakpoint the two schedules gain the one thing they were also missing: a gutter between one column of figures and the next, which is what had COST and CARBON (KGCO₂E) running together into a single head as the window narrowed.

  • The window-to-wall ratio counted every opening in the model against the wall area alone. With nothing but wall openings in the document that was right by accident; with a roof that can be glazed it would have put the rooflights into the numerator over the walls' denominator and reported a ratio about no part of the building. Openings are now sorted by the surface they are cut into, and the quantities panel reads Glazing, walls and Rooflights separately — the first of which was already mislabelled Glazing, south while summing all four elevations.

  • The model console described itself as sixteen channels while carrying seventeen, in the desk's own subheading, in the general notes and in four layout comments. It is eighteen now, and says so.

  • The console was unusable on a tablet in landscape. The desk is a column of viewport height with a fixed head and a fixed rail either side of its one scroller, so a short window took its room out of the eighteen channels and out of nothing else. Measured on an iPad at 768pt: the head was 458px, the rail 172, and the channels were left 104px — 14 % of the desk — to scroll 12,147px of controls in. A phone held sideways at 390pt got 3px.

    Three changes, in the order the room is worth taking back. The register folds on a short desk, since it was holding 323 of the head's 458px and is the one block on the desk you reach for occasionally rather than constantly — closed it still reads, carrying built to Passivhaus Classic or 5 standards, by the same rule that keeps a folded strip a reading. Below 600px tall, where no amount of folding rescues a column, the desk goes under the sheet and becomes its own index sheet — the escape the phone breakpoint already takes, since it is one question asked in two directions rather than two questions. And Widen the window is no longer said to a touch device, which has no window to widen and was losing two lines of the head to being told so.

    The channels now get 381px on an iPad 9.7 in landscape (was 104), 433 on an iPad Air (was 156), 513 on a 1440×900 laptop (was 298), and 710 on a phone in landscape (was 3). A desk tall enough to hold the register open — an iPad Pro 12.9, a desktop — is unchanged.

  • The scoreboard and the kept-schemes shelf lost their table semantics on a phone. Both fold to a block per row below 620px, and display: block on a tr drops the implicit row and cell roles in every engine — so the two tables the register is made of read aloud as loose numbers with no criterion attached to any of them, at exactly the width where the columns are gone and the roles are the only structure left. They now go through the same keepTableSemantics the results schedule and the bill do.

  • The scoreboard's Chase marker never said what chasing was. A bare verb on a small button names the control and explains nothing, and the five copies of it down the board were identical to a screen reader — five buttons reading "Chase", none of them naming the standard they belong to. The board's lede now prints the whole of it in a sentence, which is where an explanation on this sheet goes: a tooltip floats, and a hint that only exists on hover does not exist on a phone at all. The marker carries the same sentence on its title and aria-label, with the standard's name in it, and both halves flip when it is armed — Stop chasing Passivhaus Classic: take its line down from beside the drawing, rather than a tail still describing the state being left.

Changed

  • A run that fails can be downloaded. The bundle used to ride on the readings, so the download went dark at exactly the moment it became worth having: a fatal leaves no plate, no schedule and no bill, and until now it left nothing to carry off the page either — the one run nobody could debug here was the one run nobody could debug anywhere else. It now follows the run rather than the readings, and all three ways a solve can end without results — an engine that never accepted the model, a fatal, a run that came back clean but wrote no zone temperature — bundle exactly as a good one does.

    A failed bundle is not dressed up as a successful one. The button reads Download failed run, the file arrives as …-failed.zip, the manifest says so on its first line, quotes the sentence the page reported under Why it stopped, and leaves every figure the run never reached — the hours, the exit code, the error counts — as an em dash rather than a zero. It also points at results/console.log first, which is where the answer usually is: the engine echoes the whole of eplusout.err into its console, so the bundle carries the severes naming the object and the field, where the page itself could only ever show a count of them.

  • The run now follows the weather file's calendar. RunPeriod's day of week for the start day had been pinned to Tuesday since the model was written, which overrode what every weather file says about itself — TMYx declares DATA PERIODS,1,1,Data,Sunday,1/ 1,12/31 — and put every annual run on an invented year. Left empty, EnergyPlus takes the file's own start day and picks a real non-leap year to match it, so weekends fall where weekends fall. It is one field, and it is the only difference between the default IDF before this change and after.

    It matters more now that the year can be run in pieces. The field anchors to each run period's own begin date, so pinned, a January and a June would both start on a Tuesday and sit in two different calendars. Empty, they share one: measured, January begins Sunday and June begins Thursday, which is 2017.

    This moves results. Every annual run's day-of-week alignment shifts by two days, so anything scheduled by weekday — the occupancy band's weekends, the setpoint setback — lands on different dates than it used to, and any figure from a previous annual run is not comparable. Design-day runs are unaffected; a run period is not simulated without a weather file.

    It also makes the holidays true rather than approximately placed. Under the old Tuesday, Martin Luther King Day resolved to 21 January, Memorial Day to 27 May and Thanksgiving to 28 November. Following the file they are 16 January, 29 May and 23 November — the real 2017 dates, which is the year EnergyPlus picks for a Sunday start. And the weekend holiday rule finally does something real: with it on, New Year gains Monday 2 January and Veterans Day moves from Saturday the 11th to Monday the 13th, exactly as observed.

  • The run period is a year of months, not two sliders. Run from and Run to on the Run strip are replaced by a twelve-cell calendar: tap a month to take it in or out, sweep across several in one gesture, or walk the year with the arrow keys and toggle from the keyboard. Months need not be consecutive — each unbroken group is handed to EnergyPlus as its own RunPeriod, so January and July can be solved without the spring between them, and the strip says in the engine's own words how many run periods the mask makes. Below 400px the year folds to two rows of six rather than shrinking twelve cells past reading. A run needs at least one month, and the last one standing says so rather than quietly refusing.

    The rest of the sheet follows the run: a solve can now hold several weather environments, so the results schedule heads a column per run period with the months it covers, the chart letters each period's month ticks, and the run bundle's manifest names the periods instead of claiming "Annual". A weather file is no longer taken to mean a year — a bill metered over four months says so in its lede and draws no per-m² intensity, because every benchmark that figure exists to be compared against is twelve months long.

    Breaking, deliberately: the beginMonth and endMonth parameters are gone rather than migrated, and a scheme link naming either is refused. Nothing has been published yet, so there was no link to carry forward and no reason to spend a link version on one. The run period rides in a link as months=001110000000.

Removed

  • A fifth weekday is no longer sayable. 5 Fri in Dec parsed happily and would have stopped the engine dead in any year December had only four Fridays — ** Severe ** SetSpecialDayDates: … not enough Nths, a fatal error, not a warning. The nth now runs 1 to 4, plus Last. Every month has at least 28 days, so those five always exist, which makes the grammar total: every list that parses runs, under every calendar. Nothing in any published calendar was a fifth weekday.

  • Two entries starting on the same date are refused. EnergyPlus states plainly that it gives "no error message on duplicate days or overlapping days", so the second would have disappeared into the first without a word.

About this release

idfkit-shoebox is a private, continuously-deployed static site, not a published package — pushes to main already ship to production on their own. This tag exists so the sheet's own title block can cite a real revision (E-01 · Rev 0.2.0, linked to this release) instead of a build's bare commit sha, and so the run bundle's manifest and this page give readers something to check a figure against later. There is nothing to npm install.