Changelog

2026-09-10

Names that look like duplicates are now questions on the review tray, and something reads them before you do. That list used to be worked out again every time the Script page opened, and nothing checked it — which is how a card asking to combine eight unnamed scene placeholders reached a producer. The names are compared when the script is read, the same second reading that looks at every other question looks at these, and they are answered where the rest are: combine into the name you pick, keep both, or neither, with your note. Keeping two people apart still tells the next read to keep them apart, and combining two spellings now sticks. A duplicate you make by hand between reads waits for the next read.
Before a read starts, the Script page repeats back how it read your note. You can edit a line, drop it, or keep the note as you wrote it. Hedged wording stays yours. A first read waits until you answer. A re-read with the same note skips that screen. Until you confirm a line, we work from your note exactly as you wrote it.
Picture Cars and Production Design get their script work. Vehicles on camera go to Picture Cars. Sets, backings, signage and graphics go to Production Design. Existing vehicle entries move to their picture-vehicle owner; crew transport stays with Transportation.
Saving a brief whose last reading failed reads it again, even when the words did not change. The page says we read it again. A save of an unchanged brief that last succeeded still writes nothing.
A second reading that quotes the script now does the bookkeeping. A split, a fold of two of the same thing, taking something that is not an item off the list, citing the line the reading named, moving department, or putting an item on the scenes its own line names — those leave a receipt you can undo. Two names for one person still ask. A line the page never said still asks.
Sunglasses, gloves and an N95 sit on the costume and props lists both. Costume establishes them; props handles them on set. The Script page no longer asks which department owns that kind of item. A passenger mask or anything else that is one department still asks.
A count the script never gave is left blank. Carry-ons and the like stay unquantified. Nobody is asked how many. A number the page does give is kept.
Saving how this show reads its script now says if that note could not be turned into questions. The brief itself is kept. The page says so, and to save it again to retry. The standing questions stay until then.
A script longer than we can read whole is refused, with the page limit. We never read an outline instead. Nothing is read, and the Script page says so.
The re-read dialog says whether these pages are already read. When a matching read is on file it says so, and that only changed pages would be new work. When nothing matches, it says this read is new work. The count of answers that carry over is still there.
The element check says what happened to its objections. The Script page separates corrections, questions for the office and changes that could not be applied, with reasons. Older reads say when those outcomes were not recorded.
A Pulse send that fails does not stop the rest of the show's tick. The failed send is recorded. The next person, and the next kind of notice, still go out. The Pulse board and the show's HQ Pulse tab name the channel, the last error and the time. The line clears when a send on that channel gets through. A simulator is named as a simulator — the same words the rest of the product already uses.
A re-read that stops after the pages are on the board still offers the tray. The Script page says the pages are on the board and the questions on this read may not have been assessed. A failure before the board is saved still says to ask the production office. Show setup → Tools names the cause for staff.
The Script page keeps answering while a script is being read. A long read used to take the whole database for itself for seconds at a time, and anything that asked for the state of the read in that window got an error instead of an answer — four times during one script. Reads and writes no longer wait on each other, so the page keeps telling you where the read has got to while it is still going.

2026-09-09

Show setup now has a main production location. The phone list and address search use it. Empty until the office sets it. A first scheduled day can hint at a place; it is never filled in on its own. The onboarding form says which show location it is using. Address search looks there first, prefers a house and street when you start with a number, and names the region it searched. A country without a stored map box still narrows the near search to that country.
A show's paperwork now wears that show's look. Throughline staff set it from the show's HQ page — the printed name, a mark, the colours, the type, and the line at the foot. A show that has set nothing already uses its own colour and the house type. The next breakdown, crew Day Out of Days, and turnover lineup pick it up; paper already sent stays as it was.
When too many questions on a tray would not get past a professional, the Script page says so. The line is the parse team's: N of the M questions on this tray would not get past a professional; the parse team has the list. It sits under what the read landed, only when that is true.
Script review can answer what the pages already settle. With the review lane configured, supported factual answers become receipts under Answered for you, with Undo. Questions about meaning stay for a person. Different names are never silently joined, and nothing is removed from the board. The review's cost and counts are recorded; its Script-page summary and live calibration are still to come.
Library file tiles load a small picture, not the original. When a file has a stored thumbnail, the asset grid, the ingredient browser and the canvas fetch that. If the filespace picture is still on its way, the tile says Preview on its way instead of stretching the one-pixel stand-in.
Every script read now uses intelligence. Uploads, revisions and re-reads all get the full read: the cast, the questions and the second reading. A show can still switch intelligence off for the day-to-day features it uses once the script is on the board. The script itself is too important to read any other way.
A show with no model connected is told it cannot read a script. One line, in the same words on the Script page, in the activity feed and on the read itself: add a provider key under Keys in HQ, or ask us to set one up for the show. Nothing is read and nothing is charged.
A script read before this change says so, and can be read again. The Script page names the date it was read without intelligence. Reading it again is the office's click, priced like any other read; it never happens on its own.
HQ's Script intelligence tool says reads always use intelligence. The switch is for the day-to-day once the script is on the board — today, the methodology brief's own reading of the note. A show with no provider key is told it cannot read a script, in the same words as the Script page.
A running read names when it started and when it will stop itself from the first look. The remaining-time line says it is an estimate, or that the first stage is still setting the pace. It does not say the software is working the clock out. The stages speak in production English: Discovering characters, Matching names to people.
A script read stays on the Script page when you come back. Another browser, another device, or a sign-in that cleared this one still shows the running read, who started it, and when. A read that finished while nobody was looking lands the next time anyone opens the page.
Show me the twelve opens one card per place. The folded scene question becomes a receipt — Expanded into 12 cards — under Answered for you, with Undo. Each place is an ordinary scene card with its page and scene, and Break it out or Leave it inside.
The Script tray groups element questions by what they ask. Which scenes, something the list is missing, one item or two — each kind is its own sheet, so you answer one kind in a pass. A department question is a picker. A scenes question is tick-boxes. Evidence is the script page, with the scene and page above it — never a line number. The list of lines judged not to be characters names the scene and page the same way. A scene the second read already took off sits under Answered for you, with Undo to put it back. When a spelling of a person was already settled, that receipt sits on the open card as its own line, so you can see what was answered without opening a closed row.
Review evidence reads as a script page. Housekeeping cites scene and page — Sc. 22, p. 47 — never a line number. The quoted lines sit in Courier like the page they came from, with a line or two of context, and the cited line marked. When the saved draft has no page breaks, the card says so.
The second reading can make a supported correction and show what changed. A department move, a named missing item, a scene taken off an entry or a duplicate folded into another comes with a receipt and an undo. Undo puts back the saved entry, including its evidence and scene links. If someone has edited it since, the card asks for review instead of overwriting their work. Splits, removals and quantities stay questions, with the reading's proposal named.
Element cards answer the question they ask, and the obvious ones are already answered. When the second read says a scene rests on a line that does not describe the item, the scene comes off that item and the card becomes a receipt: it says what was taken off, and one click puts it back with the line it was cited to. The item never leaves the board — if that scene is the only one it has, the card asks instead. Everything else asks with answers that fit the question: one item or two names both items, which department lists the show's departments with the current one first, how many, which scenes to tick.
The review tray asks once per person, not once per spelling. Every card about the same person is now one card, whatever each was asking. A name the script spells two ways is merged and shown as a receipt you can undo. Two different names are never merged for you: that one always asks, and nothing goes on the board until you answer. And the places a script cuts to without a slugline come as one question listing all of them, each place once, instead of one card apiece.
A nickname three people hold is three questions, not one. Each holder is asked whether that name is theirs. The other people stay on the card so you can see their cues; answering one card does not decide the others.
One person reaches the tray as one card. When the read finds a gunman and a named reveal that are the same role, it asks once, with the line that says so: Same person or Keep separate. It never folds them on its own. A second card about a name that already has one folds into it. When that extra look ran, the Script page names what those calls cost, next to the rest of the read.
Connected tools can put a recent combine back. A script approver can undo a saved combine for 72 hours, provided its attached work has not changed. The activity feed records the result. The housekeeping button is still to come.
A recent combine can be put back for 72 hours. Housekeeping's confirm names that window, and Put it back reads the same offer the activity line carries. A scene, booking, or later combine that already changed the work is refused in those words.
A line of action is not a person on the cast list. If a read uses a sentence as someone's name, they stay on the board as Unnamed until you say who they are, and a card quotes the line. Real names — including ones like ATC (LA CENTER), and a slash of two names — stay as they are.
Combining character names brings their props and wardrobe with them. Items named for a character take the surviving name in scenes and department lists. Searching the old wording still finds them.
Decisions the parse already made sit under Answered for you. They do not count in the number at the top of housekeeping. Each one says when it was applied and has Undo. Accept all keeps them in one click, using whatever answer that card already offered. If one cannot be kept, the rest still are, and the tray names that card by its headline and says it stays there.
An interrupted script read gets one automatic retry. After a server restart, the Script card says the read is carrying on. Answers already saved are kept. If another restart interrupts it, the read stops and says so.
On housekeeping, an AD can combine two people even when the list didn't propose them. Pick the names, choose which one stays, and confirm. Every scene, image and file that pointed at the others will point here. The other names stay as other ways of writing the same person.
A stall dump is readable from HQ. When the event loop stalls, or a script read captures its stacks, the file lands in Diagnostics — staff sign-in, newest first, click to read. A name that walks out of that folder is refused in words. Files older than fourteen days are dropped at startup.
A one-line schedule from the AD's own software now comes in. Yellow and Goldenrod one-liners are read as one-liners, not turned away as scans. When the heading is glued to the first name on the list, that person is still on the list, and the scene still reads LEO / STELLA. A file that really has almost no text is still called a scan. A file that has text but isn't a call sheet, one-liner, shooting schedule, or DOOD is told that, without the guess.
If a show code is already taken, the refusal names the show that holds it. Creating Don't Descend as DD when an earlier Don't Descend already has that code now says which show, and whether it is on the air or wrapped. A wrapped show can release or change its code from Show setup; a show that is still on the air cannot, because mounts and folders use it. A code that was assigned from the title on an older show says so on that page, so it is not mistaken for one somebody chose.
Restarting a script read no longer doubles its bill in HQ. Saved answers count once across attempts, including failed reads. Money spent on calls that were not saved still counts.

