# Kling full closeout plan (100% closeout target) Status: active closeout plan Captured: 2026-04-01 ## Purpose This document re-frames the remaining Kling documentation/verification work from "good enough for the current implementation scope" into a stricter "close as much of the remaining residual surface as possible" standard. This is not a denial of the earlier closeout. It is the next stricter pass. The practical question is: > what exactly is still open if we insist on a 100%-closeout posture? --- # 1. Current state ## What is already effectively closed These are not the main blockers anymore: - Tier 1 top-level field-level create/query documentation for current production-relevant video routes - Tier 2 top-level element create/query/list/delete surfaces - Tier 3 top-level image-generation create/query surfaces needed for current use - current approved production route interpretation - current Omni-first operational doctrine for multi-person audiovisual work - live confirmation that `element_list + sound="on"` can succeed - generalized conversational prompting guide for current practical use ## What is still open under a stricter 100%-closeout standard The remaining items are mostly one of these types: 1. **nested documentation parity gaps** 2. **runtime semantics that docs alone do not fully settle** 3. **coverage gaps where current docs are conservative but not exhaustive** 4. **status wording that still describes some items as deferred/refinement-only rather than actually closed** --- # 2. Remaining open surfaces ## A. Nested child-row documentation perfection ### Why this is still open Earlier closeout accepted top-level strength while leaving some nested object coverage at "strong enough for implementation" rather than "fully row-complete everywhere." ### Remaining examples - deeper camera child fields - deeper mask child fields - deeper voice child fields - deeper element nested subobjects - any nested branch that is currently reconstructed from examples/structure rather than from clean standalone tables ### Closure standard Close when one of the following is true: - exact preserved child-row evidence is extracted and transcribed, or - a clearly marked "no stronger preserved child-row evidence exists" note is added and all current example-derived structure is normalized/documented consistently ### Closure mode **Documentation closeout** No live billable create is required for this class. --- ## B. Audio / voice parity closeout ### Why this is still open Audio is no longer conceptually weak, but several parity questions remain less than fully closed. ### Remaining questions - exact Text-to-Video `voice_list` row parity versus the stronger Image-to-Video extraction - stronger contract wording for native audio behavior across current Series 3 generation routes - clearer documentation of interaction boundaries between prompt speaker tags, `voice_list`, and `element_voice_id` - whether preserved artifacts expose any stronger split between speech / ambient / BGM than currently documented ### Closure standard Close when: - preserved-document evidence is re-checked and any still-available rows are extracted, and - anything not provable from preserved artifacts is explicitly marked as runtime-semantic / live-confirm-only rather than left as soft ambiguity ### Closure mode **Documentation + wording closeout**, with optional later live refinement. Live tests help, but wording can still be closed more cleanly than it is now. --- ## C. Runtime semantics that are only partly doc-provable ### Why this is still open Some questions are not really pure documentation questions anymore. They are runtime behavior questions. ### Remaining examples - exact practical limits of multi-turn dialogue density by duration - exact runtime interaction of `element_list` with native audio quality/behavior under multiple scene types - stronger continuity behavior of `video_list` under short-window chaining - `video_list + element_list` as complementary continuity + recurring-identity path - broader speaker-consistency behavior over longer clips ### Closure standard Close when each item is classified into one of these buckets: - live-confirmed - doc-confirmed only - not supported by preserved evidence - intentionally left open pending explicit future verification The key is not to pretend docs can close runtime semantics they cannot. The key is to close the *classification* cleanly. ### Closure mode **Verification closeout or classification closeout** Not every item needs to be experimentally exhausted, but every item needs an explicit closed status. --- ## D. Capability/range/exhaustiveness gaps ### Why this is still open Current docs are conservative and production-focused, but not fully exhaustive across every capability map or support range question. ### Remaining examples - model/version support ranges beyond preserved field tables - broader endpoint-level support/range questions outside current production posture - remaining non-Omni path status wording (provisional vs reusable) ### Closure standard Close when: - each remaining broader support/range question is either answered from evidence, or - formally downgraded as out-of-scope / not evidenced / not production-relevant ### Closure mode **Documentation classification closeout** --- ## E. Status wording / closeout wording mismatch ### Why this is still open Several documents still use wording such as: - complete for current scope - good enough for current closeout - refinement-only residuals That wording was correct for the earlier target, but it is incompatible with a stricter 100%-closeout objective. ### Closure standard Close when the status layer itself is rewritten to distinguish: - fully closed - explicitly residual but classified - live-confirm pending - intentionally not closed ### Closure mode **Editorial closeout** --- # 3. Strict 100%-closeout strategy ## Principle Do not force every open question into a billable live-create test. That would be expensive and conceptually sloppy. Instead, close each remaining item by the correct mechanism: - documentation extraction - wording cleanup - explicit classification - targeted live verification only where runtime semantics genuinely require it --- # 4. Proposed closure sequence ## Step 1. Close wording/status residuals first Reason: - this changes the frame from vague residuals to explicit categories - it prevents ambiguity about what "open" means Deliverables to update: - `kling-documentation-progress-board.md` - `kling-priority-todo.md` - potentially `kling-current-state-summary.md` ## Step 2. Close nested child-row/documentation parity residuals Reason: - these are document-only tasks - they are the easiest true closeout wins Likely sources to revisit: - existing tier docs - preserved qingque/full-reference artifacts - extracted deep pages already in repo ## Step 3. Close audio/voice parity wording Reason: - this is one of the few remaining substantive documentation residuals - current live testing already gives stronger footing than the docs currently express ## Step 4. Reclassify runtime semantics explicitly Reason: - some items should be closed by saying exactly what kind of confidence they have - not by pretending preserved docs prove more than they do ## Step 5. Run only the minimum new live verification that actually upgrades an open runtime class Candidates: - stronger `video_list` continuity validation - `video_list + element_list` complementarity validation - one targeted longer-form audio/dialogue runtime pass if needed for a specific unresolved classification --- # 5. Closeout rubric For each remaining item, mark it with exactly one terminal state: - **Fully closed by preserved documentation** - **Fully closed by live verification** - **Closed by explicit negative finding / not evidenced** - **Closed as intentionally out-of-scope for current repo production posture** - **Still open** 100%-closeout means reducing "Still open" to zero. It does **not** mean pretending every item has the same kind of proof. --- # 6. Immediate next task recommendation The next correct task is: > rewrite the status/TODO/progress docs so every remaining residual is converted from vague refinement language into explicit terminal-state candidates. That is the cleanest gateway to a real 100%-closeout pass.