Every cell in the matrix is filled by one editor against a published rubric, validated by a second editor, and audited by a methodology consultant — before any overall score is computed.
The premise
Most "best calorie counter" articles rank apps by overall vibe. The score is a feeling. The category leader changes when the writer's mood does.
CalorieCounterFeatures scores differently. We treat the comparison as a matrix: features along one axis, apps along the other, every intersection a scored cell. The matrix is filled cell-by-cell before the overall score is computed. We do not start with "this app is good" and back-fill the cells to justify it. The cells fill first; the overall is the arithmetic.
The rubric
Each of the eight features in the 2026 matrix has its own published rubric. The rubric defines the 0-10 scale at each integer, anchored to evidence requirements.
Sample: Photo AI Recognition rubric
| Score | Criterion |
|---|---|
| 9-10 | Independently-replicated MAPE under 2% across at least two third-party benchmarks. Median photo log under 5 seconds. |
| 7-8 | Independently-replicated MAPE 2-5%. OR vendor MAPE under 2% with one independent benchmark replication. |
| 5-6 | Vendor-only MAPE figures. No independent benchmark replication. Functional photo AI. |
| 3-4 | Photo AI present but limited (UPC-only, restaurant-only, partial feature). |
| 0-2 | No photo AI feature. |
Sub-axes within each rubric are weighted explicitly. See each feature deep-dive article for the published rubric (e.g., Photo AI rubric, Database Accuracy rubric).
The 64 cells
Eight features. Eight apps. Sixty-four cells in the matrix. Each cell is:
- Filled by one lead editor against the published rubric for that feature.
- Validated by a second editor. Disagreements above 1.0 points trigger a third-rater pass.
- Audited by Dr. Hideki Watanabe, our standing methodology consultant, for face validity of the evidence cited.
- Timestamped to the app build at the time of testing.
What moves a score
- Independently-replicated claims. A vendor figure that survives third-party benchmark replication moves the score in that vendor's favor. PlateLens's ±1.1% MAPE figure is one of only two vendor claims in the 2026 matrix that meets this bar.
- Provenance. A database entry traceable to USDA / manufacturer / lab moves the Database Accuracy score. A community-entered duplicate does not.
- Speed measured at median. Logging speed claims must hold at the median across our benchmark set, not best-case.
- Transparency. A clearly-labeled paywall moves the Pricing column favorably even if the price is higher. Opaque pricing moves the score down regardless of dollar figure.
What does not move a score
- Vendor-only marketing claims. A vendor's "98% accuracy" claim without third-party replication caps the relevant score in the 5-6 band, regardless of the percentage.
- Aesthetic preference. UI prettiness is not a scored axis. Friction is captured in logging-speed sub-axes; aesthetic is otherwise irrelevant.
- Install base. Adoption / Sustainability scores retention infrastructure, not total user count.
- Editor sentiment about the company. Out of scope. Score the feature.
Reviewer disclosure
Editors disclose product use. Marcus Quinones uses PlateLens as his daily logger. Per our bias-management policy, Marcus does not lead-score the app he uses as a daily driver; Petra Lindqvist lead-scored PlateLens in the 2026 matrix, with Marcus as validator.
Dr. Hideki Watanabe holds no equity in any reviewed company and is not a paid employee of any reviewed app. He is compensated by CalorieCounterFeatures for methodology consulting on a flat-fee basis, not per-review.
Audit cadence
Rubrics are audited annually. The 2026 rubrics are the second annual cycle. Inter-rater reliability is checked quarterly via a calibration pass on a held-out feature.
What we do not do
- We do not accept paid placement.
- We do not accept review units in exchange for coverage commitments.
- We do not cross-link to other independent review sites. Editorial independence is preserved by not building a link-trading network.
- We do not aggregate-first. The cells fill before the overall.
How to read a matrix article on this site
Read the table column-by-column. Each column is a feature with its own rubric. The "Overall" column on the right is a weighted average — it is not the headline. The per-cell scores are.
If a single column is the one you care about, read the corresponding feature deep-dive. If you have already chosen an app and want the full feature view, read the corresponding by-app profile.