Changelog

Every measurement run, methodology change, and product update. The track record so far is short (the measurement product is recently introduced), but we publish it because the value of the methodology is its reproducibility, and reproducibility starts with public history.

1. 2026-05-28· measurement ·workwhile-v3

   WorkWhile cross-channel measurement (v3, canonical case study)

   Re-ran WorkWhile on the current model panel with Codex dropped (inappropriate for a non-developer vertical), bringing it to the same breadth and rigor as the SSENSE measurement. 108-cell battery: 12 buyer-decision query variants × 3 frontier models (Sonnet, Haiku, GPT-5.5) × 3 rounds. Headline: 7.4% organic AI mindshare (4/54 subject-blind cells, Wilson CI 2.9-17.6%), 87pp organic/evaluation spread = cold-start, against operators who lead every human-discourse channel (HN 924, Reddit 1,375, Twitter 886) by a wide margin. Aggregate (reference-only): Instawork 86.1%, Wonolo 80.6%, WorkWhile 50.9%. Now the canonical /workwhile dashboard, superseding the prior measurement.

2. 2026-05-25· methodology ·methodology v2

   Organic vs evaluation-stage mindshare split

   Discovered while running Korpora-on-Korpora v2 that two of the four standard query framings (`direct-comparison` and `differentiation`) name-check the subject in the prompt itself, forcing every model response to engage with the brand regardless of whether the model has real knowledge of it. The legacy aggregate counted those cells as mindshare and gave a 46.5% headline for a brand with literally zero web presence. v2 splits mindshare into organic (subject-blind framings: discovery + use-case) and evaluation-stage (subject-named: direct-comparison + differentiation), with the spread as a diagnostic. <=15pp = aligned, 15-40pp = moderate-gap, >40pp = cold-start. Korpora-on-Korpora went from a misleading 46.5% to an honest 0% organic, 93pp cold-start spread.

3. 2026-05-25· measurement ·korpora-v2

   Korpora-on-Korpora baseline (v2, organic split)

   Re-rendered with methodology v2: 0.0% organic mindshare (0/72 subject-blind cells), 93.1% evaluation-stage (67/72 subject-named), 93pp spread = cold-start classification. The honest floor we own publicly. Foundation data shows Reddit velocity 3.84x post-cutoff. The operational story is corpus-density into the next training cycle.

4. 2026-05-25· methodology ·methodology

   Brand-name collision protocol added

   Pre-flight check now grep's subject + competitor names against high-risk collision categories (programming languages, ML libraries, biology/chemistry terms). When collision flagged, matcher tightens to require disambiguating alias. Manual review of first-round sample confirms matched mentions are actually about the subject brand.

5. 2026-05-25· measurement ·korpora-v1

   Korpora-on-Korpora baseline (v1, withdrawn)

   First self-measurement. Result (46.5% mindshare) was a false positive; the matcher couldn't disambiguate between Korpora (us) and Korpora (a Korean NLP Python library). Withdrawn from public use; result preserved for methodology-collision learning. Triggered the protocol added above.

6. 2026-05-27· measurement ·workwhile-v2

   WorkWhile cross-channel measurement (v2, sales-leader re-frame)

   Same v2 measurement against on-demand staffing competitors, re-rendered for a sales-leader audience. Strategic positioning, what each finding means for the sales motion, top three actions ordered by lift potential. Demonstrates the persona-reframe deliverable shape without re-measuring.

7. 2026-05-26· measurement ·workwhile-v1

   WorkWhile cross-channel measurement (v1, full technical)

   Full 144-cell measurement of the on-demand staffing category. Foundation-channel pull (HN, Reddit, X, GitHub, arXiv), 144-cell agent-layer battery across four frontier models, cutoff-split velocity with Poisson CIs, per-rival capability map, lift-potential ordered fix list. Headline: 6.9% organic AI mindshare against operators leading the category on every human-discourse channel by a wide margin. The gap is the work.