BUG-2234 (astrologer 2026-11-22 section 5, owner approved): migration reads houses
3/4/9/12, their lords, dasha and transits with D4 as a residence aid; D12 is the
parents chart. Education reads D1 houses 4/5/9 and their lords plus D24, not D5.
Registry, plan contract, evidence card spec, checklist, card facts, orchestrator
dispatch and the two _ROUTE_REQUIRED_LAYERS rows change together.
BUG-2235: the repair call added optional layers at about three times the main
answer's error rate and appended 「在这件事上是中性的」 lines on follow-ups. Now:
follow-ups get no layer gap-fill; first turns only get the domain's must-mention
layers that are truly absent, with full rows; added text that is neutral, not
asked for, or fails the fact guard / table checks is dropped; the writing rules
no longer ask for neutral lines and the clean-up removes them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXp2Vp62Ur3C7yRWQdSq
The chat is charged its chat.standard feature price; the plan's literal ceiling
of 1 on the model's own per-run points only stopped that from being tuned.
Positive whole numbers are still required. No stored price is changed. The
launch page estimate for chat.standard is the 2026-11-23 measured first-turn
average (about 176k input, 22.7k output tokens incl. the repair call).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXp2Vp62Ur3C7yRWQdSq
A correct 「2029 年 7 月 17 日土星进金牛,走到本命月亮上」 was compared with the
reference day's Saturn and rewritten wrong. Saturn-on-Moon, Jupiter / Saturn /
node sign and house claims and double-transit house lists are now checked for
the days the sentence (or its bullet) names; no data for that day leaves the
sentence alone; undated claims keep the reference-day reading (BUG-2033), and
BUG-2228's 「经过第 N 宫」 check uses the dated span when the clause has one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXp2Vp62Ur3C7yRWQdSq
Owner 2026-11-23: the 11-22 experiment's candidate is applied directly. Natal chat
answers (first turns and follow-ups) use one voice with fixed sections, every
delivered layer used, the timeline down to the sub-periods, houses and technique
names explained in plain words. The card carries what the backend computed and
used to drop plus precomputed tables (lordship, planets, periods, overlaps,
transit timelines, double transit, Sade Sati). One repair call per answer covers
left-out layers and sentences whose dates, durations or house lords differ from
the tables (flagged, never substituted). Deterministic clean-up only for the
steps that never misfired.
Removed: the experiment flag and every variant c1 did not run (sectioned writer,
model verifier, program substitution), the old plain voice for natal chat, the
body house rewrite (BUG-2149, reversed by the owner) and the body date
shortening (BUG-2164), the 「做哪一行问用户」 career rules. Report chats, report
generation, the no-birth-minute and declared-window agents keep their text.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXp2Vp62Ur3C7yRWQdSq
- BUG-2204: the English edition wrote the stored Chinese-first place label
verbatim, so every English edition was dropped as han_leak. English place
labels now translate country names, drop Han parentheticals and fall back
to coordinates for all-Chinese places.
- BUG-2215: the first-chart reveal's layout is inline, so the browser's
dialog padding cannot push 「跳过,查看星盘」 off screen before the lazy
stylesheet lands.
- BUG-2216: the admin create response fills kind, batch/campaign fields and
success text from what was sent.
- BUG-2217: reader-facing 「来源」 wording in data chapters and yoga status
words renamed (Python and the frontend mirror in sync).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
Owner approved the v2 screenshots on 2026-11-22: no separate expand row;
the existing top-right toggle opens the chart (own storage key), phones use
the existing bottom sheet, and the normal layout returns after the first
question. The old skip hint was aria-hidden, non-focusable 13px text.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
Synastry: sections that passed the first check get their draft back when the
rewrite breaks them, then only the sentences/reasons the gate named are removed
(same patterns as the Python gate); structural failures still reject.
Annual: text-rule failures left after the rewrite become month-level dates or a
removed sentence; 409 (another report still writing) stays in the start card;
a time chosen in rectification gets no opening sentence (BUG-2146 rule).
English: the stored englishUnavailable reason is returned (not rendered).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
BUG-2179: the plan-timing clarify reply (~70 chars) went through the smalltalk
free completion, which the database caps at 20 chars, so every save failed;
it now settles through the feedback free completion (400 chars, still free).
