Dune/docs/TRUTHTRANCE_TURN_ACTIONS.md

# Truthtrance combined current-turn actions

## Native current-turn action prototype

**Prototyped / Partial.** New raw question:
`{kind:'turnActions',target,turn,mixed}`. Action leaves are ordinary
`{kind:'payment',recipient,minimum}` or native
`{kind:'ship'|'move',territory,minimum}`. Optional current facts and full AND/OR
share sixteen logical leaves/four grouping levels. Old travel/payment question
shapes and mode gates remain. Current ordinary-lot bids, personal native Revival, final Nexus membership
and current battle decisions extend this envelope as detailed below.

## Authority and meaning

Reuse publisher November2020 FAQp8 current-turn AND/OR/compliance and supplied
Advanced physical39/printed37 authority recorded in RULE_DECISIONS. This is the
next provisional composition of existing native funded-payment and owned-stage
travel prototypes, not a printed exhaustive promise solver. Native scope and
current timing remain authoritative; optional modules retain their fences.

All facts freeze at actual answer, with the existing guaranteed aggregate
unknown interpretation. No negates the whole expression, not each action.
Alternative native routes are never fixed publicly or stored as separate
answers. The raw question is public; compiled facts/branches stay private.

Only actual ordinary payer bribes after this answer add cumulative payment
progress. Shipment fees, gifts, subsidies, earlier payments and later Mentat
receipts do not add that progress. The recipient's existing receipt/late-next-
opening mechanism stays unchanged. Each shipment leaf describes the same one
future original reserve shipment. Each movement minimum requires one physical
ground group; smaller moves never sum. Previous-turn physical events never count.

Transport closes at the respondent's actual Movement end. A whole OR answer
can retain a funded payment alternative after that close; an AND requiring a
missing transport event cannot evade it. Residual payments remain binding through
the current full turn, including Mentat. A whole Yes fulfills on actual events;
No lasts until the relevant native opportunities have closed. Constant answers
need no event. Earlier current-turn payments have separate intervals for each
answer; one future payment can satisfy compatible minima without double reserve.

## Native integration contract

The focused helper owns strict trees, signed branches and actual progress only.
Detached branch witnesses use the original scalar shipment/movement/payment
ledgers and native fee/funding/completion functions. No selected branch is
installed in the real ledgers. Existing bids, Revival, bribes, travel and battle
obligations retain their native resource protections and actual receipts.

Current funds support ordinary payments under the existing bribe prototype;
unknown future gifts/income/unowned cards are not invented. Existing owned
Charity/Revival/auction/card preparations remain the travel forecast. Mandatory
native progress or completed opposing effects can release impossibility; own
spending or closing an opportunity must leave a completion. Pending foreign
invoice/income outcomes retain original timing.

A persistent own raw aggregate history and one exact current native guide connect
human controls and four minimal profiles. Guide packets carry event/phase source
metadata, preserve actual recipients/forces/cost splits and native modal priority.
Every definite question uniformly reopens applicable concurrent preparation/
Mentat readiness, including constants, preserving the existing private boundary.
No new action rights or turn allowance is granted.

## Evidence and remaining scope

Helper: game/turn-action-promises.ts and its focused tests. Human controls:
components/turn-action-fields.tsx and components/turn-action-promises.tsx.
Native sources/tests: tests/fixture-truth-turn-action.ts and
tests/truth-turn-action-runtime.test.ts. Genuine prior Bid/Revival funding sources
are in tests/truth-turn-action-funding.test.ts.

The frozen source passes **23 files,234/234**, types/lint, production build and
three native paths with11 GameTable SSR renders, three actual answer callbacks
and five actual guide callbacks. Source27 +9 performs frozen facts, pay2,
original paid shipment1 and ground1. A second source27 +14 closes transport,
then pays in late Mentat and advances the turn. Original Advanced Mentat35 +9
pays after opening and retains the recipient's next-opening receipt timing.
All original physical inventories remain. These are local engine/SSR/callback
journeys, not browser/hydration or deployed acceptance.

