How activity date picking works: almanac rules, personal layer, explainable scores
How Astralium activity date picking works: pick an activity, score the month, expand almanac and personal evidence. Framework and testing published—not weights or mapping tables. Day-level only, not hour timing; not professional advice.
At a glance
- Activity timing returns a match score plus expandable evidence—not preset lucky/unlucky calls or fortune forecasts.
- Scoring splits into an almanac layer (same for everyone) and an optional personal layer (birth profile). Evidence is collected first; the score follows.
- Day pillars use the local calendar date at noon—day-level only, not hour-level timing.
- We publish the framework and how we test it; weights and mapping tables stay private rule assets.
Many people use a daily almanac; many apps paint days green or red without showing why. This article explains how Astralium activity timing works—for credibility, not marketing.
Why spell out the method
Almanac apps usually list today's yi/ji for everyone—they do not answer "I am moving house—which day fits better?" Other tools color days without a rule chain or a way to compare candidates.
Astralium's stance matches our charting products: checkable structure, no in-app verdicts. Date picking returns a 0–100 match score with evidence you can expand per day. The score helps you rank days in a month—it is not a promise that things will go well.
We publish the framework, evidence types, boundaries, and testing approach—not every weight, cap, or activity-to-tong-sheng table. Every evidence line in the tool can still be checked against that day's almanac text.
What we compute (inputs and outputs)
Treat activity timing as a repeatable decision-support flow, not fortune telling:
- Inputs: calendar date + activity (15 types—wedding, moving, contract, travel, and more) + optional birth profile
- Outputs: match score + layered evidence (almanac vs personal listed separately)
- Not outputs: horoscope reports, lucky/unlucky verdicts, "you must / must not" lines, or AI chart packets
Same date, activity, profile, and rule-set version → the same result, computed on the client without a black-box API. That aligns with what you want from any checkable system: repeatability and an audit trail.
Activity first, then the month
You pick what you are doing; the calendar scores the whole month. Tong sheng labels like wedding or opening a shop map internally—user-facing copy uses modern scenarios (marriage, moving in, signing, travel, and so on).
Two modes:
- General almanac: tong sheng, jian chu, and other layers identical for everyone
- Tailored for you: personal layer on top when you are signed in with a saved birth profile (entitlements as shown in the product)
The heatmap ranks days within the month; a compare basket lines up candidates side by side. Color blends monthly rank with an absolute score floor so a weak month leader does not look like a universal best day. Exact thresholds are rule details we do not publish here.
Calendar base: fix the day before you score it
Step one is not scoring—it is turning a Gregorian date into structured fields: lunar day, day stem-branch, jian chu, tong sheng yi/ji, Peng Zu, and related tags. The calendar layer shares conventions with our BaZi flow-day logic so date picking and charting do not drift apart.
- Local timezone defines "today" and month boundaries—not a UTC cut
- Day pillar anchors at local noon on that calendar date—day-level only, not hour-level timing
If you cannot verify the day pillar, match scores are meaningless. We lock calendar fields to traditional tong sheng sources before stacking rules.
Almanac layer: shared evidence for everyone
The almanac layer asks: under almanac rules, what structural factors apply to this activity on this day? The pipeline:
collect evidence → aggregate by layer → blend → apply structural caps
Evidence types today (categories only—no point values):
- Jian chu (十二建除): adjusted per activity; some activities override the default table (moving, travel, contracts, and others)
- Tong sheng yi/ji: day yi/ji matched to activity categories; primary vs secondary hits; when yi and ji conflict, net effect is constrained—not naïvely summed
- Peng Zu (彭祖百忌): matched by day stem/branch characters tied to the activity—not by substring-searching the classical couplet text (a common way to ship rules that never fire)
- Sui po / yue po caps: separate from everyday deltas so the same factor is not punished twice
Each activity has a mapping confidence tier. Low-confidence mappings weaken or disable yi/ji. The general-reference activity deliberately avoids a catch-all yi/ji map and caps how high the score can climb—it is not "good for everything."
Example phrasing (illustrative, not a real score): for marriage, a day might show "tong sheng yi: wedding (primary hit)", "jian chu: success day", "Peng Zu: no rule fired for this activity"—open the day sheet for the full list and check the almanac text yourself.
Personal layer: same almanac, extra structural checks
With tailored mode and a saved birth profile, a personal layer stacks on the almanac. It asks whether this person adds structural friction or support on top of the shared almanac—not what fate "will" do.
Categories in the personal layer today:
- Flow day vs natal branches: clash, punishment, combination—shown as structure lines (e.g. flow day clashes natal day branch), not "bad year for you"
- Flow-day shen sha: a short whitelist tied to the activity—not every star in the book
- Almanac × personal crosses: small adjustments such as ji day plus clash
- Personal caps: hard or conditional ceilings so one weak personal signal cannot silently override a strong almanac read
When birth time is uncertain, shen-sha evidence that depends on the hour pillar is discounted; clash/combine/punishment from day and year branches is less sensitive to hour error. Discount factors are private rule assets.
Almanac and personal layers aggregate separately, then blend, with a significance gate so personalization is neither cosmetic nor dominated by noise.
From evidence to score: blend, caps, display
Evidence always comes before the number. Principles we can state publicly:
- Almanac and personal layers start independently, then blend with activity-specific weights
- Structural caps apply after blending; caps and deltas are designed separately
- 0–100 maps to neutral band labels—no lucky/unlucky / must-not wording in the UI
- Rule sets are versioned (full almanac vs lite fallback) so changes are traceable
Black-box coloring vs an explainable flow:
| Dimension | Typical black-box app | Astralium activity timing |
|---|---|---|
| Rationale | Color or one line | Expandable evidence, almanac / personal split |
| Activity | Often detached from yi/ji | Pick activity, then score |
| Repeatability | Hard to verify or version | Same inputs + version → same output |
| Wording | Great luck / disaster | Match score + neutral bands |
Quality checks—not hand-painted colors
Date-picker rules have dedicated automated verification (verify:date-picker) in dev and release workflows. Examples in plain language:
- Every activity with a Peng Zu rule must fire at least once in a calendar year—guards against dead rules
- Reachability scans between yi/ji mapping categories and real tong sheng terms—guards against typos and orphan categories
- Heatmap colors and screen-reader tiers read the same effective score—no green cells that announce as low
This parallels golden-sample regression for chart engines: rule change → batch recompute → compare to baseline. Date picking verifies scores and evidence chains—not lucky/unlucky timing predictions.
What we deliberately do not do
- Hour-level timing (a separate product question if we ever ship it)
- Multi-person picking (e.g. both partners for a wedding)
- House facing, san sha direction, master-of-house selection, and other unmodeled factors
- AI chart packets or lines like "good to sign / bad to travel"
- Legal, medical, or financial advice—health and legal activities need real-world judgment
- No "guaranteed success"—match scores narrow candidates; you decide
Credibility is also clear limits—and your willingness to open evidence and check it against the almanac. For why we do not interpret charts in-app, see why we don't interpret your chart.