● Build log · execution evidence

What ships, measured weekly

This page is the public record of what Framler is and how fast it grows. We publish what we have — engine breadth, accumulated calibration data, shipping velocity — without the «100M users in Q3» slogans. User-growth cohort tables go live once the sample is big enough to mean something.

Honest disclosure. Framler launched mid-May 2026. The numbers below are real engine activity. We deliberately do not publish weekly-signup or week-1 / week-4 retention yet — sample is too thin to read as signal. Those tiles land here as soon as the threshold is meaningful.

Engine breadth

Tickers tracked
1,000+
US, EU, HK/JP/KR, select EM
Factor families
13
All peer-reviewed, all on /methodology
Confluence patterns
24
Multi-factor setups with academic basis
Behavioural invariants
59
Structural properties the engine satisfies on every request. Detail surface is admin-only — public subset on /coherence.

Calibration depth

The engine self-grades against forward returns. Every score becomes a calibration sample once the horizon closes — these are the rows the weekly calibration cron has accumulated to date.

Signal-history rows
147.6k
Daily snapshots × tickers. Drives IC and conformal halfwidths.
Scoring days logged
162
Distinct UTC dates with a published universe snapshot
Mondrian bins calibrated
143
Per-regime × tail × horizon partitions. See methodology.
Factor-weight runs
301
Bayesian posterior weight calibrations. See track record.

Roadmap milestones

Calendar-driven activations of new factors. None of these auto-flip — each goes live only when its out-of-sample measurement window clears the gate.

  1. 2026-06-01
    HAR-RV variance forecast — activated
    Corsi 2009 heterogeneous-autoregressive realised vol, blended in after the out-of-sample MSE gate cleared — a measured ~10% improvement over the Huber baseline across 56 independent months; the blend coefficient stays private. Live in the variance pipeline with a one-command rollback tag.
  2. 2026-06-15
    Analyst trend factor — activation deferred
    Original target. The Phase-3 activation gate did not clear on the measured IC — the mid-July and 2026-09-25 re-measures found no usable signal either — so the factor stays observe-only. Holding a factor back until it earns its place is the honest outcome, not a missed deadline.
  3. 2026-06-26
    30-day horizon calibration — still maturing
    The first 30-day forward-return windows have begun maturing, but only a couple of independent windows exist so far — too few to trust. The weekly accuracy-check cron keeps filling the IC + hit-rate tile as more accrue through the summer.
  4. 2026-07-01
    IVRP factor — activation deferred
    Implied-vs-realised-vol risk premium (Bali-Cakici-Whitelaw 2011). A snapshot fix on 2026-07-02 restarted honest accumulation; the 2026-09-25 read still had too few independent windows, next read ~January 2027; same Phase-3 gate as analyst_trend, not yet cleared.
  5. 2026-07-26
    90-day horizon calibrated
    The 90-day horizon received its first calibrated weights. Independent 90-day windows are still too few to judge the fit, and the Pro tier has no launch date until the live record can back it.

User cohorts

Not yet published

Weekly signups, week-1 retention, week-4 retention, and watchlist activation rate publish here once the sample is meaningful — current early-access cohort is under the threshold at which those numbers would be honest signal vs noise.

What we will publish: anonymised counts only — never per-user activity. The measurement methodology lands here at the same time as the first numbers.

Where to go next

For engine accuracy numbers, see Track Record. For the investor pitch, see Investors. For the full per-factor walkthrough, see Methodology.

Engine accuracy →Investor brief →Methodology →