Archived research surface·Last refreshed Jun 1, 2026. Not currently maintained as a daily product.
SYSTEM · ARCHITECTURE
Tracing one actor through the stacked field model
Worked example: TSM across 119 days of corpus (2025-12-09 → 2026-06-01). Each layer card below shows what the system saw for this actor at that layer over the whole window.
Point-in-time. At any time t the system only uses data available at or before t. Conviction is the model’s confidence at parse time, not a calibrated probability. Calibration, source/actor trust priors, and region reliability are planned in F-007.
Throughline
TSM’s daily expectation moved from bearish continuation (0.65, 2025-11-26) through a peak of bullish continuation 0.75 (2026-04-27) to today’s bearish continuation (0.70).
L0Event FieldBUILT
Raw signal layer. Every event becomes a point in a 1,536-dimensional semantic space at the timestamp of its occurrence, with actors linked, source, relevance, and a data-quality flag.
addsCaptures every news, filing, transcript, and social post naming TSM, point-in-time, ready for upstream clustering.
today763 events naming TSM in the 130-day window where every layer has coverage (1,568 in the full L0 corpus). Sources: NewsAPI, Finnhub news + transcripts, Reddit, X-amplification, SEC filings.
events per day naming TSM763 total · peak 69 on 01-15
Peak day sample: “TSMC Q4 profit jumps 35% to record, beats expectations”
Minor gaps9 of 130 days in this window had zero events for TSM. Some of this is real news quietness; some reflects pipeline gaps in Dec 2025 / Jan 2026 / early Feb 2026 (post-holiday weeks where fetch_today.py ran in degraded mode). A targeted gap backfill ran on 2026-05-18 to repair this; daily health checks were added downstream to prevent recurrence.
2025-11-26FIRST
0
events
2026-01-05
4
events
2026-02-11
5
events
2026-03-19
6
events
2026-04-27
3
events
2026-06-01TODAY
10
events
L1Narrative FieldBUILT / SHADOW VALIDATED
Events embed into a shared semantic space and cluster into coherent storms. Each storm has a stable ID that persists across days through F-002 lineage matching.
addsPlaces TSM inside a named narrative each day — and tracks when its center of mass shifts to a different one.
todayPrimary cluster: “intel / broadcom / apple”. Density (7d): 0.227 (momentum +0.007, novelty 0.28). Cluster labels are auto-generated from the full event neighborhood — actor membership reflects shared AI-sector coverage, not exclusive company-specific coverage.
TSM’s primary cluster over time104 distinct clusters · 117 transitions
Colored band = primary cluster label. Line = semantic_density_7d. Cluster transitions are where the actor’s narrative center of mass shifted.
2025-11-26FIRST
—
2026-01-05
AI-driven semiconductor investment outlook
3d in this narrative
2026-02-11
ASML market momentum and growth
1d in this narrative
2026-03-19
Middle East Conflict Impact on Markets
7d in this narrative
2026-04-27
AMD stock outlook and market dynamics
2d in this narrative
2026-06-01TODAY
—
L2Expectation FieldBUILT V1
Forward views per actor, embedded into the same semantic space and attached to their nearest L1 storm.
addsTurns TSM's narrative position into a directional thesis with conviction — a daily forward expectation.
todayLatest: bearish continuation at conviction 0.70. “TSM's current negative confirmation state suggests that despite strong narrative around advanced node demand, geopolitical risks are exerting significant downward pressure. This bearish trend is likely to persist over the next couple of months unless there's a significant de-esca”
TSM’s daily expectation128 days · direction = bar color · conviction = bar height
Each bar = one day’s forward expectation. Green = bullish, red = bearish, grey = neutral. Bar height = model conviction at parse time (not a calibrated probability).
2025-11-26FIRST
bearish cont.0.65
“In the next 1-2 months, TSM is likely to continue experiencing pricing pressure as strong narrative support fails to translate into price gains, similar to peers in the sector. Geopolitical risks may further exacerbate this pricing lag.”
2026-01-05
mixed rot.0.65
“As of this date, TSM's structural picture indicates a mixed outlook due to confirmed demand for advanced nodes but offset by geopolitical risks impacting sentiment. Expect fluctuations in price as market reactions to external factors continue.”
2026-02-11
bearish cont.0.70
“The next 1-2 months may see continued pressure on TSM's performance due to geopolitical risks affecting advanced node demand, with negative narrative dislocation scores indicating potential challenges ahead.”
2026-03-19
bullish cont.0.70
“The structural indicators suggest continued upward momentum in the near term, driven by strong advanced node demand despite geopolitical risks. Expect price stability and potential gains as the market digests this narrative.”
2026-04-27
bullish cont.0.75
“With a confirmed state and positive relative return, TSM appears well-positioned to maintain upward momentum in the near term, driven by strong demand for advanced nodes amidst geopolitical considerations.”
2026-06-01TODAY
bearish cont.0.70
“TSM's current negative confirmation state suggests that despite strong narrative around advanced node demand, geopolitical risks are exerting significant downward pressure. This bearish trend is likely to persist over the next couple of months unless there's a significant de-esca”
L3Expectation Lifecycle FieldBUILT V1
Each (actor, theme, direction-sign) is a persistent thesis tracked through typed lifecycle events: born / strengthened / weakened / contradicted / retired.
addsTurns TSM's daily expectation summaries into a memory system. Theses you can follow for weeks, not snapshots that vanish overnight.
today86 theses tracked over the window; 3 active today, 0 persistent (≥3 daily versions). Lifecycle events: born 86 · strengthened 0 · weakened 0 · contradicted 11 · retired 87.
