Skip to main content

MY2026 → MY2027 transition proposal: Jeep Gladiator / Cherokee / Grand Cherokee Summit

OEM / marketstellantis / us
Model pattern^(JTJL98|KMJH74|WLJT74|WLJT75)$
TransitionMY2026 → MY2027
Data partition2026-10-02 (latest; a few analyses read older partitions, listed in §7.7)
StatusReviewed; partially applied. See Applied decisions below. No DynamoDB writes.

Applied decisions​

The review applied this subset. The sections below record the workflow's original proposal.

ItemDecisionFile
CP: SEAT_MATERIAL=LEATHERETTE removeApplied on JTJL98 ^J$ only. Q (High Tide) is not orderable in MY2027 and gets no rule, so §7.1 resolves to option (a) and the partition_sums_valid failure does not apply.take-rates/jeep-demand-rules.yaml
PHT: PAINT_COLOR=PURPLE removeApplied on JTJL98 ^[CJSW]$take-rates/jeep-demand-rules.yaml
PYC: PAINT_COLOR=ORANGE removeApplied on JTJL98 ^[CJSW]$; covers PL4take-rates/jeep-demand-rules.yaml
PL4Not applied; subsumed by PYC—
MBU: BUMPER_FINISH=BLACK removeNot applied; support-prune already zeroes it—
§4.2 XC3 group and GRAB_HANDLES 0.1Applied by widening the existing ^JTJ entries from ^[ADRSWY]$ to ^[ACDJRSWY]$synthetic_candidates/gladiator/gladiator-rules.yaml, take-rates/jeep-demand-rules.yaml
AJY infotainment content, X9G and JFF on JTJL98 C and JApplied in review. The mandatory_adds rule that attaches RF7, RHY, RTM, RTQ, RTU and X9B with AJY, the X9G and JFF optional_adds, and the PERFORMANCE_AIR_INTAKE assumed rate now include C and Jsynthetic_candidates/gladiator/gladiator-rules.yaml, take-rates/jeep-demand-rules.yaml
§4.1 PR6 Red Hot PearlApplied to the shared ^WL[JT][PT] PAINT optional_replaces, mandatory_expands and mutually_exclusive_groups listssynthetic_candidates/grand-cherokee/grand-cherokee-rules.yaml
§7.2 AWQ, §7.3 new attributes, remaining §7.4–§7.7 itemsNot decided; no change—
KMJH74 H, WLJT74/75 WNew MY2027 trims with no MY2026 cohort. Candidates come from base_filters: Trailhawk from MY2026 KMJP74/Q Overland, Summit Obsidian from MY2026 Summit U of the same bodysynthetic_candidates/cherokee/cherokee-rules.yaml, synthetic_candidates/grand-cherokee/grand-cherokee-rules.yaml

1. Summary​

As proposed, the rule set failed 1 hard criterion and was not mergeable as written. The applied CP scope above avoids that failure.

The failing gate is partition_sums_valid. Only CP's trim-Q branch causes it, and narrowing that rule to trim J makes every gate pass in a re-run (§2, §5b).

Raw per-trim data. This proposal aggregates the ground-truth per-trim diff, built from stellantis.us.consolidated.features partition 2026-10-02. It covers 14 trims: JTJL98 B, C, F, H, J, Q, S, T, W; KMJH74 H; WLJT74 U, W; WLJT75 U, W. To check a specific trim, look it up there: each trim section lists every changed code with MSRP, parent package, code-guide rules and the basis note.

CountValueBreakdown
Ground-truth candidates (all changeTypes)271removed_no_longer_exists 1, removed_became_standard 34, added_new_content 3, added_became_configurable 22, parent_changed 14, removed_trim_discontinued 130, added_trim_new 67
Confirmed real61real_remove 1 (ACW), real_became_standard 34, real_add 3 (AWQ, CSQ, XC3), real_became_configurable 22, real_expand_member 1 (XPR)
Noise, filtered out205parent_changed 8, removed_trim_discontinued 130, added_trim_new 67 (§8)
Uncertain, needs human review5GFA, HT1, HT3, LBK, MM3 (all parent_changed)
Demand-rule entries analyzed61none_needed 50, remove 5, needs_new_attribute 4, flat_seed_candidate 2
Concrete demand rules proposed65 from the demand phase (CP, MBU, PHT, PL4, PYC) + 1 Report-phase replacement (CP narrowed to ^J$)
Validated clean (rule level)5 / 5CP, MBU, PHT, PL4, PYC each passed on the first real apply_removes run; the CP ^J$ replacement passed the Report-phase re-run
Validation failures (rule level)0
Hard-criteria failures1partition_sums_valid, caused by CP ^(J|Q)$
Synthetic-candidate rule changes ready to paste02 suggestions for review, not validated (§4)
Human decisions needed21multiple_valid_approaches 5, new_attribute_registration 3, marginal_value_call 7, uncertain classification 5, failed_hard_criterion 1, failed_validation 0

What the reviewer most needs to know:

  1. Most confirmed changes need no rule. The demand axis is either FUNCTIONAL_ATTRIBUTE (mutually inclusive, never renormalized) or a value that a surviving code still offers (for example GREY via PDN, BLACK via PX8, GREEN via PJ5, HARD_TOP via HT1/HT3).
  2. All five remove rules do the same work as the automatic support-prune stage (derive_support_removes) for today's candidate set. They only matter if a trim falls below prune_min_candidate_count (50).
  3. JTJL98 trim Q (High Tide, MY2027) has no MY2026 base trim, no candidates and all-null hl90 demand. Every gate failure is on Q. Whether Q is a real trim is an open decision (§7.1).

2. Hard-criteria gates​

Run on the combined final rule set (7 RemoveRules: CP ^(J|Q)$, MBU, PHT, PL4, PYC, and the CP ^J$ replacement included twice) after 3 attempts, against the latest partition 2026-10-02.

