[Maya-Kid]

Follows #995. A row that an approved rule implies (0073) takes the moment it was computed as its attestation. Since #995, a fact with no time holds at every moment (0064 decision 5), and a row materialised from a statement follows that statement's attestation. A rule's row does neither.

Problem

• Insert. imply_in_tx writes a rule's row through insert_fact_on, which stores COALESCE(attested_at, now()). That fallback is meant for a fact a person writes.
  • A phrase rule passes its statement's attestation, so an unattested statement gives the processing moment.
  • A kind-word rule passes NULL, with the same result.
• Sync. sync_typed_attestation puts a typed row back on its statements' attestation, but it has two gaps:
  • It reads only typed_fact_sources.
  • In materialize_in_tx it runs before imply_in_tx.
• Later attestation. A statement's attestation can change after its rows are computed, and a rule's row does not follow it. attested_by also stays empty. This happens two ways:
  • Time resolution attests a document's statements in its own job, then calls the sync.
  • An earlier document can say the same thing again; attest_statement then moves the attestation earlier.
• So two undated statements of the same shape read differently. Measured on dev, one statement was under a bound phrase and one under an approved rule:

row	attested_from	holds in 2000
materialised by the binding	null	yes
implied by the rule	the moment materialisation ran	no

Change

• In crates/utopia-store/src/materialize.rs:
  • Sources. sync_typed_attestation takes a row's source statements from both typed_fact_sources and implied_fact_sources. A kind-word rule's source is an entity, so it adds no attestation.
  • Value. Unchanged: the earliest attestation among the sources, with its date's name, or none.
  • Rows moved. The rows it moved before (those with a statement materialised into them), plus the rows a rule computed (implied).
  • Not moved. A row a person wrote, which only a rule's conclusion merged into, stays as it is today. attest_earlier already moved it earlier at the merge.
  • Order. In materialize_in_tx the sync now runs after imply_in_tx.
• migrations/0101_a_row_a_rule_implies_follows_its_statements.sql applies the same rule to rows written before. Like 0100, it covers every base and invalidated rows too. CURRENT_SCHEMA_VERSION becomes 92.
• 0064 decision 5 now says that a rule's row follows the statements the rule read, and names 0101.
• The comments at the two callers of the sync (graph::attest_statement and time_resolution::attest_statements) now mention rule rows.

#1007 also adds a 0101. Whichever lands second moves to 0102 and a schema version of 93.

Verification

Run in containers on a Linux host: Rust 1.98.1 and PostgreSQL 16 with pgvector.

• a_row_a_rule_implies_follows_the_statement_it_read (store tests):
  • Undated statement. The rule's row and a kind-word row have no attestation, and both hold in 2015 and now.
  • Statement attested afterwards, the way time resolution does it (attest_statement, then sync_typed_attestation):
    • the rule's row moves to that day, with the date's name, and does not hold before it;
    • the kind-word row stays unattested;
    • running materialisation again changes neither row.
  • A person's row that the rule's conclusion merged into keeps the moment it was written.
• a_row_a_rule_implied_follows_its_statements is a top-level binary, like the 0096 and 0099 tests, because the migration is not filtered by base.
  • It sets the rule rows back to the processing moment, as dev writes them, and then runs 0101 twice.
  • Undated statement: null.
  • Statement attested after materialisation: that day, with the date's name.
  • Kind word: null.
  • The person's row: unchanged.
• Control. On dev with only the two tests, plus a 0101 that does nothing so that the second test compiles, both tests fail:
  • The store test fails at its first check. The rule's row is attested at the moment materialisation ran (2026-09-30T04:09:11Z), not left empty.
  • The migration test fails at its first check after 0101 runs. The undated statement's row keeps the processing moment.
• Migration on dev's own rows. 0101 was also applied by hand to rows written by dev's code.
  • Three rule rows were at the processing moment. They became null, 2026-09-04 (签约日 2026年9月4日), and null.
  • The person's row did not change.
• cargo fmt --all --check and cargo clippy --locked --workspace --all-targets -- -D warnings are clean.
• UTOPIA_TEST_REQUIRE_PDFTOTEXT=1 cargo test --locked --workspace passes: 1306 passed, 0 failed.

[WaylandYang]

Approving. Thanks @Maya-Kid.

Reviewed at 1c599d9. The change closes the gap the table in the description shows: a rule's row now reads its statements' attestation through implied_fact_sources, the sync runs after imply_in_tx, and a person's row that a conclusion merged into is left alone because it is neither implied nor materialised from a statement. The migration follows 0100's shape (every base, invalidated rows too). The test that attests a statement afterwards, the way time resolution does, is the one that pins the behaviour that mattered.

Migration number. 0101 is also taken by #1015 (the Ark OCR reader), which is waiting on a contributor's verification with a real key. This one lands first and keeps 0101; #1015 will move to 0102 and a schema version of 93.