Everything it does, and what each part replaces

One command reports what a workbook is. How many rows and columns. Which row is really the header, as opposed to the three merged title rows above it. And for each column, how full it is, how many different values it holds, and what it is therefore doing.

A date. A flag. A status. Prose. Or a column that has been dead for two seasons.

It never reports a single cell

Not an episode number. Not a name. Not a theme. Not a status word.

The whole design is that you can learn what a grid does from counts and shapes alone. A column with six different values across a hundred and twenty rows is a status field whether or not anybody can read what those six values say.

Why this exists

The grid is the thing being replaced, and replacing it means first understanding what it does for each of the people living in it.

That conversation used to start with sending us the file. Now it starts with output that is safe to paste into a message, from a machine you control.

Headers are something you turn on

Column headings and tab names are where a grid keeps the names of the people who own each column, so they are off by default.

Turn them on and the report carries a warning at the top, where it will be read, rather than in a footnote underneath something you have already copied.

What it will not pretend to know

A spreadsheet does not record who edited which cell.

It carries one last-modified-by for the whole file. Per-cell history lives in whatever is hosting it, not in the document. The report says exactly that, rather than offering one name as though it were the answer.

The numbers across the top are about you, and they only appear when something needs doing.

An editor sees what is on their desk. A supervisor sees holds with their reasons, work nobody is on, shows close to air with picture still unlocked, and note passes that have stopped moving.

When nothing qualifies, the whole strip collapses to one green line: “Nothing needs you.”

The test a number has to pass

Three things, and a number failing any of them is furniture:

1. It has to change. A count that is the same every morning is wallpaper.
2. Somebody has to be able to act on it.
3. It has to be that person’s problem. A hold an editor cannot lift does not belong on an editor’s screen.

And it has to be right

The first version of this counted every episode with nobody on it and reported thirty. because delivered shows legitimately have nobody on them.

A metric that is wrong thirty times is noise, and noise is what got the previous strip deleted in the first place. It now counts work that exists and belongs to nobody.

That correction is the whole argument for this feature in one story: the reason the old counts went away was not that counting is bad, it is that a number nobody can act on trains people to stop looking at the top of the screen.

Thirty seconds

Sign in as a supervisor with a clean board: one green line. Put an episode on hold and the strip grows a card naming who holds it and why. Sign in as an editor and the hold disappears, because they cannot act on it.

Ask for episode 1069 and you get back every copy on the drives the facility has told Baton about. Masters, proxies, graphics, project files. Sizes and dates, newest first.

You can narrow it to one kind of file. You can get it as plain text you are safe to paste into a message.

The part that matters is what it refuses to do

A request says what to look for. It can never say where.

The drives come from a setting on the facility’s own machine, read fresh every time. There is no option that takes a path, and anything that looks like one is turned down. Ask for a folder and it tells you it wants a name, not a place.

That sounds small. It is the whole reason this is safe to hand to an assistant that is in a hurry, or to a tool. A search that can be pointed anywhere is a search that will eventually be pointed somewhere it should not go.

“I could not look” never comes back as “it is not there”

This is the rule the rest of the product is built on, applied to searching.

If a drive was not mounted, it says that drive was not searched. If a folder could not be read, it counts it and tells you. The command itself reports a failure when any part of the search was incomplete, so a script cannot read a partial answer as a complete one.

When the file genuinely is not on any drive it could reach, it says that, in those words.

A missing file and an unreadable drive look identical in a list of results. Only one of them means the file is gone.

It cannot take the facility down

There are limits on how many results come back, how deep it goes, and how long it runs. A request cannot raise them.

When a limit is reached it says so, because “here are 50 of them” and “here are 50” are different answers. A search of one or two letters is turned down, since against a shared drive that is not a search, it is a walk through everything.

It sees exactly what you see

The search runs as whoever ran it and follows the same permissions the drives already have. It does not follow shortcuts out of the folders it was given.

Nothing here gets its own key to anything.

Drag the board’s date and the whole season plays out.

An empty board before the show has opened on anything. The first Planned cards appearing. Taping. The middle stages filling up. And Delivered growing to hold every episode that has aired: 55 of them at the date the demo opens on, and 120 of 120 by the end of the season.

Delivered is a place, not a window

An aired episode is delivered, and it stays delivered.