Fourteen helper and fourteen control cases supplement eighteen native cases and
four genuine two-Truth funding conflicts. Compatible scalar/aggregate intervals
share an actual payment once; incompatible Bid/payment and Revival/payment Yes
answers reject in both acceptance orders. Earlier history/receipts survive and
the original accepted answer is fulfilled by an actual action. Current facts
freeze, movement maxima never sum, OR retains manual alternatives, whole No
survives own transport close, all four minimal profiles complete, and Basic/
Advanced late recipient receipts do not add payer progress again. A controlled
Inflation Double token variant, explicitly distinguished from original native
sources, rejects a payment-only Yes while allowing truthful No.

Initial focused feedback repairs exact stored metadata and a test fact-call
signature. Bounded review adds actual payment timing and later native Bid/Revival
resource preflight. A proposed payment gap is rejected as an incorrect test
assumption: pay9 toward a cumulative10 minimum leaves progress9 plus one funded
spice. The corrected regression proves partial9 then1, full10 and shipment routes.

Proof/manifest:
`/home/quintine/.local/share/dune/checkpoints/20261026-truth-turn-action-proof-v1/`;
journeys and HTML: adjacent `20261026-truth-turn-action-runtime/`.


Other action families, broader owned preparations/optional compositions,
unknown-fact correlations and complete rules/browser/deployed/recovery/privacy/
custody acceptance remain. Full strategy/calibration stays deferred. Preserve
all games and the hosting project; this batch performs no save/database/server/
reset/restart/deployment operations and certifies no complete mode.

## Combined ordinary bidding and Revival — 26 October 2026

**Prototyped / Partial.**

Extend action leaves with `{kind:'bid',lot,minimum}` for the current ordinary
lot, `{kind:'forces',counter,minimum}` and `{kind:'leader',leader}`. Accepted
future bids use their maximum declaration, never a sum or winning invoice.
Ordinary physical force returns accumulate; actual personal native leader
returns count once. Earlier returns, special/card/foreign-ghola returns do not.

An optional private preparation stage records the original physical lot source,
actual future bid maximum and cumulative typed returns/leader identities. Old
raw payment/transport rows stay valid without it. Stage data and virtual native
constraints remain private; own projected history retains the original whole
question and aggregate lifecycle only. All raw identities/lots are validated
eagerly, including folded alternatives.

Bidding closes at the original physical lot close. A later bid can reset a
pass, so No remains binding through reentry in that same lot. Basic Revival No lasts
to phase close; Advanced closes at the owner's actual ordered finish. Own Ready
must retain a whole completion, including a funded later payment alternative.
Future payments, shipment and ground groups keep their existing deadlines,
costs and scope. Native bids and Revival continue through original human/AI
controls and exact current-phase guides.

The frozen v2 source passes **19 files,203/203**, types/lint, production build and
four native paths with33 GameTable SSR renders, four actual answer callbacks
and15 guide callbacks. Original source22 +18 performs bid3, pay2 and later paid
shipment1; source50 +14 funds bid1, two ordinary returns and pay1 from four spice.
Personal leader source102 +8 returns Hidar plus a real free force/income and
pays2. Owned-Karama source59 +17 retains the actual foreign bidder before the
owned invoice, consumes the physical held card and pays for three returns.
All original physical inventories remain. No played source state is installed;
these are local engine/SSR/callback journeys, not browser or deployed acceptance.

