Elimination — Temerarii Media Back Office

Three Saturdays decide who stays. This is the format for them: the running order, which built scene each segment is cut on, what is on the screen and where that data comes from, who speaks, how the verdict is delivered, and what is owed to the person who is cut the moment the stream ends.

Status. DRAFT — the run of show and the scene assignments are settled enough to rehearse. What is NOT settled is listed in `not_decided` and nothing there may be improvised on the night.

Nothing here is a second copy of anything. The running order and the dates are read from engine/lib/season.py; the scenes are checked against the built collection; the pay and dignity rules are quoted with the registry that owns each one. The one thing this page adds is the production layer — and it is the layer that decides whether the night looks like a record being read or a verdict being performed.

When

Every fourth Saturday, with the last gap running five weeks because the final round is five weeks long. The dates are derived: engine/lib/season.py computes them from the round spans in governance/pay.json, so a change to the ladder moves the calendar rather than leaving two calendars.

Night	Date	Lands	Narrows
Elimination 1	Sat 07 Nov 2026	end of week 4	3 → 2 per track	Four people leave.
Elimination 2	Sat 05 Dec 2026	end of week 8	2 → 1 per track	Four people leave.
Finale	Sat 09 Jan 2027	end of week 13	1 per track	Nobody is cut. The season is settled.

The elimination stream opens at the same clock time as a weekday episode. Participants are international by requirement, so the time is fixed once and published with the contract rather than moved per round.

Before the night

The part of the format that decides whether Saturday is fair happens on Thursday. A record that is still moving, or that a participant first sees on a live scoreboard, cannot be defended against — and a rule applied without prior written disclosure cannot be applied at all.

When	What happens
Thursday, 48 hours before air	The record is FROZEN. Gate runs, registry entries, published work and producer-panel entries after the freeze count toward the NEXT round, not this one. A record that is still moving while people prepare a defence is a record nobody can prepare against. It also stops the last-48-hours scramble that rewards theatre over the thirteen weeks of work the format is supposed to measure.
Thursday, immediately after the freeze	Every participant receives their own frozen record in writing — every rule that fired, the emitter that produced it, and the running total. Nothing is shown on air on Saturday that was not in that document. This is the operational form of `no one caught off guard`. A participant who first sees a deduction on a live scoreboard has been ambushed, and a rule not disclosed before it is applied cannot be applied at all.
Thursday to Friday noon	A participant may dispute any line of their record in writing to the Executive Manager. A dispute is resolved before air or the line is REMOVED from that round's record. Machine-read rules can still be wrong — a gate can fail for a reason that is not the participant's. Resolving it on air turns a data question into a performance, and the person with the microphone wins those rather than the person who is right.
Friday	Release clearance is re-verified for everyone appearing, including anyone appearing only to be eliminated. governance/roles.json makes it a condition of every seat: nobody is on camera before counsel-reviewed releases are signed. Clearance does not carry forward between rounds, so it is checked again for this broadcast.

The run of show

Seven segments. The titles, durations and what each segment is for are read from engine/lib/season.py, which owns them. Everything else — which built scene is on air, what is on the screen, where that data comes from and who speaks — is read from governance/elimination.json and joined on the segment name. 7 of 7 segments carry a production entry.

Every scene named below is a real scene. They are checked against governance/scene-manifest.json, the snapshot of the collection obs-setup actually built. A name that is not in the manifest renders struck through — a format that invents a scene is a format the producer cannot cut, and it fails on the one night it cannot be fixed. Hover a scene to see its overlays and camera count.

Cold open

5 min · cold-open

The round in one cut. What shipped, what broke, what it cost.

Cold OpenPackage

first pill = the scene it lives on · the rest are its cutaways

On screen	Which is	Read from
The round in one cut — what shipped, what broke, what it cost	Clips assembled by the clip engine from the round's weekday episodes	`governance/pipelines.json -> clip-engine`

Who speaks

Nobody — the cold open carries no live microphone

Producer. Cold Open carries no camera and no HUD in the manifest. It is a rolled package, so the cut into The Record is the first live moment of the night and the first thing the room hears.

The record

25 min · the-record

Per track: registry entries executed, gates passed, work published. Read from the system, on screen. Not opinion — output.

