Files
zeekay 645325a641 regenerate from cloud 939b09d4
The projection was 113 commits behind the document it claims to describe, and
`generate.py --all --check` reported every one of the thirteen clients DRIFTED —
the same +19 models in each, because they all fell behind the same emission.

Regenerated in place from hanzo-inc/cloud openapi.yaml at 939b09d4
(sha256 087024f6…), which is the only description of this API that cannot name a
route the binary does not serve. `--check` is clean across all thirteen after.

.spec-lock moved with the code. The generator READS that file and never writes
it — hanzoai/ci`s lane does — so a hand run leaves every lock naming the previous
document. A lock that disagrees with the tree beside it is worse than no lock:
the next `--check` without `--spec` fetches what production serves, compares it
to a digest that was already wrong, and reports drift that is really staleness.
2026-08-24 23:32:53 -07:00

135 lines
11 KiB
C++

/**
* Hanzo Cloud API
* The Hanzo Cloud API as a customer calls it: every operation under /v1/ except the operator's admin product, relay routes, legacy spellings and capabilities still reached by flag. Tagged by product: the first path segment after /v1/.
*
* The version of the OpenAPI document: v1
*
* NOTE: This class is auto generated by OpenAPI-Generator 7.14.0.
* https://openapi-generator.tech
* Do not edit the class manually.
*/
/*
* ExperimentApi.h
*
*
*/
#ifndef HANZO_API_ExperimentApi_H_
#define HANZO_API_ExperimentApi_H_
#include "hanzo/ApiClient.h"
#include "hanzo/model/Analysis.h"
#include "hanzo/model/AnalyzeQuery.h"
#include "hanzo/model/Assignment.h"
#include "hanzo/model/CreateBody.h"
#include "hanzo/model/DecideBody.h"
#include "hanzo/model/ExperimentList.h"
#include "hanzo/model/Health.h"
#include "hanzo/model/Trial.h"
#include <cpprest/details/basic_types.h>
#include <boost/optional.hpp>
namespace hanzo {
namespace api {
using namespace hanzo::model;
class ExperimentApi
{
public:
explicit ExperimentApi( std::shared_ptr<const ApiClient> apiClient );
virtual ~ExperimentApi();
/// <summary>
/// Is every experiment in the caller&#39;s org, with its variants, status and decision, ordered by project then id.
/// </summary>
/// <remarks>
/// Is every experiment in the caller&#39;s org, with its variants, status and decision, ordered by project then id. Scoped to the org resolved from the validated principal — a distinct org is a distinct physical store, so no query here can reach another tenant&#39;s rows — and further narrowed to the caller&#39;s project scope when the credential carries one. A principal with NO project scope sees the org&#39;s experiments across all of its projects, which is the answer a reader most often expects to be filtered and is not. Requires a validated principal; refuses without one rather than answering an empty list.
/// </remarks>
pplx::task<std::shared_ptr<ExperimentList>> getExperiment(
) const;
/// <summary>
/// Is one experiment&#39;s definition and lifecycle: variants, weights, control arm, status and winner.
/// </summary>
/// <remarks>
/// Is one experiment&#39;s definition and lifecycle: variants, weights, control arm, status and winner. It reads the registry row only — the definition and the decision, never live measurements. Assignment lives in the flags plane and outcomes in analytics; this is the value that names both. Scoped to the caller&#39;s org and project from the validated principal, so another tenant&#39;s experiment of the same id is simply not found. An id that is not a legal slug is answered the same way, without a store read — the shape check and the existence check are one answer, so neither leaks the other.
/// </remarks>
/// <param name="id">ID is the experiment the URL names.</param>
pplx::task<std::shared_ptr<Trial>> getExperimentById(
utility::string_t id
) const;
/// <summary>
/// Is the variant one subject is bucketed into, and the payload that variant carries.
/// </summary>
/// <remarks>
/// Is the variant one subject is bucketed into, and the payload that variant carries. The bucketing is a deterministic hash of the subject, so the same subject gets the same arm on every call for as long as the flag definition is unchanged — and this is a pure READ: it records nothing. In particular it does NOT record an exposure. The caller&#39;s SDK must emit the experiment&#39;s exposure event itself, or the analysis has an empty denominator and every arm measures zero. An empty variant with on false is not an error — it means the flag returned nothing for this subject, so the subject is not enrolled. A flags engine that is unavailable refuses rather than defaulting to an arm. Requires a validated principal, and the experiment must exist in the caller&#39;s org and project.
/// </remarks>
/// <param name="id">ID is the experiment the URL names.</param>
/// <param name="subject">Subject is the unit to bucket — a user, org, session or audience key, matching the experiment&#39;s subjectKind.</param>
/// <param name="props">Props is a JSON object of person properties for targeting. A value that is not valid JSON is dropped rather than refused, so a malformed one changes the bucketing without saying so. (optional, default to utility::conversions::to_string_t(&quot;&quot;))</param>
pplx::task<std::shared_ptr<Assignment>> getExperimentByIdAssign(
utility::string_t id,
utility::string_t subject,
boost::optional<utility::string_t> props
) const;
/// <summary>
/// Is whether the experiments subsystem is mounted and serving in this process.
/// </summary>
/// <remarks>
