feat: Full Stack v12 — safe traffic ramp-up, variation preview & experiment exclusion by abbaseya · Pull Request #56 · convertcom/python-sdk

abbaseya · 2026-07-16T16:27:18Z

Full Stack v12 — next major SDK release

This branch delivers three major capabilities, shipped consistently across all Full Stack SDKs. All three are inert until activated on the backend/UI side, so this deploys safely in any order.

1. Anchored (ramping) bucketing layout

Makes changing an experiment's total traffic allocation safe on a running experiment.

Each variation is pinned at its own fixed offset (cumulative-weight anchor); its range width equals its allocation. Raising the total grows every arm at its own edge (even growth); lowering ejects evenly; nobody already bucketed gets reshuffled.
Gated on the experience's existing serving version property: > 11 → anchored pass (contract v12); ≤ 11 / missing → packed pass, bit-identical to today. No schema change, zero migration.
Hash, seed, stored-decision guard, forced path, and events are untouched. Still stateless, deterministic, and identical across every SDK — enforced by shared cross-SDK golden vectors.
Changing variation weights mid-flight stays unsupported (unsafe in every layout); the split lock is unchanged.

2. Experiment preview

Lets users preview any variation — including drafts and paused ones — in their own environment, with zero trace.

New debugToken config option (24h QA token) unlocks the full config, including draft/paused experiences.
Stateless preview links (web) and deep links (mobile) via convert_preview={expId}.{varId} force a single variation with no side effects: no tracking events, no visitor-state writes, no impact on reporting.
Rides the existing public serving primitive — no backend serving changes required.

3. Mutual exclusion / inclusion audience rule

Lets teams control which experiments a visitor can be in at the same time, through the normal audience-targeting rules.

New audience rule bucketed_into_experience_key, resolved against SDK-stored bucketing state.
negated: true → "NOT in experiment X" (mutual exclusion); default → "in experiment X" (inclusion).
No new application inputs; evaluates before bucketing and composes identically with both packed and anchored layouts. Old SDKs fail closed.

Notes

Backend registration of the new rule type and the version 12 stamp, plus the related convert-ui work (preview link/QR button, rule picker) and shared docs, are tracked and coordinated separately.
Full existing test suites remain green; the anchored pass adds the shared golden-vector acceptance tests.

… fixture Byte-for-byte copy from javascript-sdk feat/anchored-bucketing-layout (packages/bucketing/tests/cross-sdk-bucketing-vectors.json). sha256 9993d0e6c5719cb17a4dcbcc73af9c0cb14ee0258e7fd29cc88f44b61083dd1b 59 vectors, versions {11,12}. The fixture IS the cross-SDK contract. Beads: ai-driven-product-dev-padw Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

TDD Phase 1: pytest AC1-AC9 + 59-vector parameterized golden runner. RED confirmed: 29 failed / 51 passed / 1 collection error — ImportError/ AttributeError on not-yet-existing anchored selectors + v12 value mismatches; v11 packed vectors pass (AC6 lock intact). Beads: ai-driven-product-dev-k1vf Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…EEN) TDD Phase 2: anchored layout gated on experience.version > 11. - bucketing.py: build_bucket_ranges + select_bucket_anchored (anchor scale reuses DEFAULT_MAX_TRAFFIC; half-open anchor<=v<anchor+width; inactive arms keep weight, zero width). - experiences.py: _is_anchored_layout (float() coercion, '12' routes anchored like JS Number('12')>11) + _build_variation_allocations (full ordered list incl. inactive arms); gate wired into select_experience. Packed path (_build_buckets + select_bucket) byte-for-byte untouched (AC6). - Fixed one RED test bug: (11.9, False) -> (11.9, True), grounded on the JS oracle's literal Number(version)>11 (no flooring). Gates: full suite 949 passed; golden 59/59; mypy strict clean; ruff clean. Beads: ai-driven-product-dev-k1vf Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…unt + conftest type Documentation-only (code review R1, conf 88%): rewrote the TDD RED-phase module docstrings in the 3 new test files to describe current GREEN state; corrected the pair-count figure 18 -> 8 (verified against the fixture); fixed conftest _load return-type annotation to Union[dict, list] for the bare-list anchored fixture. No source/fixture/assertion changes. Full suite 949 passed; ruff clean. Beads: ai-driven-product-dev-k77j Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/parity/README.md + TESTING.md: list the anchored golden-vector fixture (anchored_bucketing_vectors.json, bare list, imported verbatim from the cross-SDK source) and its parity test alongside the MurmurHash3 parity proof; note it is not written by the fixture generator. Part of PR #52. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