BUG-2180: hospital/ID records saved as 「记录下来的」 store
birth_time_uncertainty_known=false (no range given); birthTimeUncertain no
longer reads that as 「说不准」.
BUG-2181: exact-day dates left in the eight body sections are shortened to the
month; data-chapter labels drop 「来源」; children-related yoga effects use the
「传统上与…有关」 form (topic map regenerated, topics unchanged).
BUG-2182: an explicit length request from the user overrides the default answer
shape in chat and in-report questions.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
redemption_code_uses is granted to admin_runtime per column, so the
table-level privilege check is false by design; check any-column and that
credit_transaction_id stays hidden. The native-layer comparison now also
drops VedAstro/official fields, whose external result and cache timing can
differ between two runs after the BUG-2172 echo check.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
Gate 8375: database-redeem-security expected the old attempt count, and
database-redeem-code-kinds-330 assumed a zero starting balance. The migration
has not run anywhere yet, so it is edited in place.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
redeem_code_v2 reports a reused single code by the same user as
already_redeemed_by_you; the legacy wrapper maps it back so the existing
redeem-security contract (and its CI database test) is unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
Owner 2026-11-22. 活动码 (campaign): one code, many users, each once, optional
start time, end time and total cap, custom text or random. 个人码: one random
code, one redemption. 批量码: N single codes sharing a batch name, CSV right
after generation. Every code may set a success title, message and one of four
drawn celebrations (default / stars / moon / gift) shown in the approved
success dialog; failures stay toasts.
Random codes drop the JYOTISH- prefix: XXXX-XXXX from an alphabet without
0 O 1 I L, crypto randomness, hashed after normalisation (upper case, no
spaces or dashes). Old JYOTISH codes keep their exact-text hash and still
redeem: redeem_code_v2 tries the exact hash, then the normalised one.
Migration 20261122140000 is additive: kind / batch / campaign / window / cap /
count / success columns, redemption_code_uses(code_id, user_id unique) as the
once-per-user record (kept on account deletion), the ledger's per-code unique
moved to that table, the code-row lock serialising a capped campaign.
redeem_code(text) keeps its signature and result and wraps v2.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
- Migration 20261122130000_release_notes.sql: release_notes (version, title,
grouped items, draft/published/withdrawn) and release_note_reads; RLS on,
no table privilege for app/service roles; users go through service-role
functions list_release_notes_for_user / mark_release_note_read (a version
published before the account existed counts as seen); the admin writes
through audited admin_save_release_note / admin_set_release_note_status
gated by the new content.release_notes.write (owner, operations,
content_editor) and reads via an admin_runtime select policy.
- APIs: GET/POST /api/release-notes (signed-in user only) and GET/POST
/api/admin/release-notes.
- Admin page 更新说明: table + editor drawer with 新功能/修复/改进 lists and a
live preview; 保存草稿 / 发布 / 撤回.
- User side: 「本次更新」 dialog once per version (知道了 records the read);
settings gets an 更新说明 pane (phone segmented nav becomes four).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
Astrologer 2026-11-22 §4 (E): only a verified cancellation says mitigation exists.
After BUG-2165 every cancellation is detector-only, but the chat card, the
evidence pack and the rewrite table still wrote the verified sentence.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
Owner bought and approved the new host on 2026-11-22 (mainland latency ~44 ms
vs ~300 ms on 118.194.235.34). The four production workflows pin the host, so
the release SHA must carry the new value. The runbook records the fresh-start
sequence (no Supabase import).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
- 依据与边界: the writer rule gives six paragraphs in a fixed order, at most
600 characters each; the last pass packs a too-long section again at
sentence ends (split long paragraphs, join neighbours, drop trailing
sentences that no longer fit) instead of falling back to computed text.
- A time that is not marked uncertain loses the writer's own
「按你提供/选择的出生时间」 prefix (self-chosen times got it in round 31).
- 判断依据: 「本方法分档」「只写星座与起止」「另一档…判为…」「不参与打分」,
a bare 「为参考」 clause and author-named methods (KN Rao profile) go;
「本方法最低线」 reads 「最低线」.