Twenty-two helper and17 control cases support18 new native cases. They cover
both bid/payment alternatives, current-lot declaration/invoice separation,
whole-AND No event orders, maxima versus summed bids, cumulative ordinary
returns, personal leader identity, prior-return exclusion, later payments,
Basic phase/Advanced own closure, exact source guards and four minimal profiles.
Three genuine cross-stage cases prove a shared bid/force/payment purse, a No
surviving pass-reset/reentry and an invoice guide waiting for actual ownership.
An additional genuine held-Karama overbid case preserves the four-spice purse
for required Revival. Zero projected payments require no additional spendable
spice; native quotes still own the actual free invoice and later return costs.
Kull modes retain their existing fences and deferred overbid boundary.

Focused feedback corrects cash-only automatic-invoice, current bidder and Basic
phase-close test assumptions. Bounded review repairs a temporary-pass expiry
and an out-of-stage invoice guide. V1 passes source checks but the private
leader smoke expects free-force income without asking for that force. The
corrected smoke asks the intended full conjunction. A separate actual overbid
probe exposes the zero-payment resource defect; its repair/regression and all
checks pass in v2.

Proof/manifest:
`/home/quintine/.local/share/dune/checkpoints/20261026-truth-turn-preparation-proof-v2/`;
journeys/HTML: adjacent `20261026-truth-turn-preparation-runtime/`.

Payment clauses retain the existing funded-current-purse envelope; wider
invoice-first payment ordering, owned/module preparations, other action
families and full-mode acceptance remain unfinished. Existing native stage
promises retain their original scope. No comprehensive rule/recovery/privacy/
custody or deployed gate opens, and this batch performs no database/save/server/
reset/restart/deployment operation.

## Combined final Nexus membership — 26 October 2026

**Prototyped / Partial.** Add `{kind:'alliance',partner}` to the existing
whole current-turn grammar, only during the current quiet ordinary native Nexus.
Every raw partner is seated and different from the respondent, even when facts
fold its branch away. Existing native consent and pairing scope remains.

Optional private `alliance:{source,partner,closed}` binds the original Nexus.
An open stage has no recorded final partner. Temporary invitations and pairs do
not fulfill a clause; actual source close snapshots final membership. Whole No
may permit a temporary pair and later require a break. Later ordinary payments
and native preparation/transport retain their original separate deadlines.

Native signed partner constraints reuse the existing reciprocal consent plans,
while original payment and physical resource quotes fund the whole branch. No
selected route or final-state witness enters real native ledgers or public
views. Missing foreign consent cannot create a definite Yes; existing native
Unknown/save/reask semantics require an otherwise funded native branch. A
detached consent-only witness can establish Unknown, never Yes or a guide; it
still runs native pairing, payment and later-stage resource checks. Known
financial impossibility, including when consent is also absent, stays No.
Every definite answer uniformly reopens own Nexus readiness, including constants.
Actual closing preflight requires a whole completion after final membership is
fixed; later actions retain any residual payment commitment. Actual ordinary
payments uniformly reopen own current-Nexus readiness for the raw alliance
history, including privately folded terminal constants. A new payment can
require a later break before Ready. An owed payment is protected before offering
its recipient a partnership; compatible alternatives remain available.

The frozen source passes **19 files,221/221**, types/lint, production build and
four native paths with37 GameTable SSR renders, four actual answer callbacks
and16 guide callbacks. Original source47 +21 forms the final pair, pays2 and
later ships1 for its real fee2. Allied source48 +29 performs own Ready, pay2,
required break and whole No through actual turn end. Source47 +24 fixes None
at the Nexus, then pays during late Mentat. No-consent source46 +6 answers
Unknown and saves the actual original Truth, retaining both physical cards and
installing no partner or promise. All original physical inventories remain;
no played state is installed. These are local engine/SSR/callback journeys,
not browser, hydration or deployed acceptance.

Twenty-eight helper and21 control cases support18 new native cases. They cover
actual final source close, both OR routes, paid temporary membership, whole No,
later payment, foreign invitation withdrawal and departure, no reactivation,
exact current own packets, uniform definite-answer readiness and four profiles.
Focused additions distinguish known unaffordability even when consent is also
missing, require renewed Ready after payment, protect an owed payee before a
prospective alliance, and keep terminal constant receipts unchanged while
uniformly reopening readiness.

