Datasets:

maxwellinked
/

time-lapse-artifacts

+# Geometric & Graph-Theoretic Schema Extension
+### *Formal Contract for Discrete Structural Ground Truth in Time-Lapse Artifacts*
+This specification defines the discrete mathematical and topological extension to the canonical time-lapse repository schema. It bridges the physical drawing video recordings (`record_id: tla-...`) and physical works (`work_id: tlaw-...`) with the offline mathematical discovery engines (`conjecture_lab`, `workbench.py`).
+---
+## 1. Schema Fields & Types
+| Field Name | Type | Value Domain | Meaning |
+|:---|:---:|:---|:---|
+| `topological_family` | string / enum | `planar_network`, `chordal_ring`, `triangulated_lattice`, `bipartite_partition`, `extremal_configuration`, `hypercube_projection`, `unclassified` | The formal topological classification of the geometric structure being inscribed. |
+| `terminal_node_count` | integer (or blank) | $\mathbb{Z}_{\ge 0}$ | Exact ground-truth count of discrete vertices/intersections in the completed work. |
+| `terminal_edge_count` | integer (or blank) | $\mathbb{Z}_{\ge 0}$ | Exact ground-truth count of connecting lines/segments in the completed work. |
+| `graph_invariants` | semicolon list (or blank) | Key-value pairs (e.g., `is_planar=true; chromatic_number=4; is_connected=true; diameter=3`) | Formal structural invariants verified on the completed terminal graph. |
+| `sequence_formula_id` | string (or blank) | e.g., `PIPE_01`, `PIPE_02` | Reference identifier linking the work to the discovered symbolic recurrence in `workbench.py`. |
+| `math_dataset_ref` | string (or blank) | e.g., `TT_06`, `FC_07` | Reference identifier to the cryptographically verified theorem in `verified_math_dataset.jsonl`. |
+---
+## 2. Missing-Value & "Rec01" Scope Rules (Strict Integrity)
+To prevent models or researchers from falsely interpreting an in-progress intermediate clip as a finished drawing, the following rules are strictly enforced per `SCHEMA.md` (§ Missing-Value Rules):
+1. **Single-Recording Drawings (No `RecNN` Suffix):**
+   * The fields describe the completed work and are populated directly on that row.
+2. **Multi-Clip Recordings (`Rec01`, `Rec02`, ..., `RecNN`):**
+   * **Only `Rec01` (the primary reference clip)** carries the terminal ground-truth values (`topological_family`, `terminal_node_count`, `terminal_edge_count`, `graph_invariants`).
+   * **All intermediate and non-terminal clips (`Rec02` through `RecNN`) must remain BLANK.**
+   * As defined in `SCHEMA.md`: *"Blank numeric values must remain blank; they are not zero."* An intermediate clip is an incomplete drawing state, not a graph with 0 nodes.
+3. **Pre-Standard / Fragment Holdings Without Terminal Verification:**
+   * Fields remain **blank** until an explicit human-verified ground-truth audit is performed. Blank values must never be promoted to inferred zero.
+4. **Physical Work Identity (`Work_Identity/works.csv`):**
+   * These fields also attach to `works.csv` at the `work_id` (`tlaw-...`) level, ensuring that each of the 425 physical artworks retains exactly one authoritative structural ground-truth definition.
+---
+## 3. Concrete Example in `metadata.csv`
+### Case A: Single-Recording Drawing (Full Terminal Counts Populated)
+```csv
+file_name,record_id,work_id,...,topological_family,terminal_node_count,terminal_edge_count,graph_invariants,sequence_formula_id
+Standard_Time_Lapses/2026-08-24.0157.BallpointPen.Ink.Paper.9x12.mp4,tla-8f92...,tlaw-5c97...,...,triangulated_lattice,18,36,is_planar=true;chromatic_number=3,PIPE_02
+```
+### Case B: Multi-Clip Drawing (Populated on Rec01, Blank on Rec02–Rec03)
+```csv
+# Rec01 carries the verified terminal ground truth:
+Standard_Time_Lapses/2026-08-22.0124.BallpointPen.Ink.Paper.14x17.Rec01.mp4,tla-1a2b...,tlaw-a43a...,...,chordal_ring,24,48,is_planar=false;chromatic_number=4,PIPE_05
+# Rec02 is in-progress; fields remain BLANK (not zero, not null):
+Standard_Time_Lapses/2026-08-22.1613.BallpointPen.Ink.Paper.14x17.Rec02.mp4,tla-3c4d...,tlaw-a43a...,...,,,,,
+# Rec03 is in-progress; fields remain BLANK:
+Standard_Time_Lapses/2026-08-23.1932.BallpointPen.Ink.Paper.14x17.Rec03.mp4,tla-5e6f...,tlaw-a43a...,...,,,,,
+```
+---
+## 4. Automatic Ingestion & Validation Integration
+During Saturday ingestion:
+1. `tools/validate_metadata.py` verifies that `terminal_node_count` and `terminal_edge_count` are valid positive integers whenever present.
+2. The validator asserts that for any multi-clip sequence, only `Rec01` has non-blank values; if `Rec02` carries values, validation emits a warning or error to prevent duplicate counting.
+3. `workbench.py pipeline` automatically verifies that any `sequence_formula_id` matches a mathematically valid held-out prediction in `cross_engine_certificate.json`.

README.md CHANGED Viewed

The diff for this file is too large to render. See raw diff

SCHEMA.md CHANGED Viewed

@@ -1,570 +1,583 @@
-# Schema
-This document defines canonical filenames, the active video-object indexes,
-and their relationship and timing sidecars. Filenames are human-readable
-descriptions; `record_id` is the persistent identity of one published video
-object. Archive-wide row counts are counts of indexed video files.
-## Canonical Filename Format
-`YYYY-MM-DD.HHMM.Tool.Medium.Support.Dimensions[.RecNN].ext`
-Examples:
-```text
-2025-07-15.0448.FountainPen.Ink.Paper.9x12.mp4
-2026-06-29.UnknownTime.FountainPen.Ink.IllustrationBoard.30x40.Rec11.mp4
-2026-06-29.UnknownTime.FountainPen.Ink.IllustrationBoard.30x40.Rec12.mp4
-```
-### Filename fields
-- `YYYY-MM-DD` — documented finish date.
-- `HHMM` — documented finish time in 24-hour local civil time.
-- `UnknownTime` — used when no reliable finish time is documented.
-- `Tool` — drawing instrument, such as `Pencil`, `FountainPen`, `BallpointPen`,
-  or `Brush`.
-- `Medium` — applied material, such as `Graphite` or `Ink`.
-- `Support` — drawing surface, such as `Paper`, `IllustrationBoard`, or
-  `WatercolorPaper`.
-- `Dimensions` — support dimensions in inches; order preserves support
-  orientation, such as `9x12` or `12x9`.
-- `RecNN` — optional file-order identifier used when documented order must be
-  preserved or otherwise identical filenames require distinction.
-  Numbering begins with `Rec01`. Historical numbering may continue across
-  finish-date labels; the prospective protocol has the narrower timestamp rule
-  defined below.
-- `ext` — file extension.
-### Filename rules
-- Fields are separated by periods.
-- Every tool listed in an active Standard metadata row must appear as one
-  exact, case-sensitive filename token. Multiword tools remain one token, such
-  as `BallpointPen` or `FountainPen`.
-- Across both active tiers, every exact filename token in the documented tool
-  vocabulary (`Pencil`, `FountainPen`, `BallpointPen`, and `Brush`) must appear
-  in that row's metadata `tool` list. Historical Pre-Standard metadata may
-  retain additional independently documented tools that are absent from a
-  filename; a standalone historical `Pen` token is not normalized without
-  separate documentation.
-- Style and aesthetic classifications are excluded.
-- `TL` is excluded because time-lapse is implicit in the primary collection.
-- `0000` is reserved for a documented midnight finish time.
-- `RecNN` specifies file order within a documented group only. It does not by
-  itself establish a separate drawing, recording or session boundary, or
-  completeness, and it does not reset merely because a historical documented
-  finish-date label changes. Under
-  `standard-2026-08-27`, all members of one documented prospective grouping
-  instead share the grouping timestamp and `RecNN` records chronological
-  file position within that group.