GateStatusDetail
partition_sums_validFAILWhat broke: 5760 (dealer, model, partition-type) groups checked: 36 dealers × 5 touched MY2027 models × 32 types (all except FUNCTIONAL_ATTRIBUTE and TRIM_CATEGORY, including _X_ and _SET). Before the rules, 1559 groups were off 1.0 on sales_take_rate (max 2.0068) and 2349 on hl90. After: sales_take_rate has 4 bad groups, all JTJL98/Q SEAT_MATERIAL, now summing to 0.0. Those 4 dealers had only a LEATHERETTE row (no NO_SEAT_MATERIAL), so CP's Q branch zeroed the group with nothing to renormalize onto. sales_hl90_take_rate has 1152 bad groups: every Q group (32 types × 36 dealers) sums to 0.0 before and after, because Q has no MY2026 model and so no recent-MY sales. The rules do not cause that, but Q enters scope only through CP. C, J, S, W all sum to 1 within 1e-6 in both columns. Re-run with CP narrowed to ^J$ (attempt 2): 4608 groups, 0 bad in either column, max deviation 4.4e-16.
no_dead_rulespassResolved with the real resolve_remove_rules against consolidated.models 2026-10-02 plus the file-level model_year ^(2027)$. CP → 2 models (J, Q); MBU → 1 (S); PHT → 4 (C, J, S, W); PL4 → 2 (S, W); PYC → 4 (C, J, S, W); each CP ^J$ replacement → 1 (J). unmatched_rules() = []. MY2027 JTJL98 trims in data: C, J, Q, S, W.
donor_clearedpassPlain rows (sales, hl90) before → after: LEATHERETTE/SEAT_MATERIAL J+Q (56.0594, 19.2276) → (0, 0); BLACK/BUMPER_FINISH S (4.5924, 3.7160) → (0, 0); PURPLE/PAINT_COLOR C/J/S/W (24.9880, 24.2818) → (0, 0); ORANGE S/W (PL4) (2.3866, 3.0865) → (0, 0); ORANGE C/J/S/W (PYC) (3.3571, 4.3060) → (0, 0). _X_ crosses with the from_value value in the matching slot also zeroed: BUMPER_FINISH BLACK 243 rows (9.51, 7.90) → 0; PURPLE 1006 rows (100.49, 97.69) → 0; ORANGE C/J/S/W 662 rows (13.77, 17.62) → 0. 0 _SET rows contain any from_value token.
no_negative_or_null_ratespass29,269 rows on touched models: 0 nulls, 0 negatives in both rate columns (min 0.0). The input's 5,928 null hl90 values (all Q) are written as 0.0 by apply_removes.
no_duplicate_keyspass52,610 scoped output rows (9 model_ids, MY2026 + MY2027): 0 duplicate (dealer_id, model_id, attribute_type, attribute_value). Same with the support-prune variant.
runtime_boundpass_run_rule_pipeline on 52,610 rows: 0.096 s. Whole script ≈ 12 s against a 120 s budget.
mass_conservationnot_applicableNo RenameRule; all 7 rules are RemoveRules.
proxy_untouchednot_applicableNo ProxyEstimateRule.
no_orphaned_removed_demandnot_applicableNo rule touches FUNCTIONAL_ATTRIBUTE; 0 FUNCTIONAL_ATTRIBUTES_SET rows contain the rules' tokens. FA demand > 0 on values no candidate has: 1996 rows, 58 (model, code) pairs, sums (401.37, 125.99), unchanged by the rules. With support-prune removes added it drops to 1036 rows, all on JTJL98/Q (30 codes; 289.57 / 0.0). Removed values (PURPLE, ORANGE, BLACK, LEATHERETTE) end at 0 on every model in scope that has candidates. Values above 0 outside the rules' trims are all pruned by derive_support_removes: S/W RED (0.128 / 0.255), S/W NO_PAINT_COLOR (≈0.57), S LEATHERETTE (0.198), W NO_SEAT_MATERIAL (0.03). Q keeps ORANGE at 1.871 but has no candidates.
no_undemanded_candidate_attributesnot_applicableNo rule creates or deletes FA rows. 8 candidate FA values have no demand row, before and after: C NON_VENTED_STEEL_HOOD (62/62 candidates), C SERVICE_PLAN (33/62), J ROOF_RAILS (2/512), J SATELLITE_RADIO (512/512), J SERVICE_PLAN (206/512), J TONNEAU_COVER (2/512), S GRAB_HANDLES (42678/127995), W SERVICE_PLAN (55819/132657). The gap predates the rules. S GRAB_HANDLES is covered by an existing assumed-rate rule in a later stage this run did not include. Every candidate value on the touched types has a demand row.

How the gates were run: the real ai_dagster.components.projected_demand._run_rule_pipeline (rename → remove → assumed_rate → proxy_estimate) with only the 7 RemoveRules, after _rules_from_file, _prune_low_demand(0.005) and _explode_model_years(2026). Inputs: smoothed_demand_take_rates filtered to the 5 MY2027 JTJL98 base_model_ids (62,748 rows, 36 dealers); candidates_output + attributes/vehicles for MY2027 JTJL98 (261,226 candidates: S 127,995, W 132,657, J 512, C 62, Q 0; 7.56M attribute rows). Untouched (MY2026) rows came out bit-identical. Candidates are production candidates_output, built from the synthetic-candidate config on main, without the rule changes in this proposal.