2026-09-08

The Brain panel says what the last automatic read actually did. A skip or a stop at the day's limit shows the same sentence the activity line does. A finished read still says it read recently.
Two spellings of one prop no longer knock the staying one off the sheet. If one record is leaving a scene and the other stays — same department, labels that differ only in capitalisation — the one that stays keeps its row. When an item moves onto a scene that already has it, the notes, the quantity and the lines it quoted travel with it rather than vanishing.
Opening material added or changed in a later draft waits for AD approval. Its script change names the source lines and asks about numbering. Approval keeps printed scene numbers and a changed opener's existing scene identity.
Two wearers no longer hide a question about two different objects. The Script result names dismissed per-person split questions. Duplicate entries and names that need an owner or description come to the review tray as their own questions.
Two people can answer different sections of one review card at the same time. Both answers and both notes stay on the card. Asking later or handing the card to another department also keeps the answer someone just gave.
A set-dressing item stays on a scene whose heading names that set. If the kept quotes are from another scene of the same set, it is on both. A costume that belongs to a character stays on a scene that character is in. Quotes from a different set, or from a scene the character isn't in, still move it.
A script read you start still runs when a spending limit cannot be set. The Script page and the activity line say so. A read nobody is watching does not start, and the message says how to run it yourself or set a limit in Show setup.
The assistant's timer no longer re-reads when it cannot read today's spending ledger. The activity line says so. A re-read you start still runs.
A name spoken over the P.A. is still that person. A delivery note in parentheses after a name — on the P.A., on the radio — still puts them on the cast list, including the cold opener.
A script read no longer stops on the first call when the text model was retired. The message says which model to pick.
Nothing is filed under a scene your board does not have. A read could hand back a bracket, a comma or a paragraph of its own notes where a scene number belongs, and items rode in on it — one read had 114 scene keys for a 97-scene board. Those replies are dropped now and the read says so, with the item count, and the commit refuses them a second time. The cold open is a scene, and it is kept.
A script read that had to be started again still shows what the whole read cost. If an earlier try stopped mid-way, the Script page names the money already spent, not only the last try — including a read that finished before this did. The spending limit counts that same figure, so a resume cannot spend past what the office approved as if the earlier try were free. The last try is named beside the total. Each part of the read shows what it actually spent next to what it was quoted.
A camera direction is not its own scene. A line such as EXTREMELY CLOSE stays inside the scene it belongs to. A jump that names a place — one already on a slugline in these pages, or a word like cabin, hall, or deck — is still its own lettered scene, and it keeps the parent's INT or EXT and time. A time of day left on a slugline (a moment later, minutes later) is not a place, and neither is shot. If a camera line was already split out, the next re-read puts it back, and its items go with it.
A prop that turns up in five scenes is one item with five scenes on it. The read used to pick one scene for it and the whole list sat there. Now it reads the lines it quoted and adds the scenes they come from. It never takes a scene away on the strength of a quote — a read only quotes an item a few times, and a hero prop is in far more scenes than it is quoted in. Where not one quote lands anywhere the item was filed, the quotes win and the item moves. And the pile the read could not place — items attached to the last scene it was asked about, quoting a line that never names them — comes to you as a card asking which scenes they are in. The card is a question, not a removal: nothing leaves a department's sheet because the read was unsure.
The person who creates a show can run its first read. The role they typed is their job, not a lockout. A show whose creator was not its administrator is fixed at startup.
The line under a running read uses your crew-card time, not a UTC stamp the browser guessed at. On a show with intelligence on, before the stop time is known it says it is still working that out. A show without intelligence, and a run that truly has no stop time, still say No deadline. Next fire times, a delivery's sent stamp, and the home feedback clock use the same zone.
Pressing Break down these pages says the note is being read. Until the save comes back, the button is off and reads Reading your note. A line says the note is being read before the pages are. A second press in that window does nothing. If the save fails, the reason is above the buttons and you can try again.
Costume opening Mara from Costume's workroom sees Costume's starting points. Other departments that have published one for that item are there to pick. Applying theirs adds their flow and leaves who owns the item alone. A starting point for a character will not apply to an asset, and the workroom list only offers starters that match what you picked. On the workroom, Apply a starting point applies the one you chose, or the button itself says why it will not. The list without a workflow opens on the work that department owes. A scene on the board that has none of this department's elements is not this department's work. Everything of this kind is the wider list.
A revision that changes a prop or a line deep in a scene files a card. A byte-identical re-upload files nothing. The first upload after this lands will surface every change the old opening-only compare had been swallowing.
The assistant now reads what the show is working on, not the first pages of the script. On a big show its picture is too small to hold everything, so it starts with the scenes on this week's shoot days, then the scenes changed in the last fortnight, then the rest in script order. What the assistant has read says what would not fit — "the script: 31 of 182 scenes" — instead of only saying that something was trimmed. The day the unit is on, its general call and the scenes up, now travel with every conversation. A short show reads exactly as before.
Ask the assistant about a day by its number and you get that day. "What's the call time on day 12?" answers with Day 12's call time and Day 12's scenes, from any screen, with your department's items on it. Ask about a day that isn't on the schedule and it says so, and says what the schedule does run. Before, it answered about today.
Every show with Intelligence on gets a second check of its element list. Nobody has to find a switch for it: the read goes over the props, wardrobe, and set-dressing list a second time and questions anything it cannot stand behind, the way it already does for the cast list. Anything it questions comes to you as a card rather than quietly changing the sheet. A show that has turned the second check off keeps it off.
The first line of a revision card names what was added and taken out. The crew-readable sentence follows it. Marking a change as having no impact is a named decision on that card, not a rewrite of the line.
A link that names a version opens that version. Pulse and the workroom Review and With the director cards carry it through. If that version isn't available to you, the page says so in one line and shows the current one.