BREAKING CHANGE: Experiences on bucketing contract v12 (experience version above 11) bucket visitors with the anchored layout, where each variation holds a fixed position derived from its cumulative traffic allocation. Changing an experience's total allocation now grows or shrinks each variation at its own boundary instead of shifting every variation range, so visitor-to-variation assignments under contract v12 differ from the earlier packed layout for the same configuration.

feat: anchored bucketing layout (contract v12) — Python SDK pass

Beads: ai-driven-product-dev-cope 19 new tests fail (missing debug_token field); 34 pre-existing pass. ruff clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-cope SDKConfig.debug_token: Optional[str] = field(repr=False) with strict validation. Config route carries debug_token + forced _conv_low_cache=1; never on track route. Full suite 968 passed, coverage 96.81% (eval 98%), mypy --strict + ruff clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-rq0l Pure helper convert_sdk.parse_preview_param; '123.456'->('123','456'), malformed->None (strict numeric-segment check, JS/Ruby sibling parity, decision I6). 18 tests, full suite 986 passed, cov 96.84% (preview.py 100%), mypy+ruff clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-lgao 11 tests fail (get_preview_decision absent). Locks full-bypass matrix incl draft/paused/ non-running/zero-traffic variation + unknown-id->None. ruff clean. NOTE: experience-status & environment bypass cases are forward-compat locks — Python's normal path does not gate on those today (decision I9; flagged for story owner). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…REEN) Beads: ai-driven-product-dev-lgao get_preview_decision(experience, variation_id) in evaluation/experiences.py: reuses _find_variation (no status/traffic filter) -> full bypass; normal ExperienceResult shape; unknown variation->None; pure/no-side-effects. 11 tests, full suite 997 passed, eval cov 98%. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-5801 8 failed + 11 errored (missing fetch_config_by_experience + module memo globals). Locks: exp=<id>&_conv_low_cache=1 (+debug_token) unconditional; 60s wall-clock TTL; thread-safe single fetch under 6-thread barrier; cross-tenant isolation; never-DataStore; failed fetch not memoized (JS parity). ruff clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-5801 HttpxTransport.fetch_config_by_experience + _build_experience_route/query: exp=<id>& _conv_low_cache=1 (+debug_token), no environment (spec Resolution clause, I16). Module-level thread-safe 60s wall-clock memo keyed sdk_key:experience_id, never DataStore, failed fetch not memoized. Test-only isolation fixture added (I18). Full suite 1016 passed, cov 96.90%/eval 98%. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-1b47 15 fail (Context.set_preview absent) + 1 diagnose regression-lock passes. Locks AC3 (?exp=-only delivery), AC4 (full-bypass matrix incl different stored decision), AC6 (isolation: concurrent non-preview context normal), AC7 (inert: unknown exp/var, fetch 503 -> warning + normal). Transport kept a public port (I22 -> capability protocol in GREEN). ruff clean; 1016 pre-existing pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-1b47 Context.set_preview(experience_id, variation_id): eager resolution (local snapshot -> ?exp= fetch via SupportsPreviewFetch capability protocol -> get_preview_decision), per-context state, precedence over stored/bucketed decision for target only, other experiences normal, inert+warn on bad input. run_experiences synthesizes fetch-only target (I27 test locks it). Transport stays a non-breaking public port. 18 targeted, full suite 1034 passed, cov 96.81%/eval 98%, mypy+ruff clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-oq4q 5/6 fail (spies see track POSTs + store writes that must be zero under preview); 1 AC10 regression lock passes (normal path unaffected). Full-lifecycle proof across core.flush() + atexit hook; AC6 isolation via shared Core; all persist/track sites swept (I34). transport + _SpyDataStore spies. ruff clean; PY-5 suite 18/18 untouched. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-oq4q Preview context leaves zero trace: track_bucketing skipped in run_experience/run_experiences (overrides enable_tracking=True), track_conversion unconditional no-op (returns DEDUPLICATED, I35), _persist_visitor_state no-ops -> zero DataStore writes. Gated at Context call sites (P2), per-context only (AC6 isolation). Full lifecycle incl flush()+atexit. Full suite 1040 passed, cov 96.82%/eval 98%, mypy+ruff clean, generated-pyi OK. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…on write Beads: ai-driven-product-dev-cszc Sweep-on-write inside _CONFIG_BY_EXPERIENCE_LOCK: closes unbounded-growth path keyed by link-param experience_ids; completes TTL-60s semantics. AC8/never-DataStore/failed-fetch- not-memoized/thread-safety preserved. ruby-sdk#41 parity (I36). 13 memo tests, full suite 1042 passed, cov 96.83%/eval 98%, mypy+ruff clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