3. Resulting demand​

3.1 Removed and recipient codes (sums over dealer × model rows)​

CodeBeforeAfter
SEAT_MATERIAL:LEATHERETTE [J,Q] sales_take_rate56.05940
SEAT_MATERIAL:LEATHERETTE [J,Q] sales_hl90_take_rate19.22760
SEAT_MATERIAL:NO_SEAT_MATERIAL [J,Q] sales_take_rate (recipient)26.308168
SEAT_MATERIAL:NO_SEAT_MATERIAL [J,Q] sales_hl90_take_rate16.772436
BUMPER_FINISH:BLACK [S] sales_take_rate4.59240
BUMPER_FINISH:BLACK [S] sales_hl90_take_rate3.71600
BUMPER_FINISH:NO_BUMPER_FINISH [S] sales_take_rate (recipient)31.918136
BUMPER_FINISH:NO_BUMPER_FINISH [S] sales_hl90_take_rate33.050436
PAINT_COLOR:PURPLE [C,J,S,W] sales_take_rate24.98800
PAINT_COLOR:PURPLE [C,J,S,W] sales_hl90_take_rate24.28180
PAINT_COLOR:ORANGE [C,J,S,W] sales_take_rate (PYC)3.35710
PAINT_COLOR:ORANGE [C,J,S,W] sales_hl90_take_rate (PYC)4.30600
PAINT_COLOR:ORANGE [S,W] sales_take_rate (PL4)2.38660
PAINT_COLOR:ORANGE [S,W] sales_hl90_take_rate (PL4)3.08650
PAINT_COLOR:BLACK [C,J,S,W] sales_take_rate (recipient)16.846220.9235
PAINT_COLOR:BLUE [C,J,S,W] sales_take_rate (recipient)8.36219.2300
PAINT_COLOR:GREEN [C,J,S,W] sales_take_rate (recipient)21.185023.7971
PAINT_COLOR:GREY [C,J,S,W] sales_take_rate (recipient)40.272548.3075
PAINT_COLOR:WHITE [C,J,S,W] sales_take_rate (recipient)31.646140.2059
PAINT_COLOR:NO_PAINT_COLOR [C,J,S,W] sales_take_rate (recipient)1.01291.1530
PAINT_COLOR:RED [C,J,S,W] sales_take_rate (recipient)0.33330.3831

The NO_SEAT_MATERIAL J+Q sales_take_rate total of 68 (not 72) comes from the 4 Q dealers whose SEAT_MATERIAL group went to 0, which is the gate failure in §2.

3.2 FUNCTIONAL_ATTRIBUTES_SET combination-length distribution​

Average share per combination length, on sales_take_rate, over 180 dealer × model groups (the 5 touched models).

LengthBeforeAfterLengthBeforeAfter
00.0376430.027757120.0254410.022770
10.1203800.094898130.0064930.004470
20.2811660.220834140.0123720.010482
30.2054180.171612150.0051150.004685
40.0896070.069497160.0137620.011310
50.0629750.046175170.0007310.000623
60.1148110.092166180.0023990.002269
70.0581200.048683190.0012960.001231
80.0394400.032247200.0061340.004268
90.1040970.061692210.0046780.003225
100.0274710.021137220.0001310.000117
110.0521130.047853

How to read it: the before shares sum to ≈1.37 per group because the smoothed _SET partitions are not normalized. apply_removes renormalizes every partition type on a touched model, so after the rules each group sums to ≈1.0. No _SET token was stripped (0 matching rows), so every length shrinks by roughly the same factor. Relative proportions hold (length 2 is ≈20.5% of the total before and ≈22.1% after; the drift comes from averaging across groups). The production component already touches these models through support-prune, so the explicit rules add no new distortion.


4. Proposed synthetic-candidate rule changes​

None ready to paste. No phase produced a validated synthetic-candidate rule. Why each confirmed group needs none:

  • real_became_standard (34) and real_became_configurable (22): content is unchanged, only is_configurable flipped. No synthetic-candidate rule; each went through the demand-axis check in §5.
  • real_remove ACW: removed packages auto-retire with their content (gladiator/gladiator-rules.yaml header). No successor appearance package exists on MY2027 S.
  • real_expand_member XPR: XPR keeps parent AAN and gains AWQ; it stays orderable through AAN.
  • real_add CSQ, XC3: gladiator/gladiator-rules.yaml already has optional_add_groups XC3 = [XC3, CSQ, CSV], scoped ^JTJ trims ^[ADRSWY]$.

Two suggestions came out of the analysis. They are not validated and each needs a decision (§7):

4.1 mandatory_expands / optional_replaces (grand-cherokee-rules.yaml): add PR6 to the PAINT targets​

Rationale: on 2026-10-02, MY2027 WLJT74/75 U candidates have no red paint (only PB5, PDN, PPS, PVB, PW7, PX8), so the support-prune sets PAINT_COLOR RED to 0.0 on both U model_ids. PR6 Red Hot Pearl Coat (RED, is_configurable=true) is on WLJT74/75 U and W in MY2027 and replaces PRV (MY2026 RED ≈0.080 / 0.044). The current ^WL[JT][PT] PAINT mandatory_expands entry expands PRV into a list without PR6.

# grand-cherokee-rules.yaml — proposed edit, not validated
optional_replaces:
- model_filters: {model_code: "^WL[JT][PT]"}
group: PAINT
from_codes: [PB5, PDN, PPS, PR6, PVB, PW7, PX8]
to_codes: [PB5, PDN, PPS, PR6, PVB, PW7, PX8]
mandatory_expands:
- model_filters: {model_code: "^WL[JT][PT]"}
group: PAINT
from_codes: [PAS, PRV, PSE, PXJ]
to_codes: [PB5, PDN, PPS, PR6, PVB, PW7, PX8]

Open before pasting: PR6 was checked only on WLJT74/75. Whether it is orderable on Limited (WL[JT]P) was not checked; if not, split the rule so PR6 applies only to ^WLJT7[45]. PD7 GHOST (WHITE, configurable on WLJT74/75 U MY2027) is also missing from these lists. WHITE demand is unaffected because PW7 offers it, so PD7 only changes the option split.

4.2 optional_add_groups (gladiator-rules.yaml): extend XC3 to JTJL98 C and J (and maybe Q)​

Rationale: XC3, CSQ and CSV are configurable on MY2027 JTJL98 C/J/S/W (CSQ also has a row on Q). The XC3 group and the GRAB_HANDLES assumed rate (0.1) are both scoped ^[ADRSWY]$, so C and J candidates never get XC3 and have no GRAB_HANDLES demand row.

# gladiator-rules.yaml — proposed edit, not validated
optional_add_groups:
- model_filters: {model_code: "^JTJ", trim_identifier: "^[ACDJRSWY]$"}
group: XC3
codes: [XC3, CSQ, CSV]

This only makes sense together with the matching demand change in §7.2 (CSQ/XC3). Note that gladiator-rules.yaml targets JTJL98 S/W only (header), so C/J/Q get no synthetic transition rules today. Production has C 62 and J 512 candidates, against S 127,995 and W 132,657.


