AFI Needle Test Lab

Remote VoteFlow backtester. V6.2.2 fixes the Completion Benchmark so only the new correction is bounded, never the race's full margin; keeps Path Benchmark at 0% direct final weight and the guarded Turning-Point system; and adds a Statewide Snapshot Battery for testing every contested statewide race at the same percentage counted. Single-race and statewide batches both expose click-to-expand full candidate projections. latest.json remains scoring-only.
https://data.americafirstinsight.com/Election2026/output/2026/AK/Governor/00/open_primary/
V3/V6.2.2 runoffs automatically use Round 1 when this is blank. Single rounds can use a comparable prior final as a turnout reference. If a remote historical reference has disappeared and returns 404/410, Needle Lab now treats it as unavailable and falls back instead of crashing the entire test.
Media Markets
Auto-loaded: engine/media_markets/ak_media_markets.js
V6.2.2 will use the FIPS mapping automatically for AK. No path needs to be entered.

Statewide Snapshot Battery

Test every contested statewide race in AK for the selected primary / initial round at one common percentage of eventual vote. Needle Lab discovers replayable true statewide races across all parties, auto-loads each party's historical library when available, and uses the state's media-market map. A 00 folder by itself is not considered statewide: U.S. House, State House/Senate/Assembly/legislative offices are explicitly excluded, and remaining races must cover essentially the state's full county footprint. Uncontested one-candidate races are also excluded so they do not artificially inflate winner accuracy.
Examples: 20, 30, 50. Each race uses its nearest stored snapshot within the normal checkpoint tolerance. The statewide battery runs as a background job, so clicking the button should return immediately instead of waiting for every race.

Historical Reference Library

Select multiple imported Open_primary references for AK. V6.2.2 combines their county turnout footprints and reweights each historical race at every snapshot using recency, office similarity, and live turnout fit. Candidate identities from unrelated historical races are never blended together. Once built, this state/party library now auto-loads on future visits.
No imported Open_primary references for AK yet. Import one or more official-results files below.

Historical Reference Importer — Official Results File

Upload a state election-results .txt, .tsv, or .csv. Needle Lab detects the county, office, party, candidate, and vote columns, then lets you choose the historical race to import. Supported formats now include Florida Division of Elections TXT, Iowa Secretary of State fixed-width detail.txt, South Carolina Election Commission fixed-width detail reports, Michigan MVIC, Oklahoma Election Board CountyResults, and generic delimited exports. Iowa and South Carolina are parsed section-by-section; SC candidate names are read from their physical report columns so odd internal spacing does not split one candidate into two.

Historical Reference Importer — Wikipedia

Paste a Wikipedia election article. The scanner lists only tables containing importable county vote totals. Democratic primary tables are preselected when detected.

V6.2.2 Historical Race Battery

Run the same V6.2.2 checkpoints across many archived races at once. The completed archive also builds state reporting-path priors (dump size, cadence, and market activity) for future races; V6.2.2 automatically excludes a race from any profile that was trained on that race. For real calibration work, keep separate tuning and holdout manifests so parameters are never judged only on the races used to choose them.
Format: {"races":[{"race_url":"...","baseline_url":"...","market_map":"...","label":"..."}],"targets":[1,3,5,10,20,30,40,50,60,70,80,90]}. Each race may omit baseline or market_map when auto-discovery is appropriate.

Saved V6 Test Reports

Reports are saved in output/reports/. This list survives page refreshes and reverse-proxy timeouts, so if Python finished writing a report before the gateway timed out, it should still appear here.
Generated Race Model Size Downloads
2026-08-29T18:36:18Z MI statewide @ 35.0% AFI_NEEDLE_V6_2_2_STATEWIDE_CHECKPOINT_BATTERY 49.5 KB
2026-08-29T18:35:34Z MI statewide @ 30.0% AFI_NEEDLE_V6_2_2_STATEWIDE_CHECKPOINT_BATTERY 49.2 KB
2026-08-29T18:32:15Z SC statewide @ 40.0% AFI_NEEDLE_V6_2_2_STATEWIDE_CHECKPOINT_BATTERY 55.9 KB
2026-08-29T18:29:46Z SC statewide @ 30.0% AFI_NEEDLE_V6_2_2_STATEWIDE_CHECKPOINT_BATTERY 55.7 KB
2026-08-29T18:28:57Z SC statewide @ 30.0% AFI_NEEDLE_V6_2_2_STATEWIDE_CHECKPOINT_BATTERY 205.2 KB
2026-08-29T18:07:21Z OK statewide @ 40.0% AFI_NEEDLE_V6_2_2_STATEWIDE_CHECKPOINT_BATTERY 69.1 KB
2026-08-29T18:06:26Z OK statewide @ 30.0% AFI_NEEDLE_V6_2_2_STATEWIDE_CHECKPOINT_BATTERY 69.4 KB
2026-08-29T18:05:10Z OK statewide @ 30.0% AFI_NEEDLE_V6_2_2_STATEWIDE_CHECKPOINT_BATTERY 248.5 KB
2026-08-29T18:03:30Z MI statewide @ 45.0% AFI_NEEDLE_V6_2_2_STATEWIDE_CHECKPOINT_BATTERY 49.6 KB
2026-08-29T18:02:45Z MI statewide @ 40.0% AFI_NEEDLE_V6_2_2_STATEWIDE_CHECKPOINT_BATTERY 50.0 KB
2026-08-29T18:00:33Z MI statewide @ 35.0% AFI_NEEDLE_V6_2_2_STATEWIDE_CHECKPOINT_BATTERY 49.5 KB
2026-08-29T17:59:56Z MI statewide @ 30.0% AFI_NEEDLE_V6_2_2_STATEWIDE_CHECKPOINT_BATTERY 49.2 KB