- A chat body with more than four exact-day dates (the user did not ask
for a day) is put at month precision by the program; 判断依据 keeps
its dates. No model call.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
The bare plan-timing starter (BUG-2161) skips classification, so classifyStarted
stayed 0 and prep.classify logged Date.now() ms; the observability schema caps
durations at 7 days, logPrep() threw, and the turn failed after the credit /
first-free reservation with nothing cancelling it. The request stayed reserved,
held first_consult_free_request_id, and the next question answered 402.
- skipped classification logs skipped, 0 ms
- logPrep drops a rejected event with a log line instead of throwing
- the reservation registers a release hook; stream-first onUnhandled and the
legacy runTurn catch both cancel it (a no-op for completed requests)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
- Debilitation: verified reads 「落陷;另有缓解条件,不代表困难全部消失」, a candidate (neecha_states)
reads 「落陷;缓解条件待核」, uncomputed / not met / self-reference only stay 「落陷」; the
chat card, evidence pack, computed fallback and writer rule R-06 follow.
- Connections: plain words for pd_range_split / pd_stability_not_assessed and for a truncated or
missing chain; an unassessed topic chart no longer hides an undetermined house; the use rule says
unfinished or unassessed is never 'no connection'.
- Chat basis completion reads references/topic_varga_routes.json (parents D12 only, children D7,
siblings D3, residence D4, education D24; D9 never a career chart).
- test_reading_facts_2126 follows the BUG-2151 wording.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
Astrologer 2026-11-22 ruling J §9. New entrypoint starter_plan_timing: the
bare starter text in a new conversation gets one fixed clarifying question on
the free smalltalk settlement (no calculation, no classification). Typed
requests follow PLAN_TIMING_RULE in the natal instructions. BUG-2158's
「未满足」 is recognised by YOGA_FORMATION_NEGATED.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
- The last pass drops a sentence-level fact error (sav_total_as_strength and nine others) instead of the
whole section; in 依据与边界 a bracketed gloss or a 「某某层:」 label is not a SAV strength verdict; the
writer rule says how to write the Ashtakavarga total.
- The computed basis names a planet once when it is also the lagna lord.
- 现在这几年: the major period is at most a short background clause; the two periods are the current and
next sub-period (notice G-23).
- Chat basis completion uses only the question's own domain routes (parents never gets the family D7).
- 判断依据 display: 保留为参考窗口, detector and 本方法 notes, 原书 wording.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
- chat-basis-completeness.ts: after the wording guards, compare the route's
on-card Narayana / MK / PiK / DK / PK / AmK / divisional charts with the
判断依据 block and append the missing ones from the card's own sentences
(no model call, body untouched)
- chat-basis-clean.ts: drop 只作辅助, 单一系统推断, 两套运程分别计算 and
undated 本轮未交付 clauses
- reading_facts.py: reader words for the self-reference witness and the
unused PD reason; contract G-22 (parameter-sensitivity once, no 激活,
no stability-evaluation line)
- personal-report-reading-wording.ts basisReaderParagraphs at display
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
Since the 2026-11-16 merge, answers on identical input dropped these four groups.
The basis rule now asks for them when the card has them and they bear on the
question; the body is unchanged. The display cleaner also drops 「不构成两种方法一致」
and calibration notes note by note, keeps a following note and the period, and a
basis written twice keeps its first copy. The backtest script defaults to the
product's single answer mode.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
PLAIN_TERMS takes the report's house rule (READING_HOUSE_RULE): the plain
voice no longer says 「说这颗星管哪几个宫、落在哪」 next to 「不写宫位编号」.
Red lines say which part of life a planet stands for instead of 哪一宫;
dates go to the month, exact days only for key turning points. After a
简明解读 answer, body lines that still write a house (question names no
house) join the existing single rewrite call (BUG-1948); a rewrite that keeps
a house, drops a date or changes the line's lead is refused and the clause
holding the house is dropped. 深度解读 is unchanged. Shared house pattern
in lib/house-mechanics.ts. Contract notice G-20.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
After BUG-2143 the model's eight reader sections carry no house numbers, but a section that fell back to the
computed text still wrote 「第 10 宫(事业)在各宫得分里是 31 分」「金星落在第 4 宫」 and 「2023-02-28 至 2041-02-27」.
The computed text now names the life area a planet looks after, the condition its sign gives it (a condition,
not a result), a plain level for the house score, and month turning points for the current and next period.
「依据与边界」 keeps houses and numbers. Tested on two public Astro-Databank charts through the same checks.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8