An experience dict with a missing or null "id" stringified to "None" and would falsely match a lookup for the literal experience_id "None" (str(None) == "None"). Reject a None id before the string comparison so malformed/id-less experiences never match — honoring the helper's documented "never mismatch on malformed bodies" contract. Adds 5 parametrized regression cases. Addresses Gemini code-review finding on #53. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-mm0v 18 new tests fail (field/method/envelope absent); 56 pre-existing pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-mm0v ContextState.bucketing field + with_bucketing (key-merge, carried through with_attributes/with_segments); 3-key envelope persist; hydrate w/ legacy 2-key + plain-attributes backward compat (empty bucketing). No eval/hash change. Full suite 1065 passed, 96.85% cov; ruff + mypy --strict clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-v0d1 5 new tests fail (sticky_bucketing param absent); 10 pre-existing pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-v0d1 select_experience gains pure sticky_bucketing param; short-circuits between qualifies() and get_bucket_value_for_visitor when stored vid resolves via _find_variation. Fresh+sticky returns share _build_experience_result helper (value-neutral). None-arg byte-identical. AC2/AC4/AC8/AC9. Full suite 1070 green, 96.87% cov; eval floor 98%; parity 214 green (AC9); ruff+mypy strict clean. Parity fixtures untouched. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-av5a 6 integration tests (AC1/2/3/5/6/8) over exp-a/exp-b 8-row fixture fail (enable_storage/read/write wiring absent). AC4 row4 added next (GREEN dispatch). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-av5a enable_storage=True kwarg on run_experience/run_experiences (threaded via _evaluate_and_record); sticky read wires self._state.bucketing into select_experience (non-preview branch); gated write-after-hash (result and enable_storage and preview is None and bucketing.get(exp_id)!=vid) -> with_bucketing + _persist_visitor_state. AC1/2/3/4/5/6/8. Full suite 1077 green 96.87% cov; eval floor 98%; parity 214 (AC9); ruff+mypy strict clean. AC6 asserts filtered to state envelope (dedup marker is orthogonal tracking artifact). set_attributes/set_segments untouched. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…oseout Beads: ai-driven-product-dev-rudj AC7: shared DataStore across two Cores -> sticky via hydration (zero hash on Core B); negative control: unshared per-instance stores do NOT share. No src change (PY-1..PY-3 mechanism already complete). AC9: parity 214 green. Full suite 1079 green 96.87% cov; eval floor 98%; mypy strict (core + serving_config_contract) + pyi-header gate clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-g4pc 6 new tests (AC11 conversion attribution x2, AC12 feature x2, AC13 diagnostics x2) fail w/ substantive assertion mismatches (sibling paths re-hash fresh); 9 pre-existing pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-g4pc Thread self._state.bucketing (pure read-only) through every variation-resolution path: run_feature/run_features (features.py), conversion attribution (conversions.py _compute_bucketing_assignments + create_conversion, tracker.py track), diagnose_experience/diagnose_feature (context.py). Read-only consumers mint no decision; write stays experience-run-only. Fixed stale conversions.py docstring. AC11 (attribution sticky both paths)/AC12 (feature sticky)/AC13 (diagnostics sticky). Full suite 1085 green 96.87% cov; eval floor 98%; parity 214 (AC9); mypy strict + serving contract + pyi-header clean. Hash contract + fixtures untouched. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…tr-coerce rehydration Beads: ai-driven-product-dev-93nf Gemini code-assist (PR #54): bucketing param + _hydrate_visitor_state return + stored_bucketing local -> Optional[Mapping[str,str]]; rehydration coerces keys+ values to str (defensive vs a DataStore persisting non-string IDs). +1 test. Type-only + one coercion; no write-path/behavior change. mypy strict clean, full suite 1086 green, parity 214. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-rloj Agent: sdk-python Additive/inert: RULE_TYPE_BUCKETED_INTO_EXPERIENCE_KEY + log_mutual_exclusion_target_not_found; no behavior change (AC7). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…data only) Beads: ai-driven-product-dev-xt69 Agent: sdk-python NamedTuple 8-row table (verbatim spec values) + build_mutual_exclusion_config(); 5 genuine sanity invariants. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-nz93 Agent: sdk-python 20 failed / 1 passed (default-ANY regression guard). Genuine RED via get_experience_by_key spy. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…udiences + bucketed_into_experience_key Beads: ai-driven-product-dev-nz93 Agent: sdk-python Audience-level exclusion resolver (read-only, presence-only, negation-last, AC8 warn) + matching_options ALL/ANY (default ANY). is_rule_matched + generic walk byte-identical (AC7 proven). 21 new pass, full 1114, cov 96.71%, eval floor 97%, parity 214. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Beads: ai-driven-product-dev-b3sb Agent: sdk-python 15 tests: 3 generic match_types (equals/contains/less) bit-identical; matching_options ALL/ANY boundary (default-ANY == pre-seam); mixed-tree fails-closed guard. src untouched. Full 1129, cov 96.76%, eval 97%. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

