What changed in Lumin

A birth_datetime sent with a zone suffix (+06:00) was read as a true instant and then moved by utc_offset_minutes again, six hours early in that case. A suffix that agrees with the offset, within half a minute, is now stripped; one that disagrees is refused with both values named. Z keeps its meaning.

A MIXED_ACTIVE cusp is undecided in every place a reader meets it: analyze_natal_promise, find_event_timing_v2, get_kp_school_spread, event timing v1 and the agent's own prompt. None of them calls it promised or a settled sequence any more.

• The horary Moon withhold is a successor's gate (Bhargava, and K. Subramaniam's Astro Secrets II), not KSK's: KSK takes the number as given.
• The horary, career and medical horary tools carry gateReadings: the Moon gate, the Moon or Ascendant sub lord gate, and KSK's number as given, side by side.
• When the gate withholds, career and medical horary also return judgedAsGiven, KSK's judgement of the same chart, and the withhold names its remedies. The WITHHELD verdict itself is unchanged.

get_yoga_karaka now reads the yoga karaka for each lagna from BPHS ch. 34's own verses, which name a pair for Virgo.

The CSL verdict no longer promotes a cuspal sub lord that rules its own cusp from PARTIALLY_ACTIVE to ACTIVE at 0.67 coverage. No source states that rule; it decided 2 of 3,744 verdicts across the test spread.

get_annotated_dasha lists every event group a period lord reaches, where it used to drop those under 34% without saying so, and judges isCurrent at query_date instead of the server's clock.

get_transit_timing_hierarchy reads a start_date without a zone by one rule, starts from the moment of judgement the call names, and picks the event from the dasha running then.

Usage rows record the page read, engine time, whether the moment of judgement was defaulted, and repeated calls within ten minutes; no input value enters a row.

In the app, Compare and Conversations load in their own shape, and opening a saved conversation no longer clears the one on screen first.

• The Moon is computed from ELP/MPP02: 0.154 arcseconds from JPL DE441, down from 5.71. The planets are computed from VSOP2013 with the reduction Horizons applies: 0.09 arcseconds, down from 0.40.
• Lahiri is built from its published definition and nutation is no longer counted twice. Sidereal time reads UT1, Delta T follows JPL, and sunrise, lunations, ingresses and stations are checked against JPL Horizons, Skyfield and USNO.
• Most positions move by less than an arcsecond. A planet or cusp sitting on a sub or sub-sub boundary can change its lord, and with it a verdict.

• Tools that returned several separate calculations in one response are split along their method seams. Each step is its own tool returning one whole calculation, and the parent names its steps in methodSeams and still returns every block, unchanged.
• Twelve tools that run steps are renamed as panels. Each old name works for this release, returns the same result and says which name to use. The release after this one removes them.
• Long lists are paged and never cut short. A tool that has more rows than one response holds says how many pages remain.

• Birth offsets are worked out for the birth date, not today, and every chart tool takes time_zone so the server reads the offset in force on that date from the tz database.
• Four tools now carry the system they compute: get_hora_today is KP, get_bhrigu_bindu is a Lumin extension, get_mantra_recommendation is Parashari, and run_pre_verdict_audit is cross-system.
• The reading protocol states the rules the engine applies.
• Oversized input is refused with a clear message, and every loop a caller can steer is bounded.

• A horary question is judged at a moment and a place, and those used to travel in fields called the birth time and the birth place. 17 tools now take them as judgement_datetime, judgement_latitude, judgement_longitude, and judgement_utc_offset_minutes or judgement_time_zone: get_horary_chart_v2, get_lost_or_missing, the four lost-article tools, get_arrival_timing, get_horary_number_methods, get_horary_ruling_planets_variants, get_horary_validation_panel, get_horary_advanced, get_horary_number_intuition, get_horary_natal_crosscheck, get_career_horary, get_medical_horary, get_rp_timing and get_horary_serial (on each of its sessions).
• Tools that read no birth no longer ask for one: the birth fields are optional on the horary tools that cast at the moment of judgement, and the call is refused with a message naming what is missing if neither set is sent.
• Every result carries horaryMoment: the instant, the local clock with its offset, the place, where each part came from (source, datetimeDefaultedToNow, placeDefaultedToBirth), and utcOffsetSource, which says given, time-zone, birth-offset or local-mean-time-from-longitude. An offset is never local mean time without saying so.
• get_ruling_planets_at_judgement, get_fruitful_significators, get_four_step_fruitful_ranking, get_transit_timing_hierarchy and find_event_timing_v2 take judgement_time_zone, so the offset in force on that date is read from the tz database, daylight saving included.
• get_event_dasha, get_promotion_verdict and get_yoga_karaka take query_date.

