Methodology & Sources
Last updated 28 September 2026
Incentive Radar is a sole proprietorship operated by Ima Kennerly in the United States.
The one rule everything else follows: no figure enters the dataset without a link that resolves to the public document it came from. A field the document does not state comes back blank, not guessed at. Anything worked out rather than read is labelled an estimate, with its reasoning shown.
Where the data comes from
Every source is a government body’s own public disclosure — a film commission’s approved-projects list, a state audit, a program report, an agency press release. Nothing is bought from a data broker, nothing is scraped from a competitor, and nothing is added by hand.
| Jurisdiction | What the state publishes | What that makes possible |
|---|---|---|
| California | A full approved-projects list with credit allocated, qualified spend, filming days, cast and crew hired, fiscal year, and allocation date | The complete list, parsed field by field, and tracked week to week: additions, removals, waitlist moves, and changed amounts |
| Georgia | Production activity — who is filming and where — plus program totals in irregular state audits. No per-project dollar figures anywhere. | Production-activity tracking and program-level totals. No per-project Georgia credit figure is published, because none exists publicly. |
| New Mexico | Production identity almost continuously; credit dollars only as monthly program aggregates | A production-identity feed and monthly program-level totals |
| NY, NJ, IL | — | Planned next. Each gets its own source study and four consecutive clean cycles before it is announced. |
How a record gets in
- Ingest. Each source page or document is fetched on a daily schedule and stored with its retrieval time. Content is hashed, so an unchanged page is not reprocessed.
- Parse or extract. Where a source is structured — California’s list is an HTML table — it is parsed by code, deterministically, with no language model involved. A language model reads only prose documents that no parser can handle, and it is instructed to return nothing rather than infer a missing field.
- Validate. A record is checked against the document it came from and its source link must resolve. Records that fail are flagged with a reason and kept out of the dataset rather than deleted.
- Snapshot and diff. The California list is snapshotted each run and compared with the previous snapshot, so changes become dated events. That comparison is the “what moved this week” section of the digest.
- Publish. Only validated records reach an issue, and every total states the period it covers.
What the numbers mean
- Credit allocated
- What the state has reserved for a project. It is not what the project has earned, been certified for, or actually claimed — those come later and can be lower. Projects also withdraw.
- Qualified spend
- The project’s own estimate of spend expected to qualify under the program, as stated in its application. An estimate at the time of application, not an audited outcome.
- Filming days
- As stated by the applicant. Where a state does not publish it, the field is blank.
- Fiscal year / program version
- California’s list spans several program generations and years. Every row carries its fiscal year, and every published total names the period it covers, so eras are never silently mixed.
- Estimate
- A figure derived rather than read. Always flagged, always shown with how it was derived.
What Incentive Radar does not claim
- It does not break news. Trade publications cover large-state allocation rounds the same day. It is not faster, and it has nothing exclusive.
- It does not yet normalize across dozens of states. Today the honest description is: the full California list, structured and tracked over time. Production-activity tracking for Georgia and New Mexico is in progress and not yet published. Wider coverage is a plan, not a current claim.
- It does not publish project locations for California. The state does not disclose them, so they are not in the dataset.
- It holds no personal contact data on anyone in the industry. Company level only.
Revisions and errors
States revise their own figures, and the dataset follows the source rather than an earlier copy. When a figure changes, the previous value stays in the time series, so the revision itself is visible. Mistakes are acknowledged in public: see Corrections.
Citing Incentive Radar
Please cite the underlying state document where you can, and Incentive Radar for the structure or the comparison. Journalists and researchers: email [email protected] to receive the underlying rows and the source links for anything you want to check.