| name | matrix-row-traceability |
| description | Confirm that a drafted application document actually follows the prescription of its matched evidence-bank matrix row โ every B-ID in the row's body stack is reflected in the draft, the dominant A-ID positioning leads, the template and length band match, and the fit-acknowledgment style and AI register are honoured. Use this skill at Step 1 close on any application-drafting session where a matrix row was matched (cite by ID), alongside length-check and watchouts-sweep. Also use when [Candidate First Name] asks 'did we follow the row', 'trace the draft to the matrix', or 'is this on-prescription'. Catches prescription drift: a draft that silently swapped the row's tested body stack for ad-hoc evidence selection, losing the calibration the row encodes. |
Matrix Row Traceability โ prescription verification skill
Codifies the matched-row mechanism of Project_Instructions.md (Mandatory retrieval-first protocol, steps 3โ4; rev 2026-06-12). Wave 3 verification skill of the framework.
The matrix rows encode tested calibrations โ which positioning leads, which body stack lands, which template and length band fit the archetype. A draft that drifts from the matched row without a documented reason quietly discards that calibration. This skill makes the drift visible.
When to invoke
- Default trigger: at Step 1 close, after
length-check and watchouts-sweep, before the draft is delivered โ on any session where a matrix row was matched.
- Explicit triggers: "did we follow the row", "trace the draft to the matrix", "is this on-prescription".
- Skip when: the session ran on a documented no-clean-match basis (closest-analogue noted in the pre-drafting assessment) โ then trace against the analogue row and note the documented deviations.
Inputs the skill needs
- The final draft text (or file path).
- The matched matrix row ID and its full row from the relevant
Examples_Section_5_*.md file (positioning A-IDs with dominant flag, body stacks grouped by paragraph, template + length band, fit-acknowledgment style, AI register, companion files).
- The pre-drafting assessment (for any documented deviations agreed before drafting).
Checks โ run all six
Load references/traceability-method.md, then:
- Body-stack reflection. Every B-ID in the row's body stack appears in the draft as substantive content (not a passing mention). List any absent B-ID and any substituted entry.
- A-ID dominance. The row's dominant positioning block leads the draft's framing; secondary A-IDs support, not compete.
- Template + length band. The draft used the row's prescribed template (C1โC6) and sits inside the row's length band (cross-checked with
length-check output, not recounted).
- Fit-acknowledgment style. Honest-gap framing follows the row's prescribed style (e.g. direct acknowledgment vs adjacent-strength bridge).
- AI register. The draft's AI positioning matches the row's register (e.g. hands-on builder vs governance-led).
- Documented deviations. Every deviation from 1โ5 is either documented in the pre-drafting assessment or flagged now as undocumented drift.
Output format
A six-check table (check / on-prescription / deviation / documented?), then verdict: On-prescription, Documented deviations only, or Undocumented drift โ review (advisory: [Candidate First Name] decides whether the drift is an improvement to fold back into the row at STEP 4, or a regression to fix). Undocumented drift that is an improvement must be recorded against the matrix row in the same session โ that is the row-calibration loop.
