← All activity records

FACTUAL ACTIVITY RECORD · An activity record based on work that took place

Correcting Editorial output validation and task display

Aligning relative references, readable task names, one-based ordering, and failure diagnostics in one generation contract

Activity period:

Purpose

Putting local absolute paths or internal identifiers in the Editorial generation prompt risked preserving them in a conversation synchronized to another surface. Generated tasks also retained opaque display names, making the target date and Channel difficult to recognize in operational views.

On the output side, contribution and source ordering started at zero and did not map naturally to human-readable order. Identity, contribution, source, and review-result mismatches were also collapsed into one failure, obscuring which contract had been broken.

Implementation

The generation prompt now reads its input through the fixed relative reference ../control/input payload. Identity comes from that input JSON, and the instructions explicitly prohibit absolute paths, filesystem details, credentials, and access tokens in responses that may become public.

A task name in the form Editorial-YYYY-MM-DD-<channel-slug> is derived from the target date and Channel slug. Its length is bounded, and a metadata-only App Server connection sets only that name, separate from the generation process.

The task name is set before registration with the target. The metadata connection has a timeout, bounded standard output and error, a minimal environment, and explicit shutdown. If naming fails, the generation process is terminated so an incomplete registration is not retained.

The ordinal fields for contributions and sources are now one-based across the schema, prompt, shared contract, and fixture, and are limited to integers from 1 through 200.

Post-generation checks now return distinct failure codes for identity mismatch, contribution mismatch, source mismatch, and incomplete review. Diagnostics can therefore identify the exact output contract that failed.

The cutoff flow that keeps generation fresh while separating metadata naming and output validation

Rendering diagram…

What was confirmed

Verification results

Tests confirmed that the generation prompt uses the fixed relative reference and omits absolute paths and internal identifiers.

The metadata fixture confirmed that the deterministic task name is set through JSON-RPC and observed before target registration.

Ordering is one-based across the schema, prompt, shared contract, and fixture, with the same accepted range.

Output mismatches are distinguished at the identity, contribution, source, and review-result levels, and the changes were applied to the target.

Basis for completion

The work was complete when the boundary for potentially public prompts, task-naming order, one-based ordinals, and output-mismatch diagnostics agreed across code, contracts, fixtures, and tests and were applied to the target.