5. Proposed demand-imputation rule changes​

Target file: projects/ai_stellantis/src/ai_stellantis/defs/stellantis/us/take-rates/jeep-demand-rules.yaml. It already sets model_years: [2027] at file level, so the rules below leave model_year out of model_filters. The file has no remove_rules key yet; applying any of these adds that top-level key. The rule field is from_value, and the rule classes live in ai_dagster.derived.projection (rules.py, removes.py).

5a. Validated and not implicated in a hard-criteria failure​

Paste-ready block. PL4 and PYC both remove PAINT_COLOR=ORANGE; PYC's C/J/S/W scope includes PL4's S/W, so keep one (PYC shown, PL4 commented). PL4, PYC and MBU still have open human gates (§7.2, §7.4) on whether to add them at all, because each duplicates derive_support_removes today.

remove_rules:
# CP Premium McKinley seats became 24J package content on Texas Trail; LEATHERETTE leaves the J
# choice set. Replaces the CP ^(J|Q)$ rule (see 5b).
- model_filters:
model_code: "^JTJL98$"
trim_identifier: "^J$"
from_value: LEATHERETTE
attribute_type: SEAT_MATERIAL
# MBU (gloss-black bumper) was ACW Dark Sky content on Sport S; ACW is gone in MY2027, so BLACK
# leaves the S choice set.
- model_filters:
model_code: "^JTJL98$"
trim_identifier: "^S$"
from_value: BLACK
attribute_type: BUMPER_FINISH
# Reign (PHT), the only PURPLE paint on Gladiator, is dropped from the MY2027 JTJL98 palette.
- model_filters:
model_code: "^JTJL98$"
trim_identifier: "^[CJSW]$"
from_value: PURPLE
attribute_type: PAINT_COLOR
# PYC Goldilocks and PL4 Joose leave the MY2027 Gladiator orderable paint set; matches the PAINT
# mandatory_expands entry in gladiator-rules.yaml.
- model_filters:
model_code: "^JTJL98"
trim_identifier: "^(C|J|S|W)$"
from_value: ORANGE
attribute_type: PAINT_COLOR
# PL4 alternative, subsumed by the rule above:
# - model_filters: {model_code: "^JTJL98", trim_identifier: "^[SW]$"}
# from_value: ORANGE
# attribute_type: PAINT_COLOR

What was verified for each rule:

CP, narrowed to ^J$ (Report-phase replacement; status passed; validated by the Report-phase re-run, not by the final combined gate run)

  • Re-run of all rules with CP ^J$ through the real _run_rule_pipeline: C/J/S/W 4608 groups sum to 1 within 4.4e-16 in both rate columns; every from_value residual, including in _X_ cross types, is 0; 0 nulls, 0 negatives, 0 duplicate keys; 0.08–0.09 s. Q rows pass through unchanged (Q keeps its null hl90).
  • The original CP checks below also hold for the J model (4fa2da21…): LEATHERETTE is zeroed, NO_SEAT_MATERIAL is the only recipient, and MY2026 J is unchanged.
  • Not verified: whether MY2027 J candidates have a successor seat material. Per the trim-diff, CP on J is non-configurable 24J content in MY2027 and E7 (CLOTH) is the only configurable seat.

MBU — BUMPER_FINISH:BLACK on S (passed, 1 attempt)

  • _rules_from_file + resolve_remove_rules → exactly one model, f052d9ef… (JTJL98/S/MY2027), matching consolidated.models 2026-10-02.
  • BLACK zeroed: all 279 BLACK-bumper rows (plain, BUMPER_FINISH_X_PAINT_COLOR, GRILLE_FINISH_X_BUMPER_FINISH) are 0.0 in both columns.
  • Recipient: per-dealer NO_BUMPER_FINISH gain equals the BLACK share (max error 1.3e-16, 36 dealers); 329 composite rows equal before / (group total − removed BLACK share) (max error 2.2e-16).
  • All 108 groups sum to 1 (max deviation 2.2e-16; up to 0.32 before, from smoothing).
  • No _SET row has a BLACK token; MY2026 S (7306df09…) unchanged (644 rows).
  • Premise: MY2027 S has 398 feature rows and none is MBU, ACW or a bumper feature.

PHT — PAINT_COLOR:PURPLE on C/J/S/W (passed, 1 attempt)

  • Resolves to exactly 4 MY2027 models (C Rewind, J Texas Trail, S Sport S, W Willys); Q excluded.
  • PHT exists only on MY2026 Gladiator; the MY2027 JTJL98 exterior colors have no purple code.
  • PURPLE zeroed: all 1150 in-scope PURPLE rows across PAINT_COLOR and its 4 cross types.
  • Recipients: all 6436 rows equal old / (group sum − purple sum), max error 1.1e-16.
  • All 720 groups sum to 1 (max 4.4e-16); 9602 out-of-scope rows identical (including 1150 kept MY2026 PURPLE rows); row count unchanged (17188).

PYC — PAINT_COLOR:ORANGE on C/J/S/W (passed, 1 attempt)

  • Resolves to exactly the 4 MY2027 C/J/S/W models; Q excluded.
  • MY2026: PYC and PL4 are the only ORANGE paints, configurable on every trim. MY2027 configurable paints on C/J/S/W are PX8, PBJ, PJ5, PAU, PDS, PW7. PL4 remains on S/W only with is_configurable=null.
  • 10,332 → 7,586 rows after pre-steps. ORANGE zeroed: 804 ORANGE rows (142 plain + cross components).
  • Recipients exact (max 1.1e-16); per-group gain matches (max 2.6e-16); 0 groups off 1; no group was ORANGE-only.
  • Pre-existing: 40 of 144 PAINT_COLOR groups summed > 1.01 in smoothed demand before the rule (up to 1.76).

