Chapter 65. Production Budget and Schedule#

65.1 Production units, cost structure, and reserves#

Estimate by production unit.

Multiplying episode count by a per-episode price hides the differences between shots. Split the budget across story, assets, still shots, motion, sound, compositing, editing, QC, distribution and management — then estimate attempts and labor per shot by risk tier.

Fixed investment and marginal cost.

Golden character images, locations, UI, music themes and automation are fixed investments amortized across episodes. Generation, repair and review per shot are marginal. A pilot's fixed investment looks heavy, and the unit cost falls once production scales.

The shot cost model.

expected shot cost = still attempts x unit price
                   + motion attempts x unit price
                   + repair hours
                   + allocated compositing / sound / review
                   + risk reserve

Use historical pass rates per shot class rather than optimistically assuming first-time success.

Labor cannot be counted as zero.

Writing, art direction, prompt compilation, selection, retouching, editing, mixing, QC, producing and legal are all recorded as hours. AI lowers the cost of execution; it does not remove the cost of judgment.

Reserves.

Keep separate reserves for creative exploration, generation failure, client change, and platform or vendor risk. Client change must not absorb the internal failure reserve indefinitely.

65.2 Dependency scheduling, throughput, daily reports, and backups#

The schedule is driven by dependencies.

Once character assets are approved, several episodes' shot design can run in parallel. Once keyframes are approved, motion can parallelize. Before picture lock, subtitles and mixing can only be temporary. A schedule lists dependencies and the critical path, not only dates.

Throughput and work in progress.

Limit the number of shots in flight. Too much work in progress backs up review and lets versions expire. Use a board showing blocked, ready, generating, review, repair and approved.

The daily production report.

Record approved seconds, attempts, cost, blockers, asset changes, the critical path for the next two days, and risks. "We generated a hundred clips today" is not progress; final usable seconds are the output.

Burn-down and forecasting.

Forecast the completion date from remaining approved shots or seconds and the current real pass rate. Update after a model or asset change; do not keep a promise that has already become false.

Backups.

Back up project files, masters, schemas, assets, licences and logs by version, in at least one independent location. Verify restoration rather than confirming the upload succeeded.

65.3 Bottom-up budgets, bottlenecks, and burn-down calibration#

Build a bottom-up budget from the shot list.

Tag each shot with its class, risk tier, target adopted seconds, expected still attempts, expected motion attempts, human selection, local repair, compositing, sound and QC. Aggregate by scene and add shared assets and management cost. The budget can then explain why a 75-second meeting scene costs more than a 75-second narrated montage.

Where historical data is thin, use ranges rather than points: optimistic, baseline and stress pass rates. One purpose of a pilot is replacing assumed pass rates with the project's own data. Before approving a season budget, you should have real cost per usable second for the principal shot classes.

The critical path and the review bottleneck.

Generation is not necessarily the critical path. With golden character assets unapproved, dozens of generation operators cannot safely start. If one director can review forty shots a day, generating a hundred only expands work in progress. The schedule records capacity and queue per station, not just task start and end dates.

When review becomes the bottleneck, reduce concurrent starts, improve the quality of submissions, automate technical checks, or add calibrated reviewers. Never fake throughput by lowering the pass standard.

Earned seconds and burn-down.

"300 seconds generated" is not completion. Only seconds that pass the current gate and can move to the next station count as earned seconds; after entering the edit, count picture-approved and master-approved seconds separately. Different gates cannot be merged into one progress percentage.

daily_production:
  date: 2026-07-05
  generated_seconds: 420
  motion_approved_seconds: 96
  picture_locked_seconds: 58
  master_approved_seconds: 0
  spend_today: 1840
  primary_blocker: Lin Yun downward-angle reference awaiting approval
  forecast_change: +1.5_days

The burn-down uses remaining approved work and rolling throughput. After a model failure or asset rework, update the forecast immediately rather than continuing to report a date that is no longer possible.

The three reserves must not be mixed.

The production risk reserve covers model failure and reasonable rework. The creative exploration reserve covers comparing directions that are not yet locked. Client change is carried by the change budget or the contract. Merge them into one contingency and the team cannot tell whether an overrun came from internal efficiency, exploratory choice, or scope increase.

Releasing reserve also needs authority. A shot exceeding its cap triggers a producer review; an operator cannot simply continue retrying. Unused reserve is not a budget that has to be spent.

Calibrating the pilot budget.

The initial budget assumed an average of four generations per shot. Pilot data showed single-character reactions averaging 3.1, document hand shots 9.4, and multi-person mediums 12.2. The team did not apply a uniform season-wide increase; it changed the shot route: key documents composited separately, multi-person scenes given more coverage, and ordinary reactions left on the original route. The revised forecast for the first ten episodes came down, with compositing hours rising while generation retries and waiting fell substantially.

A budget is not a number frozen before a project starts. It is a model continuously calibrated by shot design, pass rates and production choices.

65.4 The budgeting flow, diagnosis, and scenario models#

SOP for budget and schedule.