Bounded review finds an Unknown path that checked consent before the branch's
money. A tagged private consent scenario now runs native pairing and all other
resource checks; it can supply Unknown only, never a definite answer or guide.
V1's runtime-first proof catches a private-driver assumption: the legal No guide
may choose a break instead of Ready. V2 explicitly takes the legal manual Ready
route before the payment/break sequence. Source is unchanged by that driver
repair; types/lint, related tests and build run once after the smoke passes.

Proof/manifest:
`/home/quintine/.local/share/dune/checkpoints/20261026-truth-turn-alliance-proof-v2/`;
journeys/HTML: adjacent `20261026-truth-turn-alliance-runtime/`.

Current battle-plan and Traitor-declaration compositions, wider native
preparation/order/module combinations, unknown-fact correlations and complete
rules/browser/deployed/recovery/privacy/custody acceptance remain. All existing
preview gates and deferred full AI refinement/calibration remain. This batch
performs no database/save/server/reset/restart/deployment operation.

## Combined current battle decisions — 26 October 2026

**Prototyped / Partial.** Add atomic
`{kind:'battlePlan',battle,territory,claim}` predicates and
`{kind:'traitorCall',battle,territory,call}` to the shared whole question.
The exact current original battle and combat location bind every raw clause,
including folded ones. Plan clauses use the existing native atomic claim
parser and public printed identities; declaration clauses require an original
current revealed-battle unsubmitted voter. No extra eligibility is granted.

Optional private combat metadata holds source, evaluated original plan
predicates and actual declaration context. Future plan receipts are captured
at native reveal before cleanup; an already submitted own plan can be fixed
at answer as in the existing native question. Original declaration counts
before later Karama/prevention; a later prevented effect is not a new choice.
Closure without the original event records absence, not dialzero or Decline.
Whole alternatives and later ordinary payments retain their original meaning.

Native virtual plan/decision constraints quote original legality, preparations,
resources and physical custody. They never become real selected branch rows.
Human plan guidance fills the existing complete form for review rather than
sealing automatically; four minimal profiles can submit the native packet.
Older records remain valid without combat metadata. Public native Traitor
choices and later pure/mixed plan or Traitor questions retain a completion of
earlier whole commitments even when there is no separate bribe ledger. Pending
native income must finish before a plan or preparation guide is offered.

The frozen batch passes **20 files,249/249**, types/lint/build and four actual
native paths:22 GameTable SSR renders, four actual answer callbacks, six guide
callbacks and one human plan-fill callback followed by separate native sealing.
Source24 +9 performs frozen-spice fact, pay2 and the original physical projectile
plan. Source24 +11 fixes a false plan receipt before cleanup and completes its
payment alternative in late Mentat. Source26 +11 declares Reveal under a whole
No, honors the cumulative payment cap and closes the actual turn. Original
remote allied Harkonnen source72 +10 pays2, declares Reveal and receives actual
physical Karama prevention; the declaration receipt stays true. Original
physical inventories remain. No played source state is installed.

Thirty-seven helper and24 control cases support13 new native cases, including
four profiles. A genuine H/A/E two-Truth continuation earns an A/E battle from
source27 by native movement, paid shipment and preparation. After A pays2 under
a whole No, the second native dial question offers only the compatible No;
both actual plans and turn closure fulfill the earlier answer. Bounded review
repairs the new-only ledger preflight and pending-income guide fence. V1's
runtime passes before type errors; narrow type feedback repairs them. V2 passes
runtime, quick checks, the affected union and build on one frozen source.

Proof/manifest:
`/home/quintine/.local/share/dune/checkpoints/20261026-truth-turn-combat-proof-v2/`;
journeys/HTML: adjacent `20261026-truth-turn-combat-runtime-v2/`.