2026-09-07

A show that asks for a second check of its element list is quoted for it. The estimate before a read now includes that check, it is named among the parts a changed brief would re-buy, and the limit the read stops at is worked out from a figure that has it in. A show that has not asked for the second check sees exactly the estimate it saw before.
A stalled read is a card, not a raw error. If a script read has gone quiet, the Script page says so and how long it has been. Stop this read is there for an AD, a producer, or an admin. Anyone else sees why they cannot. Starting another upload while one is still running opens that same read.
A runner cannot start a paid script read. Upload and paste ask the same people a re-read already asks: ADs, producers, or admins. The job names who was signed in, not whoever the form typed.
An unheaded opener now raises one question, not two. If a second undeclared-scene suggestion points into the same opening lines, it is folded into the opener's card. A separate mid-scene question now names the scene it would split and the lettered scene it would create; if the line cannot be placed in a scene, the card says so plainly.
Show setup → People can keep someone out of automatic invitations. Crew who can change that page pick the person and why. They will not be added to channels on their own; someone still confirms the seat by hand. Clearing it invites them as usual again.
Show setup can change the rest hours, the working week, a department's details, and how episode codes print. People who can change those pages do that there. The week and a privacy change show what would happen before you apply them. A department's default calendar names the workflows that still follow it, so the first look is every follower, not none. If the show moved since you opened the page, it says so and asks you to reload. A department head can still turn Workflows on or off for their own department. The show code stays put.
Show setup can change the title, the phase, the colour, and the clock. People who can change The show do that there. The show code stays put: mounts and folders use it, and changing the title does not. If the show moved since you opened the page, it says so and asks you to reload.
Saving how the show reads its script now asks whether we heard it right. Each reading is Yes, No, or in your own words, with the recommended choice marked. Nothing typed is written until you confirm it. The note keeps riding the read until you say the answers cover it. A note about who is who, said in your own words, rides the character matching — not the element read — and shows next to those answers the next time you open the interview.

2026-09-06

Re-reading pages now shows last time's answers, and says what a changed brief would re-buy. The interview on that door is already filled in. Save with nothing changed writes nothing. A booking change re-buys the character matching, not the whole read; a show that never saved a new brief keeps the read it already paid for.
A new show starts ready to read its script. HQ shows the show's code as it creates it and starts script intelligence on for both a single story and a season. The production office can switch it off or on again from Show setup → Tools, which also says whether script intelligence can run right now.
Your settings and Show setup speak the house voice. The page is You. Slack, connected apps, memory, and what the assistant has read are said in production English. A no on a reach is still said once.
Reading access sits on Show setup → Access. On a series, who may read ahead is next to the keys, only for someone who can change that page. Your settings no longer points at Vendors.
Your settings is yours. The apps you've connected show once. Saying no to a reach is said once, at the foot of those cards: Throughline will do less on its own, and less for your department's team through you.
A scene whose elements were not read now shows on the tray. Ask again reads that scene's elements only. Leave it empty keeps the scene without them. A long scene on the one-scene-at-a-time path is asked in parts rather than coming back blank.
A parse now shows one line, a bar, and about how long is left. The button stays Break down these pages. The same job no longer reports itself three times. The bar does not sit at the end of a stage while that stage is still the current one. When the product already priced the read, the time left is a range in minutes, not a false countdown. You can leave the page; the read carries on. When it finishes, that same block names what landed, what it cost, and how long it took. Before the drop, a show with intelligence on says so, and that a feature takes about an hour. How the read works sits on that block, while the read runs and when it is done.
While pages are being read, the progress line keeps moving as each stretch finishes. It no longer sits still until a whole batch of answers is saved. You can leave the page; a long stretch no longer looks frozen.
How the read works is a page you can open from the parse. Seven short chapters: the pages arrive, who is in each scene, who is who, the check, what each scene needs, then it is yours. It opens when you're signed in.
The show's controls now live on Show setup. People who can change the workspace, the look, the filespace, the keys, the assistant's brief, or the Day Out of Days do that there. Your settings keeps what is yours: your look, your Slack connection, your connected apps, your memory.
Dropping pages on a show with intelligence on now asks how the show reads its script, before the parse starts. Three questions: namesakes booked as two people, one performer across roles or ages, and anything the pages will not say. Remaining room is visible as you type; past 12,000 characters on the standing rules, or 2,000 on a note, the save is refused rather than trimmed. Skip is one click and is remembered for that revision. The Script page keeps the standing brief on a card, with Edit. The brief still reaches every prompt the way it already did.
The figure before a read is priced on the lanes that will actually run. A first read says so and gives a wider band. When the read finishes, that figure is closed against the bill so the next one can learn from it.
A jump to another place inside a scene is now its own lettered scene. The parse files it before anything is written: 59 stays 59, the cockpit is 59A, and each part has its own eighths and cast. An action line that just ends in dashes stays in the scene, including a longer grab that is not a place name. A card can still put two units back together.
The Director now sits in Direction. Every show has that department, open, with the Director as its head. A cinematographer retitled Director is still not the Director. Handing a Direction step to a vendor is their own department's call, and still not another department's.
Show setup is a place of its own. People who hold a job on the show can open it and read the values they already may. Money, access, and vendors stay closed unless that job includes them. Settings in the rail is now your own page.
A Producer keeps the Director and Producers look after they can read the office's paperwork. Being able to see those records does not take that look away. A Line Producer still does not get it; the Director still does.
The Director's say is a written appointment. A credit that says Director, or a cinematographer retitled as one, does not grant it. The show's Director holds it everywhere on the show; an episode's Director holds it on that episode. Being an administrator is not being the Director.
A show that already existed when we started writing chairs down gets those chairs at startup. A Producer whose credit is exactly Producer can sign a Producer step again. A Head Production Designer is still not the Production Designer, and a chair that was taken away stays away. Fresh demo worlds keep the chairs they were seeded with.
Handing a step to a vendor is the department office's call, written down. A person given that job on the department can do it even if their credit says something else. An Associate Producer's title is not enough on its own. The people who get told about a department's steps are that same written roster, and only while they are active.
Who may change an approval room, say where a scene shoots, or re-file a scene onto an episode is written down on the person, not guessed from the credit. Holding a pass to read the office's paperwork is not enough to staff episodes. The assistant will not grant a permission; it names who to ask. Who signs a step named for a chair is written down on the show the same way: an Associate Producer is not the Producer, and an empty chair stays unsigned until someone is named.
The automatic re-read pot is $15 a day, and raising a limit stays. An administrator changing a show's daily figure is a permanent change to that show, not an increase that dies at midnight. The $50-a-day ceiling is unchanged. The settings record names the old amount and the new one.
The sweep that writes today's permission list down will not guess. Two people on one show who share a name stop the run before anything is written. A grant that cannot come back because it was revoked is named as blocked, not applied.
A 2nd AD and a production coordinator can now edit and publish the shooting schedule. The same written roster covers the paperwork tray, Pulse runs, the egress audit, and the weather watch. Who may do that is written down on the person, not guessed from the credit. A Producer's Assistant still cannot decide script truth unless someone has given them that job.

