fact-cut-runner · diff
git:20261003.3cdf089 to git:20261003.43cf982
10 added, 13 removed. Audit A to A.
---
name: fact-cut-runner
description: >-
- Compose fact-intake L1: whole-doc cut and disposition classify into
- one `_facts.json`.
+ Compose fact-intake L1: whole-doc cut and disposition classify.
---
# fact-cut-runner
- Read `$SOURCE_PATH` once. Write focus-slice `_facts.json` atoms with
- `origin.type=seed`, `derivation.upstream_ref`, and disposition. Tag each
- atom while cutting.
+ Read `$SOURCE_PATH` once and write classified seed atoms. Done when
+ `$FACTS_CTL validate --require-derivation` exits 0.
- **Must:** `$FACT_CUT_BUILD context` then cut+tag+write; pass
- `--require-derivation` validate.
+ **Must:** `$FACT_CUT_BUILD context`; cut+tag+write.
**Must not:** Confirm; Eval; `fact-store-runner`; edit source doc; write
`discovered`.
## Input
```text
REVISION_DIR: <abs revision root>
PROJECT_ROOT: <abs project root>
CYCLE_ID: <cycle id>
SOURCE_PATH: <abs intake SoT>
REQUIRE_SEED_ORIGIN: <true|false; default false>
```
## Cognition
Output terms carry the meanings defined in these units.
| Term | Unit |
|---|---|
| fact atom, `anchors`, `lens` | `../../references/cognition/fact.md` |
| `origin.type=seed`, `derivation.upstream_ref`, `disposition`, `rule_id` | `../../references/cognition/producer/provenance.md` |
| registry `intent`, `intent_boundary` | `../../references/cognition/lens.md` |
| role `consume_policy` | `../../references/cognition/profile/role.md` |
One verdict per atom, assigned while cutting:
```text
not_needed ⇐ one consume_policy rule holds (cite rule_id; omit lens)
carried ⇐ some lens intent owns it (lens = one key, chosen per lens.md)
quarantined ⇐ otherwise (omit lens)
```
One lens per fact: read its source section (`derivation.upstream_ref`) first;
if still ambiguous, take the lens whose intent is closest. Splitting is not
decided here. After the last atom, repair don't-list false kills.
## Script Macros
Contract in `--help`.
| Macro | Command |
|---|---|
| `$FACT_CUT_BUILD` | `python3 "$SKILL_ROOT/compose/fact-intake-runner/fact-cut-runner/scripts/fact_cut_build_control.py"` |
| `$FACTS_CTL` | `python3 "$SKILL_ROOT/compose/scripts/facts/facts_control.py"` |
## Boundaries
- - Control calls: `context` once; `write` once; `validate --require-derivation` once (add `--require-seed-origin` when `$REQUIRE_SEED_ORIGIN=true`). A failed fetch or validate fails the pass.
- - Evidence closure: `context` stdout, this SKILL, the Cognition units, and `$SOURCE_PATH` are the whole evidence. Read nothing outside the closure.
- - Do not list scripts, read a sibling runner, or read another cycle's `_facts.json`.
- - Write once: seed + `upstream_ref` + disposition. Do not pass `--intake-structure`.
+ - Control calls: `context` once; `write` once; `validate --require-derivation` once. Add `--require-seed-origin` when `$REQUIRE_SEED_ORIGIN=true`. A failed call fails the pass.
+ - Evidence closure: `context` stdout, this SKILL, the Cognition units, and `$SOURCE_PATH`. Read nothing outside the closure.
+ - Do not list scripts or read another cycle's facts.
+ - Write seed + `upstream_ref` + disposition.
## Execution
1. Bind Input.
2. `$FACT_CUT_BUILD context --revision-dir … --project-root … --cycle-id …`
3. Whole-doc cut; tag each atom; repair don't-list false kills.
- 4. `$FACTS_CTL write …` (no `--intake-structure`; add `--require-seed-origin` when `$REQUIRE_SEED_ORIGIN=true`).
- 5. `$FACTS_CTL validate … --require-derivation` (same `--require-seed-origin` rule).
+ 4. `$FACTS_CTL write`.
+ 5. `$FACTS_CTL validate --require-derivation`.
**Done:** validate exit 0.
## Summary
```text
Fact-cut complete.
Facts: <path>
Facts count: <N>
Disposition counts: carried=<n> quarantined=<n> not_needed=<n>
Status: ok
```