These are local engine/SSR/callback journeys, not browser, hydration or deployed
acceptance. Wider preparation/order/action/module combinations, unknown-fact
correlations and final rules/recovery/privacy/custody acceptance remain. Full
AI strategy and calibration stay deferred. Native mode/module preview fences
remain; this batch performs no database/save/server/reset/restart/deployment
operation and does not certify any complete mode.

## Printed Hajr and Ghola plays — 27 October 2026

**Prototyped / Partial.** Add one atomic named-effect leaf:
`{kind:'cardPlay',effect:'hajr'|'ghola'}`. It asks about an actual accepted use in
that printed role after this answer during the current turn. Hajr counts even
if its extra move remains unused; Ghola card use is distinct from ordinary
force or leader Revival. Earlier plays, disposals, sales, role conversions and
recovery alone do not count. A legitimate later physical replay can count.
This is an explicit composition of the existing current-turn Truth authority
and native card contracts, not a new publisher per-card promise ruling.

Per-answer optional private stages retain played/closed booleans only. Native
private counters supply detached witness baselines; earlier counts are never
installed as future answer progress. Old raw records remain valid without the
new stages. Original raw conditions and aggregate status remain the only own
history; physical card IDs and selected completion branches stay private.

Hajr's remaining opportunity closes at the owner's actual Movement end; Ghola
continues through the full current turn. Whole alternatives retain payments
and other native deadlines. A known Inflation Double opening closes payments
before Mentat without closing a remaining Ghola alternative; Ready quotes only
that payment closure. Full turn expiry still closes both families.
Basic Hajr uses its original own-turn timing;
Advanced must earn its actual immediate window through real completed movement
and required arrival decisions. Recovery supplies no extra timing privilege.

Detached native witnesses share the real purse, Tanks, card cost, leader and
movement allowance with ordinary Revival, payments, transport and current plan
constraints. No branch is stored in the real ledger. Known owned Box routes
can recover a real eligible card; unknown future draws/gifts/casualties are not
invented. Broader future preparation/funding and module combinations remain.
The current Basic Richese public replay source includes Semuta and remains
outside this whole-turn composition's existing envelope; recovery evidence
uses an original supported Advanced source instead of removing that marker.
Its actual first move and first Hajr precede the answer. The future paid Box
must precede a second real movement before Hajr can be replayed: a voluntary
Box play closes any earlier immediate Advanced window, including in detached
forecasts. Exact guide metadata is checked before passing the original native
action to strict card controls; it never becomes a Box selection field.
Native accepted printed plays, rather than their later income effects, record
receipts. Own post-cost completion is checked before foreign income/discard
responses can mask voluntary evasion. Human controls and four minimal profiles
receive the same exact native packet. All existing mode, final assurance and
AI strategy/calibration gates remain open.

The frozen source-bound proof covers22 related files and278 passing cases:
V5's234 cases in twenty unchanged files plus V6's44/44 in the two repaired
files. Six counter, ten aggregate helper, four control and sixteen native/
controlled-token cases add36 checks. Final types/lint/build pass. Seven native
journeys provide33 complete GameTable SSR renders,7 actual answer callbacks
and17 exact guide callbacks. They exercise Basic Hajr/frozen facts/payment,
Advanced actual first movement, Ghola/ordinary shared Tanks and purse, Hidar,
Advanced typed source selection, paid Advanced replay and whole No through
Mentat. The forced Sardaukar return is separately covered by the native test;
a minimal guide may choose a legal normal force instead.

The initial runtime expectation incorrectly demanded an elite guide selection;
its native packet can legally choose a normal force. The original Basic replay
source is rejected by the existing Semuta fence, so the supported Advanced
chronology is used. Targeted feedback repairs the real shared-Tanks, early
payment-closure, Box-window forecast and strict native packet issues. Type/lint
feedback is repaired. The final two test-oracle fixes change no production or
fixture behavior; the per-file manifest preserves the passing runtime evidence.