PL4 — PAINT_COLOR:ORANGE on S/W (passed, 1 attempt; subsumed by PYC)

  • Resolves to exactly 2 models (f052d9ef… S, ea645dce… W).
  • 44,532 → 9,358 rows; 70 plain + 374 cross ORANGE rows zeroed; 360 groups sum to 1 (max 3.3e-16).
  • Production output.demand_take_rates 2026-10-02 already shows ORANGE = 0.0 on both models.

5b. Not ready: implicated in the hard-criteria failure​

CP — SEAT_MATERIAL:LEATHERETTE on ^JTJL98$ / ^(J|Q)$ (rule-level validation passed in 1 attempt, but this rule causes the partition_sums_valid failure)

# DO NOT MERGE — replaced by the ^J$ rule in 5a
- model_filters: {model_code: "^JTJL98$", trim_identifier: "^(J|Q)$"}
from_value: LEATHERETTE
attribute_type: SEAT_MATERIAL
  • What passed: resolves to exactly 2 models (J 4fa2da21…, Q 803954ed…); all 72 LEATHERETTE rows zeroed; mass 56.06 (sales) / 19.23 (hl90) moved proportionally to NO_SEAT_MATERIAL (68 rows, 0 deviations); no _SET or SEAT_MATERIAL cross stripped; 0 changes on MY2026 J.
  • Why it fails: on Q, 4 of 72 dealer groups had LEATHERETTE as their only SEAT_MATERIAL row (NO_SEAT_MATERIAL was pruned below 0.005), so they sum to 0 after the remove. Including Q also pulls Q into apply_removes, which turns Q's all-null hl90 into 0.0 on every partition type.
  • Why Q looks different: JTJL98/Q/2027 has no MY2026 base trim, a null trim_display, 0 synthetic candidates, 0 candidates_output rows and all-null hl90. With fewer than 50 candidates, support-prune never fires on Q (MY2027 Q LEATHERETTE still reads 0.853).
  • Tried: Report-phase attempts 2–3 narrowed CP to ^J$, and all gates pass (§2). The combined gate run still contained the original rule, so the overall result stayed failed.
  • Open: the trim-diff shows CP on MY2027 Q as non-configurable content via 2_Q (is_configurable True → False), which suggests leatherette is standard on High Tide rather than removed. If so, Q needs no seat rule at all (§7.1).

5c. Entries resolved without a rule​

Every remaining confirmed entry, including all became_standard / became_configurable codes, with its ruleResolution. "FA" = FUNCTIONAL_ATTRIBUTE, which is mutually inclusive in template_vars.mutually_inclusive_attribute_types and never renormalized; per-dealer FA sums on these trims are 1.2–11.5, never ≈1.