It used to drop off the board ten days later. So the work of a four-month season looked like it all happened in the last week, and the one column that proves the whole thing works was permanently the thinnest object on screen.

It starts where your show starts

The left edge of the scrubber comes from your own schedule rather than a fixed number of days.

A weekly and a daily show open at different distances from air, so a single constant was wrong for both. On one show it opened seven episodes into taping, with the season already under way before anybody touched anything.

Evidence picks a column, not a board

The agent that reads your storage reports one timestamp for the whole file and none per fact. So “the masters are there” was true at every date you could scrub to, and episodes the agent knew about sat alone in Taped nine months before air.

Footage for an episode the show is not tracking yet now raises a flag naming that episode, instead of quietly putting a card on the board.

Why this matters more than it sounds

The board is the same object at every date, and it will show you the season from before it existed. That is the difference between a tool you demonstrate and a tool you can audit.

The editor presses one button and the notes appear as markers on their own sequence, each sitting on its own frame, coloured by what it is.

No file to download. No import step. No “did you get my email.”

Whose button it is

The editor’s, and not the executive producer’s.

Production management stamps every marker with whoever is signed in. So pushing on somebody’s behalf does not produce their markers; it produces yours, in their bin, with nothing recording who actually meant them.

Anyone else sees the button disabled, with the reason written on it and a pointer to the locator file, which delivers exactly the same notes under the right name. The work still gets done; it just gets done under the name of the person who meant it.

What it will not do

Overwrite the timeline.

Writing markers replaces the sequence’s whole marker representation. So this reads first, keeps every marker it did not write (including ones on other layers), and refuses outright if it could not read.

That refusal is the important part. Writing without reading first deletes somebody’s work and reports success while doing it, which is the worst possible combination: damage, plus a green tick saying everything went fine.

Dry run

Plans the write and shows exactly what would be sent, including how many existing markers it would keep.

Press Push to Avid and the markers go onto the sequence from your own browser, using the production management session you are already signed into.

Baton works out what to write. Your browser does the reading and the writing.

What this fixes

The push used to run on a server, which meant the server held a copy of somebody’s login.

A live run showed what that costs. The copy expired part way through an afternoon, every push after it failed, and nothing in the product could fix it. The repair was a person fetching a new cookie by hand.

That is not a risk to manage. It is a certainty on a schedule. There is now no copy to expire.

You see the numbers before anything moves

“2 markers onto V1, keeping 1 already there.”

It sits next to the notes it is about, with Send them and Cancel.

This write replaces the sequence’s whole set of markers. So a plan that keeps none when the sequence had some is the shape of deleting somebody’s work, and you can see that before you press anything.

A failed read never turns into an empty sequence

If the sequence cannot be read, the push stops and says so.

Carrying on would put only these notes onto the timeline and wipe every other marker on it. And it would report success.

It says what went wrong in words you can act on

Signed out is named as signed out.

A browser blocking the call is three different problems wearing one face, so it names all of them rather than picking one and hoping.

It is still only the editor’s button

Production management stamps every marker with whoever is signed in. Pushing for somebody else does not make their markers, it makes yours, in their bin.

The refusal says whose bin it is.

Every package carries a checklist of what has to be in it. The voiceover cut in. The prize copy checked against what was actually won. The legal disclaimer in the exact form legal approved.

The drawer tells you which line is still outstanding and who it is waiting on. The package cannot be approved while any of them is.

Why this one hurts

This was the part of a producer’s job with nowhere to live, which is exactly why it is the part they are most afraid of.

A bumper whose picture is finished looks finished on every screen. The person who finds the missing disclaimer is a compliance reviewer, after air.

The list is your show’s, not ours

Each production says what a package of each type has to carry.

A bumper needs a voiceover and a disclaimer. A highlights package needs neither. Another show needs three things nobody here has heard of. That is a setting, not a code change, so adopting it does not wait on us.

Finished is not the same as signed off

An item can require somebody outside the edit to approve it.

An editor can confirm a voiceover is cut in. Only compliance can confirm that a disclaimer says what the lawyers require. If the wrong person tries, the refusal names the role that can, because telling somebody “no” without telling them who to ask is a dead end.

It records which version was approved

“The disclaimer is in” is worth nothing if nobody can say which disclaimer.