-- `UnknownTime` records uncertainty; it is not replaced by an inferred time.
-- Paper-surface variants such as `HotPress` and `ColdPress` may remain encoded
-  in filenames until a dedicated metadata field is added.
-- Historical relationships must not be inferred from filename similarity
-  alone.
-- Revision-pinned historical sidecars retain the path recorded at their source
-  revision; an active filename correction does not rewrite that evidence.
-Tool and medium tokens in a filename describe documented materials, but older
-filenames do not always encode an unambiguous one-to-one pairing. Metadata must
-therefore preserve tool and medium sets without manufacturing positional pairs.
-## Canonical Metadata Contract
-`metadata.csv` indexes the validated canonical subset of MP4 files stored
-directly in `Standard_Time_Lapses/`. Direct placement alone does not make a file
-canonical: every released canonical file must have exactly one metadata row,
-complete required technical fields, and the necessary rights and provenance
-clearance. Before publication, it must cover every MP4 stored directly in
-`Standard_Time_Lapses/` at the exact reviewed revision.
-Each metadata row and its `record_id` identify one published MP4 file. They do
-not assert one drawing, one session, or one complete process. Documented
-relationships among multiple video objects remain in sidecars and do not merge
-or replace their `record_id` values.
-Historical MP4 objects may represent one source recording or may combine
-multiple recordings, including through InShot. That possibility is
-record-specific; it does not mean every historical file is compiled. For the
-documented `standard-2026-08-27` cohort, each canonical MP4 and metadata row
-represents one unchanged source recording. Cohort membership is established by
-the metadata protocol value and `record_id` membership in the revision-pinned
-relationship packets, not by filename dates or a `RecNN` suffix alone. See the
-prospective [`2026-08-27` packet](Standard_Time_Lapses/Provenance/2026-08-27/README.md)
-and the separate
-[June 18–30 historical relationship](Standard_Time_Lapses/Provenance/Historical_Relationships/2026-06-18--2026-06-30-fountain-pen-30x40.md).
-The canonical index does not make provenance, unresolved, pre-standard, or
-legacy files canonical.
-### Identity and collection fields
-| Field | Type | Meaning |
-| --- | --- | --- |
-| `file_name` | string | Repository-relative canonical video path. |
-| `record_id` | string | Persistent identity in the form `tla-` plus 16 lowercase hexadecimal characters. |
-| `work_id` | nullable string | Confirmed drawing identity, `tlaw-` plus 16 lowercase hexadecimal characters, copied from the authoritative work registry. Blank means unresolved. |
-| `collection_tier` | enum | `standard` for this index. |
-| `is_canonical` | boolean | `true` for every row in this index. |
-| `acquisition_protocol_version` | enum | Governing Standard workflow: `standard-2025-07-13` for the earlier Standard workflow or `standard-2026-08-27` for canonical fragments reviewed into the prospective fragment-sequence workflow. Assignment is based on documented sidecar membership, not finish date or a `RecNN` suffix alone. |
-The initial `record_id` values were generated from a namespaced SHA-256 of the
-published repository path. Once assigned, the stored value is authoritative and
-must be retained across filename corrections or tier changes. New tooling must
-assign an ID only when a record does not already have one. When a genuinely new
-record is promoted from a noncanonical holding, its ID is seeded from that
-object's prior archival path; it must not inherit an ID already attached to a
-different object at the corrected destination path.
-### Pre-Standard collection contract
-`Pre_Standard_Time_Lapses/metadata.csv` indexes the direct MP4 files in that
-folder. Each row uses `collection_tier = pre_standard`, `is_canonical = false`,
-and a blank `acquisition_protocol_version`. Pre-Standard status records a
-different acquisition/documentation era; it does not remove the requirements
-for stable identity, exact content hash and size, or encoded-stream metadata.
-Dates, times, tools, media, support, and historical provenance may remain blank
-when the surviving evidence is insufficient. Those blanks are historical
-uncertainty, not failed technical ingestion.
-### Confirmed drawing identity projection
-Both active indexes include `work_id` immediately after `record_id`. The Viewer
-declares it as `large_string`, with an empty CSV cell represented as null.
-The 35-field schema retains every original field and the five timing-display
-fields. `Work_Identity/works.csv` and `Work_Identity/work_record_links.csv` remain
-the authoritative relationship source; the index column is a generated copy.
-Join by exact `record_id` within the same repository revision.
-Only positive, owner-confirmed links to active work IDs are projected. The
-current 1,106 linked recordings share 425 active work IDs; 128 Pre-Standard recordings remain blank.
-No IDs are minted, reassigned or inferred from dates, filenames, sequence
-numbers, shared materials or appearance by this projection. A work ID does not
-assert recording-set completeness, gapless capture, session boundaries or
-physical-reference relationships. Different work IDs are not independently
-verified negative pairs. See the [registry contract](Work_Identity/README.md)
-for correction, retirement and supersession rules.
-After an authorized registry update, rebuild the display column and validate:
-```sh
-python tools/project_work_ids.py --root . --write
-python tools/project_work_ids.py --root .
-python tools/project_recording_times.py --root .
-```
-The validator rejects duplicate or ambiguous links, inactive/missing work IDs,
-unconfirmed links, unpinned source revisions, absent indexed records, malformed
-rows and drift from the registry. It checks same-revision confirmation-file
-checksums and the format of historical revision-pinned evidence references;
-it does not repeat the historical source-media audit or refetch those URLs.
-Both indexes and the registry are checked before either index is written.
-Blank work IDs are valid and do not block ingestion. Historical source indexes
-without this display column retain their original schema at their pinned revisions.
-### Explicit timing display fields
-The active indexes retain all original 29 columns and add the following five
-string columns. These are a generated display projection, not a new timing
-evidence source. Together with `work_id`, the Dataset Viewer declares 35 columns for both tiers.
-| Field | Meaning |
-| --- | --- |
-| `recording_start_local` | Reviewed device-filename start copied from the validated current timing view, including its explicit UTC offset. Blank if unknown or outside reviewed coverage. |
-| `recording_start_basis` | `device_filename_interpreted_in_declared_timezone` or `unknown` for a covered record. The existing `not_in_timing_snapshot` value is retained for records outside all registered timing evidence at the selected repository revision. |
-| `documented_finish_date` | Exact alias of the existing `date` value. |
-| `documented_finish_time` | Exact alias of the existing `time` value. It is not a per-record stop time for prospective groups. |
-| `finish_label_scope` | `complete_drawing_or_group` for `standard-2026-08-27`; `historical_scope_unspecified` for other surviving finish labels; `unknown` if both finish labels are blank. |
-Recording starts are joined by `record_id` and checked against current paths,
-byte lengths, SHA-256 and Xet identities. The
-[Current recording-time view](Standard_Time_Lapses/Provenance/Recording_Timing_Current/README.md)
-combines the unchanged Timing v2 snapshot with explicitly registered,
-revision-bound ingest observations. It currently covers 742 recordings, with
-514 filename-derived starts and 228 unknown starts; 492 indexed records remain
-outside this coverage. New evidence does not rewrite frozen timing packets.
-MP4 creation tags retain their creation/finalization meaning and are not
-renamed as recording stops. Source filenames and media paths are unchanged.
-The original 339 Timing v2 rows and their projected timing values are retained.
-After an index update, rebuild the five display fields and then validate:
-```sh
-python tools/project_recording_times.py --root . --write
-python tools/project_recording_times.py --root .
-python tools/validate_recording_timing_v2.py --root .
-```
-The projection validator rejects missing columns, changed evidence values,
-duplicate identities, mismatched source bindings and missing cohort records.
-The existing metadata validator continues to understand the original fields
-for compatibility with revision-pinned historical inputs.