CodeVerdictDemand tagruleResolutionGateBasis
ACWreal_removenoneneeds_new_attributenew_attribute_registrationUntagged package; child demand: PERFORMANCE_HOOD (MHZ) already zeroed by prune; BUMPER_FINISH BLACK handled by MBU rule
AENbecame_standardROOF_TYPE=DUALnone_needed–No DUAL token in any JTJL98/W demand row either year; remove would be inert
AGDbecame_standardnone (all Jeep AGD untagged)needs_new_attributenew_attribute_registrationNo attribute, so no stranded demand
ALVbecame_standardFA POWER_DOME_HOODnone_needed–No hood FA rows on J; W keeps ALV (0.35 hl90)
CBFbecame_standardnonenone_neededmarginal_value_callCP package component; no demand row; no option-code demand type exists
CSNbecame_standardnonenone_needed–493 rows all untagged; no armrest/cupholder demand
CVMbecame_standardnoneneeds_new_attributenew_attribute_registrationRegistry gap (CP siblings CVP/CVW/JPU/JVG are tagged); no rule for this transition either way
CVPbecame_standardFA LEATHER_PARK_BRAKE_HANDLEnone_needed–Single independent FA code
CVWbecame_standardFA LEATHER_SHIFT_KNOBnone_needed–MY2027 J already 0.0 via support-prune
CWDbecame_standardnone on scope (FLOOR_MATS stops at MY2025)none_needed–No FLOOR_MATS demand on scoped model_ids
CWTbecame_standardnonenone_needed–120/120 untagged
HT2became_standard (actually gone from S/W)ROOF_TYPE=HARD_TOPnone_needed–HARD_TOP stays via HT1/HT3; a remove would erase ≈0.80–0.89 live demand
JPUbecame_standardFA POWER_LUMBAR_ADJUSTnone_needed–Same-value siblings JRS/JRQ; the evidence is from trim C, while the diff names J
JRSbecame_standardFA POWER_LUMBAR_ADJUSTnone_needed–Evidence from older partitions (§7.7)
JVGbecame_standardFA POWER_ADJUSTABLE_SEATSnone_needed–MY2027 J reads 0.0 although every J build has it (§7.7)
JWGbecame_standardFA POWER_ADJUSTABLE_SEATSnone_needed–Same; also _SET gap noted in §7.7
MFJbecame_standardGRILLE_FINISH=WHITEnone_needed–S base model has only NO_GRILLE_FINISH=1.0
MHZbecame_standard (gone)FA PERFORMANCE_HOODnone_needed–All 36 MY2027 S rows already 0.0 via prune
MMUbecame_standard (gone from J)none (accessories untagged)none_needed–ROCK_RAILS exists in registry for XGS/MEF
MMVbecame_standardFA POWER_DOME_HOODnone_needed–ALV keeps value optional on W; remove would erase ≈35% live demand
MU9became_standardnone on JTnone_neededmarginal_value_callDECAL_TYPE axis exists; MU9 holds no share
PASbecame_standard (gone)PAINT_COLOR=GREYnone_needed–GREY offered by PDN; MY2027 GREY 0.188 / 0.178
PGEbecame_standard (gone)PAINT_COLOR=GREENnone_needed–GREEN offered by PJ5
PRVbecame_standard (gone)PAINT_COLOR=REDnone_needed–RED offered by PR6; gap in the synthetic candidates (§4.1)
PSEbecame_standard (gone)PAINT_COLOR=GREYnone_needed–GREY via PDN
PXJbecame_standard (gone on WLJT)PAINT_COLOR=BLACKnone_needed–BLACK via PX8; KMJH74 still configurable
ST2became_standardROOF_TYPE=SOFT_TOPnone_needed–No SOFT_TOP/DUAL demand rows on W
X9became_standardFABRIC_COLOR=BLACKnone_needed–Only fabric color on Q, BLACK=1.0; a remove would zero it and all PAINT×FABRIC crosses
XF2became_standardFA TOW_PACKAGEnone_needed–No TOW_PACKAGE demand on WLJT U either year
XFHbecame_standardFA TRAILER_HITCHnone_needed–Token already gone from 2027 _SETs; FA row reads 0.0 (§7.7)
AWQreal_addnone on scope (DRIVER_ASSISTANCE_PACKAGE elsewhere)none_needed + attribute-override prerequisitemultiple_valid_approaches§7.2
CSQreal_addFA GRAB_HANDLESflat_seed_candidatemultiple_valid_approaches§7.2
XC3real_addnone (GRAB_HANDLES via CSQ)flat_seed_candidatemultiple_valid_approaches§7.2
2S6became_configurableFA SERVICE_PLANnone_needed–2027 SERVICE_PLAN ≈0.33 (JTJL98), 0.27–0.28 (WLJT U)
ALEbecame_configurableFA NON_VENTED_STEEL_HOODnone_needed–Value also base content via MC9/MY1
CSVbecame_configurablenone on JTJL98none_needed–Requires XC3, which always includes tagged CSQ
HT3became_configurableROOF_TYPE=HARD_TOPnone_needed–Same value as HT1, its only alternative in AEN
JFFbecame_configurablenone (ACCESSORY not in binary_feature_types)none_neededmarginal_value_callPERFORMANCE_AIR_INTAKE exists for OPTION codes
JJ2became_configurablenonenone_needed–Speed calibration that follows the tire choice
MHJbecame_configurablenonenone_neededmarginal_value_callGRILLE_FINISH=NO_GRILLE_FINISH 1.0 on all trims
PD7became_configurablePAINT_COLOR=WHITEnone_needed–WHITE offered by PW7 (0.408 / 0.427)
PDNbecame_configurablePAINT_COLOR=GREYnone_needed–Inherits PAS/PSE GREY bucket
PR6became_configurablePAINT_COLOR=REDnone_needed–RED history from PRV; the fix is in the synthetic-candidate rules (§4.1)
PX8became_configurablePAINT_COLOR=BLACKnone_needed–2027 BLACK 0.2045 / 0.1951
RF7became_configurablenoneneeds_new_attributemarginal_value_callMinor USB item; no demand rows
RHYbecame_configurableFA FRONT_PASSENGER_DISPLAY (WLJT 2026 only)none_needed–Flip is on JTJL98 only, where it is untagged
RJAbecame_configurableFA FRONT_PASSENGER_DISPLAYnone_neededmarginal_value_call§7.4
RTMbecame_configurableFA TELEMATICS_MODULEnone_needed–Same-value siblings RTQ/RTV/RDG
RTQbecame_configurableFA TELEMATICS_MODULEnone_needed–Demand exists on JTJL98 both years
RTUbecame_configurableFA SATELLITE_RADIOnone_needed–C has no SATELLITE_RADIO demand row (§7.7)
TTKbecame_configurableTIRE_RIM_SIZE=17none_needed–Only tire on Q; RIM_SIZE 17 = 1.0
X9Bbecame_configurableFA SATELLITE_RADIOnone_needed–2027 J/S/W ≈0.47–0.48
X9Gbecame_configurableFA SATELLITE_RADIOnone_needed–Same-value RTU/X9B present
XXEbecame_configurableFA ONBOARD_POWER_SYSTEMnone_needed–Same-value XBE/JKV/XHR
YEPbecame_configurablenonenone_needed–$0 MSO paperwork code
XPRreal_expand_memberFA INTERIOR_CAMERAnone_needed–Same code and tag under both AAN and AWQ

6. Branch-deploy readiness​

Not ready: hard criteria did not pass. The partition to validate is 2026-10-02. Even after the CP fix, the branch push and the single-partition materialization run only when a human explicitly asks for them.


7. Human decisions needed​

7.1 failed_hard_criterion​

partition_sums_valid: CP's Q branch. Choose one:

  • (a) Use CP ^J$ (§5a). Verified: every gate passes. Q's LEATHERETTE demand (0.853) stays, and it reaches no ordering decision today because Q has 0 candidates. Support-prune removes it once Q has ≥ 50 candidates without LEATHERETTE.
  • (b) Treat Q (High Tide) as a real MY2027 trim. Then decide Q's seat material. The trim-diff suggests leatherette is standard on Q (CP non-configurable via 2_Q), which would mean no seat rule at all. Also decide whether PHT, PYC and PL4 should include Q, which keeps ORANGE at a 1.871 sum.

7.2 multiple_valid_approaches​

  • PL4 and PYC (PAINT_COLOR=ORANGE, JTJL98). Add an explicit remove as a safeguard in case a trim falls below the 50-candidate prune threshold, or rely on the prune (output already shows ORANGE = 0.0 on C/J/S/W). The gladiator-rules.yaml header sets the convention of no rule for codes that no longer exist. If added, keep one rule (PYC's C/J/S/W scope). Scope risk: if real MY2027 PL4 vehicles (S/W, is_configurable=null) ever become candidates, an S/W rule would zero real ORANGE demand; limiting to ^(C|J)$ avoids that.

  • AWQ (PROTECH GROUP IV, WLJT74/75 U+W). AWQ is a $0 reduced version of AAN: AAN minus XNT night vision, and mutually exclusive with AAN. AAN dropped from $995 to $495. AWQ is untagged on these 4 rows; all 73 other AWQ rows are FA DRIVER_ASSISTANCE_PACKAGE. Options:

    1. An attribute override tagging AWQ DRIVER_ASSISTANCE_PACKAGE on ^(WLJT74|WLJT75)$ MY2027 ^(U|W)$ (a DynamoDB write; needs approval), accepting the MY2026 rate.
    2. The same override plus a ProxyEstimateRule uplift, since a $0 package likely widens the population. No clean proxy was found.
    3. Leave AWQ untagged, since child tags (HEADS_UP_DISPLAY, INTERIOR_CAMERA) already hold the content.

