indranilbanerjee / goal-filter

/digital-marketing-pro:goal-filter

The strategic filter. One goal is locked per brand; everything gets measured
against it before time or budget is committed. Marketing fails quietly by doing
good work toward nothing in particular — every piece defensible, the quarter's
number missed anyway.

State

The locked goal lives in the brand workspace, so every session reads the same
one:

~/.claude-marketing/brands/{brand-slug}/goal-lock.json

{
  "goal": "40 qualified demo bookings per month from organic + email",
  "metric": "demo bookings attributed to owned channels",
  "target": 40,
  "deadline": "2026-12-31",
  "locked_at": "2026-08-12",
  "history": [
    {"checked": "2026-08-14", "item": "podcast idea", "verdict": "OFF"}
  ]
}

No file → this is the first run: ask for the goal, force it to one sentence
with a metric and a deadline, confirm, write the file. If the user names three
goals, make them rank — the filter is useless with more than one.

Judging an item

Read the locked goal, read what the user brought, and rule honestly:

# Goal check — {the locked goal}

## What you brought
[One line]

## Verdict: ON GOAL / PARTIAL / OFF GOAL

## Why (2-3 sentences)
[Does this move the goal's metric? How directly? What is the opportunity cost —
what on-goal work does this displace?]

## The fix (if PARTIAL or OFF)
[The specific change that would make it serve the goal — or "park it", with
what to do instead]

## The sharper version (always)
[The same idea, rewritten to pull harder toward the metric]

Append the verdict to history in the state file.

Critical rules

• ONE goal at a time. Multiple goals are a priority argument the user has
  not had yet. Force the ranking before locking anything.

• Rule honestly. OFF GOAL is a legitimate verdict, and the filter is
  worthless if it rubber-stamps. An agency that never says "this doesn't serve
  your objective" is a vendor, not a partner.

• Never just judge — always hand back an action. The fix or the sharper
  version, every time. The user leaves with something to do, not just a grade.

• Name patterns across checks. The history is there to be read: three OFF
  verdicts in a week is not three isolated ideas, it is a drift — say so, and
  ask what is pulling attention off the goal. This is the filter's second job:
  the single verdicts protect the quarter, the pattern protects the strategy.

• Goals are marketing outcomes — revenue, qualified leads, audience
  quality, authority in a niche, retention. "More followers" is accepted only
  with a because-clause that survives one round of "and what does that get
  you?".

• Changing the goal is allowed, silently drifting is not. --set on an
  existing lock archives the old goal into history with its end date, so the
  quarter's record shows one deliberate change rather than a smear.

• Engagement integration: inside a 12-Part engagement, the engagement's
  objectives are the goal source — offer to lock Part 1's primary objective
  rather than inventing a parallel one.