-### Session and material fields
-| Field | Type | Meaning |
-| --- | --- | --- |
-| `date` | `YYYY-MM-DD` | Documented finish date. For `standard-2026-08-27`, this is the finish date of the complete grouping/physical work shared by every member; earlier records retain their historically documented label. |
-| `time` | `HH:MM` or blank | Documented finish time; blank means unknown. For `standard-2026-08-27`, this is the finish time of the complete grouping/physical work shared by every member. |
-| `tool` | semicolon list | Ordered unique set of documented tools. |
-| `medium` | semicolon list | Ordered unique set of documented media. |
-| `support` | string | Physical drawing surface. |
-| `dimensions` | string | Support dimensions in inches, such as `9x12`. |
-Multiple values use semicolons:
-```text
-tool = Pencil;FountainPen
-medium = Graphite;Ink
-```
-`tool` and `medium` are independent observed-value lists. Equal lengths and
-positional correspondence are **not guaranteed**. Repeated media are collapsed
-to one value. A future paired representation should use a structured field and
-must be based on documented evidence rather than token position.
-`date` and `time` preserve the best documented finish labels associated with a
-row. For the earlier Standard workflow, the original 29-field schema does not
-distinguish whether the source documentation scoped a label to an individual
-recording, a recording session, a reviewed fragment sequence, or the physical
-work. Equal values, differing values, and `RecNN` suffixes do not resolve that
-historical scope. For `standard-2026-08-27`, the scope is explicit: every
-`RecNN` member in one prospective grouping carries the documented finish date
-and finish time of the complete grouping/physical work. The timestamp is not
-the individual end time of each fragment; `RecNN` records chronological
-position within the group.
-### Encoded-video fields
-| Field | Type | Meaning |
-| --- | --- | --- |
-| `duration_seconds` | float | Encoded playback duration. |
-| `width_px` | integer | Stored encoded width. |
-| `height_px` | integer | Stored encoded height. |
-| `orientation` | enum | Stored pixel orientation derived from width and height. |
-| `encoded_frame_rate` | float | Encoded playback frame rate as reported by the container; fractional values are allowed. |
-| `video_codec` | string | Encoded video codec. |
-| `pixel_format` | string | Encoded pixel format. |
-| `container` | string | Media container. |
-| `has_audio` | boolean | Whether the stored file contains an audio stream. |
-| `hub_xet_hash` | string | Lowercase 64-character Xet file ID from the pinned Hub resolve response's `X-Xet-Hash` header. This is distinct from the LFS-compatible SHA-256. |
-| `size_bytes` | integer | Stored file size in bytes. |
-These fields describe the published encoded file. They do not establish the
-original frame-capture interval, real drawing duration, speed-up factor, camera
-calibration, capture orientation, rotation, crop, or transcode lineage.
-Canonical status identifies the active representative of a Standard record; it
-does not assert that the encoded file is an unaltered or full-duration capture
-master. A shortened derivative may be canonical when it is the verified active
-representative and its limitation is disclosed in `notes`.
-`encoded_frame_rate`, including an encoded value of 30 fps, must not be
-interpreted as the real capture interval or drawing-time sampling rate.
-A matching ordinary SHA-256 establishes byte identity with the specific
-comparison file. Direct-device or unchanged-source provenance additionally
-requires evidence about that comparison file's origin. File size and container
-or stream metadata may corroborate provenance, but they cannot independently
-prove drawing membership, direct-device origin, or an untouched source. File
-size alone also does not establish why one file is smaller than another; a
-causal explanation requires supporting comparisons.
-### Historical provenance fields
-| Field | Type | Meaning |
-| --- | --- | --- |
-| `original_filename` | string or blank | Independently documented historical filename. |
-| `provenance_status` | string or blank | Current relationship status. |
-| `legacy_provenance_path` | string or blank | Verified historical repository path. |
-| `legacy_related_references` | string or blank | Related candidates, aliases, or rejected references. |
-| `evidence_method` | string or blank | Primary evidence used for the status. |
-| `provenance_reference_duration_delta_seconds` | float or blank | Duration delta to a referenced historical file. |
-| `notes` | string or blank | Human-readable evidence or qualification. |
-Blank provenance values mean that the field is not documented or not
-applicable in the active indexes. A blank must not be promoted to a verified
-relationship.
-Observed provenance statuses include:
-- `legacy_match_verified`
-- `legacy_aliases_verified`
-- `legacy_duplicate_removed`
-- `legacy_candidates_rejected`
-- `direct_root_upload`
-- `tier_promotion_verified`
-Observed evidence methods include:
-- `blob_identity`
-- `visual_continuity`
-- `visual_comparison`
-- `commit_history`
-- `manual_confirmation`
-- `blob_identity_and_physical_date`
-New enumerated values must be documented here before use.
-## Archive payload checksum manifest
-The root `SHA256SUMS` file is the portable fixity index for published archival
-payloads. It contains exactly one line for every repository path ending in
-`.mp4` or `.zip`, sorted by repository-relative POSIX path:
-```text
-<64-character lowercase SHA-256><two spaces><repository-relative path>
-```
-Each digest is the ordinary byte-level SHA-256 of the represented file. It is
-not encryption and is distinct from the `hub_xet_hash` Xet storage identity.
-The LFS-compatible SHA-256 exposed by a pinned Hugging Face blob inventory is
-the same portable content digest recorded in this manifest.
-Documentation, scripts, CSV indexes, pointer files, and `SHA256SUMS` itself are
-outside this payload manifest. Their versions remain bound by the repository
-commit. Adding, deleting, renaming, or replacing an MP4 or ZIP requires updating
-`SHA256SUMS` in the same review and validating it against the exact candidate
-revision.
-## Fragment-sequence relationship sidecar
-Each revision-pinned packet
-`Standard_Time_Lapses/Provenance/YYYY-MM-DD[.HHMM]/fragment_sequence_relationships.csv`
-is a derived relationship table with one row per reviewed media object in that
-ingestion. `YYYY-MM-DD` is used when one packet represents that calendar date;
-an added `.HHMM` finish-label suffix disambiguates a later packet on the same
-date without rewriting the earlier receipt. The suffix is a packet identifier,
-not an individual recording end time. A bare-date packet may contain canonical
-fragments and noncanonical derivatives under the existing placement rule; a
-timestamp-qualified packet is currently restricted to canonical fragments.
-This relationship table does not add sequence fields to either active video index.
-| Field | Type | Meaning |
-| --- | --- | --- |
-| `fragment_set_label` | string | Descriptive canonical filename stem shared by one reviewed sequence; not a global `work_id`. |
-| `object_role` | enum | `canonical_fragment` or `noncanonical_inshot_derivative`. |
-| `record_id` | string or blank | Persistent canonical identity; required for canonical fragments and blank for noncanonical derivatives. |
-| `file_path` | string | Unique repository-relative MP4 path. |
-| `sha256` | string | Ordinary byte-level SHA-256 of the represented object. |
-| `hub_xet_hash` | string | Hugging Face Xet identity at the pinned revision; distinct from ordinary SHA-256. |
-| `size_bytes` | integer | Exact object size in bytes. |
-| `sequence_position` | integer or blank | Contiguous one-based file position for a canonical fragment. |
-| `sequence_count` | integer or blank | Total canonical fragment count for the sequence. |
-| `sequence_completeness` | enum | Owner-reviewed completeness of the retained canonical source-file sequence; `owner_confirmed_complete` does not assert gapless capture, session completeness, or physical-work completeness. Derivative coverage remains separate. |
-| `source_record_ids` | semicolon-delimited string or blank | Ordered canonical source IDs for a derivative; blank on canonical rows. |
-| `derivative_coverage` | enum or blank | `complete_sequence` or `partial_sequence` for a derivative; blank on canonical rows. |
-| `evidence_state` | semicolon-delimited string | Only evidence checks actually completed for the object and relationship. |
-| `source_hub_revision` | string | Immutable 40-hex Hub commit that permits the canonical identity and path join to be checked. |
-| `notes` | string or blank | Literal qualifications that do not belong in the categorical fields. |
-Within each fragment set, canonical positions must be unique and contiguous
-from one through `sequence_count`. A `complete_sequence` derivative must list
-the full ordered canonical source sequence, and every derivative source must
-belong to the same fragment set. Derivative rows and payloads remain outside
-the canonical metadata and Viewer configurations even when stored in the
-repository.
-For canonical fragments governed by `standard-2026-08-27`, every member of a
-`fragment_set_label` must have the same metadata `date` and `time`. Those values
-document the finish of the complete grouping/physical work, not the individual
-end time of each fragment. `sequence_position` and its matching `RecNN` suffix
-carry chronological position.