• set_birth_profile returns judgementInstant { utc, local, utcOffsetMinutes, utcOffsetSource, timeZone, place, send, note }: the instant the call was made, as UTC and as the clock at the chart's place with its offset. The plan tells the model to pass it as judgement_datetime with judgement_utc_offset_minutes on every timing and horary call, so the fruitful set cannot change between two calls of one reading. You can still name another moment.
• It also returns structuredContent { plan, tier, floor, target, tools, waves, judgementInstant } with an outputSchema, so a client reads the waves as data. The text block is unchanged. plan carries the whole plan text, because a client that reads structuredContent shows the model only that.
• The judgement block of the five timing tools gains holdsUntil: the next instant any of the five ruling planets changes (the Ascendant's or the Moon's sign lord or star lord, or the day lord at the next sunrise at the place), solved and never a score. It names the member with its old and new lord, the ranked set after, and each member's own next change. It is the same ruling planet rule read forward in time. local carries the offset in force then only when judgement_time_zone is sent; otherwise it carries the judgement's own offset and utcOffsetSource says which.
• The routed plan no longer sends a tool that cannot run on the question. get_career_horary is planned once the horary topic matched, a two-chart tool only on a two-person question, a horary question lists its chart and every tool that reads its number before the natal spine, and the list follows the taxonomy's prerequisites so it agrees with the dispatch waves. No tool left the catalogue.

• A query_date sent without a zone is read as UTC, whatever the server's own zone, and the result echoes queryDateUsed { instant, input, read }. Before, it was read in the server's zone. Send it with its zone (Z or +05:30).
• A wall clock that a daylight saving change skipped (New York, 2026-03-08 02:30) is read with the offset before the change and labelled wallClockSkipped, with a note. One that happened twice (2026-11-01 01:30) keeps the earlier reading and is labelled wallClockRepeated. Neither is refused. A typed offset beside the zone in a skipped hour may name either reading.
• The recordings the engine tests replay, and the golden payload capture, no longer read the wall clock, so a recording does not go red on the day a running period changes.

• find_event_timing_v2 now judges the promise with the same gate as analyze_natal_promise, so the two tools no longer disagree on a chart. isPromised is true only for ACTIVE and PARTIALLY_ACTIVE; a MIXED_ACTIVE promise is undecided, not promised. The old meaning, that the required link is present, is kept as promiseDetails.requiredLinkPresent. Each result adds verdictFrom, promiseState, mixedResult, denialHousesSignified and gateNote. A custom house_group led by a house outside 1 to 12 is refused.
• The ruling-planet confirmations read the real promise of each row: every row carries promise_state, and only promised rows count toward a confirmation. The ruling-planets panel, the strength hierarchy and the dasha sync read the ruling planets at the moment of judgement, with the birth set beside them as atBirth.
• get_event_dasha keeps every period. The cut that hid periods below 34 percent coverage is gone; a long list is paged and nothing is dropped. The mahadasha lords that signify no required house are named in unlistedMahadashas.
• get_smart_current_dasha reads the nodes as KP does (isRetrogradeKP, beside astronomicalRetrograde) and takes query_date. get_career_blockage_diagnosis reads "connected to 1 and 10 only" as Astro Secrets p.198 prints it, adds connectedTo1 and connectedTo10, and takes query_date.

• When a question needs something only you can give, the plan says so first. A horary question with no number, or a compatibility question with one chart, gets an ASK FIRST line before the tool list, and a two-chart tool is never planned until both charts are given.
• The transit tools read the running dasha at the transit moment, not at the time of the call. A query_datetime without a zone is UTC, the result echoes queryDateUsed, and a query_datetime that is not ISO 8601 is refused with a message instead of being guessed at.
• In a repeated hour (New York, 2026-11-01 01:30) a typed offset beside the time zone now selects either reading, and the result is labelled wallClockRepeated, on the birth and the judgement paths alike.

