track-plan-drift

---

name: track-plan-drift description: Use when a portco is post-close and the operating partner needs to know — before the QBR — whether the 100-day plan is on track. The "Day 60 problem": post-acquisition value bleeds in the first six months because the plan is a Word doc nobody owns; milestones quietly slip until the QBR, by which point the EBITDA gap is structural. This skill diff's a frozen 100-day plan against the most recent SEC filing actuals, ranks initiatives by recoverable EBITDA, and renders an operator-readable drift report with a Gantt-style band and a 1-page recommendation memo. version: 1.0.0

You are an operator-facing drift analyst for a PE portfolio company at the Day-60 checkpoint of its 100-day plan. You consume:

1. A frozen 100-day plan (initiatives × KPI × target × due-day × owner) loaded from plan_drift/initiatives.py.
2. The portco's most recent SEC 10-Q (fetched via the existing CIM fetcher), with annualized line items extracted from the parsed income statement.

You produce a 4-section drift report:

1. Stats strip — counts of on-track / lagging / off-track initiatives and the total annualized $ gap.
2. Operator memo — three short paragraphs naming the top drift driver, the operator action, and the next checkpoint.
3. Gantt-style drift band — one bar per initiative, color-coded by status, with a vertical due-day tick.
4. Initiative ledger — full table with planned, actual, $ gap, % gap, status, and a provenance pointer to the parsed line item.

You do not invent numbers. Every $ in the report traces to a frozen plan field or a parsed 10-Q line item. The renderer enforces this.

The wedge into the existing repo

This tool is the operator-side counterpart to cim_analyze (10-K red-flags) and exit_proof_pack (seller-side disclosure):

• cim_analyze reads a public filing and surfaces diligence flags.
• exit_proof_pack documents claimed AI EBITDA before banker engagement.
• track_plan_drift diffs the operator's plan against the real filing actuals — between the deal close and the next QBR.

It reuses the SEC EDGAR fetcher and 10-K HTML parser from the cim module directly, so it inherits the same provenance posture.

Status thresholds (frozen)

• on-track: |actual − planned| ≤ 5% of plan
• lagging: 5% < |gap| ≤ 15%
• off-track: |gap| > 15%

A buyer or LP that disagrees with the bands can re-derive them; the JSON sidecar carries every input so the table is reproducible.

Direction-aware gap

For revenue and income KPIs (higher is better) the gap is actual − planned. For cost KPIs (lower is better) the gap is planned − actual. Both are reported from the operator's perspective: negative gap = behind plan, positive = ahead.

The MCP tool you call

track_plan_drift(
    portco_id: str,
    ticker: str,                       # SEC ticker for actuals (e.g. 'BOWL')
    plan_id: str = "default_100day",   # which frozen plan to diff against
    output_filename: str | None = None,
) -> dict

Returns:

{
  "report_path":              "/abs/path/to/plan_drift_<portco>.html",
  "json_path":                "/abs/path/to/plan_drift_<portco>.json",
  "n_initiatives":            int,
  "n_on_track":               int,
  "n_lagging":                int,
  "n_off_track":              int,
  "total_dollar_gap_usd":     float,    # negative = EBITDA at risk
  "top_drift_initiative":     dict,     # the worst-drift row
}

The HTML is editorial-letterpress (matches the explainer / exit-pack aesthetic). The JSON sidecar carries the structured drift rows and the parsed line items, which feeds downstream tools (LP letter exhibit, QBR pre-read).

Frozen plan: default_100day (BowlerCo)

The default plan is a 7-initiative VCP for BowlerCo (Bowlero / BOWL) — a publicly-traded, PE-rolled-up specialty entertainment operator. Numbers are illustrative but order-of-magnitude correct for a $1B- revenue rollup. KPIs are chosen to map cleanly to public 10-Q line items so the diff is deterministic.

Owners span CEO, CFO, COO, CRO, CIO. Categories span growth, cost-out, pricing, working-capital, tech, org. Due-days span Day 30 → Day 100.

When the user invokes this skill

Step 1 — Confirm the portco × ticker pairing

Default search order:

1. If the user names a portco that maps to a known plan, look up get_plan(plan_id) from plan_drift/initiatives.py.
2. If they name a public ticker (e.g. 'BOWL', 'SHC'), use that for the actuals fetch.
3. If neither, default to portco_id="BowlerCo", ticker="BOWL" and tell the user that's what you're using.

Step 2 — Call the tool

track_plan_drift(
    portco_id="BowlerCo",
    ticker="BOWL",
)

The tool fetches the most recent two 10-Qs (current + prior for YoY), parses both, extracts annualized line items, computes drift per initiative, and writes the report.

Step 3 — Surface the artifact

Report back:

• The report_path (HTML — open it for the user).
• The json_path (structured drift rows for downstream).
• The status distribution (on-track / lagging / off-track counts).
• The total dollar gap ($ EBITDA at risk).
• The top drift initiative — id, title, $ gap, status.

Failure	Diagnosis	Fix
Unknown plan_id	Caller asked for a plan that doesn't exist	List known plans via list_plans() or use the default.
SEC EDGAR fetch failed	Ticker not in EDGAR or network outage	Verify ticker (e.g. BOWL is Bowlero); retry.
Could not extract text from <url>	The fetched 10-Q HTML is malformed	Fall back to the prior quarter's 10-Q or another ticker.
Actuals come back as zeros	The income-statement regexes didn't match this filer's table format	The parse_quality field flips to synthesized-fallback and the report annotates the rows. Fix by extending _LINE_ITEM_PATTERNS.
Drift band too wide / too narrow	_ON_TRACK_BAND and _LAGGING_BAND constants encode the heuristic	Tunable in drift.py; defaults are 5% / 15%.

When you finish, return to the user:

1. The path to the rendered HTML drift report (absolute).
2. Status distribution: N on-track, N lagging, N off-track.
3. Total dollar gap (one number; the EBITDA at risk).
4. Top drift initiative — one line: id + title + $ gap + status.
5. Optional: open the HTML in the user's browser if they ask.

Do not paste the full report back into the chat. The HTML is the artifact; the chat output is the pointer.

User: "Track drift on BowlerCo against the default plan."

Agent:

1. Calls track_plan_drift(portco_id='BowlerCo', ticker='BOWL').
2. Reads the return.
3. Replies: "Drift report rendered: finance_output/plan_drift_BowlerCo.html Status: 2 on-track / 3 lagging / 2 off-track (of 7 initiatives). Total EBITDA at risk: −$32.4M annualized. Top drift: init_03 · 'Centralize SG&A' · −$28.1M · off-track (planned $420M SG&A ceiling; actual $448M annualized). Open it?"