HISTORICAL DIG · JUL—AUG 2026
When is the next episode allowed to start?
Production tools increasingly expose shot state, asynchronous jobs, and recovery, while public repositories still record partial failures that pull a whole batch into reruns, appended material that triggers full resets, and resumed jobs that can follow the wrong provider. Starting a batch is not the same as resuming it safely.
AI SHORT-DRAMA MOTHER WORKFLOW · NODE 09
Relation to the current subject
Node 08 passes release receipts; node 09 decides only whether one next episode may begin.
Inspect the next-episode condition written before release, the observation window, changes, owners, and episode allowance. Even a pass opens only a pilot of at most two episodes: one active and one queued.
Record state at the time
- AI short drama
- Series production
- Deep record
- Developing
- Historical E2 · maker structure and repository issues rechecked · no BitShovel series-production result
R8 CURRENT JUDGMENT
Judgment at the time
A first-release pass does not open a season; approve one episode and prove work resumes from the right place.
Open full rationale and return rules
Node 09 is not a batch-generation button; it is permission for the next episode. Approve a pilot batch of at most two episodes only after node 08 has closed its observation window, the prewritten next-episode condition is actually met, no new rights or channel conflict exists, every change has an owner, and episode spend and rework caps are explicit. Only one episode may be active at a time; the next is queued, never pregenerated. Each episode repeats nodes 02–08 and preserves its promise, asset versions, shot input hashes, provider/model/configuration snapshot, external task IDs, accepted files, cost, QC, and approval receipts. Accepted shots cannot be silently overwritten, and a failure invalidates only affected dependencies. Detailed jobs and media for a completed episode move into an openable archive package; the series overview keeps only bounded summaries and receipt links. Reaching a shot, candidate, or task limit stops intake instead of silently deleting history. Run one deliberate interruption-and-resume check during the batch. If execution identity is incomplete, accepted shots are regenerated, or cost or continuity crosses its cap, freeze new jobs and return to node 02, 03, 04, 05, or 06. Only two consecutive episodes that each pass node 08, without redoing accepted assets after interruption, can send a batch receipt to node 10. That still proves no scaling efficiency, audience stability, or profit.
- Who it was for
- People leaving node 08 with one frozen master, channel-state receipt, closed observation window, prewritten next-episode condition, and explicit episode allowance, deciding whether any follow-up episode should begin
- Why it mattered then
Open note
A first release produces one bounded observation. Opening a season too early multiplies character drift, provider mistakes, rework, spend, and channel uncertainty at once. Keeping every historical asset and job alive in one growing board also lets working memory expand without a stop.
- Where the judgment stops
Open note
This is an unrun, unfrozen two-episode pilot-batch method. BitShovel has no script, sample, release account, generation job, provider bill, real recovery drill, two-episode run, audience, settlement, or profit result. Every project, episode, digest, amount, task, and stop reason in the visual is fixed fiction.
BEFORE CREATING A BATCH
Fit check
An unmet release condition, a changing base, or a frequency-only goal stays outside series production.
This record may help
- Node 08 closed its observation window, the next-episode condition was written before release and is now met, and changes, owners, and episode allowance are openable.
- The team will keep only one episode active and stop at shot, candidate, task, spend, or rework limits instead of auto-expanding or deleting history.
Do not use it yet when
- The first release is still under review, its observation window is open, or data cannot satisfy the prewritten condition; keep waiting rather than marking unknown as pass.
- Story promise, character assets, model route, cost, or master is still changing; return to the owning node instead of hiding base-version drift inside a series board.
- The goal is posting frequency, season scale, or revenue prediction without an auditable per-episode stop line; that does not belong in this node.
HISTORICAL TRIAL · UNRUN, UNFROZEN 60-MINUTE BATCH-AND-RECOVERY DESIGN CAP
Historical trial
At most two episodes and one active; freeze accepted work, invalidate only affected dependencies, and stop at every limit.
- Time needed
- A 60-minute active cap to inspect node-08 receipts, freeze a two-episode batch, establish episode/archive structure, and design one recovery check; production and channel waits are separate, and this round remains unrun
- Likely cost
- Approve only the next episode allowance, never a season upfront. The queued episode opens only from the prior completion receipt and remaining allowance; estimate, charge, refund, and active labor stay separate
- Permission boundary
- Name owners and write permissions for script, assets, generation, editing, review, and upload. External jobs retain only safe task aliases plus provider and configuration snapshots—never credentials, accounts, or local paths
- 01
First step
Open this step
Open node 08's master, channel state, observation window, next-episode condition, change list, and allowance receipt. Create only E02 and E03: E02 active, E03 queued. Per E02 shot, record input digest, asset version, provider/model/configuration, safe task alias, candidate cap, accepted file, charge, and return node. Deliberately interrupt after partial completion and resume from disk receipts, reopening only failed dependencies. Move a completed episode into an archive package while the overview retains only episode, state, cost summary, version, and receipt links.