The RecordGate VerdictStandings

first pill = the scene it lives on · the rest are its cutaways

On screen	Which is	Read from
Per track: registry entries executed, gates passed, work published	The unified signal matrix — the one durable record of a certified week	`engine/gates/verify_unified_matrix.py`
Each rule that fired, and the emitter that proves it	The scoring registry, split on `source` — machine-read rules name their emitter, human-judged rules are labelled as judged	`governance/scoring.json`
Gate PASS / FAIL, named by week and field	The full gate suite, one verdict	`engine/goal.py`
The running standings per track	The scoreboard state pushed to the broadcast over POST /state	`governance/pipelines.json -> live-broadcast`

Who speaks

Executive Chairman reads the header
The system reads the rest — the graphics are not narrated line by line

Producer. The Record and Standings both carry the Scoreboard overlay; Gate Verdict carries only the Verdict overlay and no camera, so it is the clean full-frame cutaway when a gate result needs to be the whole screen. NOTHING here is typed into the graphics by hand — the overlays read the same state the gates wrote.

Defence

40 min · defence

Each participant gets the floor. Ten minutes: what you built, what you would do again, what you got wrong.

Hot SeatStandings

first pill = the scene it lives on · the rest are its cutaways

On screen	Which is	Read from
The participant, full frame, with their name and track	The roster	`governance/roles.json`
Their own line of the record, held on the lower third while they speak	The same frozen record served on Thursday	`governance/scoring.json`

Who speaks

Each participant in turn, uninterrupted

Producer. Hot Seat is one camera with a Lower Third and no second face. That is deliberate: a reaction shot during a defence turns a statement into a reaction, and the format is not built to harvest that.

Counterpart testimony

20 min · counterpart-testimony

Your counterpart speaks. They are not asked to be kind; they are asked to be accurate.

Two UpActive Speaker

first pill = the scene it lives on · the rest are its cutaways

On screen	Which is	Read from
The participant and their counterpart, equal size	The roster and the crossover pairing from the week spine	`engine/lib/season.py -> SPINE_A / SPINE_B, block CROSS`

Who speaks

The counterpart
The participant may respond once, briefly

Producer. Equal framing is the whole point. A testimony shot with one face large and one small has already told the audience who to believe. Active Speaker is available if a cross-talk needs following, but Two Up is the default and the cut back to it should be quick.

Deliberation

25 min · deliberation

Executive Managers and the Chairman, on air, against the published criteria. Disagreement is broadcast, not edited out.

QuadStandingsThe Record

first pill = the scene it lives on · the rest are its cutaways

On screen	Which is	Read from
The Executive Managers and the Chairman, deliberating on air	Live cameras	`obs-setup`
The published criteria being argued against, held on screen	The scoring registry and the pipeline registry's `assessed_by`	`governance/scoring.json · governance/pipelines.json`

Who speaks

Executive Managers
Executive Chairman

Producer. Disagreement is broadcast, not edited out. The one thing the producer must protect is that the criteria stay on screen while they are argued — a deliberation with no visible standard is the format failing in public, because it looks exactly like the opinion-driven judgement the whole system exists to replace.

The cut

15 min · the-cut

Named directly, told why, thanked. They speak last if they want to.

EliminationHot Seat

first pill = the scene it lives on · the rest are its cutaways

On screen	Which is	Read from
The Chairman and the participant, two cameras, with the scoreboard held	Live cameras and the frozen record	`governance/scene-manifest.json -> Elimination`

Who speaks

Executive Chairman delivers the verdict
The eliminated participant speaks LAST, if they want to

Hard rules for this segment. These are not style preferences — each one is a way the format could become something it is not supposed to be.

No music sting under the name. A sting turns a person losing a contract into a game-show beat.
No countdown, no fake pause, no ad break between the reason and the name. The reason comes first, then the name, immediately.
No reaction-shot hunt. If the eliminated participant does not want the camera, the producer cuts to Hot Seat on the Chairman.
They speak last, or not at all. Nobody speaks over them, and nobody asks them how they feel.

Producer. Elimination is the only two-camera scene in this run of show that also carries the Scoreboard. The verdict and the reason it is defensible are in frame together, which is the entire argument of the format compressed into one shot.