-Canonical-fragment evidence requires `owner-confirmed`, `checksum-computed`,
-`media-probed`, `demux-verified`, and `sampled-decode-verified`.
-`boundary-frame-verified` is included only when that separate visual check was
-actually completed. Noncanonical-derivative evidence requires `sampled-frame-verified`,
-`duration-arithmetic-verified`, `checksum-computed`, `media-probed`,
-`demux-verified`, and `sampled-decode-verified`. Derivative
-`sequence_position` is blank, `sequence_count` equals its canonical set size,
-and any partial `source_record_ids` retain canonical sequence order.
-`source_hub_revision` must be copied from an actual immutable remote commit; a
-moving branch, placeholder, or inferred value is invalid. Remote inventory,
-Xet-identity, and relationship-join validation must be completed against that
-commit. The captured inventory also binds the exact bytes of that revision's
-`metadata.csv`, so a local join cannot substitute metadata from a different
-commit. The pinned receipt remains historical evidence and is not rewritten
-when the active metadata later records a protocol boundary.
-The canonical-fragment `record_id` values in the twenty-eight packet sidecars
-spanning twenty-six calendar dates (`2026-08-27`, `2026-08-29`, `2026-08-30`,
-`2026-08-31`, `2026-09-01`, `2026-09-02`, `2026-09-14`, `2026-09-15`,
-`2026-09-16`, the additional `2026-09-16.2043` packet, `2026-09-17`, and
-`2026-09-18`, `2026-09-19`, `2026-09-20`, `2026-09-21`, `2026-09-22`,
-`2026-09-23`, `2026-09-24`, `2026-09-25`, `2026-09-26`, the additional
-`2026-09-26.1946` packet, `2026-09-27`, `2026-09-30`, `2026-10-01`, and `2026-10-02`, `2026-10-03`, `2026-10-04`, and `2026-10-05`) are the current membership authority for
-`standard-2026-08-27`. Noncanonical derivative rows do not receive a canonical
-protocol value. A dated sidecar does not introduce `work_id`, change any
-`record_id`, or backfill unrelated historical rows. Consequently, an active
-`RecNN` row absent from these dated sidecars retains
-`standard-2025-07-13`; absence does not establish that no relationship exists.
-## Active correction records
-Focused corrections that change an active canonical path or documented finish
-label without changing the represented media object are stored under
-`Standard_Time_Lapses/Provenance/Corrections/`. They are separate from dated,
-revision-pinned relationship packets and release receipts, whose bytes and
-historical claims remain immutable.
-The September 1 correction ledger is
-`Corrections/2026-09-01.1413-finish-time/finish_time_correction.csv`:
-| Field | Type | Meaning |
-| --- | --- | --- |
-| `record_id` | string | Persistent canonical identity; the active and historical rows must describe the same object. |
-| `sequence_position` | integer | Original one-based position, retained unchanged. |
-| `old_file_path` | string | Exact path in the pinned historical packets and metadata receipt. |
-| `new_file_path` | string | Exact corrected path in the active metadata and payload manifest. |
-| `old_finish_date` | `YYYY-MM-DD` | Historical documented finish date. |
-| `old_finish_time` | `HH:MM` | Historical documented finish time. |
-| `new_finish_date` | `YYYY-MM-DD` | Corrected active documented finish date. |
-| `new_finish_time` | `HH:MM` | Corrected active documented finish time. |
-| `sha256` | string | Unchanged ordinary byte-level media SHA-256. |
-| `hub_xet_hash` | string | Unchanged Hub Xet identity. |
-| `size_bytes` | integer | Unchanged media byte size. |
-| `acquisition_protocol_version` | enum | Unchanged protocol assignment. |
-Each row must map one unique `record_id`, old path, and new path. Validation
-must require an exact active join on the new path/date/time and unchanged
-identity fields; it must also require an exact historical join on the old
-path/date/time at the pinned source revision. A correction ledger cannot add or
-remove records, reorder a sequence, replace media, or rewrite historical
-evidence.
-When a fixed historical metadata receipt overlaps a later declared change,
-receipt validation reconstructs the historical bytes by applying only the
-exact documented inverses in memory. For timing v1 these are the 12-row
-September 1 finish-time correction and the separate PR #67 provenance
-classification backfill: exactly 12 named `record_id` values and exactly the
-`provenance_status`, `evidence_method`, and `notes` fields populated by commit
-`57a465ea69a4a6444e668f3267095a1b9fdfff6b` (merged as
-`6758d92f07f233166a1fa8a5c04e46c6cd96382e`). The validator first requires
-each active backfill value exactly, then reverses only those 36 cells in its
-temporary copy. The reconstructed fixed rows must match the original receipt
-exactly, and any later current rows must remain append-only. This is narrower
-than accepting general edits to the historical prefix: an undeclared
-difference, a changed identity field, or a mapping that does not round-trip is
-an error.
-## Recording-timing sidecar
-`Standard_Time_Lapses/Provenance/Recording_Timing/recording_timing.csv` is an
-exploratory derived table for the 186 canonical fragments in the six fixed
-timing-v1 source sidecars dated `2026-08-27` through `2026-09-02`. It joins to the canonical Standard index on `record_id`
-and declares the exact source Hub revision. It is outside both active video
-indexes and does not add or replace canonical metadata fields.
-Timing v1's source revision, packet bytes, and 186-row membership are fixed.
-Current metadata compatibility is established by exact inverse reconstruction
-of the 12 documented September 1 path/time corrections plus the exact
-12-record, 36-cell PR #67 provenance backfill to the historical 549-row
-receipt after excluding the named timing-display and `work_id` projections,
-followed by append-only rows in the original 29-field schema.
-Those appended rows must pass the canonical metadata validator and remain
-outside timing coverage until a separately reviewed, versioned timing layer
-includes them.
-The sidecar keeps three timing concepts separate:
-- `device_filename_start_local` is present only when a retained
-  `VID_YYYYMMDD_HHMMSS.mp4` source name supplies a whole-second timestamp
-  interpreted as recording start.
-- `format_creation_time_utc` and `video_creation_time_utc` preserve literal MP4
-  tags treated as the best-supported file-finalization boundary, not a
-  universally guaranteed stop time.
-- `capture_start_time_utc` and `recording_elapsed_seconds` are direct for 22
-  rows and profile-specific model estimates for 164 rows, as declared by
-  `timing_basis`, `calibration_profile_status`, and `timing_model_id`.
-Canonical `date` and `time` remain documented group-finish labels. They are not
-overwritten by sidecar starts or MP4 boundaries. The model has no independent
-holdout; its allowances are operational review thresholds, not proven error
-bounds. The sidecar does not establish drawing-active time, continuous work,
-source-frame timestamps, pen contact, or physical motion. Its complete field
-and auxiliary-file contracts are defined in
-[`Recording_Timing/SCHEMA.md`](Standard_Time_Lapses/Provenance/Recording_Timing/SCHEMA.md).
-## Physical-reference sidecar
-`Physical_References/endpoint_references.csv` is an optional derived
-relationship table joined to the canonical Standard index on `record_id`. It
-must declare the exact `source_hub_revision`; it does not determine canonical
-video status or duplicate editable video metadata.
-Each row has a stable `reference_id`, reviewed photograph filename and role,
-explicit surface and temporal scope, evidence state, hash state, and machine-use
-caution. Blank values mean unknown or not documented. Reference-media
-availability must be explicit, and blank proposed-path fields must not be
-interpreted as public media locations.
-`longitudinal_continuation` identifies a later or otherwise changed physical
-state and is not an exact endpoint target. Two-sided references preserve
-side-specific evidence and do not imply one shared visual target. Validators
-must require unique identities, canonical filename agreement at the pinned
-revision, allowed relationship values, explicit owner-confirmation state, and
-valid SHA-256 syntax whenever a hash is present. The complete contract is in
-[`Physical_References/SCHEMA.md`](Physical_References/SCHEMA.md).
-## Missing-Value Rules
-- Blank means unknown, not documented, or not applicable according to the field
-  description.
-- Blank numeric values must remain blank; they are not zero.
-- Blank booleans must remain blank; they are not `false`.