/// Is whether the experiments subsystem is mounted and serving in this process. It answers unconditionally. It proves exactly one thing — that this binary registered the experiments routes and is dispatching them — and deliberately no more: it reads no principal, opens no per-org registry, and touches neither the flags engine nor the analytics plane, so a 200 here says nothing about whether a given tenant&#39;s store will open or whether an analysis can run. It is the only route on this surface that needs no org. The static path is registered ahead of the /:id read, so it always wins the first-match scan. \&quot;health\&quot; is a legal experiment id, which means an experiment created under that id can never be fetched by id — pick another.
/// </remarks>
pplx::task<std::shared_ptr<Health>> getExperimentHealth(
) const;
/// <summary>
/// Registers a controlled experiment AND puts its assignment flag live, in that order, so the arms start bucketing subjects the moment this returns 201 — the flag is created active at 100% rollout, with each variant weighted as declared.
/// </summary>
/// <remarks>
/// Registers a controlled experiment AND puts its assignment flag live, in that order, so the arms start bucketing subjects the moment this returns 201 — the flag is created active at 100% rollout, with each variant weighted as declared. There is no separate start call; creating IS starting. A variant carries an opaque payload this primitive never interprets: a feature config, an ad-creative id, a subject line, a model id. Requires a validated principal, and refuses without one. The org and project are taken from that principal and the creator is stamped from the credential — none of the three is a body field, so an experiment cannot be filed against another tenant. An id already used in this project is a conflict, never a silent overwrite: re-creating would stomp the assignment flag of a run in progress. It fails closed on the flag write. An experiment whose assignment flag does not exist would assign nobody, so if that write fails nothing is registered.
/// </remarks>
/// <param name="createBody"></param>
pplx::task<std::shared_ptr<Trial>> postExperiment(
std::shared_ptr<CreateBody> createBody
) const;
/// <summary>
/// Is per-variant conversion, lift and statistical significance against the control arm.
/// </summary>
/// <remarks>
/// Is per-variant conversion, lift and statistical significance against the control arm. It reads per-subject outcomes from the analytics plane over a window, folds them into per-variant samples, and returns each arm&#39;s exposed count, conversions, rate, lift versus control, two-proportion z, two-tailed p-value and whether it clears alpha. Arms with no data still appear with zero exposed, so the read is complete over the experiment&#39;s declared arms; the control arm sorts first. The pooled-variance estimator is used and the p-value is exact; a degenerate comparison (an empty arm, no variance) answers z 0 and p 1 — not significant, never an error. Only EXPOSED subjects are counted, and each is joined to its arm by re-evaluating the assignment flag AT ANALYSIS TIME — not from what was in force during the window. That is the one rule to get right: analyzing an experiment after its winner has been promoted re-buckets every subject into the promoted arm, collapsing the control to zero exposed and making the result meaningless. Read the analysis before deciding. A subject the flag cannot place is dropped rather than allowed to poison the fold. The winner in the response is ADVISORY — the significant, control-beating arm with the highest rate, or empty when inconclusive. It promotes nothing; the decision is a separate, explicit act. Every plane read is scoped to the caller&#39;s org. Per-variant samples are also written to the research evidence plane as immutable ab rows, best-effort: the analysis is still returned if that write fails, because the samples are recomputable, and the failure is logged rather than swallowed.
/// </remarks>
/// <param name="id">ID is the experiment the URL names.</param>
/// <param name="analyzeQuery"></param>
pplx::task<std::shared_ptr<Analysis>> postExperimentByIdAnalyze(
utility::string_t id,
std::shared_ptr<AnalyzeQuery> analyzeQuery
) const;
/// <summary>
/// Promotes one variant to the whole rollout and records who decided.
/// </summary>
/// <remarks>
/// Promotes one variant to the whole rollout and records who decided. It rewrites the assignment flag so the named winner serves 100% of the rollout and every other arm 0%, preserving the flag&#39;s targeting groups and payloads, then stamps the experiment decided with the winner, the deciding credential and the time. This is a production behaviour change that takes effect immediately for every subject the flag evaluates. It requires an ORG ADMIN of the caller&#39;s own org — a stricter gate than the rest of this surface, matching the flags write plane, because promoting is a flag write. The admin check runs AFTER the experiment is found, so a caller from another tenant is answered not-found rather than forbidden and learns nothing about what exists. An experiment whose assignment flag has gone missing is a conflict rather than a silent no-op — there is nothing to promote. Deciding is NOT terminal. A second call re-promotes a different variant and re-stamps the row; the status stays decided and the previous winner is overwritten with no record that it was ever chosen. Nothing here reverts the flag to its original weights either, so an experiment cannot be un-decided through this route — restoring a split means writing the flag definition back through the flags plane.
/// </remarks>
/// <param name="id"></param>
/// <param name="decideBody"></param>
pplx::task<std::shared_ptr<Trial>> postExperimentByIdDecide(
utility::string_t id,
std::shared_ptr<DecideBody> decideBody
) const;
protected:
std::shared_ptr<const ApiClient> m_ApiClient;
};
}
}
#endif /* HANZO_API_ExperimentApi_H_ */