Next round

10 min · next-round

New rates take effect Monday. New scope. What changes.

RosterWeek CardCredits

first pill = the scene it lives on · the rest are its cutaways

On screen	Which is	Read from
Who remains, per track	The roster, minus the cut	`governance/roles.json`
The new weekly rate taking effect Monday, and the seats remaining	The round structure — seats narrow as the season advances	`governance/pay.json`
The next round's theme and its week span	The season structure — rounds, week spans and seats	`engine/lib/season.py -> ROUNDS`

Who speaks

Executive Chairman
Executive Managers on the new scope

Producer. The rate change is stated ON AIR, in the same segment as the cut, because governance/policy.json requires the ladder and its step weeks to be disclosed before signing. Saying it out loud where everyone hears it is the cheapest possible proof that it was not a surprise.

How the verdict is delivered

The Executive Chairman, in every round, without exception. governance/roles.json puts final sign-off on every elimination with the Chairman and explicitly does NOT give the Executive Manager the power to end a contract. A verdict delivered by anyone else would be the org chart contradicting itself on the one night twelve people are watching it.

In this order, every time

The reason, stated against the published criteria and the frozen record.
The name.
What they are owed, said out loud: the days worked, the prorated amount, and that week one was paid in full regardless.
Thanks, by name, for the work that is on the record.
The floor, if they want it. They speak last and are not interrupted.

The reason comes before the name. Reversed, the segment is a reveal and the criteria become an explanation offered to someone already cut. Same information, entirely different programme.

Never

Naming someone before giving the reason.
Citing a rule that was not on the frozen record served on Thursday.
Citing a rule whose point value is still a PLACEHOLDER in governance/scoring.json — the number was never shown, so it cannot be applied.
Announcing a consequence beyond elimination. Termination for cause is a separate process with its own trigger list and is not a live-television moment.

What happens to the person who is cut

This is the half of an elimination format that is usually not written down, which is precisely why it goes badly. Everything here is owed, not offered — each line names the registry that already commits to it.

When	What happens
Within one business day	The Upwork milestone is released prorated to the final day worked, with any agreed adjustments applied and the weekly cap enforced. Week one is paid in full regardless of when the elimination happened. governance/policy.json -> elimination_and_proration · governance/pay.json -> week_one_guarantee
Immediately after the broadcast ends, not during it	Operator access is revoked. Push access is revoked. Nothing is revoked while the person is still on camera. Revoking access mid-broadcast is a thing the audience can see happening, and it is a punishment the format never agreed to impose. It also risks pulling a live source out from under the producer during the show.
Immediately	They keep every DEMO clip and every showcase piece they produced, permanently. The public portfolio is part of the consideration for taking the seat, and elimination does not withdraw it. engine/lib/season.py -> SPINE_B, block DEMO
Off air, within seven days	A recorded-if-they-consent conversation with the Executive Manager: what the record actually said, what would have changed it, and what they should do next. Not broadcast unless they ask for it to be. The seat is a paid training position. A training programme that eliminates people and then says nothing has taught the last thing it teaches badly.
Permanent	Their record is not deleted from the office. It is marked as ending at that round, with the reason. A record that disappears when someone leaves is not a record. It also protects the eliminated participant: what was actually said about them stays checkable instead of becoming whatever the clips implied.

The constraints the format may not contradict

Each of these is already committed to in a registry this page does not own. They are quoted with the file that owns each one rather than paraphrased, so a format that drifts from the contract is visible here rather than discovered by someone reading their own payslip.