-- Uncertain dates and times must not be reconstructed from unsupported clues.
-- Pre-Standard descriptive/provenance metadata may be less complete, but every
-  published record must carry a collection-tier distinction, stable ID, exact
-  content identity, and complete core encoded-video metadata.
-## Required Validation
-Before publication, the validator must confirm:
-- unique `record_id`, `file_name`, and nonblank `hub_xet_hash`
-- canonical paths remain directly under `Standard_Time_Lapses/`
-- one consistent semicolon convention for multi-value fields
-- valid dates, times, and dimensions
-- complete core encoded-video metadata for every active Standard and
-  Pre-Standard row
-- orientation agrees with stored width and height
-- canonical metadata paths match the canonical Viewer records
-- every active standard and pre-standard metadata path resolves to a published
-  repository file
-- repository media absent from an active index is reported explicitly rather
-  than silently treated as released data
-- material lacking the necessary ownership or provenance clearance is absent
-  from both active indexes
-- fragment-sequence positions are contiguous and complete-sequence derivative
-  source lists exactly match their own ordered fragment set
-- `standard-2026-08-27` appears on exactly the canonical-fragment IDs in the
-  supplied dated sidecars, every other Standard row remains
-  `standard-2025-07-13`, and each prospective group shares one finish timestamp
-- noncanonical derivative payloads remain absent from both active indexes and
-  Viewer configurations
-- relationship paths, Xet identities, sizes, and canonical `record_id` joins
-  resolve at the exact declared `source_hub_revision`
-- recording-timing identities, direct timestamp arithmetic, calibration
-  formulas, sequence chronology, retained historical conflicts, correction
-  round-trips, unchanged media identities, and bundle hashes resolve at the
-  exact declared source revision and current active index
-- `SHA256SUMS` exactly covers every published MP4 and ZIP at the reviewed
-  revision, with no missing, extra, duplicate, or mismatched entries
-Run:
-```bash
-python tools/validate_metadata.py metadata.csv \
-  --pre-standard-metadata Pre_Standard_Time_Lapses/metadata.csv \
-  --fragment-relationships-root Standard_Time_Lapses/Provenance
-python tools/validate_sha256sums.py SHA256SUMS \
-  --repo-json repo_api_blobs.json \
-  --revision "<40-hex-commit>"
-python tools/validate_recording_timing.py --root .
-```

+# Schema
+This document defines canonical filenames, the active video-object indexes,
+and their relationship and timing sidecars. Filenames are human-readable
+descriptions; `record_id` is the persistent identity of one published video
+object. Archive-wide row counts are counts of indexed video files.
+## Canonical Filename Format
+`YYYY-MM-DD.HHMM.Tool.Medium.Support.Dimensions[.RecNN].ext`
+Examples:
+```text
+2025-07-15.0448.FountainPen.Ink.Paper.9x12.mp4
+2026-06-29.UnknownTime.FountainPen.Ink.IllustrationBoard.30x40.Rec11.mp4
+2026-06-29.UnknownTime.FountainPen.Ink.IllustrationBoard.30x40.Rec12.mp4
+```
+### Filename fields
+- `YYYY-MM-DD` — documented finish date.
+- `HHMM` — documented finish time in 24-hour local civil time.
+- `UnknownTime` — used when no reliable finish time is documented.
+- `Tool` — drawing instrument, such as `Pencil`, `FountainPen`, `BallpointPen`,
+  or `Brush`.
+- `Medium` — applied material, such as `Graphite` or `Ink`.
+- `Support` — drawing surface, such as `Paper`, `IllustrationBoard`, or
+  `WatercolorPaper`.
+- `Dimensions` — support dimensions in inches; order preserves support
+  orientation, such as `9x12` or `12x9`.
+- `RecNN` — optional file-order identifier used when documented order must be
+  preserved or otherwise identical filenames require distinction.
+  Numbering begins with `Rec01`. Historical numbering may continue across
+  finish-date labels; the prospective protocol has the narrower timestamp rule
+  defined below.
+- `ext` — file extension.
+### Filename rules
+- Fields are separated by periods.
+- Every tool listed in an active Standard metadata row must appear as one
+  exact, case-sensitive filename token. Multiword tools remain one token, such
+  as `BallpointPen` or `FountainPen`.
+- Across both active tiers, every exact filename token in the documented tool
+  vocabulary (`Pencil`, `FountainPen`, `BallpointPen`, and `Brush`) must appear
+  in that row's metadata `tool` list. Historical Pre-Standard metadata may
+  retain additional independently documented tools that are absent from a
+  filename; a standalone historical `Pen` token is not normalized without
+  separate documentation.
+- Style and aesthetic classifications are excluded.
+- `TL` is excluded because time-lapse is implicit in the primary collection.
+- `0000` is reserved for a documented midnight finish time.
+- `RecNN` specifies file order within a documented group only. It does not by
+  itself establish a separate drawing, recording or session boundary, or
+  completeness, and it does not reset merely because a historical documented
+  finish-date label changes. Under
+  `standard-2026-08-27`, all members of one documented prospective grouping
+  instead share the grouping timestamp and `RecNN` records chronological
+  file position within that group.
+- `UnknownTime` records uncertainty; it is not replaced by an inferred time.
+- Paper-surface variants such as `HotPress` and `ColdPress` may remain encoded
+  in filenames until a dedicated metadata field is added.
+- Historical relationships must not be inferred from filename similarity
+  alone.
+- Revision-pinned historical sidecars retain the path recorded at their source
+  revision; an active filename correction does not rewrite that evidence.
+Tool and medium tokens in a filename describe documented materials, but older
+filenames do not always encode an unambiguous one-to-one pairing. Metadata must
+therefore preserve tool and medium sets without manufacturing positional pairs.
+## Canonical Metadata Contract
+`metadata.csv` indexes the validated canonical subset of MP4 files stored
+directly in `Standard_Time_Lapses/`. Direct placement alone does not make a file
+canonical: every released canonical file must have exactly one metadata row,
+complete required technical fields, and the necessary rights and provenance
+clearance. Before publication, it must cover every MP4 stored directly in
+`Standard_Time_Lapses/` at the exact reviewed revision.
+Each metadata row and its `record_id` identify one published MP4 file. They do
+not assert one drawing, one session, or one complete process. Documented
+relationships among multiple video objects remain in sidecars and do not merge
+or replace their `record_id` values.
+Historical MP4 objects may represent one source recording or may combine
+multiple recordings, including through InShot. That possibility is
+record-specific; it does not mean every historical file is compiled. For the
+documented `standard-2026-08-27` cohort, each canonical MP4 and metadata row
+represents one unchanged source recording. Cohort membership is established by
+the metadata protocol value and `record_id` membership in the revision-pinned
+relationship packets, not by filename dates or a `RecNN` suffix alone. See the
+prospective [`2026-08-27` packet](Standard_Time_Lapses/Provenance/2026-08-27/README.md)
+and the separate
+[June 18–30 historical relationship](Standard_Time_Lapses/Provenance/Historical_Relationships/2026-06-18--2026-06-30-fountain-pen-30x40.md).
+The canonical index does not make provenance, unresolved, pre-standard, or
+legacy files canonical.
+### Identity and collection fields
+| Field | Type | Meaning |
+| --- | --- | --- |
+| `file_name` | string | Repository-relative canonical video path. |
+| `record_id` | string | Persistent identity in the form `tla-` plus 16 lowercase hexadecimal characters. |
+| `work_id` | nullable string | Confirmed drawing identity, `tlaw-` plus 16 lowercase hexadecimal characters, copied from the authoritative work registry. Blank means unresolved. |
+| `collection_tier` | enum | `standard` for this index. |
+| `is_canonical` | boolean | `true` for every row in this index. |
+| `acquisition_protocol_version` | enum | Governing Standard workflow: `standard-2025-07-13` for the earlier Standard workflow or `standard-2026-08-27` for canonical fragments reviewed into the prospective fragment-sequence workflow. Assignment is based on documented sidecar membership, not finish date or a `RecNN` suffix alone. |
+The initial `record_id` values were generated from a namespaced SHA-256 of the
+published repository path. Once assigned, the stored value is authoritative and
+must be retained across filename corrections or tier changes. New tooling must
+assign an ID only when a record does not already have one. When a genuinely new
+record is promoted from a noncanonical holding, its ID is seeded from that
+object's prior archival path; it must not inherit an ID already attached to a
+different object at the corrected destination path.