2026-09-05

The assistant's picture of a parsed scene names each person once. When the sheet has several cue variants for the same person, the read the assistant uses lists that person once, in the order they first appear. Seed shows without a resolved cast are unchanged.
An unheaded opener now asks the production office how to file it. Its card offers scene 1 with renumbering, an A-prefixed scene without it, or no scene; dialogue and shootable action lean scene 1, while a prologue, pure screen context, or any existing shoot day leans A. The setting only preselects an option — an approver must choose it — while cast, elements and mid-script lettered splits keep their existing behavior.
Tiles now load a small picture. Library cards, the canvas rail, the ingredient browser, shot plates, review strips and workroom covers fetch a stored thumbnail. The full file is served when you open it, download it, or send it to a model. Files that do not have a thumbnail yet still show the original.
A re-file plan no longer moves empty placeholder renders into a task's folder. Only the July stub signatures, and only when the file is under 100 bytes, stay on the plan as placeholder files, not moved — a small colour sidecar still files. A census script lists every version or library row that still points at one of those stubs — on disk or on the filespace — and a confirmation-gated remove deletes only what that list named. A filespace copy we cannot open is listed as unread, not removed. Filespace deletes go out as agent work; Throughline never touches the mount.
Re-reading the show now has a real daily allowance. Each show gets one pot for re-reads started by an administrator and a separate pot for automatic refreshes, reset at midnight where the show is based. Settings shows what has run and what remains. An administrator can change either amount; if a read will not fit, it stops before anything is sent and keeps the existing brief.
A vendor's delivery now arrives whole or not at all. Throughline checks and hashes the private upload, claims its retry key before writing any shared staging file, then stages and verifies every byte. A step-bound submission commits the delivery and its sign-off together before publishing the complete directory. After a restart, Throughline settles a complete or already-moved delivery, leaves an incomplete one for the same-key retry, and names genuinely missing bytes Arrival failed; none is shown as a ready file.
Vendors now have their own focused workspace. A vendor sees only steps both assigned to them and inside their live access, can leave progress notes, and can submit files for production sign-off. Their pages do not expose the production's navigation, approval buttons, date controls, pins, or internal notes. Until 1 December 2026, older submission clients that omit an Idempotency-Key still work with a dated warning; after that, the key is required so a retry cannot create a second delivery.
Who can change the show is a little tighter, in six places. A Producer's Assistant can still see colleagues' contact details, but cannot edit the crew list — that's the production office. Only an administrator can re-read the show for the assistant, or make a standing routing rule from chat (the same as changing it in Settings). A connected app can invite people and change a role, but it cannot name a head of department. A tool turned off on the Forge now shows Turn it back on, and rebuilding one is an administrator's.
The office is no longer told a read failed when it stopped at its limit. A script that hits its spending limit or its time limit still stops, and the board is still untouched, but the activity line now says it stopped — not that it FAILED. FAILED is what you retry; retrying this one spends the same money again.
A script read now stops at a spending limit instead of running away. Every read is priced before it starts and given a limit based on what reads on that show have actually cost, with room to spare. If it reaches that limit it stops and says so plainly: nothing on the board changes, and the pages it already read are kept, so running it again doesn't pay for them twice. A department second look works the same way.