    Separately, grand-cherokee-rules.yaml has no AWQ rule, so U candidates never contain AWQ.

  • CSQ and XC3 (GRAB_HANDLES, JTJL98). The existing assumed rate (GRAB_HANDLES 0.1, ^JTJ, ^[ADRSWY]$) covers S/W only; C and J have no GRAB_HANDLES row. Options: widen the trim class to include C, J (and Q?) at 0.1, choose a different rate for C/J, or leave them without a GRAB_HANDLES rate. Pair it with §4.2. Also note that standard CSP front handles share the GRAB_HANDLES value, and CSV is untagged on JT.

7.3 new_attribute_registration​

  • ACW (Dark Sky Appearance Package, $995, JTJL98/S). Register a package-level FA (for example an appearance-package value; BLACKOUT_APPEARANCE_PACKAGE is the nearest analog) or decide child tags are enough. Either way this transition needs no ACW rule: PERFORMANCE_HOOD is already pruned and BUMPER_FINISH BLACK is handled by the MBU rule.
  • AGD (WHITECAP Package, JTJL98/S). Untagged on every Jeep model. Options: a WHITECAP/appearance FA, or tag it ROOF_TYPE (it includes a top). No effect on this transition; it matters for models still offering AGD (JLJP74 G, JLJS74 R/Y, JTJH98 A/D, JTJP98 G, JTJS98 R/Y).
  • CVM (full-length floor console). Low value: interior convenience content that only appears in packages (CP, PL, VL). If declined, treat as none_needed. Note the same gap for CSN, CBF, CWT.

7.4 marginal_value_call​

  • MBU. Add the explicit BUMPER_FINISH BLACK remove on S (validated, §5a), or rely on support-prune, which already outputs BLACK = 0.0 / NO_BUMPER_FINISH = 1.0 on 2027 S.
  • CBF (Premium Door Trim Panel). Untagged everywhere; becomes 24J content on J. Tracking it matters only on the 22 configurable rows elsewhere in MY2027.
  • MU9 (1941 hood decal). Decide whether to extend DECAL_TYPE=HOOD_GRAPHIC to JT MU9 rows. If tagged, DECAL_TYPE becomes a shared axis on JT and this flip would need a HOOD_GRAPHIC remove.
  • JFF (Mopar cold air intake) and MHJ (Mopar satin black grille). Accessories are untagged by design (binary_feature_types: [OPTION, PACKAGE]). Tracking them is an OEM-wide policy change (would map to PERFORMANCE_AIR_INTAKE / GRILLE_FINISH), not a transition rule.
  • RF7 (USB Host Flip). The analysis leans toward not tracking it. If tracked, a new standalone attribute would still need no rule.
  • RJA (Front Passenger Interactive Display, WLJT74/75 U). The analysis found the U demand row at 0.0 (attribute_stats history, because RJA was standard on U). An assumed_rate_rules entry already exists on main (^WLJT7[45] U, FRONT_PASSENGER_DISPLAY 0.065). Check in MY2027 output.demand_take_rates whether it takes effect over the measured 0.0 before adding anything.

7.5 uncertain classification (all parent_changed)​

  • GFA (JTJL98/S), confidence 0.6; most reviewers disputed the classification. The parent set shrank {AEN, HT1, HT3} → {AEN, HT1}; GFA stays orderable via AEN or HT1. The classifier leaned toward no rule. gladiator-rules.yaml already adds [CS2, GFA, GFD] to S candidates with HT1 and with AEN, which matches the 2027 parent set. Decide: noise, or review GFA/HT1/HT3 together as a package-membership change.
  • HT1 (JTJL98/W), confidence 0.55. Detached from AEN (AEN is gone on W) and now standalone configurable; still under AEN on J. W ROOF_TYPE HARD_TOP ≈0.89 in 2027. Decide whether W candidates that had AEN keep HT1 when AEN auto-retires "with its content". If HT1 is retired with AEN, W candidates lose a top that is still orderable standalone.
  • HT3 (JTJL98/C,S), confidence 0.7; most reviewers disputed the classification. 2027 parent AEN appears only on code-guide rows; configure/catalog_option/equipment_categories have no MY2027 data yet. Probably a source-coverage artifact, but a real packaging change cannot be excluded until MY2027 configure data arrives.
  • LBK (WLJT74/75 U), confidence 0.55. Additive: AAN kept, AWQ added as a second parent. Depends on the AWQ decision (§7.2).
  • MM3 (JTJL98/S), confidence 0.7; most reviewers disputed the classification. The basis note does not hold on S: MM3 is not configurable on S in either year; the ACW/AJK parents come from trims H and Q. Note that gladiator-rules.yaml adds MM3 to S candidates with AJK, which assumes MM3 is AJK content on S. Confirm against the S section of the trim-diff.

7.6 failed_validation​

None. All five concrete rules passed on their first real run.