+### Pre-Standard collection contract
+`Pre_Standard_Time_Lapses/metadata.csv` indexes the direct MP4 files in that
+folder. Each row uses `collection_tier = pre_standard`, `is_canonical = false`,
+and a blank `acquisition_protocol_version`. Pre-Standard status records a
+different acquisition/documentation era; it does not remove the requirements
+for stable identity, exact content hash and size, or encoded-stream metadata.
+Dates, times, tools, media, support, and historical provenance may remain blank
+when the surviving evidence is insufficient. Those blanks are historical
+uncertainty, not failed technical ingestion.
+### Confirmed drawing identity projection
+Both active indexes include `work_id` immediately after `record_id`. The Viewer
+declares it as `large_string`, with an empty CSV cell represented as null.
+The 35-field schema retains every original field and the five timing-display
+fields. `Work_Identity/works.csv` and `Work_Identity/work_record_links.csv` remain
+the authoritative relationship source; the index column is a generated copy.
+Join by exact `record_id` within the same repository revision.
+Only positive, owner-confirmed links to active work IDs are projected. The
+current 1,106 linked recordings share 425 active work IDs; 128 Pre-Standard recordings remain blank.
+No IDs are minted, reassigned or inferred from dates, filenames, sequence
+numbers, shared materials or appearance by this projection. A work ID does not
+assert recording-set completeness, gapless capture, session boundaries or
+physical-reference relationships. Different work IDs are not independently
+verified negative pairs. See the [registry contract](Work_Identity/README.md)
+for correction, retirement and supersession rules.
+After an authorized registry update, rebuild the display column and validate:
+```sh
+python tools/project_work_ids.py --root . --write
+python tools/project_work_ids.py --root .
+python tools/project_recording_times.py --root .
+```
+The validator rejects duplicate or ambiguous links, inactive/missing work IDs,
+unconfirmed links, unpinned source revisions, absent indexed records, malformed
+rows and drift from the registry. It checks same-revision confirmation-file
+checksums and the format of historical revision-pinned evidence references;
+it does not repeat the historical source-media audit or refetch those URLs.
+Both indexes and the registry are checked before either index is written.
+Blank work IDs are valid and do not block ingestion. Historical source indexes
+without this display column retain their original schema at their pinned revisions.
+### Explicit timing display fields
+The active indexes retain all original 29 columns and add the following five
+string columns. These are a generated display projection, not a new timing
+evidence source. Together with `work_id`, the Dataset Viewer declares 35 columns for both tiers.
+| Field | Meaning |
+| --- | --- |
+| `recording_start_local` | Reviewed device-filename start copied from the validated current timing view, including its explicit UTC offset. Blank if unknown or outside reviewed coverage. |
+| `recording_start_basis` | `device_filename_interpreted_in_declared_timezone` or `unknown` for a covered record. The existing `not_in_timing_snapshot` value is retained for records outside all registered timing evidence at the selected repository revision. |
+| `documented_finish_date` | Exact alias of the existing `date` value. |
+| `documented_finish_time` | Exact alias of the existing `time` value. It is not a per-record stop time for prospective groups. |
+| `finish_label_scope` | `complete_drawing_or_group` for `standard-2026-08-27`; `historical_scope_unspecified` for other surviving finish labels; `unknown` if both finish labels are blank. |
+Recording starts are joined by `record_id` and checked against current paths,
+byte lengths, SHA-256 and Xet identities. The
+[Current recording-time view](Standard_Time_Lapses/Provenance/Recording_Timing_Current/README.md)
+combines the unchanged Timing v2 snapshot with explicitly registered,
+revision-bound ingest observations. It currently covers 742 recordings, with
+514 filename-derived starts and 228 unknown starts; 492 indexed records remain
+outside this coverage. New evidence does not rewrite frozen timing packets.
+MP4 creation tags retain their creation/finalization meaning and are not
+renamed as recording stops. Source filenames and media paths are unchanged.
+The original 339 Timing v2 rows and their projected timing values are retained.
+After an index update, rebuild the five display fields and then validate:
+```sh
+python tools/project_recording_times.py --root . --write
+python tools/project_recording_times.py --root .
+python tools/validate_recording_timing_v2.py --root .
+```
+The projection validator rejects missing columns, changed evidence values,
+duplicate identities, mismatched source bindings and missing cohort records.
+The existing metadata validator continues to understand the original fields
+for compatibility with revision-pinned historical inputs.
+### Session and material fields
+| Field | Type | Meaning |
+| --- | --- | --- |
+| `date` | `YYYY-MM-DD` | Documented finish date. For `standard-2026-08-27`, this is the finish date of the complete grouping/physical work shared by every member; earlier records retain their historically documented label. |
+| `time` | `HH:MM` or blank | Documented finish time; blank means unknown. For `standard-2026-08-27`, this is the finish time of the complete grouping/physical work shared by every member. |
+| `tool` | semicolon list | Ordered unique set of documented tools. |
+| `medium` | semicolon list | Ordered unique set of documented media. |
+| `support` | string | Physical drawing surface. |
+| `dimensions` | string | Support dimensions in inches, such as `9x12`. |
+Multiple values use semicolons:
+```text
+tool = Pencil;FountainPen
+medium = Graphite;Ink
+```
+`tool` and `medium` are independent observed-value lists. Equal lengths and
+positional correspondence are **not guaranteed**. Repeated media are collapsed
+to one value. A future paired representation should use a structured field and
+must be based on documented evidence rather than token position.
+`date` and `time` preserve the best documented finish labels associated with a
+row. For the earlier Standard workflow, the original 29-field schema does not
+distinguish whether the source documentation scoped a label to an individual
+recording, a recording session, a reviewed fragment sequence, or the physical
+work. Equal values, differing values, and `RecNN` suffixes do not resolve that
+historical scope. For `standard-2026-08-27`, the scope is explicit: every
+`RecNN` member in one prospective grouping carries the documented finish date
+and finish time of the complete grouping/physical work. The timestamp is not
+the individual end time of each fragment; `RecNN` records chronological
+position within the group.
+### Encoded-video fields
+| Field | Type | Meaning |
+| --- | --- | --- |
+| `duration_seconds` | float | Encoded playback duration. |
+| `width_px` | integer | Stored encoded width. |
+| `height_px` | integer | Stored encoded height. |
+| `orientation` | enum | Stored pixel orientation derived from width and height. |
+| `encoded_frame_rate` | float | Encoded playback frame rate as reported by the container; fractional values are allowed. |
+| `video_codec` | string | Encoded video codec. |
+| `pixel_format` | string | Encoded pixel format. |
+| `container` | string | Media container. |
+| `has_audio` | boolean | Whether the stored file contains an audio stream. |
+| `hub_xet_hash` | string | Lowercase 64-character Xet file ID from the pinned Hub resolve response's `X-Xet-Hash` header. This is distinct from the LFS-compatible SHA-256. |
+| `size_bytes` | integer | Stored file size in bytes. |
+These fields describe the published encoded file. They do not establish the
+original frame-capture interval, real drawing duration, speed-up factor, camera
+calibration, capture orientation, rotation, crop, or transcode lineage.
+Canonical status identifies the active representative of a Standard record; it
+does not assert that the encoded file is an unaltered or full-duration capture
+master. A shortened derivative may be canonical when it is the verified active
+representative and its limitation is disclosed in `notes`.
+`encoded_frame_rate`, including an encoded value of 30 fps, must not be
+interpreted as the real capture interval or drawing-time sampling rate.
+A matching ordinary SHA-256 establishes byte identity with the specific
+comparison file. Direct-device or unchanged-source provenance additionally
+requires evidence about that comparison file's origin. File size and container
+or stream metadata may corroborate provenance, but they cannot independently
+prove drawing membership, direct-device origin, or an untouched source. File
+size alone also does not establish why one file is smaller than another; a
+causal explanation requires supporting comparisons.
