# Blank working records

Version 1.0. Prepared 6 October 2026. These files are example-free templates for your own work. Their preparation date is not a source date, review date or model-run date. The worksheet records your review; it does not check sources or approve publication.

## Start a working copy

1. Download `keyword-blank.csv` or `keyword-blank.json` for a candidate-query review. Use `content-claim-blank.csv` or `content-claim-blank.json` for individual statements and metadata.
2. The single row marked `TEMPLATE` is intentionally unfinished. It contains no candidate, claim, example, source, metric or reviewer. Change its record type to `WORKING` and give each row a unique, stable ID, such as your own sequence. Replace the template row rather than counting it as completed work.
3. Fill only what you know. Leave an unrecorded date or value empty. Unknown or unavailable data are not zero. Preserve unreviewed and unresolved items when sharing the record.
4. Keep source references and limitations with the file. `reviewed` or `confirmed` describes your recorded check; it is not approval from a platform, SEO Informatica or another reviewer.

Both formats use the same ordered fields. The JSON also carries schema version, field order, allowed status values, method and limitations. Its `rows` array contains the working records. The JSON's `prepared_date` is file metadata and must never be copied into an empty row date.

## Keyword fields: 22 columns

```text
record_type,row_id,candidate,observed_query,provenance,suggested_task,reviewed_task,existing_page,ownership_review_status,decision,rationale,evidence_reference,evidence_date,market,evidence_status,exact_query_check_status,review_status,next_action,reviewer,evidence_confirmation,metric_status,metric_context
```

| Field or group | What to retain |
| --- | --- |
| `record_type`, `row_id` | `TEMPLATE` or `WORKING`, and a stable unique row ID. These are identity fields. |
| `candidate`, `provenance` | The exact query phrase and how it entered the set. A generated suggestion is not observed demand. |
| `observed_query` | The exact query actually observed in the search evidence. For related-query evidence, name that different query explicitly. Leave this empty when no query was observed. |
| `suggested_task`, `reviewed_task` | Keep the model's suggestion separate from your own interpretation. Leave the suggestion empty if no model was involved. |
| `existing_page` | The existing page relevant to the reviewed task, if any. An empty field does not prove no page exists. |
| `ownership_review_status` | `not_checked`, `bounded_inventory_reviewed`, `broader_inventory_reviewed`, or `unresolved`. Distinguish review of the supplied inventory from a broader page review; retain the scope in the reference or rationale. It does not describe search evidence or final owner approval. |
| `decision`, `rationale` | Your provisional action and why. Allowed actions: `investigate`, `existing_page_review`, `group_with_another_candidate`, `propose_distinct_page`, `defer`. Empty means no decision recorded. |
| `evidence_reference`, `evidence_date`, `market` | Exact source URL or retained reference, its relevant date, and the market/language or other needed context. Label references as `Search:` or `Inventory:` when both are retained in this compact field. Use an ISO date when known; do not fabricate one. |
| `evidence_status` | Search-observation scope only: `exact_query_observed`, `related_query_observed`, or `not_checked`. Exact-query evidence must name the candidate in `observed_query`; related-query evidence must name the different query that was actually observed. An inventory check alone leaves this field `not_checked`. |
| `exact_query_check_status` | `not_checked` or `checked`. Record the actual observation in the evidence fields; checked does not mean demand exists. |
| `review_status` | `unreviewed`, `in_review`, or `reviewed`. This is a workflow state, not a quality score. |
| `next_action`, `reviewer` | The unresolved check and the person who actually performed the recorded review. Leave the reviewer empty until known. |
| `evidence_confirmation` | `unconfirmed` or `confirmed`. Confirmation means only: “I checked that this evidence belongs to this query and decision.” Changing the candidate, task, decision, observed query, search scope, source reference, source date, market, relevant owner or rationale clears confirmation for that row. |
| `metric_status`, `metric_context` | `not_collected`, `collected`, or `unavailable`. If relevant, keep metric name, value/unit, source, settings and date together in the context field. Label each item; do not enter an invented value. |

The compact static worksheet uses one optional metric-context column. A later interface may provide separate metric source, settings and date inputs as specified in the implementation brief, while preserving all those values in its exported record. This template does not supply any keyword metric.