TSM’s expectation theses24 theses shown · 86 total
↑CoreWeave earnings volatility and outlook
●○
─Insider confidence in growth stocks
●
↓AI-driven semiconductor investment outlook
●
↑AI data centers and energy dynamics
●✕○
↓Dell Technologies Growth and Challenges
●✕
↑Semiconductor market dynamics and trends
●✕○
↓Semiconductor market dynamics and trends
●○
─Semiconductor market dynamics and trends
●○
↓Corporate filings and disclosures
●✕
↓Surge in critical minerals investment
●○○
↓Micron Technology investment outlook
●○○
↓Middle East Conflict Impact on Markets
●✕○
─Middle East Conflict Impact on Markets
●○
↑Middle East Conflict Impact on Markets
●○
↑Oil price volatility impacts markets
●✕✕
↓Oil price volatility impacts markets
●✕○○
─Oil price volatility impacts markets
●○
↑Semiconductor market dynamics and investments
●✕○
↓Semiconductor market dynamics and investments
●
─Market highs amid geopolitical tensions
●○
↑AMD stock outlook and market dynamics
●○✕
↑Market highs amid geopolitical tensions
●○✕
↓Alphabet earnings and AI investments
●○
↓Market highs amid geopolitical tensions
●○
●born▲strengthened▼weakened✕contradicted○retired
2025-11-26FIRST
quiet
2026-01-05
●born
1 active thesis
2026-02-11
○retired
●born
2 active theses
2026-03-19
○retired
✕contradicted
●born
2 active theses
2026-04-27
●born
1 active thesis
2026-06-01TODAY
quiet
L4Performance FieldBUILT V1
Slices expectations into regions of (theme × direction) and asks, for each, whether forward returns relative to QQQ moved in the predicted direction. 5d / 10d / 20d horizons. Walk-forward — every observation uses only information available on its date.
addsCloses the loop. Realized outcomes calibrate which regions of the field actually pay.
todaySystem-wide: 65 public regions (73 limited, 343 insufficient) across 2,377 signed observations. All-expectations baseline runs 48% at 5d / 50% at 20d — close to chance overall, with edge concentrated in specific regions. 14 public regions are flagged inverted (corpus hit ≤ 30%); the operating layer reads them as contrarian. TSM appears in 17 public regions and 11 limited, including 4 inverted.
TSM appears in 65 signed regions · showing top 6 by n_obs
theme · directionsample5d hit20d hitactors
↓Earnings season insights and stock analysis
n=46
43% (-4pp)
76% (+27pp)
15
↓Market highs amid geopolitical tensions
n=27
56% (+8pp)
82% (+32pp)
13
↓Corporate filings and disclosures
n=25
52% (+4pp)
36% (-14pp)
14
↑Semiconductor market dynamics and investments
n=24
46% (-2pp)
38% (-12pp)
5
↓Micron Technology investment outlook
n=24
33% (-15pp)
29% (-20pp)
4
↓AMD stock outlook and market dynamics
n=21
38% (-10pp)
20% (-30pp)
10
Hit rate = share of forward returns that moved in the predicted direction relative to QQQ. Delta vs the all-expectations baseline (48% at 5d, 50% at 20d). Green ≥ +10pp · red ≤ −10pp. Tiers: n ≥ 10 public, n 5–9 limited (shown with caveat), n < 5 insufficient (hit rate suppressed). INV badge: region flagged inverted (corpus hit_5d ≤ 30% on public tier); operating layer reads the contrarian direction (strike-through glyph shows the original L2 read, second glyph shows the effective direction).
FEEDBACKL4 measurements close back to upstream layersVISION
F-007 V1 measures region performance. The feedback loop itself is still V2: realized outcomes do not yet re-weight upstream conviction, source trust, or lifecycle thresholds. The current state machine uses static rules. For TSM, this is where the system would learn that — for instance — its winter bullish run on AI-infrastructure narratives held, or that today’s bearish read deserves more / less weight than its conviction suggests.
L4→L2
Conviction calibration
Re-weight L2 conviction by horizon, theme, and direction based on realized outcomes.
L4→L1
Source / actor trust
Re-weight L1 inputs by historical predictive value of each source and actor.
L4→L3
Lifecycle thresholds
Tune L3 Δconviction cutoffs and retirement-window length to match realized outcome dynamics.
BUILT — in production today. BUILT / SHADOW VALIDATED — computed daily, not yet promoted into the state machine. BUILT V1 — first cut shipped; V2 followups on roadmap. SCOPED / F-007 — spec written, not started. VISION — load-bearing later, no spec yet.
Rendered from public/actor_trace/TSM.json — a per-actor join of the L0 event corpus, L1 field instrumentation, L2 expectations history, and L3 lifecycle artifacts. Updates each evening pipeline run when build_actor_trace.py --all runs. For compute details: /methods. The same five-layer shape also powers the governance instance; on why this pattern recurs across domains, see a pattern for problems where beliefs must evolve.