Proof/manifest: `/home/quintine/.local/share/dune/checkpoints/20261027-truth-printed-proof-v6/`;
reused journeys/HTML: adjacent `20261027-truth-printed-runtime-v5/`.
This is local engine/SSR/control evidence. No HTTP/storage boundary changes,
server restart, reset/deletion or deployed operation occurs. Full browser,
deployed, recovery/privacy/custody and wider-combination acceptance remain.

## Native Voice and Prescience declarations — 27 October 2026

**Prototyped / Partial.** Add a source-bound action leaf:
`{kind:'battlePower',battle,territory,claim}`. Its four predicates concern native
Voice use/explicit Decline, an exact Voice category plus require/forbid command,
native Prescience use/explicit Decline, or its requested plan field. They never
describe the inspected answer or promise that the later effect survives Karama.
This composes the existing general current-turn Truth authority with original
native power timing; it is not a new publisher-specific power-promise ruling.

The original respondent must own that native power opportunity, personally or
as an eligible reciprocal ally's provider. Prescience can be pending behind
the actual original Voice decision, but cannot execute before that stage ends.
Native command quotations, No-Field restrictions and response ownership remain.
Borrowed Nexus and special powers keep their existing separate envelopes.

Whole AND/OR/No treats every predicate about this one choice as correlated.
An actual declaration closes its stage and fixes only the raw named predicate
values, before a later prevention response. An expired source with no native
declaration records absence, never an invented Decline. Own manual choices
must preserve the whole answer or a legal alternative, including later payment.
No selected route is stored. Native declaration quotations spend no spice or
card and do not invent the other player's future inspected answer.

Optional private power stages extend original combat metadata; older rows keep
their existing shape. Public own history remains the original raw question and
aggregate status. The source-bound guide forwards the real native declaration
packet; human battle plans still fill for review before separate sealing.
Four profiles retain only minimal legal participation. Full rules,
browser/deployed, assurance and AI strategy gates remain open.

The frozen batch passes18 related files and281 tests, types/lint/build. Eight
public opportunity/predicate, ten aggregate, six control and fourteen native
cases add38 checks. Eight original native journeys provide33 full GameTable
SSR renders,8 actual answer callbacks and13 exact native guide callbacks.
Original own Voice44, Prescience45, remote Voice73 and remote Prescience84
prefixes include genuine setup/deal, paid Karama1, paid reserve shipments and,
for remote providers, ordinary Worm/Nexus consent. The remote Prescience battle
uses two real paid2 Imperial Basin shipments, never the Polar Sink sanctuary.
Arrays derive every count; original physical cards and Traitors remain.

Runtime paths cover both power declarations followed by actual printed Karama,
explicit Decline, a declined power followed by late Mentat payment after battle
cleanup, both allied providers, waiting for the foreign Voice stage and whole
No through actual turn end. Native tests also cover incompatible same-choice
predicates, manual OR alternatives, exact event/phase/owner/stale guides,
requested fields without disclosed values and all four minimal profiles.

Focused feedback repairs own evasion through a newly opened foreign response:
record and check the actual declaration before giving away control. A later
native choice cannot rewrite an already fulfilled OR answer; that test oracle
is corrected. Genuine-source work rejects the prepared-marker and sanctuary
assumptions. Player-view nullability and lint feedback are repaired. These are
local engine/SSR/control checks, not browser/hydration/deployed acceptance.
Proof/manifest: `/home/quintine/.local/share/dune/checkpoints/20261027-truth-power-proof-v3/`;
journeys/HTML: adjacent `20261027-truth-power-runtime-v3/`.
No HTTP/storage boundary, database, reset/deletion, server restart or deployed
operation changes. Borrowed Nexus and special power composition, broader
future actions and all final assurance/release gates remain open.