Search evidence, page ownership and sign-off are separate. A candidate can have `evidence_status: not_checked` while its `ownership_review_status` is `bounded_inventory_reviewed`. Equally, observing a query does not prove the existing inventory was reviewed. Neither state means an owner or independent human reviewer approved the decision. Record the actual reviewer and pending approval separately from these checks.

## Claim fields: 15 columns

```text
record_type,row_id,statement,location,evidence_reference,source_date,review_date,evidence_status,qualification,editorial_action,revised_wording,reviewer,next_action,evidence_confirmation,revision_note
```

| Field or group | What to retain |
| --- | --- |
| `record_type`, `row_id` | `TEMPLATE` or `WORKING`, and a stable unique row ID. These are identity fields. |
| `statement`, `location` | The exact claim or metadata being reviewed. Locations: `body`, `title`, `description`, `structured_data`, `alt`, `other`; empty means not recorded. |
| `evidence_reference`, `source_date`, `review_date` | The supporting or contradicting source, its relevant date and when you reviewed it. Keep the two dates separate and leave unknown dates empty. |
| `evidence_status` | `unreviewed`, `supported_in_scope`, `needs_qualification`, `contradicted`, `unresolved`, or `not_applicable`. |
| `qualification` | Conditions, population, period or other scope needed to interpret the source accurately. |
| `editorial_action` | `keep`, `qualify`, `correct`, `remove`, or `investigate`. Empty means not decided. Action and evidence status are independent. |
| `revised_wording` | The proposed wording after review, if relevant. Do not overwrite the original statement with it. |
| `reviewer`, `next_action` | Who actually reviewed it and what remains to do. Neither must be guessed to finish the row. |
| `evidence_confirmation` | `unconfirmed` or `confirmed`, referring only to whether the source belongs to this statement and decision. A changed statement or source clears confirmation and returns the status to `unreviewed`. |
| `revision_note` | What changed and any previous values needed to understand the review history. |

A supported claim can be removed for irrelevance. An unresolved claim does not become verified because it was rewritten. These records do not produce a whole-draft approval, accuracy percentage, ranking prediction or compliance score.

## Original examples and later interactive exports

The blank files contain no original example columns because they contain no example. Do not add a second empty set of working columns to these templates.

The retained sample has a different job: preserve a fixed record of the exercise. Its original response and reviewed source fields remain immutable in the later interface. In particular, the keyword response's ID, candidate, likely reader task, provisional action, existing page and evidence check must survive unchanged alongside the later review. For content, preserve the original statement and location separately from revised wording and editorial decisions. Retain all dated source and review fields supplied with either sample.

The retained keyword sample also separates `observed_query`, `ownership_review_status`, `search_evidence_reference` and `ownership_evidence_reference`. Its fuller schema need not match the compact blank worksheet's column count. Do not collapse these distinctions when adding a reader's working copy.

For an interactive export that contains both the retained sample and a reader's changes, use the following explicit separation:

- JSON: stable `row_id`, an immutable `original` object containing the complete retained row, and a separate `working` object containing the reader's fields. Keep export scope (`all` or `visible`) and schema version at document level.
- CSV: prefix every retained-row column with `original_` and every reader field with `working_`, while retaining the stable row ID. Include only fields actually present in that export; do not add the prefixes to a static sample that has no reader edits. This is an export naming rule, not a claim that the interface has been implemented.

## CSV, JSON and spreadsheet safety

CSV is UTF-8 without a byte-order mark, with a header, conventional double-quoted cells and CRLF record endings. Internal quotes are doubled; embedded commas and newlines stay inside the quoted cell. Import it explicitly as comma-separated UTF-8 if your spreadsheet expects a different regional separator.

Spreadsheet applications can interpret text beginning with `=`, `+`, `-` or `@` as a formula, including after leading whitespace. A leading tab, carriage return or newline can also cause unsafe interpretation. For a generated CSV, prefix these values with a single apostrophe before conventional CSV quoting. The blank templates contain none of these payloads. Do not remove the escape merely to make a cell look cleaner. If exact machine-readable text is needed, use JSON, which retains the original characters as data.

Quoting alone does not neutralize spreadsheet formulas. Any future browser export must apply this safeguard to user input, preserve multiline content, and undergo actual download testing. The package checker tests the sanitizer logic with fixed fixtures and reads these files back; it does not test a browser or every spreadsheet application.

Keep customer information, private draft text and account details out of publicly shared evidence packs. These files do not upload or submit data themselves.