The version goes into the row and into the log. If the copy is later replaced, the sign-off on the old one is cleared rather than carried across. A signature is on a specific thing.

It says what is missing, not how many

“2 of 4” sends somebody to open the package and work out which two.

“Legal disclaimer. Post Producer has not signed it off.”

That is the sentence they were going to have to build anyway. Baton builds it.

Every handoff already names a version. This add-on checks that version against the cuts that really exist, and marks it:

• Green tick. That cut exists, and it is the newest one.
• Warning. A newer cut turned up, or the one named is not there. The message names the newer cut right in the line, so nobody has to go looking.
• Grey question mark. Nothing was checked.

It is not an Avid feature

This is the part most people assume wrong, so it is worth being blunt about.

Baton does not go and ask your production management anything. The information is pushed in, and it takes the same shape whoever sends it:

• Avid MediaCentral or Production Management
• A PilotWare export
• A nightly export from whatever you run
• A person pasting in the list of cuts

All four produce verified handoffs. Whatever holds the answer at your facility sends it, and the board compares it against what the handoff claimed.

That matters because Production Management is an enterprise system most shows do not have. It was never the requirement. Knowing which cuts exist is.

The fourth state is the honest one

A verdict is worked out fresh each time it is asked, never frozen. Nothing stores “1.7 is verified”, because a later report or a corrected handoff both change the answer, and a stored verdict would go quietly stale while still looking authoritative.

And when a source cannot be reached, that reads as unknown. Never as “this episode has no versions”. A drive that is down and a cut that does not exist are different facts, and the difference is enforced in the code rather than left to whoever writes the screen.

What it looks like with nothing connected

Handoffs still name the version, and it still shows on the card. It is just marked not checked, deliberately grey rather than red.

Nobody having looked is not a problem with the episode. A board that flashes red about its own blind spots is a board people stop reading.

Thirty seconds

Four episodes, four editors, all handing off “1.7”. One checks out. One was replaced by 1.8 forty minutes ago. One does not exist. The fourth could not be checked.

A daily show and a weekly hour do not run on the same clock, and a board that pretends they do has columns that can never hold anything.

A daily show turns around in weeks, not months

One of the demo shows carried a planned window of 130 days across a season that runs eight weeks. The pipeline was longer than the show it described, so three of the seven columns were empty at every date you could scrub to.

It tapes five weeks ahead now and finishes in days, which is what a daily show actually does.

A weekly needs bands at least a week wide

The other show had a three day lock window on a show that airs every seven days. Episodes stepped straight over it, and three columns stayed permanently empty for the opposite reason.

Spaced a week apart, each stage holds one episode and the season moves along together, which is what a weekly looks like on a wall.

Tape days belong to the block, not the episode

A daily show tapes a week of episodes in one day. One audience, one set-up, five episodes.

So five episodes share a taping date and the calendar shows five lines on that Tuesday, the way the wall schedule in the office does.

Episode keys as a production writes them

SL201 to SL240. Show code, season digit, episode number.

Three digits rather than four, because SL2001 reads as season 20 and almost no show gets there.

Why this is on a marketing page at all

Because an empty column reads as a broken product, and three of them read as a demo nobody rehearsed. Every column on both shows holds something at the date each one opens on, and that is not cosmetic: it is the difference between a board that describes your show and one that describes a shape somebody assumed.

Every post schedule starts life as offsets somebody typed once. Rough cut nine days before air, fine cut four. Nobody ever checks.

Baton has been recording when each step was picked up and when it was finished all along, as a side effect of people doing the work rather than a status anybody maintains. It now reads those records back:

You plan two days for a fine cut. The last eleven took three and a half.

What makes it worth believing is what it refuses to say

A forecast is only useful if you still believe it at 2am in week nine, so it is built around the specific ways a plausible number would be a lie.

A weekend is not work. Handed over at 4pm on Friday and finished 10am Monday is three hours of attention, not sixty-six. Count it wrong and the tool learns that your cuts take three days.

Parked time is not work. Nine days stuck waiting on a legal clearance did not cost nine days of editing. Bill it to the edit and the model decides your show is slow.

A step nobody timed is unknown, not instant. That is the most common record in any real set of data, and treating it as zero drags every estimate quietly toward nothing.