- 02
What should exist
Open this step
A bounded two-episode permit with one active episode, one episode receipt that survives interruption, and an archive index that does not keep every media object and task live in working memory.
- 03
How to tell it worked
Open this step
Two consecutive episodes each complete nodes 02–08; resumption regenerates no accepted shot; every invalidation has a dependency and owner; spend stays inside each allowance; archives open while the overview remains bounded. The only allowed claim is `completed one two-episode pilot batch`.
- 04
Stop when
Open this step
Stop when the next-episode condition is unmet, execution identity is incomplete, an accepted shot reruns, core character or story drifts, episode spend or rework crosses its cap, an archive cannot reopen, or any bounded list reaches its limit.
- 05
How to back out
Open this step
Freeze new jobs and queued episodes immediately while keeping the last accepted files, failed jobs, and cost receipts. Return story to node 02, assets to 03, execution identity to 04, spend to 05, master to 06, or release condition to 08. A repair creates a new batch revision and never overwrites the failed record.
BITSHOVEL R8 BATCH RECEIPT · FIXED FICTIONAL RECORD
Method and interfaces
Missing execution identity freezes new jobs; accepted shots do not follow the whole batch into rerun.
Open method boundary
The visual contains no real script, media, job, or bill. It shows how next-episode permission, one active episode, queueing, accepted-output freeze, an execution-identity gap, and a named return fit together.
DEMO BATCH 09 shows process structure only: E02 is the sole active episode and E03 remains queued. Missing provider identity on resume freezes new jobs and returns to node 04 while accepted shots stay frozen. It is not a BitShovel project or series-production result.
About this image and its source
Created by BitShovel on August 17, 2026 with no real script, work, job, provider, account, bill, recovery, or channel data.
Open source page ↗SOURCES AND EVIDENCE
Sources and evidence
Maker documentation establishes job and recovery entry points; repository issues bound partial reruns, appended material, and execution identity. None is a BitShovel series-production result.
Open evidence boundary
Historical E2 remains the old-page label. Revision 8 uses current Jellyfish maker documentation to establish shot readiness, asynchronous jobs, cancellation, and recovery entry points. Three public ArcReel repository issues bound risks around partial reruns, appended content, and execution-identity recovery. They are design inputs, not stability findings for BitShovel, other teams, or every version.
- 01Maker workflow description · Rechecked
Jellyfish separates shot readiness, generation jobs, cancellation, and recovery into trackable states
Open source boundary
Current maker documentation establishes these entry points and asks that task state remain recoverable after refresh. It proves no production stability, loss-free persistence, bounded memory, or repeated delivery by any team.
Open original source ↗ - 02Public repository issue · Rechecked
ArcReel #975 records a partial failure pulling a batch into rerun when version and manual fallback are absent
Open source boundary
The issue remains an open capability gap and explicitly says AI assisted its triage. It supports freezing accepted outputs and local invalidation, not a claim that every ArcReel version or batch fails.
Open original source ↗ - 03Public repository issue · Rechecked
ArcReel #1430 records appended chapters colliding with planned-content fingerprints and forcing a reset
Open source boundary
The issue gives a concrete dependency boundary between consumed ranges and appended material. It supports versioned inputs and incremental invalidation, not a claim that BitShovel implemented or verified that recovery.
Open original source ↗ - 04Public repository issue · Rechecked
ArcReel #1663 records resumed jobs polling the wrong provider when execution identity is missing
Open source boundary
The repository marks it parked and says several conditions must coincide, the job is retryable, and no data corruption is expected. It supports binding receipts to provider, model, and configuration, not inflating the case into a universal high-risk failure.
Open original source ↗
REVISION RECORD
Revision record
Revision 8 continues production Revision 7 and narrows series production to a two-episode pilot, bounded state, and one auditable interruption recovery.
Open 8 revisions
- Revision 1
Before expanding a sample into a series, named where work resumes, who owns it, and who approved it.
- Revision 2
Refined checkpoints to shot hashes, provider, task IDs, partial reruns, and an interruption-recovery drill.
- Revision 3
Made first-release validation a prerequisite for scaling and added a per-episode operating ledger.
- Revision 4
Replaced the generated lead with a released narrative frame carrying explicit authorship and an open license.
- Revision 5
Moved the lead to a project overview tied to series work, with real product UI limited to process reference.
- Revision 6
Briefly restored a real narrative frame as the lead and moved production interfaces downstream.
- Revision 7
Returned the lead to a task-aligned fixed project overview instead of letting narrative imagery stand for series production.
- Revision 8
Narrowed scaling to a two-episode pilot with one active episode, bounded archives, and one recovery drill; replaced product UI with a fixed fictional stop receipt.
WHAT THIS RECORD LEAVES
What this record leaves
Node 09 leaves one bounded claim: one two-episode pilot batch was completed.
A failure freezes new jobs and returns to a named node without overwriting accepted files or failed receipts. Only two episodes that each pass node 08, with no accepted-work rerun after recovery, can send a batch receipt to node 10.
Back to the Dig index