Skip to content
Institutional judgment memory

Your firm has perfect records — and no memory of its own judgment.

Memric reads your IC memos, post-mortems, and theses and surfaces what you believed, what got contradicted, who dissented correctly, and which live proposals echo a prior failed assumption.

IC MEMO / NVDA INITIATION / 2023-Q3
Initiating a 4% position in NVIDIA

The thesis rests on data-center GPU demand outrunning supply through at least 20251, with the CUDA software moat insulating gross margins from commoditization.2

We hold that AI-infrastructure margins are structurally durable as hyperscaler capex compounds. The principal risk is in-house silicon compressing the moat faster than the market expects.3

1
Prediction · resolved ✓
Demand outran supply through 2024-Q4. Author calibration +0.8σ.
2
Principle PRI-0042 · retired
Superseded 2026-Q1 by 3 post-mortems.
3
Dissent · J. Okafor · correct
Flagged silicon risk 14 months before it priced in.

Illustrative example. Memric runs on your firm’s own documents.

The gap

Your firm has perfect records — and no memory.

Hundreds of IC memos, post-mortems, theses, and letters. Every number preserved; every judgment forgotten. Nothing tells you what you used to believe, when it broke, or who saw it first.

IC MEMO
NVDA initiation
POST-MORTEM
TWLO close-out
SECTOR THESIS
AI infrastructure
QUARTERLY LETTER
Q4 to LPs
SLACK THREAD
NVDA trim debate
MEETING NOTE
GP offsite

What Memric surfaces

Four things your firm can't see today.

Each is a typed, lineage-traced record — not a summary. Here's what each one looks like, on real material.

01 / Predictions vs. outcomes

Calibration, per author

Calibration chart: S. Reyes is well-calibrated (Brier 0.04); D. Park is overconfident by +1.6σ at high conviction. perfect 0 100 0 100 predicted % resolved % CALIBRATION BY AUTHOR S. Reyes well calibrated · Brier 0.04 D. Park overconfident · +1.6σ at high conviction

Every prediction, scored against what actually happened. The chronically overconfident show up immediately — by name.

02 / Principle lifecycle

Beliefs that retire when superseded

42 – AI-infra margins are structurally durable
42 + AI infrastructure is a margin business
// retired 2026-Q1 · superseded by 3 post-mortems

The supersession is the record. You can always replay what you believed and why it changed.

03 / Dissent analysis

Who pushed back — and was right

2024-Q2 · J. Okafor dissents 2025-Q3 · proven correct +14 months early

Dissent is captured at the moment it's voiced, then matched to outcomes. Correct contrarians become visible.

04 / Open contradictions

Live beliefs that can't both be true

PRI-0051 · active · 2024-Q4
"Concentrate into proven winners."
active contradiction
PRI-0067 · active · 2026-Q1
"Diversify against single-name risk."

Two principles, both currently active, mutually exclusive. Surfaced before the next IC meeting — not after.

The engagement

A four-week Decision Memory Audit.

A fixed-scope engagement that turns your decision history into a defensible record. You keep the report and a Memric instance — every finding traces to a source document.

  1. 01 Intake A representative slice of your decision corpus — memos, post-mortems, theses, letters.
  2. 02 Compress Memric extracts typed, lineage-traced principles, predictions, dissent, and contradictions.
  3. 03 Report A 9-section Decision Memory Report, every claim cited back to a source document.
Decision Memory Report · v1
  1. 01 Executive summary
  2. 02 Decision timeline
  3. 03 Principle library
  4. 04 Prediction calibration
  5. 05 Retirement events
  6. 06 Dissent analysis
  7. 07 Lineage map
  8. 08 Open contradictions
  9. 09 Recommendations

Trust by architecture

Every claim carries its lineage.

Nothing Memric asserts is unsourced. Each principle, retirement, and resolved prediction traces through typed citations back to the documents that produced it — so the record is defensible, not just plausible.

The engine runs in your environment — your machine or your VPC. Your raw documents never have to leave it.

CITES COINS SUPERSEDES · RETIRES SECTOR THESIS AI infrastructure IC MEMO NVDA initiation PRINCIPLE PRI-0042 POST-MORTEM retires it

Contact

If your firm has decades of decisions and no memory of them, get in touch.

jin@memric.ai