2026-09-04

Provisioning a show's storage no longer waits on the click. Settings shows each step as LucidLink builds the filespace — creating it, waiting for it to come up, building the folders, setting up department access, minting the storage key — then Ready. A second click while that's happening is refused until it's done. Tearing storage down is still the short wait it was.
An item both readings of the script found is marked, and sorted to the front. When the breakdown reads the pages twice and both readings list the same item, the scene shows a quiet "Seen twice" next to it, and those items come first in their department and first in the review tray. It is a note of confidence on something already right, not a problem to go fix. Pages read once look exactly as they did before.
A film cannot have two scene 14s. When a later draft wants a number that's already taken, the tray asks: file it as 14A, or say it continues the one that's there. Neither leaves the lettered insert for now. A show that already has two scenes on one number still boots; those are the questions waiting.
A breakdown now records what it cost, including one that failed. Every lane the parse used is counted, not just the first, so a run on two models is not reported at a fraction of its price. A parse that stopped partway still shows the spend up to that point — the money is gone whether or not the answer arrived. Running the same pages again after a failure adds only what the retry actually buys: answers that come back from the bank are free, and they do not erase what the first attempt paid.
Re-read these pages without losing the board. A script approver can preview which scenes stay, which are new and which are no longer in the draft, then confirm. Shoot days, approvals and hand-set shooting locations keep pointing at the same scenes. Human answers and standing element rules carry over. If the pages or the board change after the preview, it asks you to look again.
When two departments name the same thing, the tray asks once. The card lists every department that proposed it. Yes puts a line under each of them; one department books and pays for it. A department can say it is not theirs and hand the card to another — including one that never proposed — and a department that already passed cannot take it back. If the last named department steps off, the question goes to the producer. Saying it is not an element still applies to the whole show.
A standing element rule now lands on every new scene, not only the parse. A split, an approved revision, a parse with the intelligence pass off, and a re-parse that reused the last answer all carry the element. A scene somebody took it off stays without it on those paths too.
A later draft that rewrites a scene's location line keeps that scene. When the pages and the board both carry the script's own numbers, the match is the number — not the slugline. A reused number over different content still comes in as a new scene. Older boards, and boards numbered by position, keep matching the way they do today until a full parse records how they were numbered.
A parse that cannot read a cue-shaped line as cast now says so. The scene and the activity line carry the count. The cue pattern is not widened; the drop is no longer silent.
A department's second look waits its turn, reads the show's own brief, and does not let an untouched card hide the next draft. One look runs at a time. An omitted scene no longer blocks a live proposal. A card nobody reviewed is replaced when the next look finds the same thing; a card someone put off stays put.
A later draft that only numbered some of its scenes no longer asks you to renumber the ones that caught a margin number. Those numbers stay put until the draft numbers every scene, and each one distinctly. A new scene that wants a number an omitted scene still holds is filed as a lettered insert (12A), so two scenes never share a number. A titled name on a later draft — Co-Pilot Dale beside a Dale you already have — lands on that person, with the other spelling kept as an alias.
The review tray names the read that produced the questions on screen. A later shadow run or a second look no longer steals those numbers. When the cards come from two reads of the script, it says so. The discarded-lines list belongs to that same read.
The schedule office can publish the whole show and keep the crew's copy as the comparison line. Preview names added, removed, renumbered and moved days plus scene reassignments. Marking a publication sent requires its real delivery receipt; the Days board then shows “was” beside dates and scenes that changed. An internal draft does not replace what crew were told, and a withdrawn sent copy is corrected by a new publication rather than erased.
A file card shows where that file has been. The latest move sits under the folder — from the old folder, or that it left — and the rest of the list opens from there. A file that has not moved looks the same as it did.

2026-09-03

An old Workflows filter link shows the full list, not an empty one. A saved In progress — or any word that is not Late, At risk, Ready, On track or No need-by — is treated as no filter. Late and Ready still filter as they do.
A look brief stays with the still it describes. Opening the same Released version after a restart does not ask for it again, and the box no longer holds every brief it has ever written in memory.
Moving a file on show storage keeps it the same library file. Its versions, tags and notes stay with it, and its record shows where it moved from and where it is now. A file that disappears waits 30 minutes before it is archived; if it returns later, the same record comes back.
A version tile loads a small stored picture, not the original file. Opening the version still shows the original. Worlds and seed media keep serving as they do.
A vendor submission that does not finish no longer hands its folder to the next one. New work lands under a name that cannot collide with a rolled-back row; existing folders stay where they are.
Asking the assistant to approve everything only talks about cards you can actually sign. A step sign-off still happens one at a time on the step. If nothing is waiting on you, it says so and stops there. A stand-in signature on the item names who stood in and for which role. The step's approval line opens that approval, not the whole list.
While the show's approvals are paused, a step that already has a sign-off card cannot be sent again. A decision that already landed still writes through.
A workflow row on the department list is Late, At risk, Ready, On track or No need-by — not a second status beside that. The cell is the same sentence the item header reads, so a past due never sits next to In progress. A missing need-by ranks ahead of On track, because an open date is less known than a flow that still fits.
An outside signature now belongs to one named person and one Approval step. The schedule office can bind that signer for a set time and revoke the binding later. Their approval key can Approve or Request changes on that step, and they can take back their own signature with a reason. A general write key and a name typed into the request can no longer sign for somebody; a turnover vendor lane cannot double as an approval lane. On Approvals, a workflow card now names the item and step it governs and links straight back to the work.
A workflow's Late / At risk / Ready / On track now comes from one reading of its steps, not a separate summary. Past a due date it names how many steps are behind; a raised hand or a last step that no longer fits the need-by reads At risk. A workflow with no need-by says so, rather than On track. An item reads the worst of its workflows, so it never says Ready while another department is still working.
A kept element that names a scan id now shows the element's name on the sheet, not the id. If the id has no name, the keep is dropped and the parse records which id and which scene.
Workflow notices stay off until the office turns them on. A sign-off waiting on you, a date that moved on work you own, a cascade that touched your department, and one daily digest — that's the set. Quiet hours hold a ping until morning rather than dropping it. Turning the notices off leaves a send that's already going out to finish, and doesn't dump a backlog the next time they come back on.
A published workflow can catch its items up, on purpose. The Workflows tab says how many sets still sit on an older version, and Update the N shows every change before anything writes. Fields a person set stay put; a step the template dropped stays unless you retire it. Put the update back from that same place. On the item, each field says whether it came from the template or was set here, and reset writes the template's value back.
One scene that cannot be read no longer ends the breakdown. That scene stays empty and the rest of the parse continues. If a second reader already finished, its answers are kept.
Saying where a folder belongs moves the files there. On Folder questions, That’s right puts the media with the named owner when that’s a different person or item. Leave it here files it under the parent it’s already in. Belongs somewhere else lets you pick any official home and moves a whole selection there at once. The files wait a minute so you can still cancel; they never go through the re-file plan.