One example is not a pattern. Below a handful of finished steps it says it cannot tell you yet, instead of quoting a confident figure with a footnote nobody reads.

An empty history is never good news. A season with no measurements behind it says two episodes cannot be forecast yet. It never says everything fits.

One editor is one queue

The calendar might show forty hours free this week. If all six remaining cuts belong to the same person, those hours get spent once, not six times.

When something misses air it names the first episode in that queue as the blocker, rather than reporting the five behind it as five separate problems.

Your hours, not generic ones

A show declares its own working week. Four ten-hour days, nine to six, a three-week hiatus. Every answer states which calendar it assumed.

A producer who disagrees with those hours needs to see them written down, not discover the problem by being wrong. A calendar nobody could actually work falls back to the default rather than poisoning every projection underneath it.

The defect this shook out

Releasing a hold used to erase it. A step row recorded what was parked now, so clearing it on release destroyed the only evidence of how long anything had ever been parked.

Which means every released hold was being billed to the edit. A step stuck nine days on a clearance taught the model that this show’s cuts take a fortnight. Holds are kept as windows now, a step can be held more than once, and the proof is a test that reports nine hours for an edit that took four the moment the fix is taken out.

What exists today, precisely

Two endpoints. What each phase has cost against what the plan allows, and whether the rest lands before air. 81 tests behind them.

There is no screen for it yet. The numbers are live and answerable now, and that is a different sentence from “this is on a dashboard”, so it is worth saying which one is true.

A month of a daily show is several hundred dated things, and a wall of them answers nothing.

So you pick the question instead, in the words a supervisor already keeps a schedule in. Eight tracks, not twenty step names:

Taping · Prepping · Editing · Reviews · Online · Delivery · QC · AIR

Just the reviews this week. Just episode 223. What is due to the caption vendor. The mix playbacks. The individual steps and vendors are still underneath, for somebody who already knows which one they mean.

Colour can follow the track, the stage, the kind of work or the person. That last one turns the month into a load chart, because every line already carries whose it is.

Vendor dates are on the calendar at all

They never were. The deliverables grid says what is outstanding per episode, and this is that grid turned on its side, which is how a supervisor plans a week.

The tracks come from your show, not from a settings page

Nothing in the product names a step, a stage or a vendor. The tracks are worked out from the kinds of work your show already declares.

A show that does not mix gets no mix track. A show whose deliveries need no quality check never sees that one. Of the two shows in the demo, one offers seven tracks and the other offers all eight, and nobody configured either.

Standards and Practices is a review, not a QC

This distinction is worth keeping because folding them together loses a week.

Broadcast standards reviews the cut, for content, at fine cut and again at lock. It is a notes pass given from outside the building. Network quality control is a technical pass over the finished master, weeks later, hunting glitches in picture and sound.

QC earns its own track because it sits in somebody else’s queue, and it is the usual reason an air date gets tight. Buried in a row of vendor names it disappears.

A day says what it is holding back

Six lines, then “11 more”, which opens.

A day that quietly clips the rest is a day somebody plans around without knowing what is under it. Finished work is hidden by default and one click brings it back.

Every filter says what it is hiding

A filtered calendar that does not say it is filtered is how somebody plans the wrong week.

Narrow it and the count line tells you what it left out and what did the leaving. A filter that matches nothing says the month is not empty, the filter is, rather than showing a blank grid that reads as a page that failed to load.

Before anything is sent, it gets sorted into one of three answers. Send now. Put it in the morning digest. Do not send it at all.

Your turn on a show close to air goes straight out, at any hour. Blocked work always goes, because somebody has downed tools and you are the way out. A delivery going out, or a show running late, waits for the morning and arrives as one line.

What this is really preventing

Not a message that fails to send.

Forty that send. A channel muted in week two. And then the most useful thing in the product is silently off for that person, and for everybody they tell.

Three things are never sent at all

You already opened it, so you know.

The step is parked, which means it is not your turn. It is the holder everybody is waiting on.

And anything that was overtaken before it went out.

A message you can safely ignore teaches you that these can be ignored. That is the expensive lesson, and it only has to be learned once.

Quiet hours belong to the show, not to the clock

A daily show that locks nine days out and a special that locks at six genuinely disagree about what “soon” means.