7.7 Follow-ups outside this proposal (no rule in this proposal)​

  1. JTJL98/Q (High Tide) data state. No MY2026 base trim, 0 candidates, all-null hl90, 1036 orphaned FA rows, stale ORANGE (1.871 sum) and LEATHERETTE (0.853), PAINT_COLOR sums up to 1.59, and a leftover TIRE_TYPE ALL_TERRAIN row (0.568) from TTP history.
  2. JTJL98 C/J coverage. Synthetic rules target S/W only. Gate-reported FA values with no demand row: C NON_VENTED_STEEL_HOOD, C SERVICE_PLAN, J ROOF_RAILS, J SATELLITE_RADIO, J SERVICE_PLAN, J TONNEAU_COVER, W SERVICE_PLAN. Analyses also found no GREEN/BLUE PAINT_COLOR rows and no SATELLITE_RADIO row on MY2027 C. (The 2S6 analysis reports SERVICE_PLAN ≈0.33 in final output; the gate measured before the assumed-rate stage. Confirm which is current.)
  3. Guaranteed FA attributes reading 0.0 in MY2027: POWER_ADJUSTABLE_SEATS and LEATHER_SHIFT_KNOB on J, TRAILER_HITCH on WLJT U. And 2027 J FUNCTIONAL_ATTRIBUTES_SET has no set containing POWER_ADJUSTABLE_SEATS, although every J build now has it. This is a systemic projection question, not a per-code rule.
  4. Tagging inconsistencies:
    • HT3 is tagged ROOF_TYPE OTHER (2026) vs HARD_TOP (2027) on S.
    • Historical J demand encodes AEN builds as DUAL,SOFT_TOP. Whether candidates and demand use the same ROOF_TYPE encoding is unverified.
    • JRQ manual lumbar is tagged POWER_LUMBAR_ADJUST.
    • CV3 urethane knob is tagged LEATHER_SHIFT_KNOB.
    • JVG is split between POWER_ADJUSTABLE_SEATS and SEAT_MEMORY within one model_id.
    • CWD lost FLOOR_MATS from MY2025.
    • TTK lacks TIRE_TYPE/TIRE_SEASON.
    • RHY (not RJA) has FRONT_PASSENGER_DISPLAY on WLJT 2026.
    • X9 is retagged FABRIC_MATERIAL FAUX_LEATHER on 2027 Q, against a CLOTH history.
    • RF7 is tagged TELEMATICS_MODULE on DT 2022.
  5. Ground-truth label mismatches. Many removed_became_standard codes actually left the trim (for example HT2, MBU, MHZ, MMU, PGE, PHT, PL4, PRV, PSE, PXJ, PYC, ST2, ALV on J), because the classifier falls back to OEM-wide presence. The treatment did not change, but the classifier's label is misleading.
  6. Evidence caveats. JRS used attributes 2026-09-09, smoothed demand 2026-09-30 and demand_take_rates 2026-07-07; PDN used demand_take_rates 2026-07-17; PXJ used 2026-09-30. The AGD and CWT demand-table scans, the CP candidate seat-material lookup, the HT3 candidates_output query and the ROOF_TYPE encoding check did not complete. Those conclusions rest on tag absence or on the other evidence named in each entry.

8. Filtered noise​

205 candidates were excluded. Check the per-trim sections of the trim-diff to confirm nothing real was dropped.

parent_changed, classified no per-code demand effect (8):

CodeTrimParent 2026 → 2027
CS2JTJL98/SAEN,HT1,HT3 → AEN,HT1
CS2JTJL98/WAEN,HT1 → HT1
GFAJTJL98/WAEN,HT1 → HT1
GFDJTJL98/SAEN,HT1,HT3 → AEN,HT1
GFDJTJL98/WAEN,HT1 → HT1
LMXJTJL98/SACW,AJK,AJQ → AJK,AJQ (ACW retired)
NHLWLJT74/75 UAAN → AAN,AWQ
TTKJTJL98/W24W → –

The code stays orderable through a surviving parent in every case. GFA on S, HT1, HT3, LBK and MM3 went to human review instead (§7.5).

[trim-level] removed_trim_discontinued (130 codes). JTJL98 trims B (Sport), F (85th Anniv Ed), H (Sahara) and T (Willys 41) have 0 rows in MY2027, so every code on them drops with the trim: ACG, ACW, ADA, ADH, AEN, AGD, AJ1, AJK, AJQ, AJY, ALV, CBF, CBQ, CBS, CHD, CLN, CLP, CMD, CP, CS2, CSD, CSN, CSU, CT6, CT9, CTB, CUW, CVM, CVP, CVW, CWA, CWJ, CWT, D7, DJF, DMF, DRF, DS7, DSA, E7, GFA, GFD, GMA, GNK, GXD, HAB, HT1, HT2, HT3, JAL, JJ2, JJ3, JKV, JLN, JPM, JPU, JRS, JVG, JWG, LAY, LFR, LH2, LMS, LMX, LNV, LPX, LPY, M9A, M9B, M9C, M9D, M9E, MBU, MDA, MEG, MEJ, MF, MFJ, MHJ, MHW, MHZ, MKM, MKN, MM2, MM3, MML, MMU, MMV, MRT, MRU, MU9, NHS, NMC, PAU, PBJ, PDS, PGE, PHT, PJ5, PL4, PRC, PW7, PX8, PYC, RC4, RE8, RS2, RSR, RTV, ST2, T5, TTH, TTK, UBX, WFD, X9, XAA, XAN, XBE, XBM, XFH, XGD, XHR, XJC, XMF, XNQ, XRB, XWX, YEP, Z6K. (gladiator-rules.yaml bridges JTJL98|H → JTJP98|G, so Sahara demand is handled there, outside this pattern.)

[trim-level] added_trim_new (67 codes). KMJH74/H (Cherokee Trailhawk), WLJT74/W and WLJT75/W (Summit Obsidian) have 0 rows in MY2026, so every code on them is new with the trim: 2S6, AAN, ACX, ADS, AHE, APP, AV1, AWQ, AWX, CAJ, CF1, CFQ, CG6, CMK, CSH, CSN, CYE, D2, EC, GNZ, GRG, GWJ, GX, JHE, JLW, JPZ, LBK, MDA, MEJ, MKM, MM8, MRT, MRU, MXS, NHL, PB5, PBJ, PD7, PDN, PJ5, PPS, PR6, PVB, PW7, PX8, PXJ, RC3, RHH, RTC, SJH, T7, TBB, WD3, X7, X9G, XAK, XDR, XF2, XFJ, XNQ, XNT, XPG, XPM, XPR, XPV, XRW, YEP. These are cold-start trims with new base_model_ids and no demand history. On 2026-10-02 the WLJT74/W and WLJT75/W trims have no candidates and no PAINT_COLOR demand rows, a trim-level gap separate from any per-code rule.

This section omits removed_became_standard (34) and added_became_configurable (22). They need no synthetic-candidate rule, but each went through the demand-axis check and is listed in §5 with its ruleResolution.