Reports

EDPA Reports — Timesheet & PI Summary Generation

What this does

Renders the human-readable EDPA artifacts from engine results by running the deterministic reports script. Per ADR-003 ("heavy compute / file generation → directly script"), this skill is a thin wrapper: it resolves the argument, shells out to .edpa/engine/scripts/reports.py, and summarizes what was written. It never hand-renders a timesheet — the script's Markdown is stable and diffable across reruns.

Arguments

$ARGUMENTS = iteration ID (e.g., "PI-2026-1.3"), or "pi " for PI-level aggregation.

Argument resolution (when $ARGUMENTS is empty)

If $ARGUMENTS is empty, blank, or "help":

1. Call MCP tool edpa_iterations (or read .edpa/iterations/*.yaml directly). PI/iteration timeline data is reconstructed at runtime from those per-PI and per-iteration YAML files — edpa.yaml no longer carries pis[].
2. Check which iterations already have results in .edpa/reports/iteration-<ID>/edpa_results.json.
3. Present options:

   Available iterations:
     PI-2026-1.1  [closed]   2026-04-06–2026-04-17   results: yes
     PI-2026-1.2  [closed]   2026-04-20–2026-05-01   results: yes
     PI-2026-1.3  [closed]   2026-05-04–2026-05-15   results: yes
     PI-2026-1.4  [active]   2026-05-18–2026-05-29   results: no (run engine first)
   
   Other options:
     "pi <PI-ID>"   PI-level aggregation across that PI's iterations with results
   

4. Default suggestion: the latest closed iteration that has edpa_results.json.
5. Ask user: "Generate reports for which iteration? [suggested-id]"
6. If .edpa/ does not exist, inform user to run /edpa setup first.

Prerequisites

• .edpa/reports/iteration-<ID>/edpa_results.json exists (run /edpa:engine <iteration-id> first)

Run the script

Per-iteration timesheets:

python3 .edpa/engine/scripts/reports.py <iteration-id>

PI-level aggregation:

python3 .edpa/engine/scripts/reports.py --pi <PI-ID>

Options: --edpa-root <path> (default .edpa), --out <dir> to override the output directory.

Never hand-render. Do not read edpa_results.json and write timesheet Markdown yourself — the script is the single canonical renderer (stable columns, capacity-override annotations, role projection). Hand-rendered output drifts and breaks diff checks.

Output artifacts

Iteration mode → .edpa/reports/iteration-<ID>/

• timesheet-<person>.md — one per person in the results: iteration, methodology (version stamped by the engine), capacity (with baseline/override detail when a capacity override was recorded), derived hours, invariant status, and an item table (Item | Level | Role | JS | CW | Score | Ratio | Hours). The Role column is a display-time projection from evidence signal types (owner / key / reviewer / consulted).
• timesheet-team.md — team rollup: methodology, planning factor, team capacity vs. team derived, one row per person (an Override column appears only when at least one override was applied).

The script prints every file it wrote with per-person derived hours — relay that list to the user.

PI mode → .edpa/reports/pi-<PI-ID>/

• pi-summary-<PI-ID>.md — per-person capacity/derived totals across all iterations of the PI that have results, plus a per-iteration breakdown (team totals, invariants).

Related artifacts (produced by /edpa:engine, not this skill)

• edpa-results.xlsx (Team Summary + Item Costs tabs) — per-item cost allocation lives in the Item Costs tab.
• Frozen snapshot .edpa/snapshots/<ID>.json — immutable, content-hashed; changed engine reruns create <ID>_rev<N>.json revisions instead of overwriting.

For a "who paid for item X" question, use the Item Costs XLSX tab, or:

python3 .edpa/engine/scripts/engine.py --edpa-root .edpa \
  --iteration <iteration-id> --explain <person-id> --explain-item <item-id>

Flow metrics

For lightweight analytics without a full engine run, use the edpa_flow_metrics MCP tool. It computes cycle time (median days from created_at to closed_at), throughput (items closed per week), and open-item age from the timestamp fields populated by sync. No edpa_results.json required — it reads backlog YAML directly.

Error handling

• Missing edpa_results.json → the script exits with an error; run /edpa:engine <iteration-id> first.
• --pi finds no iteration-<PI>.* results directories → run the engine for at least one iteration of that PI first.
• Invariant: FAIL on a timesheet → the engine recorded an invariant violation for that person; re-run /edpa:engine and investigate with --explain before distributing timesheets.
• Nothing is auto-committed. Commit the generated timesheets (and PI summary) as part of the iteration-close batch.