The rule, and who owns it	What it forces the format to do
Week one is paid in full regardless of elimination or adjustment. `governance/pay.json -> week_one_guarantee`	The first elimination lands at the end of week 4, so this is already satisfied by the calendar — but at-will elimination means it can land in week 2, and the format must not contain any beat that implies otherwise.
Pay is prorated to the final day worked, with any agreed adjustments applied. `governance/policy.json -> elimination_and_proration.proration`	It is said ON AIR in the verdict, in the order above. A payment that is only correct in a spreadsheet nobody sees is indistinguishable, to the audience and to the person, from one that is wrong.
Elimination may occur AT ANY TIME, not only at round boundaries. `governance/policy.json -> elimination_and_proration.at_will`	An elimination outside a Saturday is NOT staged. It is communicated privately, paid the same way, and reported on the next scheduled broadcast. Manufacturing a special episode for it would create exactly the ambush this framework forbids.
A rule that was not disclosed in writing before the contract was signed cannot be applied to that contract. `governance/policy.json -> _NO_ONE_CAUGHT_OFF_GUARD.THE_ENFORCING_RULE`	This is why the record freezes on Thursday and is served in writing before air. It is also why a PLACEHOLDER point value cannot be cited in a verdict: nobody was shown the number, so nobody agreed to it.
Nobody appears on camera before counsel-reviewed release forms are signed, and clearance does not carry forward between rounds. `governance/roles.json -> _shared_requirements.release`	Re-verified on the Friday. It applies to a participant appearing only to be eliminated exactly as it applies to everyone else.
Total adjustments in any one week may not exceed the published fraction of that week's rate. `governance/pay.json -> weekly_deduction_cap`	Several single rules exceed a whole week's cap on their own. The scoreboard may show the rule's full value; the PAYMENT is bounded by the cap, and the verdict states the paid figure, not the scoreboard figure.

A rule that was not disclosed in writing before the contract was signed CANNOT be applied to that contract. This is deliberately self-enforcing — it makes disclosure the company's problem rather than the freelancer's, which is the only version that actually holds. That is why the record freezes and is served in writing two days before air — disclosure is the company's problem, not the participant's.

Every scene this format needs

13scenes named, of the 40 in the built collection — 13 verified

Scene	Cameras	Overlays	HUD
Cold Open	0	none	no
Package	0	Status, Ticker	no
The Record	1	Scoreboard, Status, Ticker	yes
Gate Verdict	0	Verdict	no
Standings	0	Scoreboard, Status, Ticker	no
Hot Seat	1	Status, Ticker, Lower Third	yes
Two Up	2	Status, Ticker, Lower Third	yes
Active Speaker	1	Roster, Status, Ticker, Lower Third	yes
Quad	4	Status, Ticker	yes
Elimination	2	Scoreboard, Status, Ticker	yes
Roster	0	Roster, Status, Ticker	no
Week Card	0	Week	no
Credits	0	Card	no

Every scene this format calls for is already built. Nothing here needs a new scene collection — the format was written against what the switcher actually has, not the other way round.

Not decided

4 questions the format does not answer yet. They are listed because the alternative is improvising one of them live, and every one of them lands in the first round if it is not settled first.

Cannot be improvisedThe Executive Manager track is itself being eliminated. Which Executive Manager deliberates on the Executive Manager track?
A candidate deliberating on the seat they are competing for is the same conflict that keeps the stop switch with the Chairman. Unresolved, it lands live in round one.

OpenWho testifies about whom, and is the pairing published in advance?
The week spine names CROSS as the crossover block but does not fix pairs. An unpublished pairing revealed on air is a surprise, which the disclosure rule forbids.

OpenThe Final ends week 13 and eliminates nobody. Does it use this run of show at all?
Six of the seven segments still apply; `the-cut` does not. Running the format unchanged would build to a cut that never comes.

OpenDoes chat affect the outcome in any way?
governance/scoring.json already carries chat_vibe_check as a HUMAN-judged rule and records why chat sentiment cannot be machine-read. Whether that rule can move an elimination — as opposed to a week's points — has not been decided, and it is the difference between a record and a vote.

Where this page comes from

Section	Source
Segment titles, durations, what each is for	`engine/lib/season.py` → `ELIMINATION_RUN_OF_SHOW`
The three dates and what each round narrows to	`engine/lib/season.py` → `saturdays()`, itself derived from `governance/pay.json`
Scenes, on-screen content, who speaks, the aftermath	`governance/elimination.json`
Whether a named scene is actually built	`governance/scene-manifest.json`
Pay, proration, the week-one guarantee	`governance/pay.json` · `governance/policy.json`
Who may deliver a verdict, and release clearance	`governance/roles.json`
What each rule scores, and which are machine-read	`governance/scoring.json`, rendered per seat on the role pages

The show · roles and what each seat is measured on · operations policy · broadcast architecture