## Original Storm and fresh-blow card plays — 27 October 2026

**Prototyped / Partial.** Add a source-bound action leaf:
`{kind:'windowCardPlay',effect:'weather'|'atomics'|'harvester',source}`. It asks
about an actual future printed-role play after this answer in that original
Storm or fresh-blow window. It does not predicate the selected Weather distance
or resulting spice amount. Earlier play, disposal, another role and recovery
alone do not count. Existing Hajr/Ghola shapes and counters remain unchanged.
This composes existing general current-turn Truth authority and native card
contracts; it is not a new publisher-specific card-promise ruling.

The original Storm window stays open through dial reveal and closes when actual
movement starts, including continued Storm decisions in phase0. Each newly drawn
fresh blow gets one stable public identity, retained across Harvester plays;
older windows use a stable turn/territory/sector fallback. A subsequent actual
blow has a different source, including the second Advanced blow in phase1.
Harvester retains its existing stacking/replay and destroyed-window policy.

Private per-answer stages retain exactly the raw named effects, sources and
played/closed booleans. Public own history remains raw question/aggregate only.
Detached witnesses execute actual effects and known owned recovery costs before
sharing resulting pieces, terrain and purse with other obligations. Original
Weather selection0..10, Atomics qualifying fighters/casualties and Harvester's
fresh-blow arithmetic remain native. No branch or speculative card is installed.

Definite window answers uniformly reopen own concurrent readiness. Own Ready
quotes closure without closing the real source early and must retain a whole
completion, including legal later alternatives. Actual source expiry closes
only that source's stages. No remains binding while the real window is open.
Guides use exact own native card/Box/paid-selection or window-Ready packets.
An owned paid Box preparation reopens its respondent's window acceptance so
its cost cannot leave an owed recovered play behind an earlier Ready.
Minimal profiles can accept a No-Harvester window; automatic Harvester never
reopens their already accepted blow. Complete rules, browser/deployed,
assurance and AI strategy gates remain open.

Final proof covers18 files/303 passing cases, types/lint/build. Eight source,
eight aggregate, seven control, fifteen native/controlled-view and one shared
availability case add39 checks. Nine original native journeys provide37 full
GameTable SSR renders,10 actual answer callbacks and14 exact guide callbacks.
Original Weather6/8, Atomics36 and Harvester14 prefixes preserve33 Treachery
and15 Traitors. Atomics actually kills the two prior paid4 Wall counters.
Weather's manual10, late Mentat payment, foreign Weather, whole No and the
second actual Advanced blow retain their actual original source semantics.

The first native source check catches the missing first Storm open marker;
setup now opens it as well as later turns. A two-question regression catches
a guide tagged to an unrelated earlier Hajr row: source-specific card/Box/
Ready guides now select a live row naming the actual original window. An
unopened paired Truth/Hajr deal makes this real: A binds Hajr/pay2, E binds
Weather, payment leaves Hajr live, the correct Weather callback fulfills E's
answer, then native Movement/Hajr fulfills A's earlier answer. Neither branch
is stored as a separate answer. Types feedback is repaired.

V2's earlier20-file/324 run precedes this view-only guide correction. V3 reruns
the affected18 files and adds the paired case. The two unchanged early-travel
files'22 cases remain earlier-source evidence and are not counted in303.
The No-Harvester policy additionally uses a controlled public bot view with one
original counter rearranged to the fresh territory; the real native Game and
army are unchanged. Native profile actions still use the actual Ready packet.
Bounded known-Box review finds no recursive whole-question search or double
fee; actual window Box-combination acceptance remains open.
Proof/manifest: `/home/quintine/.local/share/dune/checkpoints/20261027-truth-window-proof-v3/`;
journeys/HTML: adjacent `20261027-truth-window-runtime-v3/`.
This is local engine/SSR/control evidence. No HTTP/storage boundary, database,
reset/deletion, server restart or deployed operation changes. Broader card,
recovery and module combinations and all final assurance/release gates remain
open.