Tell Whether Culture Investments Are Working

Budget season is the week after the quarterly review. Each year, leadership asks whether last year's culture investments moved anything. This guide shows how to read driver-score change across the snapshots that straddle an initiative's completion date. It shows how to ground the delta in engineer voice and organizational benchmarks. It also shows how to assemble a readout that distinguishes good, mixed, and failed investments before the next budget cycle.

Prerequisites

Complete Demonstrate Engineering Progress first. That guide covers how to install Landmark, validate standard data, confirm your roster, and run the health view. The steps below assume Map's activity layer holds data. They also assume you have at least two GetDX snapshots. Ideally one snapshot comes from before the initiative began and one from after it completed.

You also need to know which initiatives you ran and when they completed. Landmark does not surface initiatives directly through the CLI. Bring the initiative name, owner, completion date, and intended driver. Take them from your GetDX workspace or your team's planning notes.

Match each initiative to its intended driver

You hire a culture investment to move a specific outcome. Before you read any data, write down what each initiative was supposed to change:

Initiative	Owner	Completed	Intended driver
init_007 Deep Work remediation	you	2025-02-28	deep-work
init_029 One BioNova	you	2025-08-15	code-review

Initiative IDs and completion dates come from GetDX. Driver IDs come from your drivers.yaml. List them with npx fit-pathway driver --list if you need to look them up. If you cannot name the intended driver for an initiative, the readout has no place for that initiative. Pause and resolve that before you continue.

Read the driver trend across the initiative window

For each initiative, read the driver's score across the snapshots that straddle its completion date:

npx fit-landmark snapshot trend --item deep-work --manager you@example.com

  Trend: deep-work (Your team)

    2024-12-15   58
    2025-03-15   71
    2025-06-14   74
    2025-09-13   76

init_007 completed 2025-02-28. That date falls between the 2024-12-15 and 2025-03-15 snapshots. The driver moved from 58 to 71 (+13) across that boundary. It then held above 70 through later snapshots. That is the shape of an investment that landed. The step change across the completion window persisted. It did not revert.

Repeat for the second initiative:

npx fit-landmark snapshot trend --item code-review --manager you@example.com

  Trend: code-review (Your team)

    2025-03-15   76
    2025-06-14   78
    2025-09-13   77
    2025-12-12   78

init_029 completed 2025-08-15. That date falls between the 2025-06-14 and 2025-09-13 snapshots. The driver moved from 78 to 77 (-1). That change sits inside the noise of the prior quarter's variation. That is the shape of an investment that did not move the outcome you hired it for.

Pass --format markdown on either command to produce output you can paste directly into a planning document or a VP-facing slide.

Compare against the organization to rule out drift

A driver score can rise across a window because the organization as a whole moved. The initiative can have no part in that rise. Use snapshot compare to check whether the change is specific to your team:

npx fit-landmark snapshot compare --snapshot NzE4MmRk --manager you@example.com

  Snapshot comparison: NzE4MmRk (Your team vs organization)

    Driver          Team   p50   p75   p90
    deep-work         74    65    73    82
    code-review       78    70    80    88
    incident-response 65    68    76    84

If the organization-wide median moved with your team, the initiative may not be responsible for the gain. Environmental factors lift everyone. If the team's percentile rose relative to the organization across the snapshot boundary, the investment is more credibly the cause. Use the snapshot ID that immediately follows the initiative's completion date. Find it with npx fit-landmark snapshot list.

Ground the delta in engineer voice

A score change is more defensible when engineers say the system changed. Surface comments for the driver in question:

npx fit-landmark voice --manager you@example.com

  Voice: Your team (latest snapshot)

    focus       4 comments
      "No-meeting Wednesdays actually stuck this quarter"
      "Deep work blocks make a real difference"
      "Fewer interrupts during the afternoon stretch"
      "Meeting load is more reasonable than it was"

    Below-50th driver alignment:
      incident-response (48th percentile) — 3 incident comments

When themed comments line up with the intended driver, the qualitative evidence backs the quantitative shift. An example is focus comments that cluster after a Deep Work initiative. When comments cluster on a different theme, or do not mention the driver at all, the readout should say so.

Assemble the readout

For each initiative, write one of three verdicts grounded in what you saw:

• Worked. The driver moved across the completion window. The team's percentile rose relative to the organization. Engineer voice aligns with the intended driver. Example: "init_007 (Deep Work remediation, completed 2025-02-28) tracked with deep-work, which moved from 58 to 71 across the Q4→Q1 snapshot boundary. The team's Q1 percentile placed it above the organizational median. Engineer comments that quarter clustered on focus and meeting load. The recommendation is to continue the policy."
• Mixed. The driver moved, but the organization moved with it. Or engineer voice does not corroborate. Surface both the score change and the caveats.
• Did not land. The driver did not move across the completion window. Or it moved within the prior quarter's range of variation. Engineer voice also does not align. Recommend no renewal of the spend without a redesign.

The VP-facing version of each verdict is two sentences. The first names the initiative and its intended driver. The second gives the observed change and the confidence in the evidence. Avoid stronger language than the data supports. Culture investments interact. Landmark surfaces correlation across a snapshot boundary. It does not surface causation.

Verify

You have a defensible culture-investment readout when you can answer these questions:

• Did the intended driver move? You ran npx fit-landmark snapshot trend --item <driver> for every initiative. You can name the snapshot pair that straddles its completion date.
• Is the change specific to your team? You ran npx fit-landmark snapshot compare --snapshot <id> for the snapshot immediately after each initiative. You can say whether the team's percentile rose or moved with the organization.
• Does engineer voice agree? You ran npx fit-landmark voice --manager <email>. You can point to themed comments that align with the intended driver. You can also note their absence.
• Have you classified each initiative? Every initiative on your list carries a worked, mixed, or did not land verdict. The verdict includes the score delta and the qualitative evidence behind it.

If a verdict rests only on a score change, with no comparison and no voice to corroborate it, treat it as provisional. Say so in the readout.

What's next