2026-09-02

A workflow's formal sign-off and its Approval card now stay together. Sending the step for sign-off opens the show's Approval; Approve, Request changes, Reject, reopen, or unapprove there updates the step in the same save. The named role holder, delegate, or recorded stand-in can sign — being an admin is not a signature — and a signed step links back to Approvals with the deciding hand and date. Reopening keeps work that was already released moving while the department decides what to redo. Asking for that approval again starts from its Approval card, so the step cannot move behind it; a withdrawn submission leaves no false “waiting on you” card on Today and returns there when it is sent again.
A department can write its own workflow templates. On the Workflows tab — not a new tab — the department lists the starting points it already uses, edits the steps in a list with a live diagram beside it, and publishes only when every problem is named. A coordinator or producer in that department, the head, someone they've named on the contact card, or an admin can edit. Anyone on the show can save a real workflow as a template; if they can't publish it, it stays a private draft until the department does. Lead times copy as they are set. Taking them from this item's dates is a separate step that shows the numbers first. Nine steps is about as long as a workflow stays readable — more than that is a warning, not a refusal. If two people save at once, the second sees what it says now.
Putting a workflow change back is only offered when it can happen. Starting a step still puts back. Applying or replacing a starting point, or customising the steps, says so on the trail instead of showing a button that always refuses. Replacing a starting point now names the one that was there.
A department can take its own second look at the script. Choose which departments should look again after the first breakdown. Anything new comes to the review tray with a quote checked against the page; it doesn't change a scene until somebody says yes. A repeated item, or one the production has already said isn't an element, stays out. If the four-hour deadline arrives first, the result names which departments finished and which still need a look, while the committed breakdown stays in place.
The date and sign-off work now finishes on the item. Change a workflow's need-by and the review shows every date that follows it before anything moves. Pin one step and that promise saves on its own; the other dates stay where they are and come back as a counted list to review. The same page now runs the department's own sign-off from submission through Approve, Request changes, or Reject. If somebody skips a sign-off with a reason, it stands on the Sign-off overrides register until it is signed or the department head or schedule office marks it seen. Changes names who acted, what was there before, and who can put it back.
The Episodes board tells you when nothing is assigned to you yet. On a season that already has episodes, a crew member with no seat no longer reads "No episodes yet" or gets sent to set the season up. It says none are assigned to them, and to ask the production office about that.
On a series, the Script page lists each episode's scenes together. Scene 3 of episode 1 no longer sits next to scene 3 of episode 2. Each episode's run has a heading you can open. A scene that isn't in an episode yet sits under its own heading. A feature film's list is unchanged.
An element that didn't make the list can wait on the tray. If the parse finds something it can't place, you get a card: yes, add it; no, not an element; or ask me later. Yes puts it on the scene as an item for that department, never as a person. No means it won't ask again.
Notes on a version come back to the person who made it. When a director or a department head writes notes, that version shows up on the submitter's Today page, grouped so they can open each one and mark them seen. The notes stay on the version either way.
Folder questions now go to the department head. A wardrobe folder waits on costume, a VFX folder on visual effects — not on a random admin. The person who owns it sees one card on Today and a page of every open question, with pictures they can click and a way to answer a handful at once. Settings still shows the whole list, and names who each one went to.
A department's Workflows tab lists every workflow, and every item without one. Search, sort, and filters live in the address. Tick the rows the filter shows and the count names both units — items and workflows — when they differ. Timeline draws only dates somebody typed: a hollow start tick and a runway only where Start by and due are both there. Type a Start by on a step; it saves on blur, or it tells you it didn't. If that day has passed and the step still hasn't started, the card offers Start it, Move the start, or Mark it seen. The item page can draw the steps as a graph; more than thirty live steps says so instead of guessing a drawing. The old Flow tab is now Turnovers.
You can customise a workflow on the item, or start one from a department's starting point. On a set, a character, a location, or a shot: add a step, duplicate one, reorder, draw a wait, or add a sign-off — review the change, then apply it. Starting points sit under the steps. Every starter says how far this release can take it before you apply, and a banner if it has not been reviewed for this show yet. Apply, merge, replace, or add another workflow. A change can cover fifty workflows at a time.
A shot, a set, a character, and a location now have a page. Open one from the graph and you get its steps — every department that has a workflow on it, side by side, with who signs each step and what's waiting. The trail is on the same page; if a change can come back, Put it back is right there. A shot also shows its plates and versions. An asset still opens on the page it always had, and that page now carries the same steps and trail underneath. Vendors on a shot wait for a later release.
Type S1E5 in search and the episode comes up. On a series, the palette now finds an episode by its code — S1E5, s1e5, 1x5, or the studio's own production code — and opens that episode. A code that isn't on this show is simply no match. Feature films stay as they were.

2026-09-01

Picture descriptions have a daily limit, and HQ sets it. Every picture that lands gets a description and search tags; now each production has a daily allowance for that work, and HQ → Models is where the switch and the amount live. When a show reaches its limit, files still land and file as normal, the descriptions simply wait for the day to roll over, and the activity feed says so once. The tab shows each show's spend today, how many pictures were described, and how many were skipped.
A numbered picture sequence arrives as one item. Drop fifty EXRs, DPXs, or numbered stills in one folder and Throughline shows one row with the frame range and a picture from the middle. Folder questions count it once, the filing plan moves it once, and the looking pass reads that one picture. Every frame is still checked during the move. A missing frame stays part of the same sequence; an unreadable frame leaves the blank stand-in and never holds up the arrival.
A task's folder is laid out like everyone else's. Renders go in renders/, attachments in reference/, note frames in notes/ inside the task's room. Earlier today a task's files sat loose at its root; planning the re-file again puts them where every other folder keeps them.
A photo dropped on the filespace now shows the picture itself. The agent on the mount makes a small JPEG from the file and the card uses that. The original never leaves the drive. A clip uses one frame when the machine can pull one; if it can't, the card stays on the blank stand-in and ingest keeps moving.
A wardrobe task files with the character, not under casting. Link what the task is for — a character, an asset, a location, a scene, or a shot — on the task's own page. Costume work then sits in that person's wardrobe folder; hair and makeup sit in theirs. A prop task stays with the character's main folder. Nothing moves later just because the department on the card changed. Three costume tasks already on the pilot — Mara's slicker, Teddy's uniform, Wren's rehab-tech build — pick up that link the next time the show boots.
Storage setup and teardown finish clean, even when the provider is slow to answer. Setting up a show's drive, syncing folders, checking health and tearing one down no longer step on each other when a request runs long. A teardown the provider refuses stays marked “tearing down” and holds off any new setup until a retry finishes it, instead of failing quietly. If the provider answers late with an account we had already given up on, that account is cleaned up and its key replaced, nothing stale left behind. A drive name stays taken until its cleanup is done, so a new show can't grab a name still being cleared and an old cleanup can't delete the show that took the name after it. Moving an approved file is now crash-safe: nothing gets overwritten, and a file can't end up stranded outside the show's drive if the machine dies mid-move. Each show's storage key is listed in Settings, but Throughline manages it; it can't be typed over or cleared by hand.
Pictures now arrive with words of their own. After an upload, a confirmed proposal, a real filespace preview, or a finished render, Throughline queues a background look. A successful look adds a short plain-English description plus useful lowercase production tags. The description appears on the file card, and a word found only in those tags finds the picture in Canvas and in the jump-to box. A clip uses its sampled frame when one exists. A failed or unreadable look leaves existing words untouched. Shows without a connected visual provider — or with the pass switched off — keep working with empty fields; there is no surprise backfill or failed upload.
A photo dropped on the filespace shows up in the library. Put a still in a character's folder — or a location's, or any other home the show already has on the drive — and it lands on that card. The name comes from the file; a tiny stand-in holds the place until the mounted agent sends its bounded preview. Search finds it by that name and by the description and tags the looking pass writes from that preview. Searching from Canvas or the jump-to box stays inside this show.
A scene, a shot, and the work under them now get folders. A scene that already holds a photo or a render lives under breakdown, named like a person would say it — scene number plus the place. A shot lives under VFX; a real shot code keeps the caps the vendor wrote. A task never gets a folder of its own: it sits under the scene or shot it belongs to, and its files drop straight in. Camera, editorial, and DIT one-offs still stay put until there's a home for them.
The newest model lane has to match the shape we asked for. A parse sent down that lane now carries the same contract every other lane does — extra fields don't sneak through. Cache and thinking-effort controls stay off until we've proven them.
Another model lane is available. HQ → Models can send a task down the new lane the same way it already sends one to the others. That lane bills a flat rate instead of by the call, so a dollar total can't catch a runaway; our own testing counts calls against a ceiling and stops there instead.
The crew can make useful folders without fighting the filing system. Familiar additions such as old, archive, wip, fittings and dated folders are kept where the crew made them. reference and notes are folders Throughline makes itself, so they stay off that list. If Throughline cannot tell what a folder means, Settings asks in plain language and leaves every file in place until an admin answers. “It shouldn’t be here” adds the files to the same reviewable filing plan as every other move — it never moves them immediately. Wardrobe can also use 01_prep/wardrobe/_inbox/: clear character matches join the next plan, while uncertain filenames come back as questions instead of being guessed or forgotten.
A file that comes back after it's been filed shows up on the next plan. If a vendor drops the same path again, or two older files now want the same name, the Filing card says so — a stay with a reason, never a silent skip. The mover also refuses a camera-roll landing that hides behind a folder symlink, and it will not treat a stray file at the destination as "already moved." Its durable move journal now also recovers the exact approved file after a restart if a destination folder was renamed during placement; the journal stays until Throughline receives the ack.
Approving a plan that was already replaced says so. If someone plans again or answers a storage question before the last plan is approved, Approve names the miss instead of quietly approving a changed plan. Two admins can never add unseen files under a plan somebody already reviewed. A file the agent reports twice under the same identity is counted as a repeat, not an error.
A costume, a prop, a location still now gets its own folder. Characters, locations, and shoot days already had homes on the filespace. Anything on the board that already holds a photo or a render gets one too, in that department's prep area — wardrobe and hair items sit beside their character folders.
A reused access-key number no longer wedges the usage count. If an old Throughline access key was deleted and the database reused its id, today's calls stopped counting. That leftover row is cleared, and the counter keeps moving.
There's a page for when a key or a machine looks compromised. Contain the credential first, then look. The runbook lives with the rest of the engineering docs (docs/SECURITY-INCIDENT-RESPONSE.md).