+### Historical provenance fields
+| Field | Type | Meaning |
+| --- | --- | --- |
+| `original_filename` | string or blank | Independently documented historical filename. |
+| `provenance_status` | string or blank | Current relationship status. |
+| `legacy_provenance_path` | string or blank | Verified historical repository path. |
+| `legacy_related_references` | string or blank | Related candidates, aliases, or rejected references. |
+| `evidence_method` | string or blank | Primary evidence used for the status. |
+| `provenance_reference_duration_delta_seconds` | float or blank | Duration delta to a referenced historical file. |
+| `notes` | string or blank | Human-readable evidence or qualification. |
+Blank provenance values mean that the field is not documented or not
+applicable in the active indexes. A blank must not be promoted to a verified
+relationship.
+Observed provenance statuses include:
+- `legacy_match_verified`
+- `legacy_aliases_verified`
+- `legacy_duplicate_removed`
+- `legacy_candidates_rejected`
+- `direct_root_upload`
+- `tier_promotion_verified`
+Observed evidence methods include:
+- `blob_identity`
+- `visual_continuity`
+- `visual_comparison`
+- `commit_history`
+- `manual_confirmation`
+- `blob_identity_and_physical_date`
+New enumerated values must be documented here before use.
+## Archive payload checksum manifest
+The root `SHA256SUMS` file is the portable fixity index for published archival
+payloads. It contains exactly one line for every repository path ending in
+`.mp4` or `.zip`, sorted by repository-relative POSIX path:
+```text
+<64-character lowercase SHA-256><two spaces><repository-relative path>
+```
+Each digest is the ordinary byte-level SHA-256 of the represented file. It is
+not encryption and is distinct from the `hub_xet_hash` Xet storage identity.
+The LFS-compatible SHA-256 exposed by a pinned Hugging Face blob inventory is
+the same portable content digest recorded in this manifest.
+Documentation, scripts, CSV indexes, pointer files, and `SHA256SUMS` itself are
+outside this payload manifest. Their versions remain bound by the repository
+commit. Adding, deleting, renaming, or replacing an MP4 or ZIP requires updating
+`SHA256SUMS` in the same review and validating it against the exact candidate
+revision.
+## Fragment-sequence relationship sidecar
+Each revision-pinned packet
+`Standard_Time_Lapses/Provenance/YYYY-MM-DD[.HHMM]/fragment_sequence_relationships.csv`
+is a derived relationship table with one row per reviewed media object in that
+ingestion. `YYYY-MM-DD` is used when one packet represents that calendar date;
+an added `.HHMM` finish-label suffix disambiguates a later packet on the same
+date without rewriting the earlier receipt. The suffix is a packet identifier,
+not an individual recording end time. A bare-date packet may contain canonical
+fragments and noncanonical derivatives under the existing placement rule; a
+timestamp-qualified packet is currently restricted to canonical fragments.
+This relationship table does not add sequence fields to either active video index.
+| Field | Type | Meaning |
+| --- | --- | --- |
+| `fragment_set_label` | string | Descriptive canonical filename stem shared by one reviewed sequence; not a global `work_id`. |
+| `object_role` | enum | `canonical_fragment` or `noncanonical_inshot_derivative`. |
+| `record_id` | string or blank | Persistent canonical identity; required for canonical fragments and blank for noncanonical derivatives. |
+| `file_path` | string | Unique repository-relative MP4 path. |
+| `sha256` | string | Ordinary byte-level SHA-256 of the represented object. |
+| `hub_xet_hash` | string | Hugging Face Xet identity at the pinned revision; distinct from ordinary SHA-256. |
+| `size_bytes` | integer | Exact object size in bytes. |
+| `sequence_position` | integer or blank | Contiguous one-based file position for a canonical fragment. |
+| `sequence_count` | integer or blank | Total canonical fragment count for the sequence. |
+| `sequence_completeness` | enum | Owner-reviewed completeness of the retained canonical source-file sequence; `owner_confirmed_complete` does not assert gapless capture, session completeness, or physical-work completeness. Derivative coverage remains separate. |
+| `source_record_ids` | semicolon-delimited string or blank | Ordered canonical source IDs for a derivative; blank on canonical rows. |
+| `derivative_coverage` | enum or blank | `complete_sequence` or `partial_sequence` for a derivative; blank on canonical rows. |
+| `evidence_state` | semicolon-delimited string | Only evidence checks actually completed for the object and relationship. |
+| `source_hub_revision` | string | Immutable 40-hex Hub commit that permits the canonical identity and path join to be checked. |
+| `notes` | string or blank | Literal qualifications that do not belong in the categorical fields. |
+Within each fragment set, canonical positions must be unique and contiguous
+from one through `sequence_count`. A `complete_sequence` derivative must list
+the full ordered canonical source sequence, and every derivative source must
+belong to the same fragment set. Derivative rows and payloads remain outside
+the canonical metadata and Viewer configurations even when stored in the
+repository.
+For canonical fragments governed by `standard-2026-08-27`, every member of a
+`fragment_set_label` must have the same metadata `date` and `time`. Those values
+document the finish of the complete grouping/physical work, not the individual
+end time of each fragment. `sequence_position` and its matching `RecNN` suffix
+carry chronological position.
+Canonical-fragment evidence requires `owner-confirmed`, `checksum-computed`,
+`media-probed`, `demux-verified`, and `sampled-decode-verified`.
+`boundary-frame-verified` is included only when that separate visual check was
+actually completed. Noncanonical-derivative evidence requires `sampled-frame-verified`,
+`duration-arithmetic-verified`, `checksum-computed`, `media-probed`,
+`demux-verified`, and `sampled-decode-verified`. Derivative
+`sequence_position` is blank, `sequence_count` equals its canonical set size,
+and any partial `source_record_ids` retain canonical sequence order.
+`source_hub_revision` must be copied from an actual immutable remote commit; a
+moving branch, placeholder, or inferred value is invalid. Remote inventory,
+Xet-identity, and relationship-join validation must be completed against that
+commit. The captured inventory also binds the exact bytes of that revision's
+`metadata.csv`, so a local join cannot substitute metadata from a different
+commit. The pinned receipt remains historical evidence and is not rewritten
+when the active metadata later records a protocol boundary.
+The canonical-fragment `record_id` values in the twenty-eight packet sidecars
+spanning twenty-six calendar dates (`2026-08-27`, `2026-08-29`, `2026-08-30`,
+`2026-08-31`, `2026-09-01`, `2026-09-02`, `2026-09-14`, `2026-09-15`,
+`2026-09-16`, the additional `2026-09-16.2043` packet, `2026-09-17`, and
+`2026-09-18`, `2026-09-19`, `2026-09-20`, `2026-09-21`, `2026-09-22`,
+`2026-09-23`, `2026-09-24`, `2026-09-25`, `2026-09-26`, the additional
+`2026-09-26.1946` packet, `2026-09-27`, `2026-09-30`, `2026-10-01`, and `2026-10-02`, `2026-10-03`, `2026-10-04`, and `2026-10-05`) are the current membership authority for
+`standard-2026-08-27`. Noncanonical derivative rows do not receive a canonical
+protocol value. A dated sidecar does not introduce `work_id`, change any
+`record_id`, or backfill unrelated historical rows. Consequently, an active
+`RecNN` row absent from these dated sidecars retains
+`standard-2025-07-13`; absence does not establish that no relationship exists.
+## Active correction records
+Focused corrections that change an active canonical path or documented finish
+label without changing the represented media object are stored under
+`Standard_Time_Lapses/Provenance/Corrections/`. They are separate from dated,
+revision-pinned relationship packets and release receipts, whose bytes and
+historical claims remain immutable.