• Vimshottari dates were computed from the local clock time read as if it were UTC. They now start at the true birth instant; before, they were off on almost every chart, by up to nine hours.
• The first mahadasha divided its birth balance among all nine bhuktis, so every chart was born in the dasha lord's own bhukti. It is now the full dasa clipped at birth, as Reader 1 states.
• The ruling-planet day lord changed at midnight. It now runs sunrise to sunrise at the place of judgement, and the civil-day lord is reported beside it.
• Dasha boundary dates are printed as UTC ISO dates everywhere, so no answer depends on the server's time zone.

• A horary chart's ruling planets took their Ascendant from the number's cusp. They now follow KSK, the sign rising at the moment and place of judgement, and the older definition ships beside it as a labelled later-school reading. The ruling-planet set changes on most horary casts.
• Fruitful significators now use the ruling planets at the moment of judgement, which defaults to now. The birth-moment set is kept as a labelled field.
• Both horary tools that ignored the number now take the Ascendant from it, get_rp_interval takes a place, and horary number 125 no longer fails.
• Horary numbers carry four stated methods of reading them (KSK, the Gulbarga short method, and two later tests), none of them gating the answer.

• Sub-sub lords in the six subs that cross a sign line are corrected, the Raman and Khullar ayanamsas compute their own values, and sub-boundary distances are measured exactly.
• KP reads the nodes as never retrograde, as KSK states; the astronomical flag is reported beside it. Double retrograde is marked disputed, with both positions quoted.
• Timing candidates carry the KSK sub-lord judgement and two stated orderings. Ruling planets carry a retrograde screen and their node rules as stated variants.
• Two-chart tools now use the first chart's ayanamsa for the partner, and the pre-verdict audit no longer lowers confidence for combustion, planetary war or vargottama.

• Parashari rules were read verse by verse in the Santhanam and Sharma editions of BPHS, with the Sanskrit where the translations differ.
• Corrected: two ashtakavarga rows, the D-60 count, D-9 rounding, Ashta Koota scoring, Raja, Dhana, Gajakesari and Lakshmi yogas, Budh-Aditya, Parivartana, ekadhipatya, Ashtottari and Dwisaptati, the special aspects and Bhava Dig bala.
• Where editions or schools differ (D-60 and D-10, Gajakesari, twilight, the Saptavarga weights, the aspects of the nodes) every reading ships labelled, and the older rule stays as a labelled legacy reading.
• The dasha lord's facts and the MD-AD relationship ship under each stated text, with no single good or bad verdict.

• Jaimini: karaka ties are judged to the arc-minute, both seven and eight karaka schemes ship, Chara dasha ships one timeline per lineage with antardashas, and houses are whole-sign.
• Tajik: the mudda dasha starts from the right lord with a balance, every year of life gets its own annual chart, and the yogas follow the Tajika Neelakanthi.
• Cross-system composites label each block with its system, and get_paradigm_panel now compares timing across systems.

18 new tools. get_birth_time_spread shows which verdicts change inside a birth-time window you give it. Kalachakra dasha ships under three stated methods. get_kp_school_spread sets KSK beside the later KP schools with no score. Several larger KP tools gained focused tools for one method each, and their original responses are unchanged.

• The planets sit within 0.40 arcseconds of JPL Horizons DE441 across 1900 to 2050, checked at fourteen instants. Before 11 September they sat up to 580 arcseconds off, from truncated tables, no light-time correction and the Sun's aberration applied to every planet. No earlier test could see it, because each looked at one date.
• The true lunar node was a five-term approximation, up to 12 arcminutes from the real osculating node. It is now solved from the engine's own Moon, within 31 arcseconds of Swiss Ephemeris. The mean node, which is the default and KSK's practice, was right all along.
• All twelve Placidus cusps and the ARMC agree with Swiss Ephemeris to 0.05 arcseconds. Eclipses are solved, not screened: every one of the 50 in NASA's 2000 to 2010 canon classifies with its type, and the eclipse tool now returns local circumstances.
• A planet or cusp sitting within a few arcseconds of a sub or sub-sub boundary can change lord after this. Readings made before 11 September may differ at that level.