One number would wake one show’s people far too often and the other show’s people far too late. It comes from each show’s own cadence.

Every decision is written down, including the silences

The question people actually ask is “why was I not told”.

That has an honest answer only if the reason was recorded at the time. So it is, including for the messages that were deliberately not sent.

Everything that was not worth interrupting you for arrives as one message at the start of your next working day.

“3 things waiting. The soonest airs in 2 days.”

Then a handful of lines, and a way back to the board.

What it leaves out is the point

Overnight, a good share of what was waiting stops waiting. Somebody picked it up. Somebody finished it. A revision moved it to another person.

Every item is checked against the board again at the moment of sending, and anything that is no longer yours is dropped. It is also marked as settled, so it does not come back tomorrow morning and every morning after that.

A digest that opens by telling you to do something you did last night is not one small mistake. It is the reason the next forty go unread.

It cannot say the same thing twice

Run it again ten minutes later, because a timer fired twice or somebody restarted something, and nobody is told anything a second time.

That is what makes it safe to leave on a schedule.

It will not arrive at three in the morning

A summary that lands before the working day is exactly the interruption it was batched to avoid. One that lands at four in the afternoon is just a late message.

It holds until the morning, and it says that it is holding.

What it replaces

The digest used to be decided and never sent.

There were three answers. Send now, hold for the morning, do not send. The middle one wrote a row that nothing ever read again. In practice the system had two answers and a slow way of saying nothing.

Issuing a delivery revision opens a box that asks two things, and will not go ahead without either. What was wrong with the master, and the new delivery date.

It also tells you the cost before you commit:

“3 endpoints already took delivery and will have to take it again. Re-online, re-caption, re-deliver.”

Why the reason is required

“REV 2” on a card told the next person nothing. The only person who knew what happened was whoever pressed the button, and a week later finding out meant asking around until somebody remembered.

Why the date is required as well

The original delivery date has passed, so every date worked out from air is now wrong for this episode.

A calendar still showing the overtaken deadline is worse than one showing nothing, because it looks like something somebody agreed to. That is the same rule as everywhere else here. A stale fact presented confidently does more damage than a gap.

Turning over a new cut asks for neither

No box at all. Nothing has left the building, and it costs another pass rather than a re-delivery. Making a routine offline turnover feel like an incident would teach people to avoid the thing they should be doing.

A second version of an episode (an ALT, an INTL, a network-specific cut) starts as a copy of the first one.

It brings over the parent’s packages except the ones not cleared for that version, and it tells you what it left out instead of quietly dropping it.

A promo that must not go to another network cannot get there by accident.

The right-click menu shows what you can do. It also shows what you cannot, and why, instead of hiding it.

The message says “this is Harper’s to finish”, not “permission denied”. Hiding the option teaches somebody the software is broken. Naming the person tells them who to go and ask.

The machine at the facility runs one script, and the output answers everything anybody was going to ask.

What this machine has on it. What shipped. Whether the agent passes its own self test. What shape the data is in. Who ends up with a full board. And whether the drop frame arithmetic still lands on its known boundaries.

What it replaces

A list of questions, answered one at a time, typed up, and then matched back against the questions.

Most of the cost was never the checking.

It will not print a single name

The output goes into a conversation, and pasting something into a conversation is publishing it.

So it carries counts, shapes and yes-or-no answers. Never a show, a person, an episode number, a path or a theme. A show is “show 1”. A season’s range is a span rather than its numbers. A gap is a reason rather than the people in it.

The server enforces that instead of trusting the script to remember, and a test fails if an identifier ever gets through.

It is not a way to run commands remotely

That is deliberate.

It takes nothing that reaches a filesystem or a shell, and it runs only what is written into it. That is the same posture the on-site agent takes, for the same reason.

It found two faults in its own bundle on the first run

A line-ending check whose pattern had collapsed to nothing, so it matched every file and reported a problem that did not exist.

And a self test wired to one of the three shipped binaries, so it silently never ran on the other two.

Nothing is ever really deleted. Everything keeps its history, along with a note saying why it was put away and who did it.

“What was in that show?” stays answerable, including after the people who knew have moved on to another production.

Initials sit right on the episode card.

A board up on a wall answers the most-asked question in the building without anybody opening anything, clicking into anything, or getting up.