Technical protocol · 2026
Tian Mira Benchmark 2026.1: a technical protocol for comparing BaZi engines
Twenty pseudonymized cases, a documented technical score, and a private external pilot with no automatic publication.
Introduction
Two BaZi calculators can diverge for several reasons: time normalization, historical timezone, true solar time, year and month boundaries, hour pillar calculation, school conventions, derived structures, and Five Element weighting.
The Tian Mira Benchmark 2026.1 is an attempt to make these differences observable, documented, comparable, and reproducible. It does not seek to impose a single school.
1. Why create this benchmark?
- •Different results without explanation: two BaZi engines produce different pillars or percentages for the same birth date, without the user understanding why.
- •Confusion between technical calculation and interpretation: an engine may display a "stronger" or "weaker" Day Master reading without documenting the coefficients used.
- •Absence of displayed conventions: many calculators do not specify their default timezone, solar time handling, or weighting method.
- •Percentages presented as absolute: elemental percentages are sometimes displayed without error margins, warnings, or methodological context.
- •Inability to distinguish an error from a methodological difference: without a common protocol, a divergence may be misinterpreted as a calculation fault.
2. What the protocol contains
- •20 real pseudonymized cases: no names, no direct identifiers; birth data has been transformed into structured JSON.
- •Variety of technical situations: different timezones, hour boundaries, leap years, hours without an obvious earthly branch.
- •Reference answers kept separately: Tian Mira's calculations are not shared with the participant.
- •Results submitted as JSON data: a single submission.json file, with no program or archive.
- •Automated comparison after validation: the submission is checked (size, structure, security) before any comparison.
3. Main technical score
| Catégorie / Category / 类别 | Points |
|---|---|
| Inputs and normalization | 10 |
| True solar time | 15 |
| Four Pillars | 40 |
| Derived structures | 25 |
| Completeness and transparency | 10 |
| Total | 100 |
Inputs and normalization — 10 points
Date, time, timezone, coordinates, and data normalization.
True solar time — 15 points
Longitude, historical timezone, solar correction, and the time actually used.
Four Pillars — 40 points
Year, month, day, and hour pillars. This is the core of BaZi calculation.
Derived structures — 25 points
Day Master, hidden stems, Ten Gods, Na Yin, interactions, cycles or luck pillars, depending on supported fields.
Completeness and transparency — 10 points
The engine must clearly declare calculated, unknown, not calculated, unsupported fields, missing data, and warnings.
4. What does not automatically change the main score
- •Five Element percentages: the Wood, Fire, Earth, Metal, Water distribution depends on the chosen weighting.
- •Day Master strength: the "strong", "weak", or "balanced" assessment may vary by school.
- •Roots and support/pressure ratio: their interpretation is tied to the method's conventions.
- •Favorable elements and confidence level: these fields reflect analytical choices, not standardized calculations.
- •Narrative interpretation: it is not compared within this technical protocol.
- •A difference in these fields should therefore not be automatically turned into a technical error.
5. Technical difference or methodological difference
Technical difference
Two engines claim to use the same conventions but produce different day pillars. This type of divergence is tracked and counts toward the score.
Methodological difference
Two engines calculate the same Four Pillars but use different weighting for the Five Elements. This divergence is documented but does not penalize the main technical score.
6. Privacy and security
- •Cases contain pseudonymized birth data.
- •Names and direct identifiers have been removed.
- •Indirect re-identification may remain theoretically possible.
- •Any attempt at re-identification is strictly prohibited.
- •The package is not public and must not be redistributed.
- •Tian Mira's answers are not provided to the participant.
- •The submission is a single JSON file.
- •No received program or script is executed.
- •Size limits and JSON complexity limits are enforced.
- •No publication is automatic.
- •Any potential public report requires the participant's prior consent.
7. Pilot process
- 1Private invitation
- 2Participant agreement (signed form)
- 3Transmission of the pilot package
- 4Execution of the 20 cases by the participant's engine
- 5Return of a single submission.json file
- 6Security and structure validation
- 7Private comparison with reference answers
- 8Private report delivered to the participant
- 9Possible aggregated publication with explicit consent
8. Status at time of publication
The technical pilot infrastructure is ready. No real external engine has yet been evaluated in this framework. No independent results are therefore published to date. The project is now entering its phase of seeking a first participant.
9. Limitations
- •Only 20 cases: the protocol does not claim to cover every possible situation.
- •No absolute BaZi truth: the benchmark compares calculations, not metaphysical "truth".
- •Protocol centered on documented choices: it does not evaluate the quality of human interpretation.
- •Possible school conventions: differences between legitimate schools are not counted as errors.
- •No generalization to all engines: a given engine's results are only valid for that engine, in that version, with those parameters.
- •No scientific certification: the protocol does not issue any label or certification.
- •No measurement of human interpretation quality: only technical calculation is observed.
- •No claim that Tian Mira is superior: Tian Mira's answers serve as a documented technical reference, not as an absolute verdict.
10. Invitation to participate
Do you develop or operate a BaZi engine and wish to participate in a first private technical pilot? The initial pilot batch uses 20 pseudonymized cases, before progressive expansion of the benchmark. Participation does not entail automatic publication, and any potential public result remains subject to your prior consent.
FAQ
Does the benchmark claim Tian Mira is the best engine?
No. Tian Mira does not claim to be the best BaZi engine. The benchmark compares technical consistency, not superiority.
Is this a certification?
No. The protocol does not constitute an official certification and does not issue any label.
Are the 20 cases public?
No. The cases are reserved for the private pilot and are not publicly downloadable.
Are the cases anonymous?
They are pseudonymized, not declared anonymous. Indirect re-identification remains theoretically possible.
Is a percentage difference automatically an error?
No. Methodological differences are separated from the main technical score and are tracked without penalty.
Can a participant refuse publication?
Yes. Publication is disabled by default. No results are published without the participant's explicit consent.
Has any external engine been tested yet?
No. The protocol and pilot are ready, but no real external evaluation has yet been published.
Citation
Tian Mira. "Tian Mira Benchmark 2026.1: a technical protocol for comparing BaZi engines." Tian Mira, 2026.
This citation refers to the public protocol documentation. It does not constitute a citation of independent results or a public dataset.
Conclusion
The Tian Mira Benchmark 2026.1 provides a documented framework for comparing the technical consistency of BaZi engines. It does not claim to establish absolute truth, does not certify any engine, and publishes nothing without consent. The pilot is technically ready; no external evaluation has yet been performed.