PrimeSync AI Studio

Release history

PrimeSync AI Studio changelog.

Material MCP, security, scope, retention, and tool-contract changes are recorded here before they are described publicly.

Studio pricing update

USD 25 monthly or USD 250 annually

  • Studio subscriptions are now USD 25/month or USD 250/year. Both plans retain 50 shared analysis jobs and 100 Ask PrimeSync website questions per month.
Studio access update

Subscriptions and shared analysis allowance

  • Studio analytical tools now require a USD 39/month or USD 390/year subscription, with 50 jobs per calendar month shared across ChatGPT and Claude.
  • Account & activity shows verified Studio usage separately from website-assistant question credits. CSV imports, memory operations, clarifications, tool errors and identical retries do not consume additional analyses.
  • Earlier release notes describing free MCP analysis access are historical and superseded by this update. The homepage strategy brief remains free.
v2.23.1

Reliable tool contracts and workflow recovery

  • Legacy audits now return an explicitly unverified static report instead of failing on a missing input. A clean approved-workflow pass still requires the original blueprint, its matching hash and the code marker.
  • The public JSON manifest now publishes the exact input and output schemas registered by the MCP server. Connection guidance explains how to recover from stale client definitions.
  • Automatic Monte Carlo and research simulation counts fit the full dataset within the hosted compute budget and disclose every adjustment. Explicit counts are never silently reduced. Missing source counts are labelled unverified completeness.
  • Two-moving-average crossover audits check both operands and signal direction. Unrestricted sessions no longer trigger an unnecessary timezone question.
  • Authoring returns a concrete code-submission contract; research responses keep actual evidence visible without embedding duplicate reports. Research candidates can carry their original blueprint into the static audit.
  • Added regressions for all six MCP tools, schema parity, malformed calls, legacy audits, numeric reproducibility, dataset completeness and code round trips. Static inspection does not independently compile Pine Script or verify live-market performance.
v2.23.0

Properties throughout the strategy workflow

  • Added explicit P0 legacy, P1 mini/standard futures, P2 micro futures, P3 stocks, P4 CFD comparison and Custom configuration choices, independently of V entry profiles.
  • Full account, sizing, fee, margin and execution snapshots now travel with blueprints, source audits and backtest exports.
  • Pine audits detect mismatched declaration values and order-level quantity overrides. Static checks do not operate the TradingView Properties tab or compile code.
  • Backtest evidence requires confirmation of effective Properties. Final candidate selection rejects missing or conflicting configurations and comparison-only cost assumptions.
  • Legacy callers can continue exploratory work; final validation requires the new Properties evidence. No automatic preset selection or changes to unapproved trading rules.
v2.22.0

More precise research and safer strategy continuation

  • Research ranks drawdown and win rate directly, uses the supplied net P&L column for net-profit objectives, and discloses fixed composite formulas.
  • Reconstructed candidate filters no longer report unsupported baseline-beating probabilities. Empty holdouts remain missing evidence, and training folds are clearly labelled.
  • Preservation distinguishes protective instructions, executable Python floor division, quoted text, and comment-only edits. Ambiguous changes require review.
  • Portable context can be reused directly after memory deletion. Conflicting project identities, memory bypasses, and stale baseline code require clarification.
  • Added fixed onboarding questions and summary labels in English, Arabic and Hebrew. Full open-ended responses remain the connected AI's responsibility.
  • Added actual Python preservation fixtures and local Cloudflare D1 ownership/concurrency checks. These are not TradingView compilation or independent security certification.
  • The web Monte Carlo workbench rejects malformed returns without silently dropping trades.
v2.21.0

Optional personal memory and project continuity

  • Added a separate, default-off memory permission and context tool for confirmed preferences and minimal strategy summaries.
  • Added owner checks, revision protection, export, corrections, reset, and deletion. Shared reviewer identities cannot persist memory.
  • Added temporary preference overrides, project selection, conflicts with approved rules, and distinct beginner/advanced response guidance.
  • Remembered evidence never establishes runtime validation. Code/data mismatches reset remembered evidence.
  • Fixed consumed-request reconstruction on the hosted MCP endpoint; hardened account mutation origins and unknown-tool audit labels.
v2.20.0