feat(preview): experiment preview — debug_token + set_preview (qs-02)

feat: sticky-bucketing persistence (qs-03) — per-visitor decision map + full read-back parity

feat: mutual-exclusion audience rule bucketed_into_experience_key (qs-04)

gemini-code-assist

Code Review

This pull request implements several key features for the Convert Python SDK, including the anchored bucketing layout (contract v12), experiment preview capabilities with zero-trace suppression and process-wide memoization, sticky-bucketing persistence via a 3-key envelope, and mutual-exclusion audience rules (bucketed_into_experience_key). It also includes comprehensive integration and parity tests. Feedback on the changes highlights a robustness issue in the _sole_rule_item helper, which lacks defensive type-checking for nested configuration structures and could crash on malformed inputs; implementing the suggested type-validation block will ensure the SDK fails closed gracefully.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

Wire qs-02 preview into demo/run_demo.py via a new CONVERT_DEMO_PREVIEW env var ("{experienceId}.{variationId}"): parses with parse_preview_param, calls context.set_preview, and prints the forced decision. Adds a dedicated paused (non-running) preview-target experience to config_fixture.json with numeric ids (900210001 / 900210101 / 900210102) so a real convert_preview link parses and the demo proves the two headline guarantees — a paused experience's variation is force-rendered (bypassing status + bucketing), and the preview context is zero-trace (track_conversion with force_multiple=True still yields tracked=False, distinguishing real suppression from ordinary dedup). A different experience on the same context is unaffected; malformed/unresolvable ids degrade to the AC7 inert-with-warning path. demo/README.md gains a "Testing preview links" section and .env.example documents the new vars. 5 TestPreviewFixture tests added. demo/ + fixture + docs + tests only — no SDK source change. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

usmanabbas7 · 2026-07-22T17:00:31Z

Final release review — sign-off ✅ (with one non-blocking follow-up)

Reviewed the consolidated branch first-hand in an isolated worktree at 0d4dd38, focusing on how the three capabilities compose (each PR was already approved in isolation; the interactions across the overlapping files — bucketing.py, experiences.py, context.py, rules.py — were not).

Verdict: approve for merge.

Quality gates (CI-pinned deps)

Check	Result
Full test suite	1,134 passed
Ruff (`E/W/F/B/SIM/RUF`)	clean
mypy `--strict` (shippable package)	clean, 47 files
Coverage	97% total; bucketing 100%, experiences 99%, rules 95%
Merge integrity	clean linear stack `anchored → preview → sticky → exclusion`; no conflict markers
JS anchored-bucketing parity (72 vectors via real `select_experience` seam)	pass

Capability behavior (empirically probed, not assumed)

Safe Traffic Ramp-Up — stopping an arm in a 3-way split leaves the other arms' visitors exactly in place (A→A, C→C, stopped→idle), distribution byte-identical pre/post. The old packed v11 path reshuffles the same scenario. The version > 11 gate routes legacy configs to the old path and v12 to the safe one. ✔
Variation Preview — zero-trace suppression is comprehensive and per-context: bucketing events, conversions (track_conversion short-circuit), and visitor-state persistence all suppressed; forced decision bypasses qualification/bucketing/exclusion by construction; a concurrent non-preview context sharing the tracker/store is unaffected. ✔
Experiment Exclusion — works on the default path; qualification checked before sticky read-back (can't resurrect an unqualified visitor); read-only against stored bucketing state. ✔

One finding — non-blocking, recommend follow-up

run_experiences(enable_storage=False) silently disables mutual exclusion — returns ['exp-a', 'exp-b'] where the default returns ['exp-a']. Cause: exclusion reads the accumulated self._state.bucketing map, but that map is only populated by the write-after-hash in _evaluate_and_record, which is gated by the same enable_storage flag. enable_tracking=False is unaffected.