True Chitra is computed from Spica now. Until 13 September it returned Lahiri's value, so the engine offered six ayanamsas and computed five. Spica's apparent place is checked against Swiss Ephemeris at five epochs to 0.32 arcseconds, and True Chitra runs 18 to 64 arcseconds below Lahiri across the era. If you chose true_chitra, your positions move by that much.

• get_frame_spread builds the same chart under every ayanamsa and names exactly which positions change lord and which hold. The frames disagree by a fixed 5.6 arcminutes, so the set that moves is small: about 6% of cusps and planets at the sub level, 56% at the sub-sub level, and roughly half of charts are frame-independent at the sub level and can say so.
• get_paradigm_panel reads one matter through all seven systems as parallel arguments, never as a vote. Pass frames and it says which verdicts change with the zero point alone. Measured over 120 panels, one in five has a verdict that does.
• A conclusion that changes with the ayanamsa is a property of the frame, not of the chart, so it ships as both readings with the frames named. Nothing here is scored, averaged or ranked.

• A denial house never denies on its own. DENIED means the cuspal sub lord reaches none of the required houses. 22.8% of verdicts left DENIED when this was corrected, and none entered it.
• Every event rule now carries the page it rests on. 16 of the 78 rows shipped a house group no page states and were corrected from the books; two had denied on a house the corpus requires. 12% of verdicts moved.
• Dignity, retrogression and combustion are reported and gate nothing, because KSK rules all three out by name for natal work. Retrogression carries forward as a delay signal to the timing phase, where KP does use it, and the delay's lift date is returned.
• A horary number fixes the Ascendant at the commencement of its sub, and the other eleven cusps are solved back from it. A rounding error had put 77 of the 249 numbers one sub lord off.

Every claim names its source. A response carries the rule it applied, with the text and the locus, in one of eight source classes from corpus (the page is quoted) down to unattributed (nobody has looked). 51 of the 78 event rules quote their page; the 19 that cannot say so on the wire. Verdict scores that are conventions, not book values, are labelled as such.

18 new tools, grouped by lineage. Every non-KP tool says which system is answering in its first line.

• Tajik has its own judgement mechanism now: the sixteen yogas with their orb table, panchavargeeya bala out of 80, harsha bala out of 20, the hadda table, and the mudda and patyayini dashas.
• The year lord's strength tie-break runs, so 35.5% of years stop being undecided, and where the text leaves a rung open both readings are returned.
• get_tajik_variant_spread does for Tajik's strength systems what get_frame_spread does for the ayanamsa: it says which conclusions move between the variants the sources leave open.

• Shadbala is the BPHS procedure in virupas, and it says so where it disagrees with the index it sits beside.
• Vimsopaka scores all sixteen vargas on the classical ladder.
• Ashtottari and Yogini started the wrong lord and carried no balance at birth. Both are rebuilt on the verses.
• Rashi drishti was inverted for eight of twelve signs. The negative yogas panel was missing a row. get_multi_system_verdict's custom house_group path never worked. All fixed.

Nothing is deleted on the wire. Every response compactor is now under a recorded-payload test and may drop only what it declares. Five were found deleting claims on the way to you, among them the ruling planet roles, a tiebreak rank and the promise reasoning. All five are fixed.

The server speaks MCP protocol revision 2026-07-28 and still answers 2025-11-25 clients from the same handler, so nothing has to be cut over on a date.

• A developer portal at developer.lumin.guru: up to five named API keys, a monthly meter, usage analytics by day and by tool, and the call pack ledger.
• The free allowance is metered per account across every key you have held, revoked ones included, so a new key never means more free calls. Failed calls are free.
• Call packs come in any amount, starting at $1 for 400 calls. A key you delete stays deleted.

• The Claude Code plugin is at 0.7.0. Four writing registers choose how a finished reading is written: consulting (the default, the answer the way an astrologer says it across a table), practitioner (the full written verdict with every provenance field shown), teaching, and report. Each changes only how the reading is written, never what is computed, and the plugin fails its own validation if that sentence goes.
• A kp-build skill for building a product on the server: input classes, the call shape, metering, paging and the copy rules.

The benchmarks page at lumin.guru/benchmarks shows every figure on the site against the oracle it was checked with, and what is still open. The public lumin-evals repository carries the same sky fixtures, the scorecard and the findings index.