First, decompose by station and shot class. Second, separate fixed from marginal. Third, apply historical pass rates. Fourth, add labor hours and reserves. Fifth, build dependencies and the critical path. Sixth, limit work in progress. Seventh, update daily reports and forecasts using usable seconds. Eighth, route changes through budget approval. Ninth, test restoration regularly.

Fault tree.

Symptom: generation spend is within budget and the project loses money. Labor, failures and overhead were not counted. Use true cost.

Symptom: everyone is busy and progress does not move. Too much work in progress, plus a review bottleneck. Limit WIP and clear the blockers.

Symptom: the schedule slips every week. Ideal pass rates were used. Roll the forecast from historical data.

Checklist, exercises and deliverables.

Check that the budget is built per shot; that fixed investment is amortized; that labor is included; that reserves are separate; that the schedule carries dependencies; that WIP is limited; that daily reports use usable seconds; and that backups can be restored.

Exercise one: build a shot-level budget for E001. Exercise two: forecast a ten-episode schedule from pass rates. Exercise three: identify one review bottleneck and adjust WIP.

Deliverables for this chapter: the production budget, the shot cost model, the critical path, the WIP board, the daily report, the burn-down forecast, and the backup plan.

Three-scenario budgeting.

Build optimistic, baseline and stress scenarios for the same project. Variables include shot count, still and motion pass rates, human review minutes, the share of complex compositing, client change and vendor pricing. Fixed asset investment stays the same while marginal cost moves substantially with the failure rate.

scenario_budget:
  pilot_E001_E003:
    optimistic: {motion_pass_rate: 0.55, total_index: 82}
    baseline: {motion_pass_rate: 0.38, total_index: 100}
    stress: {motion_pass_rate: 0.24, total_index: 146}
  stop_trigger: stress_forecast_exceeds_approved_reserve

The index here replaces currency deliberately, to emphasize method. When the stress scenario triggers, reduce shots, change route or re-approve — do not keep promising delivery against the baseline budget.

The resource capacity table.

Each station records weekly capacity, first-pass rate, rework volume and effective available capacity. An editor who can rough-cut ten episodes but fine-cut five is planned around five. If QC can review forty shots a day, the generation queue should not submit a hundred and twenty.

The capacity table also accounts for leave, training, model queuing and client feedback delay. Utilization sustained near 100 percent removes the ability to absorb rework and surprises; critical stations need buffer.

65.5 Capacity, variance attribution, unit economics, and cash flow#

Cost variance must be attributed.

Differences between actual and budget divide into quantity variance, price variance, efficiency variance, scope variance and time variance. More shots is quantity or scope. A model price rise is price. A falling pass rate is efficiency. Staff idle while waiting for a client is time. Different variances belong to different owners.

"We are 12 percent over this week" cannot direct action. "Multi-person shot pass rate fell from 35 to 18 percent, projected to affect the next 22 shots" triggers a shot redesign or a route change.

The first-ten-episodes budget calibration meeting.

After the pilot, recompute E001–E010 using real shot class counts, attempts, hours, rework and approved seconds. Decide per class: continue on the current route, build a template, redesign shots, raise the reserve, or stop a high-risk set piece. Cost reduction must not sacrifice the story promise or the blocking standards.

The meeting outputs a new budget, critical path, weekly capacity, risk list and the pricing boundary for the scale gate. Without calibration, do not commit to a fixed total price for sixty episodes from a three-episode sample.

Unit economics use approved seconds.

Generating a minute of material through an API is not obtaining a minute of finished work. Unit cost uses approved seconds, approved shots or accepted episodes as the denominator, and includes failed attempts, human review, rework, storage, tooling, management and rights. Compute it per shot class; an average conceals the real risk in multi-person and lip-sync shots.

Unit economics also connect to commercial outcomes: cost per approved minute, cost per qualified view, contribution per paying user, and payback period. Creative quality must not be dictated by one ROI figure, and a project still has to know which route cannot be sustained.

Schedule risk simulation and confidence intervals.

Critical task durations and pass rates are not fixed values. Build optimistic, baseline and stress distributions from historical batches, simulate the critical path and the delivery date, and report an 80 percent probability of completing by a date rather than a single promised day. Without history, state that expert estimates are being used, and update after the pilot.

Simulation is most useful for finding the variables that dominate delivery: review waiting, multi-person pass rate, client feedback or music licensing. Preparing an alternative route for the sensitive variables beats adding a uniform 20 percent buffer to everything.

Cash flow and profit are different statements.

A project can be profitable on paper and run out of cash because a client pays late while procurement is up front. The cash flow statement manages model credits, vendors, staff and tax by actual payment dates. The profit statement judges economics by project revenue and cost. Milestone payments should cover irreversible spend and the next stage's operating needs.

The daily production report shows only the budget consumption and forecast it needs to, without exposing salaries or client-sensitive information. Financial and creative permissions stay separate, and anomalies still enter project risk promptly.

A note on sources#

Costs published as "per episode" usually omit failures, labor and overhead. Budgeting by production unit and reporting in earned seconds is what reveals whether a workflow scales — and where it stops scaling.