1. Reproducibility in RodentMetrics
RodentMetrics has been used repeatedly during roughly three weeks of internal development and testing. Across that internal testing, when the same source video was analyzed again with the same arena/zone geometry, the same assay-specific configuration and the same analysis parameters -- on the same software version -- the reported analysis result was observed to be the same each time.
That is the specific, bounded condition this page documents:
| Same source video | + |
| Same arena/zone geometry | + |
| Same assay-specific configuration and analysis parameters | + |
| Same RodentMetrics software version | = |
| Same reported analysis result, observed during repeated internal testing | |
This describes an observed engineering property of the software under a specific, unchanged condition -- it is not a guarantee, and it is not a substitute for independent scientific validation (Section 7).
2. What stays fixed
The following inputs and configuration fields are what the tested condition holds constant. Each is stored with the completed experiment so it can be checked or reproduced later:
- The source video file -- the exact recording being analyzed.
- The analysis interval -- the configured start time and duration of the analyzed window.
- Arena geometry -- the calibrated boundary (and, where applicable, physical dimensions) used to transform tracked positions into arena coordinates.
- Assay-specific zone/region geometry -- for example Elevated Plus Maze arm/center regions, Y-Maze arm/center regions, Novel Object Recognition object placement, or Morris Water Maze pool/platform/quadrant geometry.
- Tracking settings -- contour/detection thresholds and related tracking configuration.
- Assay-specific parameters -- for example confirmation-duration thresholds, movement/immobility thresholds, or object exploration distance and orientation settings.
- The RodentMetrics analysis engine version -- persisted on every completed experiment.
A completed experiment's stored configuration and result together let the same condition be reproduced or checked later, rather than relying on memory of what was configured at the time.
3. Deterministic metric computation
Once tracking evidence exists for a frame, the metric calculations RodentMetrics reports from it do not involve randomness. There is no random-number generation, random seeding or probabilistic sampling anywhere in the metric-calculation or tracking code -- the same tracking evidence and the same configuration are processed through the same fixed arithmetic every time. Examples, verified directly from the active implementation:
- Open Field zone/center-periphery time, distance, speed and immobility are calculated from the tracked trajectory and the configured thresholds.
- Elevated Plus Maze open-/closed-arm time and entries are calculated from confirmed zone-transition events.
- Y-Maze spontaneous alternation is (correct overlapping triads of distinct arms) ÷ (total overlapping triads) × 100, from the confirmed arm-entry sequence.
- Novel Object Recognition's discrimination index is (novel time − familiar time) ÷ (novel time + familiar time), from confirmed exploration bout durations.
- Morris Water Maze escape latency is the confirmed platform-reach entry time minus the analysis start time, from the confirmed reach event.
The underlying video-tracking pipeline shared by Open Field, Elevated Plus Maze, Y-Maze and Morris Water Maze is also a fixed, non-probabilistic detection algorithm -- not a machine-learning model with run-to-run variability. Automated tests re-run this tracker against the same recorded video and assert that the frame-by-frame tracked positions and detection records come back identical.
Novel Object Recognition's discrimination index depends on nose/body pose keypoints produced by an external pose-inference service, not the shared tracker above. The arithmetic RodentMetrics applies to those keypoints is deterministic and is verified the same way -- but this page does not claim that the external pose-inference step itself is proven to return identical keypoints on every call. See Section 6.
4. Repeat-analysis observations
During internal development and testing over approximately three weeks, RodentMetrics was used repeatedly, including re-analyzing the same source video with unchanged configuration. In that internal testing, the reported analysis result was observed to match across those repeated analyses.
This is internal repeat-analysis testing -- ad hoc, practical engineering observation made during development, not a controlled study. RodentMetrics does not record a specific count of how many videos, repetitions, animals or software builds this covered, and this page does not state one. It should not be read as, and RodentMetrics does not describe it as, independent validation of any kind.
5. What can change a result
Because the tested condition in Section 1 depends on holding several things fixed, changing any of them is expected to change the reported result -- that is a feature of a configuration-driven analysis, not an inconsistency. Real factors that can change a result include:
- Changing the arena boundary, physical dimensions, or assay-specific zone/region geometry (arms, objects, platform, quadrants).
- Changing the analysis interval (start time or duration).
- Changing tracking settings or assay-specific thresholds (for example a confirmation duration, a movement/immobility threshold, or an exploration distance).
- Changing the source video itself -- a re-export, re-encode, trim, or a different recording of the same session.
- Changing the RodentMetrics software version, which can include changes to the tracking or metric-calculation code between versions.
- For Novel Object Recognition, a change in the pose evidence returned by the external pose-inference service for the same frames.
Comparing results across an experiment where any of these changed is comparing two different configurations, not evidence against reproducibility under a fixed configuration.
6. Assay-specific validation caveats
Novel Object Recognition and Morris Water Maze are supported workflows with scientific validation and scope limitations that this page does not change or upgrade:
Novel Object Recognition's discrimination index depends on an external, pose-based, validation-only detector for object-exploration evidence -- a different, less-tested layer than the shared tracker used by the other four assays. The arithmetic applied to that pose evidence is deterministic and tested (Section 3), but this page does not claim the same reproducibility confidence for the full Novel Object Recognition pipeline that it claims for the tracker-based assays, because no test in the codebase establishes that the external pose-inference step itself returns identical output across repeated calls.
Morris Water Maze currently covers single-trial analysis only, and its measures remain validation-only. It does not currently compute multi-trial learning curves or cross-session group comparisons. The escape-latency, path-length and quadrant calculations described on this page are deterministic per trial (Section 3); this page does not extend that to any claim about learning-curve or multi-trial reproducibility, because RodentMetrics does not currently perform that kind of analysis at all.
Reproducibility, in the bounded sense this page describes, is a separate question from experimental status. A metric can be reproducible under a fixed configuration and still be experimental, validation-only, or limited to single-trial analysis.
7. Reproducibility is not validation
This page deliberately keeps two different questions apart:
| Question | What it asks |
|---|---|
| Repeatability | Given the same input and the same configuration, does the software report the same output? |
| Validation | Does that output accurately correspond to the animal's actual behavior, or to a trusted independent reference measurement? |
This page provides evidence only for the first question. RodentMetrics' internal repeatability observations -- however consistent -- do not, on their own, establish that its reported metrics are scientifically accurate, that they agree with manual scoring or another tracking system, or that they have been independently validated. RodentMetrics has not been independently validated against a reference standard, and this page makes no such claim.
A researcher's own validation of RodentMetrics' outputs against their protocol, manual review, or a reference method remains a separate and necessary step, not something this page's repeatability evidence substitutes for.
8. Reporting for reproducible analysis
To let another researcher -- or a later version of yourself -- reproduce a specific reported result, RodentMetrics recommends recording:
- The assay analyzed
- The RodentMetrics software version used
- Source-video identifying details (filename, duration, resolution, frame rate)
- The analysis interval (start time and duration)
- Arena and assay-specific zone/region geometry
- Assay-specific configuration and thresholds used
- Tracking/quality review notes for the result
- Validation status of pose-derived or single-trial measures, where applicable (Novel Object Recognition; Morris Water Maze)