Stage A.1 - Liberator/kti 20-Pair Parity (BLOCKED)

Run date 2026-06-07 · Companion to Liberator validation + kti pilot · Anchor: Pull-In Research Hub

BLOCKED: Liberator minute_bars dataset is currently not entitled

The 20-pair Liberator/kti parity test could not run because kti liberator query --dataset minute_bars consistently times out with "context deadline exceeded" on every (symbol, date) pair tried. Root cause: the user account's Liberator entitlements have dropped from 27 datasets to 10 sometime between 2026-06-05 15:00 ET and 2026-06-07 19:55 ET. The remaining 10 entitlements are fis_* (lending data), s3_partners, and wonda - no market_data datasets including minute_bars are currently accessible.

This is a real, structural blocker. The 20-pair parity test cannot complete until minute_bars entitlement is restored (probably requires re-auth with CloudQuant or a renewed token). Once Liberator is back, re-running the test takes ~5-10 minutes.

Stage B is BLOCKED on Liberator restoration. The hybrid source recommendation from Stage A cannot be locked in until parity is verified at scale.

1. Entitlement state evidence

1.1 Before (2026-06-05 15:00 ET, during initial Stage A pilot)

$ kti liberator datasets
Found 27 entitled dataset(s) across all environments:

Name: arca_imbalance         Category: market_data
Name: daily_bars             Category: market_data
Name: daily_bars_adjusted    Category: market_data
Name: minute_bars            Category: market_data    <-- AVAILABLE
Name: minute_bars_raw        Category: market_data
Name: trades_v2              Category: market_data
Name: fis_borrower_activity_rating_b1    Category: fis
... (17 more)

1.2 After (2026-06-07 19:55 ET, attempting parity test)

$ kti liberator datasets
Found 10 entitled dataset(s) across all environments:

Name: fis_borrower_activity_rating_b1
Name: fis_loan_rate_dispersion_r2
Name: fis_loan_rate_eod_n2
Name: fis_loan_rates_p2
Name: fis_retail_loan_rates_p5
Name: fis_short_data
Name: fis_utilization_isin_u1
Name: fis_utilization_percent_u2
Name: s3_partners
Name: wonda

# minute_bars NOT in list

1.3 Query attempt evidence

$ kti liberator query --dataset minute_bars --date 2026-06-04 --symbols NVDA --format json
Error: failed to query data: failed to read stream: context deadline exceeded
(Client.Timeout or context cancellation while reading body)

real    4m0.260s

Every minute_bars query times out at the 4-minute deadline. This is consistent with the server rejecting the query at the entitlement layer but the kti CLI not surfacing that as a fast error - it waits for the stream that never comes.

2. Partial results - 1 verifiable pair

Only one (symbol, date) pair from the Stage A initial pilot has BOTH kti coverage AND a previously-cached Liberator response. That's NVDA 2026-06-02, which was fetched at 2026-06-05 15:32 ET before entitlement changed.

MetricValue
Symbol / dateNVDA / 2026-06-02
kti RTH bars390
Liberator RTH bars390
Close max diff$0.0000 (exact)
Close median diff$0.0000
OHLC spot-check (3 bars at 09:30, 10:00, 15:55)Identical to penny
Volume median diff5,251 shares (Liberator higher by 1-6%)
Volume max diff230,456 shares (at 09:30 open burst)
kti VWAP at session closecompute later (not done here)
Liberator native VWAP at close~226.17 (from per-bar field)

This single pair confirms that when Liberator IS available, OHLC matches kti exactly. But n=1 is not the 20-pair statistical baseline the user requested.

3. What I tried

PairDateLiberator status
RGTI2026-06-04TIMEOUT (120s)
QQQ2026-06-04TIMEOUT (120s)
QBTS2026-06-04TIMEOUT (120s)
MRVI2026-06-04TIMEOUT (120s)
IGV2026-06-04TIMEOUT (120s)
IBIT2026-06-04TIMEOUT (120s)
... (rest not attempted)process killed after 6 timeouts—

The parity script stage_a1_parity.py is queued; it will reuse the kti cache and only fetch the Liberator side. Re-running once entitlement is restored is ~5-10 minutes for the full 20 pairs.

4. Pairs queued for the test (when Liberator returns)

The script selected 20 distinct symbols from the kti cache, sorted by recency (most recent dates first), one per symbol where possible:

#SymbolDateWhy selected
1RGTI2026-06-04Most recent
2QQQ2026-06-04ETF
3QBTS2026-06-04Small-cap
4MRVI2026-06-04Biotech
5IGV2026-06-04Tech ETF
6IBIT2026-06-04Crypto-adjacent (the leak-symbol)
7-20Various2026-06-03 and olderDiversity across slugs and dates

5. Stage B status: BLOCKED

Stage B cannot proceed until 20-pair parity is verified

The Stage A initial pilot's tentative recommendation (hybrid source) was based on a single overlapping pair (NVDA 6/2) plus 4 Liberator-only samples. The user explicitly requested 20-pair verification before Stage B. That bar has not been cleared.

If Stage B were forced to proceed now:

Stage B is correctly BLOCKED.

6. Remaining risks

RiskSeverityMitigation
Liberator entitlement may not be restored quicklyHIGHContact CloudQuant; check whether re-auth or token refresh is needed
Even with parity verified, volume drift (~5-10%) may affect RVOL-derived signals across source boundaryMEDIUMCharacterize on 20 pairs; if uniform, adjust Stage B pipeline; if symbol-specific, may need per-symbol calibration
Liberator native VWAP cross-check (Stage A.1 §3) untested without live accessMEDIUMAdd to 20-pair test once Liberator returns
kti window may shrink further with time (today: 14d effective, was 60d earlier in the project)LOW-MEDMonitor kti retention; document in Stage B blocker if window drops
Pre-market policy not yet decided (see PREMARKET_POLICY_RECOMMENDATION)LOWIndependent decision; not gated by Liberator

7. What this report did NOT do

8. Updated Stage B recommendation

Stage B: BLOCKED pending Liberator entitlement restoration

Specifically:

  1. Restore Liberator minute_bars entitlement (CloudQuant action; re-auth or token refresh likely)
  2. Re-run stage_a1_parity.py for the 20 queued pairs (~5-10 min once Liberator works)
  3. Confirm OHLC parity (expected: match exactly, as NVDA 6/2 already showed)
  4. Characterize volume drift (expected: ~5-10% Liberator-higher, uniform; verify it's not symbol-specific)
  5. Run Liberator native VWAP vs recomputed VWAP cross-check (see VWAP_SD_RECOMPUTE_METHOD)
  6. Then commit to hybrid (recommended) or Liberator-only source for Stage B

Estimated time to clear once Liberator is back: under 30 minutes of compute. Estimated time to wait for Liberator restoration: unknown (depends on CloudQuant entitlement action).

Links