2026-08-31

Existing footage no longer becomes a cast booking just because the script transcribes its dialogue. A named public figure in identifiable news, broadcast or released footage stays a clearance item; a fictional video call or screen performance made for this production stays cast. If the breakdown's own note points both ways, the review tray now asks “clearance or cast?” before the role can silently reach the cast board.
The show's filespace starts organizing itself. Settings gains a Filing section on the storage card: every character, location, shoot day, and item that already holds a photo or a render gets its own folder there, created automatically the moment the person, place, or item is added. A Plan the re-file view lays out where each stray file on the drive should move and why, grouped by reason, with an honest "not sure" pile kept separate from the moves it's confident about — and nothing moves until the admin presses Approve. After that, the card reads back what's queued, what's moved, and what didn't move. Underneath, an agent now watches the filespace directly: it picks up a new file within seconds of it landing, and comes back clean after a crash mid-write instead of missing the file or counting it twice.

2026-08-30

A Day Out of Days lands on the board as its own document. Once the production office approves it, a DOOD PDF reads onto the board one cast line at a time, instead of sitting as a file nobody could see next to the schedule. See Call sheets and day-outs.
Archival footage named in a script stays a clearance item on the breakdown now, instead of showing up as a phantom cast member.

2026-08-29

Crew get their own Day Out of Days. The cast DOOD ships for every production; a crew DOOD is now a per-show switch a production can turn on. Once it's on, a crew member's own held days show on Today, and their contact card gets a booking editor for holding and releasing days. See Call sheets and day-outs.
A professionally numbered shooting script keeps its own scene numbers. When a draft carries a complete set of margin numbers — lettered inserts, OMITTED gaps and all — the breakdown now files each scene's cast and elements against the number printed in the margin instead of its position in the document. An unnumbered spec draft parses exactly as before.
The schedule board tells the truth about dates, and about who's reading it. Every shoot day now carries a real calendar date beside the label the office typed, instead of guessing one that was never stated. Booking terms and open questions on a day stay with the crew office, the budget office, and admins; a crew member reading their own line sees it without the contract attached.
The duplicates review now asks its questions in plain production language, and a straight "no" is taken as an answer.

2026-08-28

Two characters named in the same line no longer merge into one. A joint cue like "AL AND ROSE" used to get claimed by whichever character the pass committed first, silently folding the other person's lines onto their card. Both people now keep their own identity on the board.
Breakdown elements get more departments to land on. Picture Vehicles is now its own department in the registry, Specialty Units is renamed and redescribed for armory and animals, and the element pass can now name virtual-production, greens and specialty-units work that previously had nowhere to file.
Grok joins the model providers a production can connect its own key to. See Your provider keys.
When a model declines to answer a question about a script, the breakdown now tries once more with a more capable model before giving up on the run.

2026-08-27

A shooting script with no sluglines still breaks down. Numbered headings that aren't standard INT./EXT. lines — era stamps, OMITTED markers, plain caps headings — are now read using the script's own margin numbers, instead of going unrecognized by the parser.

2026-08-26

The board stops showing its own workings. Stray PENDING markers next to a role that's already booked are gone, printed page numbers no longer leak into scene text, and a production can renumber its scenes outright — a cold open becomes scene 1, with every lettered scene riding along.
A production can brief the breakdown before it runs. A methodology note and typed casting declarations — "these two Bobs are cast separately," "one performer plays all four, de-aged" — now live in Settings until they're signed off, and travel into every prompt the parse makes. The review tray shows when a brief guided a run.
A scene description now always ends on a full sentence or a clean word, never mid-clause. An exported breakdown is named for the draft it breaks down, not the day someone happened to export it.

2026-08-25

An oversized scene can be split right from the review tray. Accepting a scene-boundary card now splits it out into its own lettered scene (59 becomes 59, 59A, and so on), re-measuring page-eighths and moving cast and elements onto the right sheet.
The review tray's cards ask sharper questions. A twin card folds two names into one on "same person," or books both people on "two different people"; a performer-count card asks how many performers are involved and which scenes each covers; the duplicates panel stops offering a re-scan once a real intelligence run has already answered.
The breakdown sheet no longer prints a raw candidate id where an element name belongs. Approving paperwork while a breakdown review is still open on it now says so, instead of quietly approving something nobody decided.

2026-08-24

Uncertain cast now shows on the board, marked provisional, instead of waiting out of sight in the tray. A principal who reads as ambiguous no longer disappears until someone confirms them by hand — they're on the board with a provisional badge, and the review tray catches up around them.
The review tray now batches the easy calls: confirmations that don't need a second look sit together under "Needs your eye," and Confirm all clears the whole batch in one action with a per-row undo.

2026-08-22

The breakdown reads a script the way a line producer does. A model now judges every candidate character and element with the whole story in view, resolves who's who as casting roles, and sends anything it isn't sure of to a human review tray — rather than guessing.
The script drop zone now warns, in red, when no model is connected — parsing degrades a lot without one, and it says so before a script goes in, not after.

