How Astralium keeps charts accurate: golden samples and regression

Chart structure, not the reading

Charting sites run into two different problems that often get talked about as one: charts that are miscalculated—wrong hour, wrong calendar conversion, star positions that don't line up—and readings that are disputed—different traditions, useful gods, timing methods. Mixing the two makes any claim of accuracy meaningless. Our automated checks address only the first: for a fixed birth or cast input, the engine's structural output must stay stable and reviewable.

Specifically:

• Calendar math: Gregorian/lunar, solar terms, zi-hour day change, true solar time, time zones, daylight saving
• Chart structure: palaces, stars, stems/branches, six relations, shi/ying, Qi Men nine palaces
• Packet shape: the AI-copyable text stays aligned with the on-screen chart

Outside that scope: preset lucky/unlucky judgments, useful-god picks, timing calls, auto-written readings, and any “was this reading right” test. See our philosophy and why we don't interpret.

Golden sample → recompute → locked baseline → mismatch holds the release

Fixed input (golden sample) → engine recompute → compare to locked baseline → mismatch holds the release

Golden samples: for each discipline, we keep deliberately chosen birth or cast inputs plus expected output snapshots—calendar math, palaces/stars, stems/branches, and similar structural fields. They hit edges that break easily: zi-hour day boundaries, true solar hour shifts, solar-term handoffs, time zones, daylight saving, Jyotish nakshatra borders.

Automated regression: whenever a calculation rule or piece of logic changes, the full golden-sample set reruns. If output drifts from the locked baseline, the change does not ship until we know why, update the baseline, and document what changed. Same principle as our online charting picker post: rule changes must be checkable and documented.

Pre-release checks go beyond the wheel. The packet—the text you copy for AI—is held to the same standard in regression: field completeness, stable sections, the same facts as on screen. Packet format changes follow the same rule as engine changes: compare against fixed samples, update with a record. We don't publish the full template, but you can copy the same birth and compare side by side—see our AI-readable chart packet post and Part V below.

Why we don't publish the full golden-sample library: the full set and line-by-line rule tables are internal regression assets—publishing them hands over implementation detail and still wouldn't replace checking your own chart. Most charting sites keep this internal for the same reason—that alone isn't evidence either way. Spot-check evidence we do publish: three reproducible public cases (Part III), documented conventions (Part IV), and a three-step self-check (Part V).

Three cases you can recast right now

These three use public birth dates and places—no real user data. Pick one, enter the same details here, and check whether the fields below match. Further reading: true solar time; Zi Wei chart vs birthday-only AI: accuracy comparison.

Which calculation rules each discipline uses

After you cast, the packet states which convention was used. If your practitioner follows a different rule set, treat theirs as authoritative—but you should still know which one this site assumes, so you can tell whether a difference is a matter of method or an actual error. When we change a convention, we note it in product release notes, rerun golden samples, and update the locked baseline.

• BaZi: default midnight rule—day pillar changes at 23:00; true solar time is longitude correction only, no equation of time
• Zi Wei: longitude and equation of time
• Qi Men: default chai bu; zhi run optional in the tool, labeled in the packet
• Jyotish: Lahiri (Chitrapaksha) ayanamsa at birth time; whole-sign houses; mean Rahu/Ketu
• Western: tropical zodiac; house system stated in the packet and settings

Runtime basis, regression reference, and cross-checks per discipline:

How to check a chart on your own

Read the wheel and rule notes in the packet → copy the full text and check completeness → cast the same birth elsewhere and compare side by side

1. Wheel and rule notes: the packet should state true solar time, the zi-hour rule, the Qi Men ju method, and similar. If your lineage differs, use your own convention
2. Copy the packet: check field completeness and whether it matches the screen—gaps or drift are usually obvious
3. Side by side: cast the same birth on another site and paste both texts; or try the three public cases in Part III. Whether a gap comes from the input or the charting is usually easy to tell

These three steps work on any charting site, not just this one. Golden-sample regression is how we maintain structural consistency; whether to trust a chart still comes down to checking it yourself—don't take “most accurate” at face value, and don't dismiss a site just because it keeps its full golden-sample library private.

What “42 automated checks” actually means

You may see “42 automated checks” on the homepage or product pages. That is the current total of fixed comparison items we maintain—chart output, copyable packets, true solar time, pattern flags, and other steps that affect what you see. Roughly three groups:

• Chart output: golden samples per discipline (see table below), including tricky hours and edge inputs. After a rule change, they rerun in regression and must match the locked baseline
• Copyable packet: the text you copy for AI is checked on its own in regression—field completeness, stable sections, alignment with the on-screen wheel
• Settings and derived fields: true solar time, star brightness, pattern labels, and similar—so the page doesn't drift from what was actually computed

The number grows as we add checks. It reflects regression coverage we maintain—not a guarantee on any one chart. Use Part V to check your own.

What's worth checking when you compare charting sites

Many charting sites are known for stacking disciplines side by side—Zi Wei, BaZi, astrology, Liu Yao, all in one place. How many disciplines a site offers and how carefully it maintains chart data are two different things. When comparing any site, including this one, two dimensions matter: whether the wheel follows a stated convention, and whether the text you copy away matches what's on screen. The table below lists common patterns on both.

The packet is worth checking on its own too—after casting, the structured text you copy should match the wheel; missing fields or drift usually show up quickly once you look. You can check this yourself: cast the same birth on another site and paste both copyable outputs side by side. What the packet includes: AI-readable chart packet. How to pick a site type: online charting picker.

Adding more disciplines and maintaining chart data quality are two separate jobs.

What this checking doesn't prove

A few things sit outside this checking entirely:

• Interpretation: even when structure checks pass, traditions, themes, and timing still depend on lineage and the reader. We don't ship preset lucky/unlucky judgments.
• Lineage conventions: early/late zi hour, Qi Men chai bu vs zhi run, house systems, ayanamsa choice—traditions don't fully agree on these. We document our convention and run golden-sample regression against it. If your tradition differs, treat yours as authoritative.
• Input accuracy: a wrong birth time, bad coordinates, or a mis-recorded coin throw—the engine reflects what was entered, not unstated intent.
• Test completeness: regression catches drift from a change early; it is standard engineering practice, not proof of zero errors. Fixes, new samples, and baseline updates are routine when something is off.
• AI reading: whether a model invents details or mixes systems is a usage question. See our accuracy comparison and full workflow.

What this means for you

When we say a chart “passes verification,” we mean structure matches the locked baseline, copyable text, and a documented convention—not that the reading is correct, and not a guarantee of your birth hour. Automated regression shows we maintain structural consistency seriously; whether to trust your chart still comes down to Part V or the public cases in Part III.

Astralium provides charting and structured export; themes, timing, and lucky/unlucky calls are for you, AI, or a practitioner you trust. Whether this fits your needs is something you decide from what you check, not from this post.