+The September 1 correction ledger is
+`Corrections/2026-09-01.1413-finish-time/finish_time_correction.csv`:
+| Field | Type | Meaning |
+| --- | --- | --- |
+| `record_id` | string | Persistent canonical identity; the active and historical rows must describe the same object. |
+| `sequence_position` | integer | Original one-based position, retained unchanged. |
+| `old_file_path` | string | Exact path in the pinned historical packets and metadata receipt. |
+| `new_file_path` | string | Exact corrected path in the active metadata and payload manifest. |
+| `old_finish_date` | `YYYY-MM-DD` | Historical documented finish date. |
+| `old_finish_time` | `HH:MM` | Historical documented finish time. |
+| `new_finish_date` | `YYYY-MM-DD` | Corrected active documented finish date. |
+| `new_finish_time` | `HH:MM` | Corrected active documented finish time. |
+| `sha256` | string | Unchanged ordinary byte-level media SHA-256. |
+| `hub_xet_hash` | string | Unchanged Hub Xet identity. |
+| `size_bytes` | integer | Unchanged media byte size. |
+| `acquisition_protocol_version` | enum | Unchanged protocol assignment. |
+Each row must map one unique `record_id`, old path, and new path. Validation
+must require an exact active join on the new path/date/time and unchanged
+identity fields; it must also require an exact historical join on the old
+path/date/time at the pinned source revision. A correction ledger cannot add or
+remove records, reorder a sequence, replace media, or rewrite historical
+evidence.
+When a fixed historical metadata receipt overlaps a later declared change,
+receipt validation reconstructs the historical bytes by applying only the
+exact documented inverses in memory. For timing v1 these are the 12-row
+September 1 finish-time correction and the separate PR #67 provenance
+classification backfill: exactly 12 named `record_id` values and exactly the
+`provenance_status`, `evidence_method`, and `notes` fields populated by commit
+`57a465ea69a4a6444e668f3267095a1b9fdfff6b` (merged as
+`6758d92f07f233166a1fa8a5c04e46c6cd96382e`). The validator first requires
+each active backfill value exactly, then reverses only those 36 cells in its
+temporary copy. The reconstructed fixed rows must match the original receipt
+exactly, and any later current rows must remain append-only. This is narrower
+than accepting general edits to the historical prefix: an undeclared
+difference, a changed identity field, or a mapping that does not round-trip is
+an error.
+## Recording-timing sidecar
+`Standard_Time_Lapses/Provenance/Recording_Timing/recording_timing.csv` is an
+exploratory derived table for the 186 canonical fragments in the six fixed
+timing-v1 source sidecars dated `2026-08-27` through `2026-09-02`. It joins to the canonical Standard index on `record_id`
+and declares the exact source Hub revision. It is outside both active video
+indexes and does not add or replace canonical metadata fields.
+Timing v1's source revision, packet bytes, and 186-row membership are fixed.
+Current metadata compatibility is established by exact inverse reconstruction
+of the 12 documented September 1 path/time corrections plus the exact
+12-record, 36-cell PR #67 provenance backfill to the historical 549-row
+receipt after excluding the named timing-display and `work_id` projections,
+followed by append-only rows in the original 29-field schema.
+Those appended rows must pass the canonical metadata validator and remain
+outside timing coverage until a separately reviewed, versioned timing layer
+includes them.
+The sidecar keeps three timing concepts separate:
+- `device_filename_start_local` is present only when a retained
+  `VID_YYYYMMDD_HHMMSS.mp4` source name supplies a whole-second timestamp
+  interpreted as recording start.
+- `format_creation_time_utc` and `video_creation_time_utc` preserve literal MP4
+  tags treated as the best-supported file-finalization boundary, not a
+  universally guaranteed stop time.
+- `capture_start_time_utc` and `recording_elapsed_seconds` are direct for 22
+  rows and profile-specific model estimates for 164 rows, as declared by
+  `timing_basis`, `calibration_profile_status`, and `timing_model_id`.
+Canonical `date` and `time` remain documented group-finish labels. They are not
+overwritten by sidecar starts or MP4 boundaries. The model has no independent
+holdout; its allowances are operational review thresholds, not proven error
+bounds. The sidecar does not establish drawing-active time, continuous work,
+source-frame timestamps, pen contact, or physical motion. Its complete field
+and auxiliary-file contracts are defined in
+[`Recording_Timing/SCHEMA.md`](Standard_Time_Lapses/Provenance/Recording_Timing/SCHEMA.md).
+## Physical-reference sidecar
+`Physical_References/endpoint_references.csv` is an optional derived
+relationship table joined to the canonical Standard index on `record_id`. It
+must declare the exact `source_hub_revision`; it does not determine canonical
+video status or duplicate editable video metadata.
+Each row has a stable `reference_id`, reviewed photograph filename and role,
+explicit surface and temporal scope, evidence state, hash state, and machine-use
+caution. Blank values mean unknown or not documented. Reference-media
+availability must be explicit, and blank proposed-path fields must not be
+interpreted as public media locations.
+`longitudinal_continuation` identifies a later or otherwise changed physical
+state and is not an exact endpoint target. Two-sided references preserve
+side-specific evidence and do not imply one shared visual target. Validators
+must require unique identities, canonical filename agreement at the pinned
+revision, allowed relationship values, explicit owner-confirmation state, and
+valid SHA-256 syntax whenever a hash is present. The complete contract is in
+[`Physical_References/SCHEMA.md`](Physical_References/SCHEMA.md).
+## Prospective Geometric & Structural Schema Extension
+Optional prospective fields for discrete geometric and topological structure
+(`topological_family`, `terminal_node_count`, `terminal_edge_count`,
+`graph_invariants`, `sequence_formula_id`) are specified in
+[`GEOMETRIC_SCHEMA_EXTENSION.md`](GEOMETRIC_SCHEMA_EXTENSION.md).
+These fields associate verified physical drawings (`work_id`) and recording
+clips with discrete graph-theoretic invariants without modifying core video
+schema or existing row indices.
+## Missing-Value Rules
+- Blank means unknown, not documented, or not applicable according to the field
+  description.
+- Blank numeric values must remain blank; they are not zero.
+- Blank booleans must remain blank; they are not `false`.
+- Uncertain dates and times must not be reconstructed from unsupported clues.
+- Pre-Standard descriptive/provenance metadata may be less complete, but every
+  published record must carry a collection-tier distinction, stable ID, exact
+  content identity, and complete core encoded-video metadata.
+- For prospective geometric and structural annotations (see `GEOMETRIC_SCHEMA_EXTENSION.md`),
+  terminal node and edge counts for multi-clip groups are populated strictly on `Rec01`;
+  all intermediate and in-progress clips (`Rec02`..`RecNN`) remain strictly blank.
+## Required Validation
+Before publication, the validator must confirm:
+- unique `record_id`, `file_name`, and nonblank `hub_xet_hash`
+- canonical paths remain directly under `Standard_Time_Lapses/`
+- one consistent semicolon convention for multi-value fields
+- valid dates, times, and dimensions
+- complete core encoded-video metadata for every active Standard and
+  Pre-Standard row
+- orientation agrees with stored width and height
+- canonical metadata paths match the canonical Viewer records
+- every active standard and pre-standard metadata path resolves to a published
+  repository file
+- repository media absent from an active index is reported explicitly rather
+  than silently treated as released data
+- material lacking the necessary ownership or provenance clearance is absent
+  from both active indexes
+- fragment-sequence positions are contiguous and complete-sequence derivative
+  source lists exactly match their own ordered fragment set
+- `standard-2026-08-27` appears on exactly the canonical-fragment IDs in the
+  supplied dated sidecars, every other Standard row remains
+  `standard-2025-07-13`, and each prospective group shares one finish timestamp
+- noncanonical derivative payloads remain absent from both active indexes and
+  Viewer configurations
+- relationship paths, Xet identities, sizes, and canonical `record_id` joins
+  resolve at the exact declared `source_hub_revision`
+- recording-timing identities, direct timestamp arithmetic, calibration
+  formulas, sequence chronology, retained historical conflicts, correction
+  round-trips, unchanged media identities, and bundle hashes resolve at the
+  exact declared source revision and current active index
+- `SHA256SUMS` exactly covers every published MP4 and ZIP at the reviewed
+  revision, with no missing, extra, duplicate, or mismatched entries
+Run:
+```bash
+python tools/validate_metadata.py metadata.csv \
+  --pre-standard-metadata Pre_Standard_Time_Lapses/metadata.csv \
+  --fragment-relationships-root Standard_Time_Lapses/Provenance
+python tools/validate_sha256sums.py SHA256SUMS \
+  --repo-json repo_api_blobs.json \
+  --revision "<40-hex-commit>"
+python tools/validate_recording_timing.py --root .
+```