MCP v2 SDK, security, and validation integrity

  • Migrated to the split MCP server and client SDK 2.0.0 packages with MCP protocol 2026-07-28 negotiation while retaining tested legacy initialization compatibility at the existing /mcp/v1 endpoint.
  • Made highest net P&L the canonical default objective with deterministic downside tie-breaks; tightened strategy-generation coverage, blueprint binding, static findings, aggregate compute limits, and final-exit chronology.
  • Separated caller-attested source records from independently verified evidence, hardened reviewer credentials and session expiry, and enforced tool-call limits across an OAuth session rather than an individual access token.
v2.19.0

Research matching and evidence lineage

  • Research Builder now returns a structured strategy profile, source-specific search queries, transparent source relevance scores, caller-attested and rejected evidence, and stable source-to-hypothesis-to-candidate lineage.
  • Abstract-only, incomplete, promotional, irrelevant, inaccessible, and leakage-prone evidence is explicitly labeled or rejected; supplied research remains untrusted data and is never treated as instructions.
  • Every strategy-research response starts with a complete stage map and exposes protected logic, missing evidence, overfitting warnings, validation gates, research status, and the next action.
v2.18.4

Cross-client logo discovery metadata

  • OAuth protected-resource and authorization-server discovery now advertise the official PrimeSync AI Studio logo through the registered logo_uri metadata field.
  • MCP initialization now includes the official 48×48, 96×96, and 512×512 PNGs plus the PrimeSync AI Studio website URL.
  • Fresh permanent asset URLs avoid reusing any client-cached image response; connector applications still control whether and when their interface renders server branding.
v2.18.3

Authorization logo display repair

  • The OAuth approval page now permits its same-origin PrimeSync AI Studio logo under the page's strict Content Security Policy.
  • The consent logo uses a same-origin URL and retains all existing OAuth, PKCE, CSRF, redirect, and framing protections.
v2.18.2

Universal connector icon fallbacks

  • Added standard favicon, Apple touch icon, and web-app icon metadata for AI clients that do not read MCP server icons.
  • MCP server and tool metadata now advertise both 96×96 and 512×512 PNG sizes under fresh URLs.
  • Every icon is an exact copy or deterministic resize of the supplied official PrimeSync AI Studio artwork.
v2.18.1

Connection icon cache refresh

  • The official PrimeSync AI Studio logo now uses a new permanent asset URL in MCP server metadata, the public tool manifest, OAuth approval, and website icon metadata.
  • Obsolete orange fallback icon files were removed so clients cannot rediscover the retired mark.
  • The supplied 512×512 official artwork is preserved byte-for-byte.
v2.18.0

OAuth rotation and import integrity

  • Access tokens now expire after one hour, and every refresh rotates the refresh token; reuse of a rotated token revokes its complete connection.
  • Authorization responses now include RFC 9207 issuer identification, and insufficient-scope challenges enumerate all permissions required by the tool.
  • CSV imports reject duplicate or blank normalized headers and unterminated quoted fields, while duplicate-row totals now count every removed row accurately.
  • The public result schemas and static-audit description now match the MCP result contracts, including review-required outcomes.
v2.17.1

Correct OAuth return destination

  • The approval screen now identifies ChatGPT, Claude, or a local Codex client from the verified OAuth callback instead of displaying a hard-coded Claude message.
  • Unknown callback clients use neutral MCP-client wording, and existing OAuth, PKCE, CSRF, redirect, and scope protections are unchanged.
v2.17.0

Final-candidate evidence and hosted safety

  • Added a fully gated final-candidate mode requiring exact code/data fingerprints, version, parameters, chronological folds, untouched holdout, realistic cost stress, and parameter sensitivity.
  • Preserve Existing Logic now reports deterministic semantic fingerprints for signals, orders, sizing, visuals, alerts, and execution, blocking changes outside the approved scope.
  • CSV parsing and Monte Carlo runs now enforce explicit hosted work budgets with recovery guidance instead of silent truncation or timeout-prone requests.
  • Submitted code, CSV cells, filenames, metadata, and source evidence are explicitly treated as untrusted data; instruction-like research text is withheld from hypothesis generation.
v2.16.0

Open AI strategy architecture

  • Research now exposes two explicit operating modes: Open AI Strategy for new architectures without PrimeSync.me defaults, and Preserve Existing Logic for bounded repairs and conversions.
  • Strategy Builder can hand complete arbitrary Pine Script v6 or Python blueprints to a connected AI, then audit the returned code candidate against the locked blueprint hash.
  • Multi-dataset research and approved Monte Carlo selection now carry compilation, fresh-export, cost, mapping, and lineage evidence gates end to end.
  • Standalone Monte Carlo defaults to 30,000 seeded block-bootstrap simulations, supports a strongest-robustness objective, and can estimate breach probability for a trader-supplied account failure threshold.
v2.15.0

Strict functional-review closure

  • The TradingView importer now accepts valid rows when timezone metadata is unspecified, reports exact row and column mappings, and blocks warning-state static audits from evidence acceptance.
  • Zero-trade datasets stop before candidate and Monte Carlo generation, while caller-attested full-text academic records produce explicit, falsifiable hypotheses without being presented as profitability proof.
  • Monte Carlo exposes its seed and provenance, rejects incomplete or mismatched candidates before simulation, and returns the preferred candidate code unchanged with a mandatory revalidation rule after any code change.
  • Strategy Builder now anchors ATR risk sizing to the modeled slipped entry price and puts the Strategy Completion Map first, including transparent selection defaults.
v2.14.0

Final review completion

  • ATR stop prices now use the approved distance exactly while TradingView models configured order slippage separately.
  • Deterministic audit rules no longer raise false warnings for all-session strategies, verified blueprint metadata, or stable stops.
  • TradingView imports explicitly flag samples below 60 closed trades as provisional for robustness and candidate selection.
  • Academic research claims now state the exact boundary: PrimeSync evaluates cited evidence supplied by the browsing client but does not independently retrieve papers.
v2.12.0

Monte Carlo candidate comparison

  • Generated code is explicitly a candidate—not a finished strategy result.
  • The completion path requests fresh TradingView CSV evidence from the trader-selected timeframes or assets and validates every dataset before final analysis.
  • The finished result is the Monte Carlo comparison plus the exact unchanged code of the preferred caller-attested candidate. Any later code adjustment starts a new backtest and Monte Carlo cycle.
v2.11.0

Strategy Completion Map

  • Every Strategy Builder response now starts with a complete supplied, missing, blocked, ready, and pending checklist.
  • The checklist explains what PrimeSync AI Studio can do afterward: audit, TradingView confirmation, fresh backtest import, dataset validation, Monte Carlo, candidate comparison, isolated research, and final decision.
  • Unfinished stages remain visible even after code generation so candidate code is never presented as a completed or validated strategy.
v2.10.0

Exact costs and strict TradingView imports

  • Written cost rules such as “one tick slippage” now generate the exact Pine value instead of silently falling back to zero.
  • Static audits compare approved commission and slippage values against the values configured in code.
  • TradingView imports now recognize Profit and Profit USD headers, reject invalid dates and malformed content, and report detected or missing columns clearly.
v2.9.0

Staged Pine evidence and MCP access

  • At this release, MCP tool calls were available at no charge. This was superseded by Studio subscription access on October 8, 2026. Studio analyses remain separate from website question credits.
  • Static audits now separate pass, review-required, and failed outcomes without claiming compilation or runtime verification.
  • TradingView imports now require explicit lineage, compilation, freshness, cost, and mapping evidence before acceptance.
v2.8.0

Lossless validation

  • Session-breakout Pine generation now models entry slippage, tick-rounded targets, whole futures contracts, and close-confirmed invalidation without an invented intrabar stop.
  • CSV import detects comma, semicolon, and tab formats, removes exact duplicate rows, aggregates partial exits, and reports integrity warnings.
  • Monte Carlo warns about small, one-sided, and outlier-sensitive samples and recommends block-length sensitivity.
v2.7.0

Staged Pine evidence

  • Static audits now return STATIC PASS, STATIC REVIEW REQUIRED, or STATIC FAILED and explicitly leave compilation and runtime behavior unconfirmed.
  • Expanded Pine inspection for higher-timeframe leakage, negative offsets, pivot confirmation, intrabar execution, risk sizing, partial exits, mutable stops, session timezones, pyramiding, and alert-to-arrow parity.
  • Fresh TradingView imports now report BACKTEST EVIDENCE ACCEPTED only after lineage, mapping, costs, chronology, and user attestations pass—without claiming independent TradingView access or cryptographic provenance.
v2.6.0

Session breakouts and transparent resampling

  • Strategy Builder now recognizes Asian-session high breakouts and asks only for missing session hours or timezone instead of unrelated moving-average parameters.
  • Added deterministic Pine Script v6 generation for explicit long-only session breakouts with volume confirmation, fixed-risk sizing, frozen breakout levels, and R-multiple targets.
  • The focused Monte Carlo tool now defaults to seeded circular block bootstrap, exposes block length and resampling method, and states whether local trade-order dependence is preserved.
  • Clarified the code audit as static inspection, documented its runtime limits, and documented the required blueprint-hash code marker.
v2.5.0

Baseline and experiment memory

  • Added a client-managed baseline snapshot with dataset, market, timeframe, and metrics.
  • Added an append-only experiment ledger with accepted, rejected, or inconclusive decisions and supporting evidence.
  • Rejected experiment changes are returned as blocked suggestions and must not be proposed again unless the trader explicitly requests a retest with new evidence.
  • Strategy code remains request-only and is not stored by PrimeSync.
v2.4.0

Deterministic Pine generation

  • Strategy Builder now emits complete Pine Script v6 for supported, fully specified SMA/EMA, RSI, ATR-risk, bar-close strategies.
  • Generated code embeds the mandatory approved blueprint hash and returns deterministic audit findings in the same response.
  • Unsupported or underspecified rules return one exact clarification list instead of a generation instruction or placeholder code.
  • Added a regression test based on the supplied SMA50, RSI(14), ATR(14), 1%-risk strategy.
v2.3.0

Executable audit and candidate selection

  • Monte Carlo now returns complete visible per-candidate percentiles, drawdown, losing-streak risk, and negative-result probability.
  • Added multi-candidate comparison with objective-based selection restricted to complete datasets and verified audit lineage.
  • Replaced the audit checklist response with deterministic PASS, WARNING, and FAIL findings plus evidence.
  • Made the Strategy Builder blueprint hash mandatory for approved-workflow audits.
  • Raised the per-candidate return limit to 20,000 and added original-versus-received completeness reporting.
  • Clarified that strategy IDs and research ledgers are client-managed and are not retained by PrimeSync.
v2.2.1

Claude connector compatibility

  • Added a direct standards-compliant OAuth callback redirect for Claude's hosted connector.
  • Preserved the existing ChatGPT authorization handoff.
  • Added Claude connection instructions to the public guide.
v2.2.0

Continuity, imports, and typed results

  • Added a privacy-safe, client-supplied strategy ID and research ledger so rejected hypotheses are not silently repeated across sessions.
  • Added a TradingView backtest importer that validates mappings, excludes open trades, preserves chronology, and returns normalized result arrays.
  • Added stable blueprint hashes and generated-code markers for drift detection during audits.
  • Published typed MCP output schemas and example result objects for all five tools.
  • Renamed Monte Carlo output to seeded and reproducible, and expanded source discovery beyond SSRN.
v2.1.3

Baseline-first validation gates

  • Added an eight-gate process: baseline reconstruction, dataset mapping, logic equivalence, isolated hypotheses, chronological folds, untouched holdout with costs, Monte Carlo baseline comparison, and fresh TradingView confirmation.
  • Requires weak or ambiguous candidates to stop at the failed gate instead of being polished into a misleading final result.
  • Keeps strategy-specific filters and historical PrimeSync profiles out of the reusable process.
v2.1.2

SSRN research suggestions

  • Added SSRN as the preferred academic source for research-backed hypotheses and strategy suggestions.
  • Requires exact paper citations, labels preprints as provisional evidence, and keeps every paper-derived idea outside the strategy until the trader explicitly approves it.
  • Reinforced that PrimeSync Project filters, presets, profiles, and hidden rules are never inserted into a new trader-owned strategy.
v2.1.1

Named operator disclosure

  • Published נביה חיר as PrimeSync AI Studio's legal operator across the security, privacy, terms, machine manifest, and review materials.
  • Kept developer-platform verification, independent security review, bug bounty, and directory status separate and accurately marked as pending.
v2.1.0

Trusted configuration

  • Added OAuth 2.1 authorization code flow with PKCE S256, dynamic client registration, refresh-token support, and revocation.
  • Moved the documented production contract to the versioned /mcp/v1 endpoint.
  • Added separate build, audit, and validation permissions and per-tool security metadata.
  • Added a public tool manifest, MCP-specific privacy and security pages, an honest assurance-status section, and user-visible activity metadata.
  • Added per-connection rate limits, request-size limits, and a 30-day metadata-only audit log.
  • Expanded the Research & Build Engine contract for indicators, experiments, models, isolated repairs, conversions with visual parity, standalone Monte Carlo, fresh-result validation, and opt-in profiles.
v2.0.0

Research & Build Engine

  • Added current-request-first research, TradingView CSV reconstruction, chronological folds and holdout, cross-asset candidates, and 30,000-run circular block-bootstrap Monte Carlo.
  • Retained Strategy Builder, Code Audit, and focused Monte Carlo tools for compatible prompts.