# Kling documentation progress board

Status: strict-closeout reclassification in progress
Captured: 2026-03-29
Updated: 2026-03-31

## Purpose
Show what parts of the Kling documentation effort are:
- fully closed by preserved documentation
- fully closed by live verification
- closed by explicit classification/out-of-scope handling
- still open under the stricter 100%-closeout standard

This is a progress board, not a replacement for the detailed TODO or gap log.

---

# A. High-level completion view under strict closeout

## 1. Documentation architecture
Status: **fully closed by preserved documentation**

Completed:
- source/reference layer established
- field-level layers established
- implementation guidance layer established
- policy/guardrail layer established
- external blog-note layer established
- README navigation updated repeatedly
- doc-set map created

Meaning:
- the document system itself is organized and navigable strongly enough to support strict closeout work

---

## 2. Tier 1 field-level reference
Status: **closed at top-level; nested-parity residual still open**

Strongly covered already:
- Omni-Video create/query family
- exact Text-to-Video create rows
- exact Image-to-Video create rows
- exact Text-to-Video query single/list path/query/response rows from preserved capture
- exact Image-to-Video query single/list path/query/response rows from preserved capture
- `multi_prompt[]`
- `image_list[]`
- `video_list[]`
- `element_list[]` downstream usage
- `voice_list[]` existence and key rules on image-to-video create
- `mode`, `duration`, `aspect_ratio`, `watermark_info`, and `final_unit_deduction` working picture

Still open under strict closeout:
- deeper nested child-row perfection for camera / mask / voice branches
- explicit reclassification of runtime semantics that preserved docs alone do not fully settle

Meaning:
- Tier 1 top-level documentation is strong enough for implementation
- what remains is not top-level shape uncertainty, but nested parity and runtime-classification cleanup

---

## 3. Tier 2 elements reference
Status: **closed at top-level; nested/semantics residual still open**

Strongly covered already:
- full section map exists
- downstream `element_list[].element_id` documented
- exact Create endpoint path documented
- exact Create top-level request rows documented
- exact Multi-Image / Video Character example bodies documented
- exact Query Custom single/list endpoint paths documented
- exact Query Presets list endpoint path documented
- exact Delete endpoint path and body field documented
- exact top-level response metadata family documented (`element_id`, names, reference type, voice info, tags, owned_by, status)

Still open under strict closeout:
- nested-object child coverage still partly reconstructed from structure/examples rather than standalone child-row tables for every nested object
- preserved docs still show slight single-query selector ambiguity (`{id}` route vs `task_id` / `external_task_id` wording)
- top-level lifecycle is now live-confirmed for create/query success, but broader semantics wording still needs final classification cleanup

Meaning:
- Tier 2 is no longer open in the old Phase 1 sense
- it remains open only in the stricter closeout sense of nested parity / selector wording / final semantics classification

---

## 4. Tier 3 image-generation reference
Status: **closed for current preserved-document surface; support-range residual still open**

Strongly covered already:
- full section map exists
- strategic role in frame-anchor generation documented
- relation to element workflows documented
- exact create-body rows transcribed for Omni-Image and Image Generation
- exact query single/list result rows transcribed for both families
- quality/resolution/result-type fields transcribed

Still open under strict closeout:
- some capability-map-dependent support ranges remain outside the preserved field-table evidence
- no richer nested element attachment field was recoverable beyond `element_list[].element_id`; this needs explicit negative/limit-state wording rather than soft residual wording
- live support verification for model/version-specific feature ranges remains separately classified, not silently deferred

Meaning:
- Tier 3 top-level field documentation is closed strongly enough for current use
- what remains is exhaustiveness/range classification rather than core payload-shape uncertainty

---

## 5. Audio / voice reference
Status: **partially closed; parity/semantics residual still open**

Strongly covered already:
- standalone audio endpoints now field-level visible from preserved artifacts
- custom voice management surfaces now field-level visible
- Create Element surfaces expose `element_voice_id`
- Image-to-Video now has exact preserved `voice_list` request-body structure
- Text-to-Video and Image-to-Video both preserve prompt-level `<<<voice_1>>>` guidance
- strongest current native speaker-binding pattern is now source-grounded as:
  - speaker tag inside `prompt`
  - sibling `voice_list`
  - `sound: "on"`

Still open under strict closeout:
- exact Text-to-Video `voice_list` row with the same extraction strength as Image-to-Video
- exact runtime contract for native audio inside all Series 3 video-generation paths
- exact interaction rules between element-bound voice (`element_voice_id`) and generation-time `voice_list`
- exact dedicated fields, if any, for speech vs ambient/BGM separation in Series 3 native generation payloads

Meaning:
- standalone audio is now materially documented and no longer conceptually weak
- however, audio/voice remains one of the main true residual surfaces under a strict 100%-closeout target
- remaining work splits cleanly into documentation parity extraction and runtime-semantic classification/verification

---

## 6. Pipeline policy / guardrails
Status: **fully closed by preserved documentation + current operating decisions**

Completed:
- billable create guardrails
- SOT-before-create rule
- one-scene <= 15s policy
- approved payload patterns
- field status register
- current-production route narrowing (`text2video`, `image2video`, Omni only)
- legacy/non-current route downgrade handling for routes such as `reference2video`

Meaning:
- current policy/guardrail guidance is explicit enough to be treated as closed for the present repo posture

---

## 7. Code synchronization
Status: **narrow current-scope closeout achieved; broader sync still open**

Completed:
- model policy lock to 3.0 / 3.0 Omni
- builders/validators for first-frame, video_list, element_list scaffolding
- validator enforcement of live-confirmed Omni/text multi-shot `multi_prompt[].index + duration + prompt`
- README / status wording updated so Phase 2 is not described as wholly pending

Still incomplete:
- broader builder/validator cleanup outside the Omni-first narrow scope
- full helper retirement / quarantine pass
- full alignment with future field-level docs
- stronger provenance comments for every doc-derived vs live-confirmed branch

Meaning:
- enough sync now exists to honestly mark the current narrow Phase 2 target as done,
  but not to claim full pipeline-wide code/document convergence
- specifically, this closeout must not be interpreted as a resolved non-Omni model-policy decision or as approval for broad non-Omni defaults

---

# B. Progress summary by priority under strict closeout

## Documentation-first goal
Current status: **top-level closeout achieved; residual parity/semantics work remains**

## Code-sync-second goal
Current status: **narrow-scope closeout achieved; broad convergence not claimed**

## Live-verification-later goal
Current status: **still open where runtime semantics cannot be honestly closed by preserved docs alone**

---

# C. What strict closeout now requires
Under the stricter closeout target, documentation is not treated as fully closed merely because it is implementation-safe.

Instead, every remaining item must end in one of these states:
- fully closed by preserved documentation
- fully closed by live verification
- closed by explicit negative finding / not evidenced
- closed as intentionally out-of-scope for the current production posture
- still open

The current doc set is past the old implementation-safe threshold.
The remaining work is to reduce `still open` items toward zero without pretending preserved evidence proves more than it does.

## 2026-03-29 closeout audit result
- preserved deep-page artifacts do contain exact Text-to-Video and Image-to-Video query envelopes, including `task_result`, `watermark_info`, and `final_unit_deduction`
- preserved deep-extra artifacts do contain the Tier 2 element create/query/list/delete top-level rows and examples
- preserved deep-extra artifacts do contain Tier 3 Omni-Image / Image Generation create/query rows at the current field-level target
- remaining gaps after that audit were real but narrower than before: capability-map ranges, nested child-row perfection, and live/runtime semantics not fully closed by preserved evidence alone