2026-08-21

Uploading a script no longer freezes the page. Ingest now runs as a background job with visible progress, instead of a request that could run for minutes and time out the upload.
Character names that are really the same person now fold together. An inline aside like "DALE ON THE PHONE" or an occupational title like "CO-PILOT DALE" now folds onto the character it's describing instead of minting a second card — and it refuses to guess when a fold could be wrong.
Every item in a one-liner's Cast or category cell gets its own line inside the cell now, instead of a comma list that wraps mid-word. The script page names the model that will actually read the pages, instead of a generic nudge to go connect one. The Pulse send log moved from Settings to the Jobs page, where "what did Pulse do" already lives.

2026-08-20

The breakdown leaves as paperwork. A one-line schedule as XLSX — a script-order sheet always, a shooting-order sheet whenever the show has shoot days — or the same one-liner as CSV, or the long-format element interchange as CSV or JSON. One door: GET /api/breakdown/export?fmt=xlsx|csv|elements_csv|elements_json, signed-in and clamped to what the caller may already see. Cast carries board numbers assigned by first appearance the way every AD one-liner already does, the header block names the revision the pages actually carry, and an episodic export carries its episode in the rows and in the filename so two episodes exported the same day don't collide.
Scenes carry their own length, in eighths of a page. A scene reads 1 4/8, not a rounded page count, and every draft before the shooting script locks is a dated White page instead of a color collision. A length read from anything but a PDF's own page breaks says so — it's marked estimated rather than presented as measured.
A crew member meets Sign in with Slack on their own first morning. Connecting a Slack account is something a person does for themselves, from their contact card or the offer waiting on Today — not something they have to go find. When Slack meets an account it doesn't recognize, it now points at that one-click fix before falling back to the production office. See Slack.
The assistant answers from the same graph the screen is drawing. It matches a scene number with or without leading zeros, groups a character's scenes by episode without dropping any past the cap, and names a node's other connections instead of going quiet. It no longer calls data absent when it never looked for it.
Settings says who each control affects. Everything that shapes the whole production — message routing, Pulse, reading access, department appearance, channels — sits in one Production section, closed by default. A person's own controls, including their own connected apps, stay in their own column.

2026-08-19

Slack moved to where each half belongs. Everything that is a fact about the SHOW — install, mode, health, disconnect, the simulator and the coordinator queue — is now one panel in Settings → Slack, and Slack's own consent screen returns you there, so the handshake begins and ends on one screen. Which Slack account belongs to a person is a fact about that PERSON, so it moved onto their contact card in Crew, beside their email, phone and time zone. The separate Channels screen is gone; every API route it used stays exactly where it was. See Slack.
Slack connects in one click, and leaving really revokes. The install starts in Settings, goes to Slack's own consent screen, and comes back to Settings. Disconnecting by any door — from here, from Slack's admin screen, or by uninstalling the app — drops the tokens for good. See Installing the app.
One question gets one answer, in the thread it was asked in. The events door acknowledges Slack immediately, remembers what it's already seen so a retry doesn't answer twice, and replies in the thread rather than the channel.
api.leeme.co and mcp.leeme.co are the addresses. Both were provisional in these docs until now. They're not — build against them.

2026-08-18

Throughline is its own authorization server. An MCP-aware client can now connect with nothing but a URL: protected-resource and authorization-server metadata, dynamic client registration, PKCE, and a consent screen you sign in to as yourself, so what the connection reaches is exactly what you reach. Refresh tokens rotate, a replayed one is detected and kills the grant, POST /oauth/revoke does what RFC 7009 says, and deactivating a person reaches their tokens rather than waiting for them to expire. Access keys are unchanged — this is a third way to hold a credential, not a replacement. See Discovery.
Connected apps, in Settings. Every grant standing on the show, who holds it and what it can reach, and a revoke that bites on the next call.
The audit trail answers every question it promised. A complete event taxonomy, 400 days of memory, and alerts that reach a person instead of a log file.
A failed ingest says so at the drop zone, where the file was dropped, rather than only in a banner at the top of the page.

2026-08-12

Assignments go down to the scene, and sideways as a grant. A person can be assigned particular scenes rather than a whole episode, and a may-see grant can widen what somebody reaches. A grant only ever opens a door — it can never close one the show already opened — and the same width governs script, media, and Slack channel membership alike, so those three can't drift apart.
The episode page tells the truth about who's on it, and someone granted a read in from outside leaves notes that carry their name.

2026-08-11

A day's paper opens where the day is, and leaves as one zip. Call sheets, sides and maps are readable on the day itself instead of downloading one at a time, and the whole day's packet comes down together. See Call sheets and day-outs.
Episode budgets get a door, and a series' board and spend reports backfill themselves rather than starting empty.
Department appearance collapses to one default and a list of exceptions, instead of a setting per department. See Per-department display names.
WhatsApp inbound reaches the pilot. Twilio's own request signature is the gate, so a crew member's reply arrives without a shared secret in the URL.
The docs learn to be read by machines as well as people. Every page is served as raw Markdown beside its HTML, and the site publishes llms.txt, llms-full.txt, sitemap.xml, and the API's OpenAPI document.

2026-08-10

MCP server. POST /mcp — the same REST v1 surface over Streamable HTTP JSON-RPC, authenticated on the same access keys, with zero authorization logic of its own: every tool and resource calls the identical function its matching REST endpoint calls, so a key gets the same clamped answer and the same refusal on either door. See MCP.
Context bundles, turnovers, media, and a vendor money read. /v1 grows a second shape of endpoint — one call that assembles everything about a scene or a character, a vendor's own turnovers and submissions, media read by id, and a vendor's own purchase orders — all clamped to the calling key. See One round trip, clamped to the caller.
Engagement-bound access keys. A Throughline access key can now be minted against one vendor organization's engagement instead of the whole production — a vendor's own automation, reaching exactly what a person seated on that engagement would reach. See Keys.
Slack. Install a workspace for a production, bind crew accounts, and let the show provision its own channels off the crew list. Notifications route by urgency — interrupt, ambient, digest — and an approval's whole lifecycle lands on one message, updated in place, rather than a fresh post every time. See Slack.
Forge builds behind a kernel jail for vendor and API access. Building or running generated code through a vendor's own seat or an access key now requires the deploy's kernel-level sandbox where the host provides one; the production's own crew is unaffected.
Public docs launched (pilot preview).

2026-08-09

Access key metering. Every authenticated /v1 request is now counted per key, per day — visible next to the key in Settings as a request count and a last-used timestamp. Nothing is gated or billed on it yet; access stays seat-priced. See Keys.
Two new scopes. submit:turnover and read:money join read and write — see Scopes.
Episodic productions. Seasons and episodes, per-episode budgets and schedule filtering, now reflected across the read surface: an episodic show's scenes and schedule strips carry episode_id/episode_code, and /schedule and /script-changes both take an episode_id filter.