Severity: low-medium. Default path is safe; the break needs an explicit opt-out of persistence while relying on exclusion in the same call, and cross-request exclusion requires a DataStore anyway.
Not covered by tests — integration tests only exercise two sequential singular run_experience() calls with default storage.
Parity note: the port cites parity with JS BucketingAttributes.enableStorage, but that field isn't present in the JS repo on main/fetched branches; JS _retrieveBucketing writes the in-memory bucketing map unconditionally on a fresh bucket (separating in-request accumulation from persistence in a way this port conflates). Could not verify the exact enableStorage branch semantics against the JS oracle.

Suggested follow-up (post-merge): add a test for run_experiences(enable_storage=False) + exclusion, and confirm intended semantics against JS — likely fix is to decouple the in-request exclusion-input map from the persistence flag.

_{🤖 Generated with Claude Code}

abbaseya and others added 30 commits July 4, 2026 02:55

Merge pull request #52 from convertcom/feat/anchored-bucketing-layout

466ab51

feat: anchored bucketing layout (contract v12) — Python SDK pass

wf(wf-pysdk-qs02): PY-1 debug_token — tests (RED)

4496fbb

Beads: ai-driven-product-dev-cope 19 new tests fail (missing debug_token field); 34 pre-existing pass. ruff clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

wf(wf-pysdk-qs03): [PY-1] bucketing substrate — tests (RED)

ad772e6

Beads: ai-driven-product-dev-mm0v 18 new tests fail (field/method/envelope absent); 56 pre-existing pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

wf(wf-pysdk-qs03): [PY-2] sticky read-back seam — tests (RED)

c577181

Beads: ai-driven-product-dev-v0d1 5 new tests fail (sticky_bucketing param absent); 10 pre-existing pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

abbaseya and others added 8 commits July 15, 2026 22:27

wf(wf-pysdk-qs04): [PY-3] tests (RED) — seam AC1-AC9

ddbd679

Beads: ai-driven-product-dev-nz93 Agent: sdk-python 20 failed / 1 passed (default-ANY regression guard). Genuine RED via get_experience_by_key spy. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Merge pull request #53 from convertcom/feat/experiment-preview

44f4d69

feat(preview): experiment preview — debug_token + set_preview (qs-02)

Merge pull request #54 from convertcom/feat/sticky-bucketing-persistence

d5f9120

feat: sticky-bucketing persistence (qs-03) — per-visitor decision map + full read-back parity

Merge pull request #55 from convertcom/feat/mutual-exclusion-rule

303ddb1

feat: mutual-exclusion audience rule bucketed_into_experience_key (qs-04)

abbaseya self-assigned this Jul 16, 2026

abbaseya requested a review from usmanabbas7 July 16, 2026 16:27

gemini-code-assist Bot reviewed Jul 16, 2026

View reviewed changes

Comment thread src/convert_sdk/evaluation/rules.py

github-code-quality Bot found potential problems Jul 16, 2026

View reviewed changes

abbaseya changed the title ~~Full Stack v12: safe traffic ramp-up, variation preview & experiment exclusion~~ feat: Full Stack v12 — safe traffic ramp-up, variation preview & experiment exclusion Jul 16, 2026

abbaseya closed this Jul 16, 2026

abbaseya reopened this Jul 16, 2026

usmanabbas7 approved these changes Jul 22, 2026

View reviewed changes

abbaseya merged commit 1f4bd82 into main Jul 24, 2026
27 checks passed

abbaseya deleted the feat/fullstack-v12 branch July 24, 2026 17:00

Navigation Menu

Uh oh!

feat: Full Stack v12 — safe traffic ramp-up, variation preview & experiment exclusion - #56

feat: Full Stack v12 — safe traffic ramp-up, variation preview & experiment exclusion#56
abbaseya merged 39 commits into
mainfrom
feat/fullstack-v12

abbaseya commented Jul 16, 2026

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

usmanabbas7 commented Jul 22, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

abbaseya commented Jul 16, 2026

Full Stack v12 — next major SDK release

1. Anchored (ramping) bucketing layout

2. Experiment preview

3. Mutual exclusion / inclusion audience rule

Notes

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

usmanabbas7 commented Jul 22, 2026

Final release review — sign-off ✅ (with one non-blocking follow-up)

Quality gates (CI-pinned deps)

Capability behavior (empirically probed, not assumed)

One finding — non-blocking, recommend follow-up

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants