Updates

What changed in the application recently, what each change was for, and what has explicitly not been done yet.

Rebranded to IndicaView, and repositioned as a hosted service

The product was renamed to IndicaView and repositioned. It is a hosted SaaS web application: nothing to install, nothing to self-host.

What changed in the product and in these docs:

Nothing about billing, checkout, plans or credit consumption is implemented. What exists is the seam: a canonical product/commercial configuration (app/js/brand/product-config.js), a credit-balance accessor (app/js/commercial/credits.js) that reports “unavailable” rather than inventing a number, and reserved (nullable, unpopulated) credit columns on the account profile. See Commercial model — or, in the repository, docs/commercial-model.md.

The retired name and the retired positioning have been removed from every user-facing surface, and scripts/check-product-config.cjs fails the build if either comes back.

Four substantial features

The largest changes since this documentation was last written. Each has its own page.

Authentication

Optional Supabase sign-in integrated directly into the static workspace, plus a My Account area with Profile and Security tabs. Email and password only; no OAuth, no MFA, no account deletion.

The important design point is that a single constant decides whether sign-in is required, and the default is 'optional' — guests reach the chart, the optimizer and the analysis without an account, and a missing or unreachable auth backend degrades to that state rather than breaking.

Profile data is protected by row-level security rather than by the routing flag. Details, the RLS model and the unverified email-delivery path are in Authentication.

NSGA-II

A seventh optimizer, for when two objectives genuinely conflict — profit against drawdown, say. Trials become vectors rather than scalars, and the population is sorted into Pareto fronts.

It ships alongside GA rather than replacing it, and shares GA’s operators and non-dominated sort. A follow-up commit closed four correctness and honesty defects, and forced the legacy contract back on the new code so the two algorithms stay comparable.

Result-card export

A client-rendered PNG summarising an optimizer run — drawn with canvas, at 2160 × 3840, with no network request. Your Pine source and the winning parameter values are excluded by a whitelist rather than a blacklist.

It is reachable from the optimizer results, the Analysis toolbar and the detached window — though on desktop the results control now lives in a kebab rather than on the toolbar. See Sharing & export.

Responsive layout

A phone layout below 768 px, built as a phone-first enhancement rather than a desktop reduction: a two-tab pane switcher, bottom sheets replacing selects and dropdowns, kebab menus replacing button groups, measured column pins, and 44 px touch targets achieved by growing hit areas rather than painted boxes.

Most of the recorded before/after numbers exist because the first attempt looked plausible in code and was visibly wrong on a device. See Mobile.

Supporting changes

• Visual language — drawdown redrawn throughout, the Run control reduced to a play glyph and one word, and an Apple-style sign-in treatment.
• New in-repo docs — docs/authentication.md, docs/SHARE_CARD.md and docs/mobile-ux-audit.md. All three are more detailed than this site; two of them are themselves slightly stale, and this site records where.
• A smoke aggregate runner — scripts/run-smokes.cjs, which times out per file and records TIMEOUT rather than skipping quietly.
• Auth test suites — a unit tier and a Playwright tier, both against mocked Supabase HTTP.
• RLS migrations under supabase/migrations/, with rollback-only SQL checks under tests/auth/.

Where the documentation itself disagrees

Worth knowing before you trust any internal design note:

• Three files under docs/ are ported upstream reports from a different host toolchain. They describe an astra.cmd harness and a chart backend this repository no longer uses. Treat them as provenance, not as current state.
• The repository’s own README.md and docs/architecture.md describe an earlier phase of the application. Where they conflict with this site or with the code, the code is right.
• One consequence you can verify yourself: the charting library was swapped from ApexCharts to Apache ECharts 6.1.0 in 790cfbb, because ApexCharts requires an OEM licence once embedded in a third-party product. Any note still describing ApexCharts is stale.
• docs/SHARE_CARD.md is close but has drifted in three places: it says the result card hands the file to the OS share sheet on a phone (that path is unreachable), it shows the desktop control as a visible button (it is now in a kebab), and it describes the footer as configuration-driven (that one string is hardcoded).

Recent changes

Grouped by theme rather than by date.

Optimizer ownership and data contracts

The optimizer was progressively moved from a detached results window onto an in-workspace surface, while the canonical data layer was kept intact.

Correctness fixes

Tooling and documentation

Phase F

Phase F is the most recent delivery sequence, and it is unusually well documented because each commit carries a detailed body.

The optimizer redesign

The redesign was a presentation and control-surface change, explicitly not an engine change. Engine algorithms, scoring, metric calculation, search-space rules, Pine sources, market-data behaviour, database names and database version were all left alone — and the engine-versus-synthetic distinction between real and reconstructed results was preserved throughout.

The shipped architecture is a deliberate four-way split:

The redesign document closes by superseding itself and pointing at the analysis implementation notes, so treat it as a historical record of a decision rather than current behaviour.

What is in flight

Everything here is a real, sourced “not done”.

Explicitly deferred

• Live email delivery is unverified. Signup acceptance, confirmation, reset and a real authenticated browser save/reload have not been confirmed end to end. Dashboard redirects, email templates and SMTP are manual steps no tooling can check.
• The NSGA-II Pareto archive is never consumed. The search maintains it; only the current generation’s front is displayed.
• The result card’s OS share path is unreachable. The code exists and the UI never passes the option it needs, so every path ends at Download PNG. There is also no copy-to-clipboard.
• No mobile smoke test. Every responsive claim rests on manual measurement in Chrome device emulation; safe-area insets, non-Chromium engines and real devices are untested.
• Live market integration and Python hyperopt are deferred. The Python optimiser work was neither run nor repaired as part of this sequence.
• Walk-forward / out-of-sample validation is a later phase. Until it exists, every sweep result is a candidate, not a validated edge. Extending to prop-firm simulation does not fix this — in-sample variance is still in-sample.
• Zero-copy worker transport is designed and disabled; it needs headers a plain static deployment cannot set.
• Multi-asset support is blocked: adding a field to the market descriptor activates a module-level, order-populated cache inside the vendor bundle.

Superseded conclusions

• “Parallel evaluation is not worth it” — retracted. The correct statement is 3.33× at the default depth, gated by depth.
• The detached optimiser-results window — superseded by the in-workspace analysis surface. Legacy links redirect.
• ApexCharts — superseded by ECharts in this repository.

Retained but unowned

The detached window surface and its page are still present and still openable, but are explicitly not yet owned by the replacement. Retiring them is an open decision, not a pending task.

Verification gaps

• The browser smoke skips rather than fails when its automation dependency is missing.
• The end-to-end run-to-charts assertion was explicitly deferred out of the step that would naturally have covered it.
• Five seeded browser trials cover startup sampling — not mature convergence, and not every algorithm or strategy. No large stress run has been performed in the browser.
• Storage-quota exhaustion and delayed-write ordering were tested with controlled adapters, not by exhausting real storage.
• No screen-reader session, no physical touch-device test, no blocked-popup test, and no exhaustive malformed-record test. Narrow-layout checks used desktop viewport emulation.

Next

• Architecture decisions — the invariants these changes preserve.
• Limitations — the consolidated list of what the product does not establish.